research report 136 - health and safety executive · exposure assessment section of ease. only one...
TRANSCRIPT
HSE Health & Safety
Executive
Evaluation and further development of the EASE model 2.0
Prepared by the Institute of Occupational Medicine for the Health and Safety Executive 2003
RESEARCH REPORT 136
HSE Health & Safety
Executive
Evaluation and further development of the EASE model 2.0
JW Cherrie University of Aberdeen &
Institute of Occupational MedicineEdinburgh
J Tickner, J Friar
KS Creely, AJ Soutar, G Hughson Institute of Occupational Medicine
Edinburgh
R Rae Artificial Intelligence Application Institute
Edinburgh University
ND Warren, DE Pryde Health and Safety Laboratory
Sheffield
This study examined the underlying structure and philosophy of the EASE (Estimation and Assessment of Substance Exposure) model version 2.0, developed by the UK Health and Safety Executive (HSE). Our aim was to provide a critical assessment of the utility of its performance to date. On the basis of this review, recommendations for the structure of a revised exposure model are outlined.
This report and the work it describes were funded by the Health and Safety Executive (HSE). Its contents, including any opinions and/or conclusions expressed, are those of the authors alone and do not necessarily reflect HSE policy.
HSE BOOKS
© Crown copyright 2003
First published 2003
ISBN 0 7176 2714 4
All rights reserved. No part of this publication may bereproduced, stored in a retrieval system, or transmitted inany form or by any means (electronic, mechanical,photocopying, recording or otherwise) without the priorwritten permission of the copyright owner.
Applications for reproduction should be made in writing to:Licensing Division, Her Majesty's Stationery Office, St Clements House, 2-16 Colegate, Norwich NR3 1BQ or by e-mail to [email protected]
ii
SUMMARY
This study examined the underlying structure and philosophy of the EASE (Estimation and Assessment of Substance Exposure) model version 2.0, developed by the UK Health and Safety Executive (HSE). Our aim was to provide a critical assessment of the utility of its performance to date. On the basis of this review, recommendations for the structure of a revised exposure model are outlined.
The EASE model, which is a general model to predict workplace exposure to a wide range of substances hazardous to health, has been under development and in use since the early 1990s. As neither the development of the model’s structure, nor the ideas and principles that underlie it, have been previously published in the open literature we have sought to do this. This information has been documented by gathering surviving evidence and interviewing personnel involved in its creation and development. The HSE’s National Exposure Database (NEDB) was used as the principal data source and a description of the development of the output ranges of exposures provided within the model is described. A number of problems and limitations of the model have been identified by users and the description of the model’s development provides some explanation as to why these are present.
The EASE model has been widely used in risk assessments for new and existing chemicals and has also been distributed to over 200 people in Europe, North America, Australia and Asia. To gain useful, constructive critic ism of the model, 27 stakeholders participated in structured interviews to determine their views on issues such as the limitations of the model, suggestions for improvement and accuracy and precision. Stakeholders were also asked to comment on what outputs they would like to obtain from an ideal model.
Stakeholders’ were classified as academic, industrial and government users. Although their experiences and use of EASE were very variable, and a broad spectrum of views was obtained, general themes were observed. Overall, stakeholders felt that the model should be updated and exposure ranges “tightened up”. It was considered that this should partly involve updating the data set on which the model is based, as the link to measurements was particularly valued. Stakeholders were very critical of the dermal model and felt that this required priority attention. More accurate and precise exposure estimates were considered desirable, although a consensus view about the degree of accuracy and precision was not reached. Independently, however, we have suggested that in a given workplace situation an improved version of EASE should aim to predict the median exposure to within a factor of three times the true value. The stakeholders also stressed that in updating the model the simplicity and usability of the software should not be compromised to any great extent. It was also emphasised that the model’s terminology and accompanying literature must be clear, unambiguous and concise. A subset of stakeholders’ were asked for their views on the types of outputs they would wish from an ideal model for the purposes of regulatory risk assessment. The general consensus was that a good reliable estimate of exposure was required, which can then be taken forward and discussed by professionals. Although the inclusion of information such as uptake and systemic dose was thought useful, stakeholders were more cautious about their inclusion in a revised model.
Some of the stakeholders reported conducting validation studies of EASE and a literature search using bibliographic databases identified several others. Some of the authors of this report had also previously completed research on the validity and consistency of the EASE model and these studies were also included in the scope of this work. All of these studies were reviewed, both in terms of the overall accuracy and precision of the exposure assessments, and to identify limitations of the logic structure. Six reports or papers investigating the validity of the inhalation exposure assessment section of EASE were identified and reviewed, along with two studies investigating the validity of the dermal
iii
exposure assessment section of EASE. Only one study was identified where the consistency of EASE was investigated. This is an unpublished investigation carried out by the HSE and for completeness a summary of this work has been included as an Appendix within this report.
There are three comprehensive validation studies available that included more than 17,000 measurements of inhalation exposure each covering between 41 and 70 EASE end-points. Review of these produced the fairly clear view that, for inhalation exposure EASE tends to either predict close to the measured exposures or to overestimate. There were very few instances where EASE was seen to underestimate inhalation exposure. Although EASE is conservative it is also highly variable in its exposure predictions and it is this unpredictability that is unacceptable. As with the inhalation model there were considerable overestimates of actual dermal exposure, however it should be noted that there is a much more limited knowledge base to validate the dermal model and any conclusions about the performance of this aspect of EASE are much less robust.
A conceptual model of exposure was developed to investigate whether the structure of EASE is appropriate. This analysis showed that while EASE has a number of characteristics that describe exposure it is a great simplification of what takes place when people are exposed to chemicals; and this simplification does not incorporate all of the important exposure determinants. More importantly, it is believed that EASE does not produce estimates of exposure that are unambiguous or complete. The conceptual model may provide a more rational basis for developing an improved version of EASE.
Before arriving at a recommendation for an improved version of EASE it was also considered important to review other available occupational and non-occupational exposure assessment models, partic ularly with respect to their underlying structure and usability in order to identify other alternative approaches. Approximately 50 to 60 computer exposure assessment programs were identified and a small number were selected for further consideration and review. These were the Wall Paint Exposure Model (WPEM), Multi-Chamber Concentration and Exposure Model (MCCEM), Consumer Exposure Model (CONSEXPO) and Bayesian Exposure Assessment Toolkit (BEAT). In addition, the exposure reconstruction method devised by Cherrie (1999) was considered. When reviewing the model implementations, a number of factors were considered. These included issues such as usability, routes of exposure, appropriate situations for use, substances covered, amount of knowledge required by the user, strengths and weaknesses.
All of the types of models used have their advantages and disadvantages. In order to use a particular model properly, the user should fully understand the basis of the model and also its strengths and weaknesses, with the assumptions and appropriateness of the input parameters being considered fully before use. The ease of use of a model package is important and the implementations considered varied in this respect. WPEM was the most straightforward to use, leading the user through step-by-step. CONSEXPO, which considers all routes of exposure and uptake, requires considerably more knowledge to use it properly. It is also important that the exposure estimates obtained are reproducible for a particular scenario when different assessors use the model. With the exception of the exposure reconstruction method, no attempt has been made to address this issue in any of the other model implementations considered. Given its complexity, it is anticipated that more problems in this respect will be experienced with CONSEXPO, than with the simpler and more transparent models such as WPEM and EASE. Overall, none of the existing models provided a better basis for predicting occupational exposures than EASE, although many had ele ments that could be advantageous in any future development of EASE.
On the basis of these reviews, recommendations for the structure of a revised exposure model have been suggested. We consider that a deterministic model is appropriate for the further
iv
development of EASE and could provide more accurate exposure estimates. Models and measurements are complementary and we believe that combining both sets of data will maximise the possible accuracy and precision of any regulatory exposure assessment. Many stakeholders also value an exposure prediction system based on measurements. It is envisaged that any available exposure measurements to be used in a risk assessment will be held in a database along with appropriate contextual information. The relevant information in the database and the associated model predictions could be combined using a Bayesian statistical framework, similar to that in the BEAT model. Rather than having predefined ranges as at present, an updated version of EASE should present the user with a confidence interval for the mean predicted exposure. The overall complexity of the proposed model for occupational exposure is greater than the present system and this makes it an absolute necessity to develop a completely self-contained computational package that does not require the user to have any mathematical expertise. The final integrated model should be disseminated to end-users through the Internet and must have its own self-contained help system.
Before a decision can be made on the best way to model and estimate exposure, the developers of the model, the regulators and industry need to discuss the purpose and intended use of a successor to EASE. Both of these will have a strong bearing on the form of the model and acceptance of any new scheme will be dependent on all relevant stakeholders being committed to the proposed approach.
v
vi
CONTENTS
SUMMARY v
1. INTRODUCTION 1
2. THE DEVELOPMENT OF THE EASE MODEL 1992-2002 3
2.1 Introduction 32.2 The starting point 32.3 Technical development of the model 62.4 Software development 132.5 Discussion of the model and its limitations 182.6 Assessment of EASE – how well does it reflect real 20
Exposure in the workplace? 202.7 Summary
3. STAKEHOLDER INTERVIEWS 21
3.1 Introduction 213.2 Methodology 213.3 Results 223.4 Discussion 363.5 Follow up stakeholder interviews 403.6 Conclusion 42
4. REVIEW OF EASE VALIDATION AND CONSISTENCY 43STUDIES
4.1 Introduction 434.2 Evaluation of the structure of the EASE model 444.3 Identification of relevant studies that assess the 50
validity or consistency of EASE 4.4 Summary of studies of EASE validity and consistency 514.5 Overview of the validity and consistency of EASE 62
5. REVIEW OF ALTERNATIVE EXPOSURE MODELS AND 65ADVANCES IN MODEL IMPLEMENTATION AND COMPUTER SOFTWARE
5.1 Introduction 655.2 Classification of models 655.3 Identification of alternative exposure models 685.4 WPEM v3.2 695.5 MCCEM 1.2 (Beta version) 725.6 CONSEXPO 3.0 755.7 BEAT 1.13 825.8 Exposure reconstruction model 855.9 Discussion on alternative exposure models 865.10 Review of advances in model implementation and 87
computer software5.11 Advances in artificial intelligence 875.12 Advances in knowledge management 885.13 Advances in computer technology 90
vii
5.14 Advances in computer hardware 915.15 Discussion on implementation and software considerations 92
6. A DISCUSSION OF THE WAY FORWARD 95
6.1 Introduction 956.2 What should an improved EASE deliver? 956.3 The way forward 966.4 Updating EASE 996.5 Concluding remarks 100
ACKNOWLEDGEMENTS 103
REFERENCES 105
APPENDICES 109
Appendix 1: Personnel involved in the development of EASE 109Appendix 2: Consistency study 111Appendix 3: Stakeholder interviews questions and prompts 125
viii
1. INTRODUCTION
European Directives and Regulations require risk assessments of notified new and existing substances used in industry. These risk assessments, which must consider both the toxicity of the chemicals used and the potential levels of occupational exposure, are designed to evaluate the potential for adverse health effects for the workers who will handle the chemicals. The results of the risk assessment may have important economic consequences for industry.
The level of occupational exposure to a chemical is known to depend on the physical properties of the substance being handled, the tasks being undertaken, and on the controls in place within the workplace, although the nature of these relationships are complex. Other factors such as the volume of the workspace and the general ventilation in the area may, in certain circumstances, be an important determinant of inhalation exposure (Cherrie, 1999). When assessing new or existing substances regulators use a knowledge-based model called EASE – Estimation and Assessment of Substance Exposure, which was developed by the UK Health and Safety Executive (HSE). EASE is described in the Technical Guidance Documents in support of the Directive and Regulation (European Commission, 1996). The model is currently in its second version for Microsoft Windows. The EASE model categorises occupational exposure with reference to historical data collected in the UK’s National Exposure Database (NEDB), although how these data were used to derive the EASE categories is unclear.
Recent work to validate the predictions of the EASE model, carried out by the Institute of Occupational Medicine (IOM) (Hughson and Cherrie, 1999) and others has highlighted weaknesses in the model when assessing inhalation exposure. Experience with EASE, both in actual risk assessments and in comparisons with published exposure data, indicates a clear need for improvement in several respects. The numerical estimates of inhalation exposure often appear to be positively biased and this may undermine the credibility of EASE in the eyes of the relevant stakeholders, particularly as the risk assessment may have important societal resource implications. Some work has also been undertaken to investigate the reliability of EASE predictions for dermal exposure (Hughson and Cherrie, 2001). This research found a similar tendency for positively biased exposure estimates from the model and other structural problems with the dermal exposure sections of EASE. HSE have already recognised that “the EASE model, particularly, is limited in this area” (Risk Assessment and Toxicology Steering Committee, 1999).
This study aimed to examine the underlying structure and philosophy of the EASE model version 2.0 and provide a critical assessment of its utility and performance to date, taking into particular account the experiences of European and other stakeholders. On the basis of this review recommendations for the structure of a revised exposure model are outlined in this report.
The origins of the EASE model have been documented by gathering information from HSE files, scientific publications and interviewing personnel involved in its creation and development. This work is described in Section 2. This section provides detail on the original concepts and ideas surrounding the model, the current state of the model, how it was developed, its distribution and use. To gain useful, constructive criticism of the EASE model, government, industry and academic stakeholders were consulted. Through structured telephone interviews, issues such as previous use of EASE, limitations and suggestions for improvement were discussed and an attempt was made to gain a consensus view of what stakeholders would consider adequate in terms of the accuracy and precision of EASE predictions. This work is described in Section 3.
1
A review of studies concerning the validity of EASE was undertaken and is described in Section 4. An attempt to draw common experiences, both in terms of the overall accuracy and precision of the exposure assessments and the limitations of the logic structure identified as part of these investigations, was made. Studies in the “grey literature”, such as internal reports, were included, along with work published in the open peer-review scientific literature. Other models used for assessment of occupational and non-occupational exposure were identified and reviewed to identify possible alternative approaches that could be used in the future development of EASE. A small number of key models were reviewed in depth to assess their underlying structure and usability. These evaluations are described in Section 5. Recent advances in model implementations and computer software are also reviewed in this section.
On the basis of these reviews, the project team provides recommendations for the structure of a revised exposure model and these are outlined in Section 6. This provides a clear plan for the future development of EASE so that it can provide appropriate reliable estimates of chemical exposure for regulatory risk assessments. Further work is necessary to develop these plans into a fully functioning software package.
2
2. THE DEVELOPMENT OF THE EASE MODEL 1992-2002
2.1 INTRODUCTION
The EASE model has been under development and in use since the early 1990s. The Health and Safety Executive (HSE) developed EASE in collaboration with the Health and Safety Laboratory (HSL) as a general model to predict workplace exposure to substances hazardous to health, which would be applicable to a wide range of substances and circumstances of use. Short biographical details of those personnel involved in the development of EASE are given in Appendix 1. At the time of reporting, the second Windows version was in use and a third version had been prepared. However, neither the development of the mode l structure, nor the ideas and principles that underlie it, have been published in the open literature.
This section aims to describe the evolution of EASE and covers:
• the fundamental thinking that led to the development of the current model;
• the development of the detailed structure of the model; and
• the process used to determine the exposure ranges that appear in the output predictions made by the model.
Although much of the detail of the development of the model was recorded at the time, not all of the documentary record has survived. Therefore, in preparing this section, recollections of HSE and HSL personnel involved in the development were heavily relied upon.
2.2 THE STARTING POINT
It is not easy to determine the starting point for an idea or innovation and the development of EASE is no exception. The concept of a system for predicting exposure to hazardous substances arose in the late 1980s and early 1990s, during the development of new legislation in the European Union (EU). The European Commission Directive 93/67/EEC (European Commission, 1993) set out the procedures for Competent Authorities in Member States to assess the risks of new substances to man and to the environment before they are placed on the market. A complementary measure (Commission Regulation (EC) No. 1488/94) covered risk assessment for a priority list of existing substances already in use (European Commission, 1994). Wider initiatives such as the Organisation for Economic Co-operation and Development (OECD) Existing Chemicals Programme also began to seek exposure information, in particular for High Production Volume (HPV) chemicals (OECD, 1993). These initiatives were concerned with exposure estimation as an aid to priority setting in the risk assessment of chemic als, primarily for regulatory purposes.
A chapter by Shillaker in “Risk Management of Chemicals” (Royal Society of Chemistry, 1992) set the scene for future developments. Shillaker’s paper describes a scoring system in which toxicological data, physical properties (e.g. boiling point, vapour pressure), the way in which the substance is used and production tonnage are combined to produce an overall priority score on which risk assessment decisions can be based. The scoring system in itself does not generate any information concerning exposure, but the introduction of use patterns and tonnage quantities showed the development of new thinking in risk assessment and risk management to cover the effects of using the chemicals, as well as their intrinsic toxic hazard.
The event, which gave the initial impetus to the development of a general-purpose model to predict occupational exposure, was an OECD Workshop, which took place in Orlando,
3
Florida in 1992. The summary of the monograph reporting the workshop sets out the ideas, thoughts and needs for estimating human exposure (OECD, 1993):
“The measurement of occupational and consumer exposure contributes to the data set on which to base the overall assessment of the potential hazard posed by chemicals to human health. Often such measurements are lacking. Therefore, in order to screen chemicals so that resources can be directed towards those which present the greatest problems, realistic estimations of human exposure need to be made.
This document offers a choice of possible approaches for assessing occupational and consumer exposure, by both the dermal and inhalation routes, primarily to chemicals produced in high volumes and for which only limited data are available. For some of these approaches, programs on computer diskettes are available.
The use of actual measured data is recommended, but where these are not available the approaches presented here can be adapted to using calculated values or data from analogous or surrogate chemicals.
The purpose of the exposure assessment is to obtain the Estimated Human Exposure level. In combination with the assessment of health effects based on toxicity studies, the Estimated Human Exposure level can be used to judge whether further action in relation to the chemicals is required. It is recommended that exposure to, and toxicity of, the chemicals be analysed together: thus the amount of detail required for the one will be dictated, to a large extent, by the severity of the other. The approaches presented encompass the range from qualitative to sophisticated quantitative assessments.
In the occupational setting, exposure by the inhalation route has received by far the most attention. For many chemicals, atmospheric standards exist, either on a national or internationa l basis, which quantify both the permitted levels and duration of exposure. The situation is not so well defined for the dermal route of exposure. The Workshop recommended that further research be conducted into methods for assessing such exposures, and that the results of this work be widely disseminated.”
Murray Devine then with HSE’s Health Policy Division, chaired the Workshop sessions on occupational exposure assessment. He produced a paper on the estimation of occupational exposure to chemicals, which set out a generalised approach for developing a method to predict workplace exposure (Devine, 1993). In essence, this paper focused on the factors that determine exposure in the workplace and suggested that the concentration of a substance in the workroom atmosphere might be predicted by analogy with similar situations where the concentration has been estimated by exposure measurements. It was recognised that “…. the overt use of subjective judgement without adequate codification or validation…has rendered the approach less than fully satisfactory in the “regulatory context”” (Devine, 1993), i.e. that subjective judgement and expectation alone, however “good”, can never provide a basis for a predictive method or model on their own. However, if such judgements can be “calibrated” by reference to a body of measured exposure data that is sufficiently precise, representative and comprehensive, then a model based on these judgements might be acceptable as a generalised predictor of exposure.
Devine (1993) went on to explore “….whether it could be possible to develop the empirical approach for exposure estimation, and then (indicated) in general terms how it could be applied….”. It was postulated that a simplified model might use two broad parameters to generate an intuitive banding structure, which for any given situation would classify exposure
4
somewhere in the spectrum “high—medium—low”. The parameters would be derived from a measure of the “containment” of the process and an indicator of the tendency of the substance to become airborne (related to volatility for liquids and particle size for solid materials). In this manner, exposure could be seen both as a function of containment level and physical properties of the substance. Table 1 provides some outline definitions for containment levels and Figure 1 overleaf provides a graphical representation of the concept of a basic model.
Table 1: Possible definitions for levels of containment (Devine, 1993)
Level Process type
A Fully contained plant
B
C exhaust ventilation (LEV)
D Spray application
Small quantities - good local exhaust ventilation Larger quantities - local exhaust ventilation
Large evaporation areas - no containment or local
The overview of the workshop acknowledged the UK proposal as:
“a possible approach based on a structured logic tree using analogous exposure data. Its future development into a simple, validated "expert system" for screening purposes is attractive, particularly as the ultimate choice of exposure databases could be adapted to national needs. In the first instance, the UK National Exposure Database (NEDB) could be a relevant tool in the development of this approach. The classifications based on physical and (apparent) toxicological properties and containment le vels will need careful definition for more general future use” (OECD, 1993).
The crystallisation of the concept of using a structured logic tree as the basis for an expert system to predict exposure appears to have been developed during the discussions that took place at the seminar. Devine’s paper detailed in OECD (1993) made only tentative suggestions on how the fundamental idea of a model based on physical properties and containment level might be transformed into a practically useful tool. Indeed, the paper was more concerned with establishing a sound conceptual basis for a suitable model than with the details of the model itself, which would require a considerable quantity of development and evaluation work to establish.
The theoretical approach that was established at the Orlando seminar was used to develop the model that eventually became EASE.
5
Figure 1: Exposure as a function of containment level and physical properties (redrawn from OECD (1993))
1 2
3 4
0
20
40
60
80
Notional exposure (ppm)
Containment level A Level B Level C Level D
Physical properties (relative vapour pressure)
2.3 TECHNICAL DEVELOPMENT OF THE MODEL
2.3.1 The inhalation model
A supplementary paper by HSE (1992a) began the detailed development of the inhalation model. The paper proposed the definition of fields for three factors that needed to be considered. These were physical properties, toxicological properties and containment level, reflecting Shillaker’s priority-setting scheme (Royal Society of Chemistry, 1992). The outline structure proposed for the model is shown in Figure 2.
Figure 2: Model for assessing inhalation exposure (HSE, 1992a)
Liquid, high vapour pressure Physical properties Liquid, medium vapour pressure
Liquid, low vapour pressureSolid, finely dividedSolid, granular
Toxicological propertiesVery toxic Toxic Harmful / irritant Low toxicity
A Containment B (see Table 1 for definitions)
CD
Output (exposure ranges for NUMBER OF FIELDS = 5 x 4 x 4 = 80
analogues)
6
Substances could be assigned to one of the 80 exposure scenarios, referred to as “fields”, that the model created and it was postulated that “exposure can be estimated by analogy to substances in the same field for which well-validated exposure is available” (HSE, 1992a). Two essential requirements were identified for the successful operation of the model. These were:
• the development of an expert system capable of assigning a substance to the correct field and identifying appropriate chemical analogues and
• the availability of a comprehensive data set of well-validated exposure for a wide range of substances.
The HSE’s National Exposure Database (NEDB) was seen as being an appropriate and vital exposure data set. The inclusion of toxicity as a classification at this stage is interesting, given that it was later dropped from the model. HSE (1992a) states that:
“In considering the influence of toxicity on exposure, it is reasonable to assume that rapidly-occurring acute effects are more likely to influence exposure than long-term insidious effects, which although serious, may not be apparent at the time of exposure. Thus, in assigning a substance to a toxicity classification, there is a need to consider the “perceived” toxicity among both manufacturers and end-users…It follows that in assigning toxicity classifications, there may be a divergence between whether the model is to be used for predicting what exposures are likely to occur in the workplace or for estimating what might be considered acceptable exposure levels.”
Toxicity can be seen as a measure of the hazard that the substance presents to users, and is not directly related to assessment of exposure, which is concerned with factors, such as, the potential to become airborne, how the substance is used and how it is controlled. An overall risk assessment contains two elements – the hazardous nature of the substance (toxicity) and the probability and intensity of exposure. In developing a useful model that is able to predict exposure it is therefore important to leave aside considerations of the hazard and concentrate on the physical situation in the workplace. For this reason, toxicity was removed from the development of the model at this early stage.
In parallel with the OECD initiative, the EU was beginning to consider exposure assessment as part of the 7th Amendment to the Dangerous Substances Directive. This was initially aimed at, and restricted to, new substances brought onto the market. The amendment introduced the concept of risk assessment; prior to this assessments were based on toxicological principles and did not consider practical exposure in the workplace. A model or prediction procedure was needed to determine likely exposure in the workplace, and the aim of the development process was to generate a means of estimating exposure using the knowledge and experience of occupational hygienists. This was similar to the ideas discussed at the OECD workshop, which is not entirely surprising as the impetus for the proposals came from the same source, that being HSE.
It was therefore decided to base the model on three parameters covering the overall situation in the workplace. These were:
• the tendency of the substance to become airborne; • the means of controlling exposure or of preventing the substance from entering the
workroom atmosphere and • the way in which the substance is used.
7
To align the model with the common format established by the EU Council Regulation 793/93 for existing substances on the priority list (European Community, 1993b), some of the definitions in the Harmonised Electronic Data Set (HEDSET) used for risk assessment by Competent Authorities were also employed as “categories of use” in the new model. These categories were:
• used in closed systems; • inclusion on a matrix; • non-dispersive use; • wide dispersive use.
This has resulted in some ambiguity in interpretation and a “stretching” of some of the terms to accommodate situations for which they were not originally intended. The first category is quite straightforward, as it refers to those systems where the hazardous substance is essentially being held in a vessel or sealed container whilst being used and the probability of direct exposure is therefore very low. The third category covers the use of the substance in circumstances where the process is not enclosed, but there is some degree of containment of the substance. “Wide dispersive use”, the fourth category, is intended to reflect processes where there is little or no provision for containing the substance.
The category that has caused the greatest problems is “Inclusion on a matrix”. It was originally intended for use as an indicator of consumer exposure where the substance of concern was inhibited in some way from immediate release into the atmosphere, for example, if it were bound in a greasy substrate. In the industrial setting, it refers to a range of situations where the hazardous substance is inhibited from entering the workroom atmosphere by virtue of its physical form. This includes pelletised powders, waxy and greasy formulations. The problem of interpretation arises because the term “inclusion on a matrix” is not commonly used in industry or in occupational hygiene, and consequently is not well understood. The manuals that have been provided to users of EASE (HSE, 1999) contain some explanatory text on the categories of use to assist in deciding which category to select, but the difficulties have remained.
Using this structure, it was decided to produce an analogue model and to use NEDB as the principal source of exposure data. NEDB was set up in 1986 to store and record exposure measurements made by HSE’s field occupational hygienists (Burns and Beaumont, 1989). One of the main purposes of the database was to provide reliable information that could be used to inform decisions on the setting of occupational exposure limits in support of the Control of Substances Hazardous to Health Regulations (COSHH). By the early 1990s, NEDB contained about 100,000 exposure measurements and was seen as an invaluable resource that would be an ideal starting point for developing a predictive exposure model.
Personnel responsible for NEDB began to retrieve and classify information from the database so that ranges of exposures could be determined for various groupings of the three parameters of the model (tendency to become airborne, method of use and method of control). Exposure data was retrieved from NEDB for a wide range of solvents and other substances in liquid form and sorted by industry, job and process. Retrieved data sets were presented as box and whisker plots, from which inferences were made about the likely ranges of exposure for generic scenarios. An example of one of these box and whisker plots for acetone exposure levels is shown in Figure 3.
8
Figure 3: Box and whisker plots for generic scenarios (scanned from original)
ACETONE
CONC. (ppm)
Key to processes
Number Processes 58 Degreasing 98 Spray painting 109 Degreasing (hot) 113 Shoe production 120 Cleaning 208 Vapour degreasing 211 Assembly 271 Coating 427 Casting 431 Painting 473 Plant maintenance 492 Polyurethane components production 513 Gasket production 532 Hand tool production 538 Motor repair 590 Powder handling
Eight-hour time weighted average exposure to acetone is illustrated in Figure 3 for approximately 16 industrial processes. The data are presented in a single vertical line for each process, with the characteristics of the plots being as follows:
• the upper and lower limits of the box are the inter-quartile range, i.e. the upper limit is the 75th percentile and the lower limit is the 25th percentile;
• the solid vertical line represents the extremes of the 90th and 10th percentile; • outliers beyond the 10th and 90th percentile are shown as individual points and• the median (50th percentile) is represented by the solid horizontal line within the box
9
A coded variable was allocated to the data for the processes at the time of input, for example, 113 refers to shoe production and 431 to painting. Plots such as those illustrated in Figure 3 were generated for a wide range of substances for which NEDB contained exposure data.
An iterative discussion and refinement process followed which led to the structure of the first version of EASE in 1994. The challenge was to assess and organise this data so that it could be grouped into a manageable number of similar exposure groups. The size of this task is apparent when considering that the data covered 400 substances, 150 industries, 750 process and 750 jobs. Not all of these ranges were independent, but nevertheless there was a potential for 3.4x1010 combinations of just these four factors. The process of examination and discussion reduced all the data to a series of 10 exposure ranges into which most samples could be placed.
Data from processes considered to have similar potential for exposure were then aggregated and grouped together. The use of occupational hygiene knowledge and experience became crucial at this point, and extensive discussion took place to allocate all the exposure data into a small number of ranges. The inhalation model contained 170 possible combinations of the three parameters, tendency to become airborne, use and control. An exposure range was assigned to each of these, based on expert opinion and interpretation of the box-whisker plots produced for the exposure data retrieved for each combination. The predictions were tested initially by comparing them with further searches of the NEDB data. In some cases the ranges were adjusted to agree with the NEDB data. It is acknowledged by those who developed the model that there is a degree of circularity here - NEDB was used both in the determination of the ranges and as the source of validation data. This is a fundamental flaw in the objectivity of the model; however there was no other database of exposure information that could be used to validate the model. Perhaps more importantly, EASE was always intended to be a flexible model, where the output ranges could be adjusted, as better and more reliable information became available. This was undertaken for later versions, particularly for the work to produce the third Windows version of EASE.
The original data used was restricted to 8-hour time weighted average information, usually calculated from task-specific exposures and knowledge of the work patterns, breaks etc. However, this approach was later abandoned as it was recognised that the model would work better if it predicted task exposure, which could then be adapted to any required pattern of work and include meal breaks, preparation time, different tasks, etc. This is not part of the model itself, but is a feature of the way in which it was intended to be used. It was always considered by HSE that the outputs given by EASE should be regarded as broad estimates. It was also considered that the model would perform best if used by experienced occupational hygienists, who could adapt the results in the light of experience and factors not covered by the scope of the model. In any event, because of the extent of the ranges and the large degree of variability in the exposure data, it was considered that the ranges encompass task specific and time weighted averages equally well. The model is not sufficiently precise (nor was it intended to be) to accommodate these differences.
2.3.2 Development of ranges for dusts
The early work in developing the detailed structure of the model and output ranges was restricted to liquids. Vapour pressure was used as the indicator of volatility and hence the tendency to become airborne. It was necessary to produce a parallel model for solid materials, related in some way to their dustiness. This was undertaken using similar methodology as for liquids, i.e. using expert opinion and NEDB data to develop a suitable model structure. For fibrous dusts, information was also obtained from HSE’s Guidance Note EH35– Probable concentrations of asbestos dust (now out of print).
10
Determination of the output exposure ranges for dusts had three elements. Firstly, using the vapour ranges a series of “equivalent” ranges for dusts (e.g. 50-100ppm equivalent to ?5-10 mg/m3) were postulated. Secondly, as it was not considered possible to introduce an “intrinsic dustiness” equivalent to liquid volatility, the model for non-fibrous dusts was based on materials handling and control. For fibrous dusts, subjective value judgements were made of the ability to become airborne, based on knowledge of fibre size distributions and practical experience with asbestos exposure. Finally, during external consultation, it was realised that some particulate materials, including pharmaceutical preparations, have a tendency to aggregate together. This might be due to absorption of water vapour, the addition of a waxy substrate or surface electrostatic effects. However the implication was that these substances would not generate the same level of dust as dry, powdery materials which do not have this propensity to aggregate. It was therefore decided to include a factor into the model that relates to the ability of particles to aggregate.
For the pattern of use category, the intention was to group processes according to the amount of physical energy used in the process, the intuitive idea being that high-speed mechanical processes will produce more dust than when the same substance is handled manually, without significant disturbance.
2.3.3 Development of the dermal exposure model
In the early 1990s, there was increasing interest in looking at exposure in terms of the overall burden on a worker. Exposure by routes other than inhalation began to attract increasing attention. To reflect this emphasis, it was decided to include a dermal exposure model so EASE could provide an indication of overall exposure for risk assessments of hazardous substances.
At the time of conception, there was little dermal exposure data available and there had not been any extensive research in this area. There were also no standardised measurement or assessment techniques available. A number of measurement techniques had, however, been investigated, including skin patches to absorb the contaminant, wipe samples, hand rinsing and visualisation with fluorescent tracers, although the data were sparse. There was little, if any, correlation between the different techniques, and the information available was only for a very limited range of substances, largely pesticides. NEDB contained no dermal exposure data. Some efforts to develop procedures for dermal exposure assessment were available from the US EPA (Fehrenbacher and Macek, 1992), and this information allowed a simple dermal exposure model to be developed.
The structure of the model is similar to the inhalation model in that it starts with three parameters – physical state, pattern of use and pattern of control - however a fourth was added to reflect the amount of contact with the substance of interest. For gases and vapours and also for non-dusty solids, dermal exposure was assumed to be “very low” – in most assessments, this would be considered negligible.
For the patterns of use and control, the categories selected for the inhalation model were also used although simplified given the lack of reliable data for dermal exposure. The dermal model is illustrated in Figure 4, and this shows that the use categories “inclusion on a matrix” and “non-dispersive use” have been combined, and pattern of control is restricted to two categories “direct handling” and “not direct handling”.
The contact level is intended to give an indication of frequency and duration of direct exposure to the substance. Again, as there was no standard way of assessing contact, the scale in Table 2 was suggested as a first attempt to quantify the degree of potential exposure.
11
Fig
4: D
eter
min
atio
n o
f der
mal
exp
osu
re in
EA
SE
(HS
E, 1
999)
De
term
ina
tio
n o
f d
erm
al
ex
po
su
re
(fro
m c
ha
rt 2
)
Ph
ys
ica
l s
tate
Ga
s o
r
va
po
ur
Ve
ry l
ow
Liq
uid
Pa
tte
rn
of
use
Clo
se
d
sy
ste
m
Ver
y lo
w
Inc
on
to
ma
trix
o
r n
on
-dis
p
Pa
tte
rn
of
co
ntr
ol
No
t d
irec
t h
an
dli
ng
Ve
ry l
ow
Dir
ect
ha
nd
lin
g
Co
nta
ct
lev
el
No
ne
Ve
ry l
ow
Inc
ide
nta
l
0-0
.1
mg
/cm
/d
ay
Inte
rmit
ten
t
0.1-
1 m
g/c
m
/da
y
Ex
ten
siv
e
1-5
m
g/c
m /
da
y
Wid
e d
isp
u
se
Pa
tte
rn
of
co
ntr
ol
No
t d
irec
t h
an
dli
ng
Ve
ry l
ow
Dir
ect
ha
nd
lin
g
Co
nta
ct
lev
el
No
ne
Ve
ry l
ow
Inci
den
tal
0.1-
1 m
g/c
m
/da
y
Inte
rmit
ten
t
1-5
mg
/cm
/d
ay
Ex
ten
siv
e
5-1
5
mg
/cm
/d
ay
So
lid
Is s
oli
d d
us
ty?
Ye
s
No
Ve
ry lo
w
2 2
22
22
12
Table 2: Contact level criteria for dermal exposure
None No contact
Incidental 1 event per day
Intermittent
Extensive >10 events per day
2-10 events per day
As there was no generally agreed methodology for assessing dermal exposure, it was necessary to decide what quantitative parameters would be used to express exposure in the model. It was assumed that the hands and forearms would be the anatomical areas most commonly exposed. The skin area of the hands and forearms is approximately 2000 cm2. It was also assumed that exposure could be assessed as the quantity of contaminant in milligrams that was deposited on this area of skin during a working day. This would lead to the estimation of overall exposure, with no attempt being made to assess how much of this exposure would penetrate the skin. No account was therefore taken of the effect of hand washing, or evaporation or any other loss of contaminant from the skin, and no allowance was made for the use of personal protective equipment.
The resulting model is very rudimentary and was principally intended as being an initial attempt at producing a dermal exposure model, which could later be refined and modified in the light of developing knowledge. As dermal exposure is still very much under investigation, and no firm consensus of exposure methods and measurement emerged before the end of 2000, the dermal model has remained unaltered since its initial development.
2.4 SOFTWARE DEVELOPMENT
2.4.1 Prototype system
The project resulted in a prototype model for dusts and vapours, which was implemented as an expert system using the Leonardo expert system shell. (HSE, 1992b). The occupational hygiene knowledge principles outlined in Murray Devine’s two papers (OECD, 1993; HSE, 1992a) were developed into a more formalised logic system which was refined in successive stages as the project progressed. In parallel to the structure of the model evolving, the computer model was also being actively developed.
Given that the original purpose of EASE was that it would be used as a decision support tool, precise exposure levels were not required. Indeed, a precise exposure level was unlikely to be a realistic target given that the detailed properties of a new substance are relatively unknown and the circumstances under which it will be used are not specific. These conditions were appropriate for employing expert systems software technologies that were developed during the 1970s and 80s, particularly those of rule -based systems.
Basically, a rule -based system consists of an engine, a set of rules and a working memory that contains “facts” in the form of object-attribute-value entries. Each rule has a set of conditions that may or may not be satisfied by entries in the working memory and a set of actions that follow as a consequence of its conditions being satisfied. The engine then takes the set of rules and matches them against the entries in the working memory to produce a subset of the rules whose conditions are satisfied. One of these rules is then selected and its actions are activated, possibly adding new derived facts and modifying the working memory. The engine then matches all the rules against all the entries again and so on, until an action terminating the system is activated.
13
A project was therefore established to convert the model into a coherent system of facts and rules and to create an operational expert system with a suitable inference engine. Essentially, the development followed an iterative process, where the theoretical model was converted into a system of rules with explanatory text and the resulting software model was assessed. EASE was developed into a deterministic system, always producing the same result when given the same data. No heuristics were involved, with the system always operating strictly in accordance with its rules and the facts available to it.
After discussion, the model was modified to produce an improved version. This process produced more than 20 distinct iterations of the software system and a great deal of discussion and comment. It is not possible to reduce all of this information, however Figures 5 and 6, which were prepared at the end of the development process, show the increase in sophistication from the initial early concept, to the system used in the first prototype model.
Fig 5: Early logic structure (HSE, 1992)
EXPOSURE
PATTERN OF CONTROL PATTERN OF USE TENDENCY TO BECOME AIRBORNE
Aerosol Formation
Physical State
Particle Size Volatility
Inference Engine
14
Fig 6: Advance logic structure used to construct the first version of EASE (HSE, 1992)
EXPOSURE
PATTERN OF CONTROL PATTERN OF USE
USER CHOICE OF USE PATTERN
Equation
Existence of Barrier
Dilution
Nature of Barrier
Segre-gation
Bulk Handling with RPE
Inference Engine
Sampling
TENDENCY TO BECOME AIRBORNE
Aerosol Formation
Vapour Pressure
Antoine
Chemical Type
No. of Carbons
Process Temp-erature
Melting Point
Boiling Point
Physical State
Particle Size
Correction Factor
Volatility
LEV
Algorithmic Computation
2.4.2 Public system
The EC accepted the general ideas and concepts in the prototype model; however, it was requested that the model should be freely distributed to industry without any copyright charges. Consequently, HSE was contracted by the EC to develop a version that could be used without restriction. The contract also required user documentation to be prepared and the model to be checked for system integrity and robustness. The Artificial Intelligence Applications Institute (AIAI) at Edinburgh University and HSL were subcontracted by the HSE to do this work and the following sections provide a brief synopsis of the work carried out between 1993 and 2001.
The first implementation of the EASE model, based on the HSL prototype, was targeted at a lowest common denominator computing environment, specifically an IBM PC with 640 Kb of main memory running under the DOS operating system. This imposed restrictions on the implementation, constraining the size of the running system and restricting the ways in which the user and the system could interact.
The system was designed as two separate modules, the inference engine and rule base, and the user interface. The inference engine was built using the CLIPS expert system building tool. CLIPS was selected for two reasons: first, it was a mature tool (developed originally for NASA) that implemented a similar rule -based technology to that of Leonardo; second, it was available in the public domain so copies could be distributed to third parties without any associated royalty or other licensing constraints.
15
This resulted in the first DOS version of EASE and this was accepted by the EC and published in the official technical guidance in support of the EU legislation on new and existing substances and was widely used for this purpose (European Commission, 1996).
This first version contained three vapour pressure ranges for classifying exposure to vapours, and it was soon apparent that this was too crude a system to give useful predictions of likely exposure. A second version of the model was planned, in which the number of vapour pressure categories was increased to six to improve the precision of the model.
Follow ing the general adoption of the Windows operating system in the mid 1990s and the consequent increase in power of the hardware required to run Windows successfully, both the system size limitation and the user interface restrictions could be removed when EASE was updated to version 2 for Windows. A typical output screen from the model is shown in Figure 7. For some time the model was also available in a DOS version, however, after 1997 only the Windows version was distributed. The model was distributed freely to all bona fide enquirers. As well as this being a requirement of the EC, there was also a strong desire to publicise the model so that it could be further refined and improved. Up until 2002, the model had been distributed to over 200 users in Europe, North America, Australia and Asia.
Figure 7 – Typical Output from EASE
Version 2 of EASE was used extensively by HSE’s Occupational Hygiene Unit until 1999. EASE predictions were routinely used in the preparation of reviews of exposure to substances for which occupational exposure limits were being considered by the Health and Safety Commission (HSC) Advisory Committee on Toxic Substances (ACTS). This experience identified a number of further problems and limitations of the model. These included the following:
16
• For low volatility liquids, such as isocyanates, it was considered that exposure to liquid droplets was likely to be at least as important as exposure to vapour given the high molecular weight and boiling point of these substances. It was not possible to transfer from the vapour section of the model to the section dealing with particulates and consequently it was decided to introduce a link using further analysis of the NEDB exposure data.
• In the non-fibrous dusts section, where low dust techniques are used, there is little in the way of definition and assistance to the user. Furthermore, the section of the model concerning low dust techniques and the use of local exhaust ventilation (LEV) results in a single end-point range of 0-1 mg/m3. This was judged as being too large a range to accommodate, with any precision, the conditions that exist during, for example, pharmaceutical manufacture where product purity is important. It was therefore decided to re-assess the data in this area, with the aim of producing a more detailed spectrum of ranges within the overall context of “low dust techniques”.
• For many applications for which EASE was being used, the physical properties of the substance were well known and recorded. In version 2, the user still has to manually input the substance vapour pressure. A database of substance properties was created and added to the EASE software to assist the user, so that the model could read in the appropriate vapour pressure directly.
• When assessing the exposure to gases such as chlorine and hydrogen chloride, which are highly water soluble, it was not possible to determine a single value for the equivalent vapour pressure of the gas in aqueous solution. For some of these substances, empirical information on vapour pressure as a function of solution strength is available, and the model could be extended to include this.
These modifications and changes to the user interface which were implemented in Visual Basic, were incorporated into version 3 of EASE for Windows during 2000. The development work was completed, but problems arose with the interpretation of the model during user trials and version 3 has not been distributed.
2.4.3 Engineering considerations
Building a knowledge-based system is not trivial, however the most difficult part is obtaining and codifying the “knowledge” that is required to define the set of rules. This “knowledge acquisition” is a specialised task that distinguishes the building of knowledge-based systems from other types of software system.
The quality of the system's decisions depends not only on the knowledge that has been captured and embodied in the rule set, but also on the process used to capture this knowledge. The process of capturing and encoding knowledge was a major concern of the knowledge engineering community, particularly in Europe, during the period immediately before the development of EASE. EASE therefore was in a position to make use of the KADS methodology (Schreiber et al, 1993) which was then, and still is, the leader in this field. Through using the KADS methodology to capture and structure the knowledge of the EASE system, it proved possible to document the knowledge involved. The resulting models were very complex and Eric Pryde from HSL presented these as flowcharts so that they could be understandable by professional occupational hygienists. The EASE flow charts could only be produced through gaining this better understanding of the knowledge involved by this engineering exercise.
17
Despite the knowledge acquisition requirements, EASE is much like any other software system in that it needed to be designed, built, documented and maintained.
As the EASE system was classified as being a safety-critical system, its development had to comply with the TickIT interpretation (BSI, 2001) of the requirements of the ISO 9001 quality standard (ISO, 2000). Work was subject to external audit, which was carried out by Lloyds Register. The decision that EASE was safety critical is debatable given that it is a decision support tool, not a decision taking tool. However developing the system in accordance with quality standards had real advantages, the most important of these was the emphasis on testing. Although the amount of effort that went into testing the first version of EASE seemed disproportionate at the time, the test suite produced was then reused by the subsequent developments. This gave a level of confidence in the correct functioning of the modified systems and their consistency with the original system, which would have required considerably more effort to produce at the time.
Each version of EASE required significant software changes, even though the changes to the EASE model were not themselves major. Completely new implementations of the user interface were included in each version. This was not as difficult as it might have been given that the original system was designed to run under both DOS and Unix operating systems. To achieve this, a cross-platform tool was used to implement the user interface; therefore a set of standard communication facilities needed to be defined. Reimplementing the user interface involved reimplementing the same set of facilities using different tools. The popularity of Windows led to a lack of interest in other platforms, therefore the cross platform capability was lost in later versions, particularly in version 3, for which Microsoft's proprietary Visual Basic tool was specified. The proposed gain from moving to Visual Basic was that the addition of special purpose code, such as that required to handle water-soluble gases, would be easier.
The other major software change with version 3 was the introduction of a link to a database of vapour pressures. This allowed values from the database to over-ride the rules for producing approximate vapour pressures for new substances. This facility had been investigated previously, but it had proved difficult to locate useful data that could be redistributed as part of the EASE system without incurring royalty payments.
Versions 1 and 2 of EASE were released on a floppy disk, from which the files could be simply copied across to the host machine. The effect of using Visual Basic and providing a database in version 3 resulted in a system that was very much larger than before, so much so that it needed to be distributed by CD-ROM. It was also substantially more complicated to install on a host machine as it required the three system components, the user interface, the inference engine, and the database, to be linked together using proprietary technology. This unfortunately appeared to be dependent on the version of the operating system involved, with behaviour under Windows NT, 2000 and XP being markedly different from behaviour under Windows 95 and 98.
Modifying the existing implementation could just accommodate the changes that were required to the EASE model, but it proved hard to adapt what had previously been a relatively simple tree structure to accommodate the linkages between the vapour and the particulate sections. This, in particular, illustrates the future need to revise the EASE model substantially.
2.5 DISCUSSION OF THE MODEL AND ITS LIMITATIONS
The legislative background and the technical information available have shaped the design of EASE over the past 10 years. It is probable that if the development of the model had started from scratch without these influencing factors, then its outcome would have been rather
18
different. However, the desire to accommodate both theoretical concepts of contaminant generation and the empirical practical knowledge of what actually happens in the workplaces has produced a model that is generally applicable across the enormous range of conditions that exist in industry. Because of this reliance on practical knowledge and experience, the model outputs were selected as ranges rather than as predictions of a specific concentration with confidence limits. It was recognised that conditions vary widely within industry and that a range of likely concentrations would give some idea of the upper and lower limits to be expected.
The analysis used to assign the ranges to the model consisted of a mixture of NEDB data analysis using box-whisker plots, supported by expert discussion and judgement. This combination of processes was relatively unstructured in the sense that it seemed not to have followed any formal procedure. This is a weakness in the development of the model and the absence of any comprehensible documentation makes it impossible to reconstruct the basis for the model’s end-point ranges.
The NEDB data itself is not wholly comprehensive. The database contains over 100,000 samples of exposure data. Most of this information was largely gathered between 1986 and 1993, after which the rate of data collection reduced significantly. The EASE output ranges have not been updated since 1992 (except for the analysis undertaken in 2000 to produce the modifications for the undistributed version 3 for Windows). Furthermore, NEDB itself has an inherent bias, in that HSE Specialist Occupational Hygiene Inspectors as part of their enforcement duties obtained approximately 90% of the samples. Consequently, a tendency towards high levels of exposure would be expected, as companies with no perceived problems were gene rally not sampled. Even so, NEDB still contains many samples indicating low exposure (<25% of the appropriate occupational exposure limit), so the actual bias is not as large as would be expected. Whether or not NEDB should be considered as containing worst-case data is debatable, but it cannot be regarded as being truly representative of occupational exposure in Great Britain given that it does not come from a random selection of workplaces and circumstances.
The early statements about EASE and the early analysis were restricted to 8-hour time-weighted average data. In later versions, the predictions have come to be regarded as task-specific exposures, from which full–shift averages could be calculated if the pattern of work and exposure is known or can be predicted. From this viewpoint, exposures for any pattern of work can be predicted, which gives the model more flexibility. Although this shift appears significant, the precision of the EASE estimates is in fact sufficiently relaxed for the differences to be accommodated without difficulty. In essence, EASE outputs are no more than approximate estimates of exposure, and should be used as such to give broad indications of the likely range of airborne concentrations rather than anything more precise.
As EASE was developed in the UK, and used only NEDB exposure data from the late 1980s and early 1990s, it necessarily reflects only the situation in the UK during that period. It is known that industrial practices vary from country to country, therefore the application of EASE to situations in other EU or non-EU countries should be treated with caution.
The decision to adopt the pattern of use categories from the HEDSET system (European Commission, 1996b) was taken in order to expedite the completion of the first version of the model. It is not always easy to assign industrial processes to one of these categories. A revised grouping for the “use” parameter would benefit those users of the model who find that limitations of the HEDSET categories are at odds with their understanding and terminology.
A major omission in the model is that it takes no account of the quantity of substance being used. This limits its applicability, as there is no distinction between using gram quantities in a laboratory and kilotonnes in a large processing plant. It would not be scientific to include a
19
simple multiplier to apply to the quantity used, as the quantity used does not always reflect the degree of exposure. For example, there will only be very low exposure to benzene in a fully enclosed and automated petrochemical plant while dangerously high concentrations could result from the uncontained handling of few millilitres on an open bench. As there is no simple way to link exposure with the amount of the substance that is used or the rate of use, no allowance was made during the development of EASE for any quantitative indicator of substance use. This permitted the model to be based on the initial 3-parameter concept of Figure 1 and maintained its fairly simple structure. It may be possible to revise the model to take account of the quantity of the substance in the future.
The dermal exposure model is very rudimentary; however, nothing better was available when EASE was first developed. Consequently, a simple model was constructed with the aim of giving a first approximation of dermal exposure. As knowledge and experience increased, it was foreseen that the initial model would be overtaken by something more rigorous which reflected a greater level of understanding. Dermal exposure is rapidly becoming better understood and the EU-wide RISKOFDERM project (RISKOFDERM, 1999) is focussing on this problem. When the project is concluded it is expected that more information will be available on which to base an improved exposure model. In the interim period, however, the outputs of the dermal model should be regarded with great caution.
2.6 ASSESSMENT OF EASE – HOW WELL DOES IT REFLECT REAL EXPOSURE IN THE WORKPLACE?
A number of validation studies have been undertaken in which the predictions of the model are compared with actual measurements and estimates of exposure. These studies have been reviewed in Section 4. In addition, as part of previous validation work commissioned by HSE in 1999, an assessment of the degree of variation between different users of the model was undertaken to determine whether it was capable of being used in a consistent fashion. A series of 15 descriptive scenarios were developed and 34 test subjects were asked to run EASE for each, using the description information to inform their decisions on potential for exposure, process type and degree of control. In an attempt to relate the scenarios to real life situations, where hard information may be lacking, some of the scenarios contained only very broad superficial statements about the processes. A short report of this study, which has not been published elsewhere, is provided in Appendix 2 and this work is also discussed in Section 4.4.3.
2.7 SUMMARY
The EASE model has been under development and in use since the early 1990s. It is a general model that is able to predict workplace exposure to any substance hazardous to health. Using surviving documentary evidence and through discussions with key personnel, the development of the model’s structure and the ideas and principles behind the process have been described. The current EASE model has been widely used in the risk assessment process for new and existing chemicals and has also been used extensively by HSE’s Occupational Hygiene Unit. EASE has also been widely distributed to over 200 users in Europe, North America, Australia and Asia and so it may also have been used for other purposes. A number of problems and limitations of the model have been identified and the description of the model’s development provides some explanation as to why these are present. The development of EASE has been one of continual evolution. For example, the prototype of version 3 of EASE for Windows was developed in response to limitations highlighted by ACTS though this version has not yet been distributed because of problems that arose during user trials. When developing EASE further it is important to consider fully the opinions of actual users in order to gain useful constructive criticism and suggestions on what improvements need to be made.
20
3. STAKEHOLDER INTERVIEWS
3.1 INTRODUCTION
The aim of the stakeholder interviews was to determine the views of a number of key individuals who have either used EASE 2.0 or have been involved in using the output from the model in regulatory risk assessments. The exercise aimed to gain an insight into the validity and reliability of the model, as well as their opinions on how the model could be improved. It was originally anticipated that 30 interviews would be conducted by telephone and these would include stakeholders involved in regulatory risk assessment, practising occupational hygienists, academics and others to ensure that a full spectrum of views would be obtained.
3.2 METHODOLOGY
3.2.1 Selection of shortlist of EASE users
As mentioned earlier, EASE 2.0 has been distributed to over 200 users. A database of the names and contact details of those who had been sent a copy of EASE was forwarded to the IOM by the HSE. The data was reviewed for consistency and duplicates. Four duplicates were removed and missing contact details were updated where possible. From the review of the data, 137 potential contacts were identified.
This database was then shortened and amended by the project team to a usable list of approximately 45 contacts. These were broadly categorised as being members of ‘industry’, ‘academia’ or ‘government’.
3.2.2 Development of structured interview and proforma
A semi-structured interview proforma was prepared to obtain information from the stakeholders (Appendix 3). This was intended to be administered via telephone and to act as a prompt for the interviewer. The interview was designed to elicit the following information:
• stakeholders’ understanding of the purpose for which EASE was intended; • specific scenarios in which the stakeholders had used the model; • details of any EASE validation work carried out by the stakeholder; • limitations of the model and suggestions for improvements; • details of other exposure assessment models used and • stakeholders’ views on the accuracy and precision of EASE predictions.
A proforma was developed to help record systematically the information obtained from the interviews.
3.2.3 Contacting stakeholders
Stakeholders were contacted by either email or fax, which contained information regarding the aims of the study and requested their involvement. Where necessary, several attempts were made to contact stakeholders over the study period. In instances where they had left an organisation, steps were taken to either trace them at their new employer or to identify another suitable stakeholder for interview. The database was updated accordingly to take account of this new information.
21
3.2.4 Conducting the interviews
commencing the interview, the researcher again discussed the main aims of the study and confirmed that the stakeholder was happy for the conversation to be recorded.
RESULTS
3.3.1
All except two were conducted by
minutes). The remaining stakeholders requested a paper copy of the questions to complete at their convenience. A list of questions was forwarded and these were completed and returned to the researcher.
Despite several attempts, nine stakeholders could not be contacted. Five others had left their It was interesting
to note that four individuals who had received copies of the EASE program had never used it
to participate with no explanation provided.
The survey participants belonged to a range of countries including EU member states, Norway and the USA. Table 3 indicates the number of stakeholders from each country that were interviewed.
Table 3: Participating countries
Country No. of interviews Finland 2 Germany 4 Hungary 1 Netherlands 4 Norway 1 Ireland 1 UK 10 USA 4
As mentioned in Section 3.2.1, stakeholders were coded a priori as academic, government or industry. industry users accounted for just over half of the interviews conducted.
Table 4: Number of stakeholder interviews by category of EASE user
Group Number of interviews Academic 5
8 Industry 14
To ease the stakeholder gently into the interview, background information concerning job titles and their role within the organisation were sought. These were summarised, where
To help maintain consistency, the interviews were undertaken by the same researcher. Before
3.3
Stakeholder characteristics
Twenty-seven stakeholder interviews were completed. telephone and recorded, with the interviews lasting on average 34 minutes (range 15-50
respective organisations and no suitable alternative person was identified.
and were also not involved in the regulatory risk assessment process. One individual refused
Table 4 details the number of stakeholders in each category and it is evident that
Government
possible, into five broad categories – practising occupational hygienist, regulator or consultant
22
working on the regulatory risk assessment process, occupational epidemiologist, occupational toxicologist and other (Table 5). It should be noted that some stakeholders fell into and were
Table 5: Summary of stakeholders job category
Job category Number Practising occupational hygienist 14
Regulator / consultant working on regulatory risk assessments
Occupational epidemiologist 3
Occupational toxicologist 3
Others 1 Researcher / lecturer 1 Occupational hygiene
manager 1 Analytical chemist
It was anticipated that the responses and views of academic, industry and government stakeholders might be different therefore the interview responses were summarised accordingly.
3.3.2 Previous experience with EASE
Purpose of EASE and extent of use
Stakeholders were first asked to state what they thought the purpose of EASE is and to provide details of the extent of their use of the model.
Academics
stated that EASE was a tool for assessing exposure in situations where workers are exposed and little or no exposure measurement data is available. Only two stakeholders made reference to the fact that it is a tool used for regulatory purposes,
physical properties. These same stakeholders also made reference to the fact that EASE is increasingly being used by occupational hygienists in actual workplaces to obtain estimates of exposure. One stakeholder had a slightly different way of describing the purpose of EASE, stating that it is a way of obtaining a “risk number” for conditions and compounds which are unknown or for which they have little information.
use was generally fairly low. One stakeholder had used the model once, another once or
stated that they used the model more extensively e.g. on average more than five times per year.
Industry
Over half of the industry stakeholders were aware of the regulatory role of the model both for
friendly tool for doing exposure assessment in the regulatory context”. One stakeholder
counted in more than one category.
12
All the academic stakeholders
one of whom added that it predicts possible exposure levels based on their chemical and
All of the academic stakeholders had been aware of EASE for some time but the frequency of
twice a few years ago, and another 4 or 5 times over the past few years. Two stakeholders
new and existing substances. According to one stakeholder, EASE is “designed to be a user
23
expanded on this, stating that the tool provides the user with some idea of possible risk and is based on measurements from the 1960s and 1970s held by HSE. Another stakeholder emphasised its use in assessing exposure to downstream users where little or no data is available.
Just under half of the stakeholders saw the model as being something an occupational hygiene professional could employ, using it as an aid to risk assessment and helping them make a quick and easy decision on where further investigation or measurements needed to be made. “It helps individuals model exposure without taking expensive measurements”, “useful tool when you have no idea what exposures are like”, “need a model to help you”. One stakeholder also thought it was a useful tool for smaller companies, with limited funds available, to help them prioritise activities of concern. One occupational hygienist said that he could see how the tool could be useful for the purposes of regulatory risk assessment given that it provides an overall general estimation of exposure for a large population group. However, he felt it wasn’t suitable for practical occupational hygiene purposes given that it is not specific enough to deal with particular situations.
As with the academic users, most industrial stakeholders had been aware of EASE for some time (most 3-5 years), some of who had used the previous version 1. Use of the EASE model with these stakeholders was very variable. One stakeholder had not actually used the model (though he had seen the outputs); three had used the model only once and four used the model on a few occasions a number of years ago. Another user stated that he used the model on a few occasions and used it as the basis of developing his own model, which was then used extensively. Four mentioned that their use of the EASE model was very variable, using it extensively for short periods of time and very occasionally at others. Only one stakeholder reported using the model extensively as a routine measure, when carrying out risk assessments for clients.
Government
All bar one of the government stakeholders noted that EASE was used for the notification procedure under European legislation for both new and existing chemical substances. The remaining stakeholder simply noted that EASE helps give a simple figure for exposure assessment. Two stakeholders also referred to the use of EASE in the workplace by practising hygienists, with one commenting that he felt the use of the model for this purpose was inappropriate. He was currently involved in a research project aimed at assessing the use of EASE for this purpose. The second stakeholder felt that providing there was good guidance available, the model could be a useful tool for occupational hygienists.
Most stakeholders had been aware of the model for several years, and again, experience of using the model was very variable. One American stakeholder had not actually used the model, the reason being that she had helped develop and used mathematical models that fulfilled her particular requirements. She mentioned that she was interested in the model but that it was different to those she would normally use. She mentioned that she didn’t have the time needed to understand it and develop “a degree of comfort with using the model”. One stakeholder had very rarely used EASE over the past two years whilst another had used the model extensively over the past few months for research purposes. Two stakeholders had used the model a couple of years ago for particular substances in the regulatory risk assessment process, whereas another two used it several times a year for this purpose. Only one stakeholder could be truly classified as being an experienced user, using the model on almost a daily basis.
24
Installation and usability
As with all software tools, seamless installation and compatibility with the computer hardware used is essential, as is a “user friendly” interface and explanatory text. Stakeholders were therefore asked to comment on these factors.
Academics
Only one stakeholder reported problems with the installation of the software, although he was only able to provide very limited information. He mentioned that during a workshop for occupational hygienists on the use of the model, some computers were missing files that were needed for the program to work correctly. He mentioned that he had contacted HSE regarding this problem at the time and a new package was issued. No other problems were experienced on receipt and use of this.
All of the stakeholders found the computer interface easy to operate though one did comment that they didn’t like the program opening into three windows and the need to “scroll down” for information.
With regards to the explanatory text, there were some differences in opinion. One stakeholder felt that he couldn’t comment given that his experience was extremely limited. One stakeholder stated that for those with many years’ occupational hygiene experience, it would be very clear, though perhaps not so for those with more limited knowledge. One stakeholder stated that it was a good idea to include explanatory text but that the terms and words presently used were not always understandable, though when prompted for examples he was unable to provide further information. Another stakeholder said that during a workshop, users went through several of the scenario options and combined the ranges obtained from each in an attempt to avoid the category selection problem. Although he recognised that this was a less than satisfactory way of using the model, the users were not aware of how else to deal with this problem. The stakeholder also reported that the workshop participants had particular problems with understanding the terminology “non-dispersive” and “wide dispersive” use. It was also not clear to them when closed system / full containment applied.
Industry
Only one stakeholder experienced problems with installing the software, although was unable to provide any further information given that their IT department resolved the problem. Again few problems were reported regarding the user-interface, with most stating that it is easy to use. However two respondents did mention that what with the developments with software packages over the years, steps should be taken to make the package look more professional and “slick” (it was felt at present that it looks “old and clunky”). The stakeholder who was unhappy with the interface couldn’t remember specifically what he didn’t like (it had been a number of years since the model was last used) but thought it needed improving. One stakeholder did comment that you could obtain an exposure estimate for anything “e.g. a house or a cat”. As he explained, the name of the substance is typed in and once you have been through the process an exposure estimate is given. However any text can be entered at the start and it is not clear that the exposure estimate relates to the characteristics selected and not the name typed in at the start.
This stakeholder group was divided in their opinions regarding the explanatory text. One couldn’t comment on this given that he hadn’t actually used the model. Six found the explanatory text clear enough, with one stating that it didn’t use terms unfamiliar to an occupational hygienist. Another stated that once they realised the explanatory text was available they found it useful. Lastly, one stakeholder thought
25
that the text was too simplistic and stated that it was akin to “modelling for 12 year olds”. However, a number of stakeholders had problems with the jargon and felt that it needed better explanation. One stakeholder stated that the vocabulary would be better explained by examples. Although he now understands what “wide dispersive” use means, he didn’t when he first encountered it, and a second stakeholder also thought this term was confusing. Two stakeholders were particularly unimpressed with the terminology, both stating that it was difficult to understand what was actually meant by the different classes and that clarification is desperately needed. One stakeholder also stressed that it is vital that the terms are as simple and understandable as possible given the global use of the model.
Government
None of these stakeholders experienced any installation problems and those who could comment on the user interface found it easy to use. However one stakeholder did mention that improvements are always possible and thought that the input of the user name at the start was an unnecessary step.
One stakeholder was particularly unhappy with the explanatory texts, finding them very vague. He thought that a collection of exposure scenarios, possibly accompanie d by photographs would help make the distinctions clearer. A similar idea was also suggested by a second stakeholder, who thought that it would be useful to develop an application manual containing practical examples aimed at helping users deal with more complicated situations. A third stakeholder also thought the explanatory text required improvement and mentioned that even though he had used the model extensively over the past four years, “non”- and “wide dispersive use” still gave rise for discussion, as did “inclusion onto the matrix”. One non-native English speaker did experience problems with the terminology but once a native English speaker had explained this he was able to understand it fully. The other three stakeholders who had used the model all found the explanatory text acceptable.
Examples of previous use of the model
Stakeholders were asked to provide details regarding the purpose and scenarios for which the EASE model had been used and to provide an indication as to whether the model fulfilled their requirements.
Academics
One academic stakeholder had not actually used EASE, stating that he had browsed through so to get a feel for the model. The other stakeholders had used the model for a variety of different purposes.
One stakeholder reported holding a one-day workshop for occupational health and safety experts concerning the use of EASE in the profession. This workshop was entitled “Occupational Health Assessment of Exposure to Substances at the Workplace without Measuring: the EASE Expert System” (Van Rooij and Jongeneelen, 1999) and led to a validation of the EASE model, which is reported in Section 4. Extensive exposure assessment is also currently being undertaken for a number of compounds, including chloroprene and intermediates. As part of this assessment the EASE model is being used, with its outputs being compared with actual measurements wherever possible. Although the work is still ongoing, the stakeholder did report that the model is a very good tool, though not perfect. Another stakeholder focussed on the dermal model to assess exposure during a spray painting task. The purpose of this exercise was to compare the estimates obtained from the EASE model with those from a model he had developed. He mentioned that EASE overestimated by a factor of 2.
26
The EASE model had also been used in retrospective exposure assessments. In the first instance, the inhalation model was used to obtain estimates of exposure to carbon black for various tasks in the workplace. Task-based estimates of dust exposure were obtained, with the stakeholder finding it relatively simple to place real-life work situations in terms of the EASE categories, although some broad assumptions had to be made. When questioned about the ranges obtained, the stakeholder thought that they were too wide, and although were probably accurate, were not very precise. However he felt that it would be difficult to tighten up the dust ranges given that this would require more information to be entered into the model. He also felt that it would be difficult to develop a model that provides very accurate estimates. Although EASE fulfilled his requirements, he felt that, overall; it didn’t provide him with any information that he didn’t intuitively know already. In the second instance, EASE was used to retrospectively assess solvent exposure in a variety of situations between 1970 –1990. The respondent was very enthusiastic about the use of the model stating that, as far as he was concerned, there was no better model for assessing exposure.
Industry
As mentioned previously some users had very limited experience of using the EASE model. One such stakeholder, who had simply browsed through the package to see what it could do, felt that the tool was conservative in the sense that it overestimated exposures. He stated that this did not concern him given that he would prefer to over-protect workers and saw EASE as being a very quick decision-making tool, highlighting areas where further action is necessary. He also mentioned that the tool was not routinely used by his organisation as they felt that it required further validation and others within the company were concerned by the degree of overestimation because of possible cost implications associated with implementing recommendations on the basis of these estimates. Another stakeholder stated that he did not use EASE, as he felt it didn’t provide him with useful information. He felt that as an occupational hygienist it was better to look at an activity and use professional judgement to assess the situation rather than rely on a model.
Two industry stakeholders used EASE for retrospective exposure assessment. Using only the inhalation model, exposure to a mixed mineral substance from a steel operation was assessed as they were required to use a “rough and ready approach which was perceived as being more authoritative than a subjective, albeit an informed, assessment”. They did mention that the EASE outputs seemed realistic when compared to their own subjective opinions. The second stakeholder used the inhalation model in a retrospective cohort study where estimates of exposure to benzene were required. He felt that as EASE is based on data from the 1960’s and 1970’s it would be a useful tool for this purpose. The dermal model was not used, with the stakeholders stating that it was “of no beneficial use whatsoever”.
Concerns regarding the dermal model were evident throughout this stakeholder group, with many not using this component of EASE. One stakeholder said that he did not place a lot of value in the dermal model compared to his own “back of the envelope” calculations. One stakeholder who used the dermal model compared the outputs with actual measurements and found that it overestimated exposure. He also mentioned that in his study, most of the dermal exposure occurred on the face and neck, which the EASE model does not even consider.
Other situations where the EASE model has been used include:
• Worst-case estimates were obtained for chlorinated alkanes and these were found to overestimate exposure. The stakeholder was unsure as to whether the model was over predicting or whether it was due to the information being fed in. (The stakeholder had not personally used the model and had relied on a colleague passing on the outputs).
27
• EASE was used to predict task-specific exposures, for example draining liquids from a pump over a 15-minute period for a wide range of single solvents i.e. MEK, butanol, naphtha and hexane. In most instances it was felt that the exposure estimates were reasonably precise and accurate, with EASE fulfilling his requirements in that it produced a range he could actually work with. As the stakeholder had enough information about the processes, he was able to place the situations into the EASE categories with no difficulty.
• The American stakeholder used the model to obtain task specific and worst case estimates for both dermal and inhalation exposure to Primene 81R (mixture of isomeric amines) as he was reporting to the European Regulatory Body and wanted to “speak the same language”. The stakeholder thought the outputs were “right on the mark” compared with monitoring data and other modelling outputs, once he had adjusted this to take account of the fact that EASE deals with pure chemicals. He mentioned that he found it straightforward to place real life situations in terms of the general categories although he had to make some assumptions when dealing with downstream users.
• Task specific exposures, which were then used to calculate 8-hour time weighted averages, were obtained for a Styrene oligomer (a low volatility product). The stakeholder felt that the predictions were quite high compared to what they anticipated, and what was gained from other models.
• One stakeholder used EASE for a range of substances including pesticides, fungicides, inorganic substances, metal compounds, zinc compounds and solvents. He always looked at task-specific exposures and in his experience, EASE always over-predicted exposure. He did not feel that the predictions were sufficiently precise and accurate, however, he thought that the estimates obtained for vapours were reasonable, although conservative.
Government
With two exceptions, the government stakeholders had used the EASE model as part of the regulatory risk assessment process. In most instances, EASE had either been used a number of years ago or was used currently on a fairly extensive basis. Details of previous use were sketchy, as stakeholders could not remember much of the detail.
In the instances where EASE was used for only one substance, the following comments were obtained. The dermal model was used as part of the risk assessment for acrylonitrile. The inhalation model was not used as they had extensive data on this aspect of the substance. Various scenarios were reviewed, focussing primarily on rubber production. As the stakeholder had no dermal monitoring data he was unable to comment on the accuracy and precision of the outputs though he felt they were very conserva tive. No problems were experienced in placing real-life work situations in terms of the EASE categories. Another stakeholder noted that he had used EASE for single solvents but no further information was provided on his findings. He did however state that in some situations he found it difficult, and in some cases impossible, to place real-life work situations in terms of the EASE categories, although he did also note that generally he had to make broad assumptions about the task to feed into the model.
For those stakeholders who had used the model for several substances over many years, the following comments were obtained:
28
• Both the inhalation and dermal model had been used. The stakeholder thought the dermal model was very poor, imprecise and needed improving for future use. He noted that dermal exposure is very complex and felt that the ‘RISKOFDERM’ project would be useful for the development of such a model and would provide answers as to what factors need to be considered when assessing dermal exposure. RISKOFDERM is a multi-centre Pan-European project aimed at improving understanding of the nature and range of dermal exposures to hazardous substances throughout the EU (RISKOFDERM, 2001). Based on his experience, EASE overestimates when considering powdered chemical intermediates. For example, his research suggests values of 1 mg/m3 whereas EASE estimates 2-5 mg/m3. The stakeholder felt that expert judgement is needed to place real-life work situations into the EASE categories but that it was sometimes difficult to assign categories. He therefore felt that more precise decision boundaries are required.
• Another respondent who used both parts of the model felt it was impossible to discuss fully, given that he had used EASE for approximately 50 new substances per year and also all the existing substances the organisation had to deal with. Points raised were that scenarios involving dumping, where large amounts of solid substances are added to a process, led to underestimates both for inhalation and dermal exposure. He also thought that EASE wasn’t accurate enough for liquids with low vapour pressures or for substances formed during a process. He also thought it was relatively easy to place real life work situations in terms of the EASE categories but did acknowledge that he had used the model for over four years and so was a highly experienced user.
The model is also being used as part of a validation exercise to assess the potential use of EASE by practising occupational hygienists. This work has recently finished and is currently being written up for publication.
3.3.3 Future applications of EASE
Steps were taken to establish other applications that stakeholders envisage using EASE for in the immediate future.
Academics
Only one of the academic stakeholders was currently using EASE and anticipated using it in the immediate future. As mentioned previously, EASE is being used to help predict workers’ exposure to chloroprene and other various intermediates. As part of the exposure assessment, actual measurements will be compared with the EASE predictions. It is hoped that the results of this validation exercise will be available April – June 2003.
Industry
Most of the industrial stakeholders had no future plans to use the model. One stated that he had put use of the model “on hold”, as he was aware it was being upgraded. He did state however that he would consider using the model as part of the design criteria when developing new plants. One stakeholder is still using the model as part of a validation exercise, with another two still regularly using the model, either to check competent authority submissions or as part of their consultancy work.
Government
One stakeholder stated that he is constantly using the model with another two stating that it is being used by others within their organisations for regulatory risk assessment.
29
3.3.4 Validation of EASE against measured exposures
Academics
One of the stakeholders reported conducting a previous validation exercise of the inhalation model of EASE (Van Rooij and Jongeneelen, 1999) and this is discussed in detail in Section 4. At time of reporting, a validation exercise of EASE outputs and actual exposuremeasurements for chloroprene and intermediate products was being undertaken as part of a larger research project in the USA. Preliminary spot checks have shown that EASE underestimates considerably for high molecular weight compounds and intermediates and a full report is anticipated mid 2003.
All of the respondents were aware of validation work undertaken by other organisations. Research mentioned included that undertaken by the Institute of Occupational Medicine (Hughson and Cherrie; 1999, 2001) and the University of Birmingham (Mark, 1999).
Industry
Only two of the stakeholders had attempted to formally validate the EASE predictions with actual measurements, and most were not aware that validation studies had been undertaken. There is an ongoing validation study for chlorinated alkanes, and previous studies attempting to validate dermal aspects of the model were identified (Hughson and Cherrie, 2001). Although not a formal validation, good agreement with actual monitoring data was observed for MEK and Primeme – 81R. Other validation studies that were mentioned were work undertaken by BauA (Bredendiek-Kamper, 2001) and HSE involving lead industries (Wheeler, personal communication).
Government
One stakeholder had just completed a validation study sponsored by the Ministry of Social Affairs and Health, Finland, in which actual exposure measurements were compared with EASE estimates. As this work is soon to be published and the results are presently confidential, no further information was provided in this report.
Three stakeholders were aware of previous validation work including studies by IOM (Hughson and Cherrie; 1999, 2001), BauA (Bredendiek-Kamper, 2001) and also INRS (Vincent et al, 1996).
3.3.5 Strengths and limitations of EASE
Academics
Stakeholders liked the fact that the exposure assessment process was put into a clear framework and that assumptions made had to be clear and explicit. One stakeholder also commented that he liked having the log session window, as this allows the user to review what categories have been chosen. He also liked the fact that the model is easy to use, not too sophisticated and provided an output after entering only a few pieces of information. One important point raised was that EASE can never replace actual measurements, and problems will occur if people attempt to do so. It was also stated that individuals should be aware of its limitations when using the model.
Many limitations of the program were raised and these are summarised below: • the ranges are wide and crude; • EASE overestimates substantially in some scenarios;
30
• the user has to enter in the vapour pressure manually; • amount of substance is not a determinant of the assessment process; • no distinction is made between pure substances and mixtures and • the distinction between task-related assessment and 8-hour time weighted averages
should be explained more.
Two stakeholders had not actually used the dermal model and so were unable to comment on this. However one did state that he hadn’t used this aspect of EASE because there isn’t any data or information on the underlying model and he therefore doubted the quality of the model. One respondent who had used the dermal model was unhappy with the fact that it focussed solely on exposure to hands and forearms, instead preferring that the whole body be accounted for.
Overall it was felt that EASE should be used to provide an indication of what the likely exposures should be, thus providing some sort of ranking and prioritisation for further actual measurements.
Industry
As with the academic sector, some of the industry stakeholders’ experiences with EASE were fairly limited. In some instances the model was used simply to obtain a feel for what it could do. Others had only used certain aspects of the model and many had not used it for dust scenarios or dermal exposure assessment. This again will impact heavily on what the perceived strengths and weaknesses of the model would be.
Stakeholders reported the strengths of the model as including the following. They liked the fact that it was easy to use, simple and that each route could be tested quickly and easily, with an estimate being obtained after just a few steps. Many saw it as being a “quick and easy decision making tool”. Other strengths mentioned included the fact that as it was based on actual workplace measurements (‘cuts through a lot of problems associated with deterministic models’), it is less subjective, can be used consistently by others and is useful for retrospective exposure assessment purposes. One stakeholder mentioned that he liked the fact that it is well documented in its branches and decision logic.
As with the academic users a number of limitations of EASE were highlighted:
• the model is very simplistic, with scenarios not always able to be categorised correctly;
• EASE does not include short-term exposures and workers very rarely carry out the same tasks for 8-hour days;
• the choice of only four different categories of use scenarios was felt insufficient; • model needs to be updated with more recent measurements to keep predictions in line
with new technology; • model doesn’t consider how close worker is to the source, and general ventilation; • no clear guidance provided on how to deal with the problem of mixtures (this is a
particular problem when dealing with downstream users); • model doesn’t differentiate between respirable and total inhalable dust (although this
is implied) or different dust control techniques; • dermal model “loose”, focuses only on the hands and forearms when it should include
chest, legs and possibly also the torso, and doesn’t give any consideration to the use of PPE;
• the quality of the software, on-line instruction and written material need improving; • the “science” behind EASE isn’t transparent and • the model can easily be used incorrectly or inappropriately.
31
Government
Simplicity was again a key strength mentioned by all stakeholders. One stakeholder mentioned that it was so surprisingly easy to use that she thought she was doing something wrong. She was also surprised that there were so few steps required to reach the end point. One stakeholder commented that in most instances EASE was realistic or overestimated, which was good given that it erred on the side of caution. Another stated that it is the only exposure model that can be used for all situations and feared that no money would be invested into developing other models for regulatory risk assessment. Lastly, one stakeholder mentioned that the model is sufficiently reliable in most cases and felt safe in the knowledge that a good level of expertise and knowledge was used to develop the model. However, another stakeholder did question whether it was sufficient to base a decision on the output of the EASE model alone as it is quite subjective in how users interpret the different use patterns.
The stakeholder who was the most experienced EASE user felt that the limitations of the model were as follows:
• the dermal model wasn’t valid and shouldn’t be used (however he did recognise that he believed there weren’t any alternatives);
• the actual measurement data on which the model ranges are based are too old and more up to date data from a variety of UK and non-UK sources should be included;
• the development of the model was not transparent; • the amount of substance used is not taken into account by the model; • the model cannot be used to assess exposure to substances that are emitted as a by-
product from a process; • the terminology is not always clear and further explanation is required and • lastly, consideration is not given to substances of different particle sizes.
Other limitations mentioned by stakeholders included the following:
• the model cannot be used to provide estimates for spray application; • it is not clear how to handle partial vapour pressure; • doesn’t include many parameters that in practice would impact on exposure, for
example, ventilation rate, temperature changes and duration; • one stakeholder felt that effectiveness of control measures should be considered and • the same stakeholder was also of the opinion that the terminology for the pattern of
control – “direct handling” and “not direct handling” - were not good parameters for dermal exposure assessment, as they don’t provide any information on the degree of exposure.
Similar comments to those listed above were obtained from the stakeholder who had validated the EASE model for practical use by occupational hygienists. This stakeholder also believed it was incorrect that only extensive use contributes to dermal exposure and that the dermal model considers only the hands and forearms. The stakeholder was also concerned that the model couldn’t be used to assess exposure to mixtures, and that the quantity of material used was not taken into consideration.
3.3.6 Suggestions for improvements
Academics
Most stakeholders suggested that steps should be taken to deal with the limitations of the model as highlighted. Some did provide additional comments, suggesting that the model
32
could be more focussed by making it more specific to various tasks and processes and by considering the fact that preparations, rather than substances, are mostly used in industry. One stakeholder also suggested the possibility of enlarging the model in order to combine several tasks and thus obtain an 8-hour time weighted average. However this stakeholder recognised that if such improvements were made it could result in a more cumbersome and complicated model, thus detracting from many if its strengths.
Industry
Additional suggestions for improving EASE included developing a model that has a probability distribution rather than a range as its output. From the perspective of a small to medium sized enterprise (SME), it was deemed beneficial to have a tool which is simple, quick, conservative (but not too conservative), and possibly linked with COSHH Essentials. It was also felt that the underlying data in the model should be expanded so that it is both updated and includes European wide information. Again the need for improving the dermal aspect of the model was stressed.
Government
Most of the suggestions for improvements have already been discussed. A second stakeholder also suggested using a COSHH Essentials approach to modifying exposures in the model, taking account of the amount of substance used. It was also suggested the dermal model should either be dramatically improved or removed and that more databases should be attached to the model. One stakeholder felt that EASE should be ‘self-learning’ model like the BEAT model claims to be. (The BEAT model has been developed by HSE for dermal exposure to biocides and is discussed in Section 5.7). One stakeholder stated that “EASE demands improvements and a guide for unified, harmonised application” and felt that it could be improved by basing the model on more measurements. They also felt that comparing EASE with other models and data exchange with experienced users would help identify what further improvements are needed.
3.3.7 Accuracy and precision of exposure estimates
To aid in the development of an updated version of EASE, stakeholders were asked to comment on how accurate and precise they felt exposure estimates should be and whether an improved version of EASE is actually needed for the purpose of regulatory risk assessment.
Academics
Three stakeholders had no problem with responding to the question on how accurate exposure estimates should be; stating that exposure prediction should be within a 2-fold, 100% or an order of magnitude of true exposure values. The other stakeholders had difficulty in answering this question. The first stated that it all depends on the remit of use of the tool and again emphasised that a model will never replace actual measurements. The other stated that he simply did not know and felt that as the model is used primarily as a ranking tool, the relative accuracy of the situations was important.
None of the stakeholders were able to comment on whether an improved version of EASE is needed for regulatory risk assessment given that they had not used the tool for this purpose.
Industry
Again some stakeholders felt it was difficult to suggest how accurate and precise exposure estimates should be if EASE was updated, stating that it depends heavily on what the scope of the model is. However, it was considered that it should be accurate and precise enough to
33
support decision-making. Most felt that the degree of overestimation needed to be reduced to reflect current work practices. Factors of 2-fold and 10-fold and also 20% of the true average were suggested. However it was again mentioned that in order to obtain a high degree of accuracy the model would become very difficult to use, with the exposure estimate perhaps being based on only a few data points. One argued that there is no point investing a large amount of time and money making answers only slightly better as, at the end of the day, it is a model.
On the whole, stakeholders believed that an improved version of EASE is needed, making statements such as: “feel it is absolutely necessary, always room for improvement” and “need to make it the best we can”. One stakeholder however felt that the model probably couldn’t be improved with the data that it currently contains. Another stakeholder voiced this sentiment stating that more data would be necessary to support these changes and that the data should always remain up to date.
Government
One stakeholder simply commented that an improved version of EASE is urgently needed for regulatory risk assessment. However, stakeholders found it very difficult to comment on how accurate and precise exposure estimates should be. One stakeholder, who had not used the model for the regulatory risk assessment purpose, simply said “the more accurate the better”. Another stakeholder expanded on this, stating that plus or minus 10% of the true average would be good. However they were not sure how this could be done, and there was also concern that this would lead to a cumbersome model. Another stakeholder stated that the degree of accuracy and precision required is dependent on the goal one is trying to achieve. She felt that risk assessment should involve a tiered approach. This sentiment was also echoed by another stakeholder, who stated that screening level models should purposefully over predict and that they would be comfortable with it being within an order of magnitude of the true values. For a more detailed risk assessment, the exposure estimates should obviously be as accurate as possible and in this area actual measurements would be essential.
3.3.8 Other exposure models used
Stakeholders were also asked to provide some basic information on other exposure models they had used.
Academics
One stakeholder stated that he didn’t use any other “off the shelf” models and had developed his own statistical models when undertaking exposure assessments. Another stakeholder had also developed his own model for assessing exposure and uptake of solvents in sprayers. He had compared the outputs of this model with EASE and, although his were more accurate, more detailed information was actually needed to obtain the assessment.
Another organisation had developed an in-house chemical model that included parameters such as time, wind speed and direction, workers’ habits and temperature of the process. They also mentioned that they had tried to use models developed by the US Environmental Protection Agency (though couldn’t state which), but that these were quickly dropped because they were too complicated. Another stakeholder also mentioned that they used US EPA models such as the Multi-Chamber Concentration and Exposure Model (MCCEM), which estimates average and peak indoor air concentrations of chemicals released from products or materials in domestic residences. This is available from the Internet site http://www.epa.gov/opptintr/exposure/docs/mccemd/.htm and is discussed in Section 5.6. CONSEXPO, a model used to assess consumers’ exposure to chemicals by inhalation, dermal and ingestion exposure developed by The National Institute of Public Health and the
34
Environment in the Netherlands (RIVM) was also reported to have been used and this model is discussed in Section 5.7. Overall the stakeholder felt that these models weren’t as broad as
In comparison to these, the stakeholder concluded that EASE was the best. The final stakeholder in this category used several exposure models although these were all for the purpose of assessing environmental
The stakeholder said that all were useful but as is the case with all models, careful interpretation must be given of the outcomes.
Table 6: Environmental exposure models used
Name of model Purpose Developers Lifeline 1.0
exposure, dose and risks associated with the use of a pesticide active ingredient
The Lifeline Group Inc.
Clea (Contaminated Land Exposure Assessment)
Uses a variety of Monte Carlo simulations
humans become exposed to soil contaminants
UK Environment Agency
Predictive Operator Exposure Model
Generic database of monitored operator exposure studies on plant protection products in Europe.
European Expert Working Group
Industry
Half of stakeholders had not used other exposure models, with one stating that this was because use of EASE was necessary for the purpose of the European Risk Assessment Framework.
model. For his purposes he preferred to obtain a point estimate which he could compare with limit values, thus judging acceptability of exposure. The model required additional
of process, dust potential and time of exposure for each task. The exposures for each task are used to give a daily exposure over the whole shift. This information is then used to judge
data, for example, actual measurements. Another stakeholder also developed a model (although before he had used EASE), which enabled him to consider multiple sources and
However, he acknowledged that this was more complicated than the EASE model.
The other stakeholders had used a variety of exposure models, although had not always compared these with EASE.
Lifeline (“very ambitious”), Wall Paint Exposure Model (WPEM) (“easy to make mistakes”, “comparable values with
by the US EPA’s Office of Pollution and Toxics (OPPT) and is discussed in Section 5.5.
Government
Other exposure models were reported although most of these were for purposes other than regulatory risk assessment. Acrylonitrile (dermal exposure) reported using the SkinPerm model, which predicts uptake rather than exposure. He was not very complimentary about this model, finding it
EASE and that they were difficult to use and even less relia ble.
rather than occupational exposures. These are listed in Table 6.
Evaluation of aggregate and cumulative
to examine different pathways by which
EUROPOEM –
Given that EASE was not fulfilling his requirements, one stakeholder developed his own
parameters such as quantity handled, general local ventilation, vapour pressure, temperature
acceptability with limit values: below 10% of OEL, stop assessment, above 10% collect more
effectiveness of ventilation rates.
These included the saturated vapour pressures model, box model and eddy-diffusive models, which give deterministic values.
EASE”) and MCCEM and EUROPOEM were also mentioned again. WPEM was developed
One stakeholder who was carrying out the risk assessment for
35
complicated and not really applicable. In this instance he felt that EASE was more realistic and fulfilled his purposes better. The American stakeholder reported using a variety of models developed by the US EPA such as MCCEM. As she had not used the EASE model she was unable to compare the various models.
3.4 DISCUSSION
Stakeholder interviews were completed for 27 individuals from a variety of backgrounds. As discussed in Section 2, EASE was intended primarily as a model to assess exposures for new and existing chemical substances as part of the regulatory risk assessment process under European legislation. As expected, all the government stakeholders (except one) were aware of this purpose, although only half of the industry and academic stakeholders were (or at least when asked to define the purpose of EASE this was not mentioned). Overall, stakeholders viewed EASE as being a tool to assess exposures to substances in the absence of actual exposure measurements, with several viewing it as a tool that can be used by practising occupational hygienists, although others felt that the model was not suitable for this purpose.
Although stakeholders had been aware of the EASE model for several years, use of the model was very variable. Of those surveyed, three stakeholders had not actually used the model, approximately seven reported using it extensively, with the majority of respondents using the model either once or twice or for short periods of time with many months or years between use depending on work being undertaken. Such variability in use and experience with the EASE model will impact heavily on the comments and suggestions made as stakeholders’ terms of reference will be different.
As with all software tools, seamless installation and compatibility with the computer hardware used is essentia l, as is a “user friendly” interface and explanatory text. Only two stakeholders reported problems with the installation of the software although neither could provide any additional information as to what the problems were. As these stakeholders were in the minority, with their problems being quickly overcome it may be felt unnecessary to make any major changes to this aspect of the program. However, it would be advisable to ensure, as far is reasonably possible, that any new software distributed can run on all currently available computer systems. Overall, stakeholders were happy with the computer interface, finding it simple and easy to operate. However three stakeholders did make minor comments on this aspect of the model. For example, they didn’t like the fact that the program opened into three windows and they needed to “scroll down” for information; they felt the interface could be updated to make it look more professional and the input of the user name at the start was an unnecessary step. The user interface is a legacy from the very first Leonardo prototype, with the first DOS version of EASE having a similar interface layout. The Windows versions have also kept the same general layout. The stakeholders thought that it is essential that the software is updated and has a more polished feel in order to convey the right impression. They also felt that the implementation of such changes would be fairly simple and inexpensive to achieve given that they would not affect the actual structure of the model itself. However, this is not necessarily true, with previous experience indicating that development of the user interface is very time-consuming. For example, at least half of the project time and cost was spent on the user interface in the last revision of EASE.
There was a difference in opinions with regards to the explanatory text. Whilst many of the stakeholders found the text easy to use, with exposure categories clearly defined, many experienced problems with the terminology used. One stakeholder did state that for people with many years occupational hygiene experience it would be very clear, though he was not sure whether that would be the case for other users. He felt that it is vital that the terminology is clearly understood by all qualified users of the model, with all jargon removed, so that it can be easily used across Europe, if not globally. In other words the program should include terms that an occupational hygienist or risk assessment specialist (though not necessarily a
36
production operator) would understand - it should befit the general characteristics of the end user. Particular problems were identified with “non-dispersive” and “wide dispersive use”. It was also not clear when “LEV” and a “closed system / full containment” applies despite the explanatory text that was provided. Clarification of the terminology is needed, and the stakeholders’ suggestion of including a collection of exposure scenarios, possibly accompanied by photographs, is one that could be considered further. It must be emphasised that unless future models deal with characteristics in terms that users know and understand, they are unlikely to be generally accepted.
EASE has been used in a variety of circumstances, in addition to its original purpose. For example, academic stakeholders have conducted validation studies, comparing the outputs of EASE with actual exposure measurements. The model had also been used for retrospective exposure assessments for epidemiological studies and industry stakeholders have also used the model for retrospective exposure assessment. EASE has also been used as a practical industrial hygiene tool, helping stakeholders assess exposure in the absence of monitoring data and helping them prioritise where further action or actual measurements are needed. Stakeholders’ opinions regarding the accuracy and precision of the model varied, depending on the purpose for which they used the model and the substances assessed, and so it was again difficult to summarise these succinctly. As expected, most of the government stakeholders had used the model for regulatory risk assessment. However details obtained about previous use were sketchy as stakeholders were unable to remember a lot of the details.
Only those government stakeholders involved in the regulatory risk assessment process reported that they would be using the model in the future. Few of the other stakeholders anticipated using the EASE model in the future. One of the academic stakeholders did report that they were validating EASE outputs with actual measurements as part of a larger research project and that this would be reported mid 2003. As part of the ongoing review of EASE it is important that this and other validation work is considered in due course. Another industry stakeholder is also using the model as part of a validation exercise and the results of this should also be considered in due course. Two other industry stakeholders reported that they would still regularly use the model, either to check submissions to a competent authority or as part of their consultancy work. In some instances it appeared that stakeholders were disenchanted with the model and would perhaps be more inclined to use it in the future if it was updated and improved. For example, two industrial stakeholders commented that factors such as quantity of substance handled, general / local ventilation, time of exposure, position of worker relative to source, were all important factors which should be addressed when assessing exposure. They themselves had taken steps to develop their own models capable of addressing these factors. Perhaps if EASE were updated to consider such factors more users would be attracted to using the model and would continue to do so.
Some stakeholders had either undertaken or were aware of studies that had validated the EASE model with actual measurements and these are reported and discussed in more detail in Section 4. As mentioned earlier, it is important to be kept aware of future validation studies and to consider their conclusions when updating and amending the model.
A general theme from stakeholders’ comments was that they placed greater value on actual workplace measurements than computer predictions or models. The fact that stakeholders, particularly in industry, do place greater emphasis on actual measurements can perhaps explain some of the poor uptake and use of the model within this sector. Those industrial stakeholders interviewed generally had the resources available to undertake extensive monitoring however; it is felt that this is not indicative of industry as a whole.
The main strengths and limitations of the EASE model as highlighted by the three stakeholder categories are summarised below. As previously mentioned, use of the EASE model varied; with the model also being used for different purposes. It is felt that particular attention should
37
firstly be given to the ‘government’ stakeholders’ comments, given that they have actually used the model for the regulatory risk assessment process. However, this should not detract from the value of the comments from other stakeholders, given that the model is being used for other purposes (and increasingly so, particularly by practising industrial hygienists), with a lot of their comments being very relevant and applicable in the broader sense.
Overall, stakeholders felt that the strengths of the EASE model were as follows:
• substance assessment is put into a clear framework with clear and explicit assumptions being made;
• the model is easy to use; • each route can be tested quickly and easily; • the model is based on actual workplace measurements; • EASE tends to overestimate exposure, thus erring on the side of caution; • the model is sufficiently reliable in some cases and • the log file allows users to review the categories chosen.
These comments are encouraging and give a clear indication of what stakeholders like and want from an exposure model and it is vital that these are maintained in any future developments. It also suggests that the original aim of producing a model that could predict workplace exposures to a wide range of hazardous substances had been successfully achieved. However stakeholders were prompted to identify limitations and concerns they had with the model and some of the more important are listed below:
• Wide, crude ranges, with EASE consistently overestimating, substantially so in some scenarios. Although stakeholders valued the fact that EASE overestimated, the degree by which it did so was of concern and indeed prevented some from continuing to use the model. The vast majority of stakeholders requested that the output of a range of exposures should be maintained as this reflects real-life work situations, though these should be tightened up somewhat.
• The amount of substance is not a determinant of the assessment process. The ability to assess quantity of substance used could also help tighten up the exposure ranges obtained.
• The distinction between task-related assessment and 8-hour exposure should be explained more. It was also suggested that the model be enlarged in order to combine several tasks and obtain an overall 8-hour time weighted average. The development of the model in this way would indeed make the model more applicable to industrial use.
• No data or information on the development of the underlying models is provided, leading to doubt on their quality and science. The publication in the open literature of the development of EASE’s structure and the ideas and principles that underlie it, such as detailed in Section 2, will help stakeholders understand the model more fully and should help alleviate some of their concerns. It may also be useful to include a short summary of the development of EASE in any future guidance that accompanies the software.
• The dermal model is very limited and basic. Many stakeholders were very critical of the dermal model and reported that they did not use it because of this. It was felt that steps should most certainly be taken to improve the model, though they did recognise that this is hampered by a lack of information on dermal exposure and modelling within the scientific community as a whole.
38
• Only four different “pattern of use” options to choose from e.g. “used in closed systems”, “inclusion onto matrix”. This was felt to be insufficient given the differences and availability of processes used within workplaces. It was also noted that it was not possible to differentiate between good and bad control options, which can affect workers’ exposure quite dramatically. The ability to differentiate between good and bad control measures was not originally included given that it was felt unnecessary for the purposes of the notification of new substances (NONS). Also there is explanatory text available concerning effectiveness of LEV. For example, for LEV to qualify as being present it must be appropriate for its intended purpose and operating at or about its design effectiveness. If not, one of the other patterns of control should be chosen. However the ability to include more control options and to somehow assess more fully their effectiveness would obviously help increase the applicability (and hopefully accuracy) of the model to the given scenario and would also help address the first limitation mentioned.
• The model needs to be updated with more recent measurements from a variety of data sources. A number of stakeholders felt that the model might be overestimating partly because it is based on fairly old measurements. (It should be noted however that one stakeholder thought that EASE was based on measurements from 1960’s and 1970’s when in reality the data came from NEDB collected from early eighties to early nineties). The inclusion of new updated measurements from mid to late nineties onwards should help increase the robustness of the model by allowing the ranges to be tightened up. It may also be possible to implement more of the suggested recommendations if there is more data on which to base the exposure estimates. It is noted that the collection and inclusion of new data would require agreement from other organisations and industries, as well as a review of the data to ensure its appropriateness. However if this was to result in the production of a tool which industry, academia and government organisations were more happy with and more willing to use, then it is hoped that this would not be too difficult to achieve.
• The model doesn’t consider how close to the source the worker is and should consider general ventilation rates.
• No clear guidance is provided on how to deal with the problem of mixtures. This is a particularly important limitation given that in most instances mixtures rather than pure substances are used. Again this could account in part for the over-estimation and also represents a problem regarding the application of the model in real-life work situations.
• An inexperienced user can use it incorrectly and inappropriately. As mentioned previously, proper instruction and easy-to-use terminology with appropriate worked examples and scenarios will help alleviate some of these problems. However, it must be emphasised that EASE is not a tool to be used by individuals not experienced in occupational hygiene or the interpretation of exposure data or those who haven’t received proper training.
• The majorit y of stakeholders felt the log file was a very useful element of the model. However one stakeholder stated that it wasn’t clear whether the exposure estimate related to the characteristics selected or the name input at the start. Text is included in the program, which states that the name of the substance is not used in the modelling process but is used to keep a record of the session. However, this should be made more explicit in order to avoid any confusion.
39
With regard to accuracy and precision of exposure estimates, it was again emphasised that in the stakeholders’ opinions a model will never replace actual workplace measurements. However it was considered that the model should be accurate and precise enough to support decision making, but always overestimating to a certain extent so as to err on the side of caution. It was also emphasised that to achieve a high degree of accuracy it would probably require a very cumbersome and complicated model with the resulting estimates being based on only a few measurements within the data set. It would appear that certain compromises would have to be made in terms of accuracy and precision and usability of the model. Indeed this was also evident from some of the stakeholders’ comments on other models used. For example, US EPA models were used although stakeholders felt they were too complicated and reported that this put them off using them again. In most instances stakeholders had not used any additional models. However it was interesting and reassuring to note that in those who had, several felt that in the scenarios used, EASE was the better model.
3.5 FOLLOW UP STAKEHOLDER INTERVIEWS
3.5.1 Remit
A subset of stakeholders were contacted on a second occasion to determine their views on the types of outputs they would like to obtain from an ideal exposure assessment model for the purposes of regulatory risk assessment. The purpose of this was to help develop ideas and recommendations on how the EASE model and indeed any model used for the purpose of regulatory risk assessment should be developed in the future. Such issues are explored in more detail in Section 6 of this review.
3.5.2 Methodology
Nine industrial and government EU stakeholders were contacted by email explaining the purpose of the follow up interviews. In addition, one individual from the HSE not involved in the preliminary interviews was also contacted for his opinions.
Stakeholders were asked to consider the following questions and were informed that the researcher would contact them by telephone for their comments.
In an ideal situation, what outputs would you like to obtain from an exposure assessment model?
For example, for inhalation exposure would you prefer to have the estimates expressed in terms of an 8-hr Time Weighted Average, taking into account all the possible variations in task and exposure times?
Also, would you like such things as the breathing rate of the individual to be considered so that an estimate of uptake can be obtained?
For dermal exposures, should the output provide information on the surface loading on the skin e.g. expressed in mg/cm2? Which anatomical areas should be included? Would it also be useful to consider skin penetration and include dermal uptake?
In theory it may be possible to combine the dermal and inhalation estimates of uptake to obtained a measure of systemic dose. Do you think this would be a useful feature of a revised EASE model? If so, would you also think it could be useful to include exposure by the ingestion route?
It was anticipated that these interviews would last approximately 5 to 10 minutes, and so they were not recorded.
40
3.5.3 Results
Six follow-up interviews were undertaken; one of the responses was provided by email.
Stakeholders were first asked to comment on whether they felt a range or a point estimate of exposure was the most appropriate output from an ideal exposure model. Only one respondent stated that he would like a point estimate. The remainder stated that they would like to obtain a range with further information such as the average, various percentiles and standard deviations also being included. One stakeholder stated that they would also like to obtain a distribution of exposures though was not sure how realistic this would be.
When asked if they would prefer task specific or 8-hour time weighted average outputs, stakeholders were divided in their response. Three stakeholders stated that task specific exposures were best, given that the majority of workers do not undertake the same task for a full working shift. Two respondents stated that they would like 8-hour TWA exposures. The final stakeholder stated that both task specific exposures over a working day and estimates of exposures that may result in significant variations above the background average are necessary. They felt that what was really needed was something which could help give an accurate description of the exposure pattern during the working day.
Opinion was also divided on whether estimates of uptake should be incorporated into an exposure model. Two stakeholders stated that they felt this aspect should not be included given that uptake is a very complicated process, which requires a lot of variables and professional judgement to determine. They felt that incorporating such variables was taking the model too far and what they really wanted and needed first and foremost were reliable exposure values. The remainder of the stakeholders thought it would be useful to obtain an assessment on uptake but emphasised that clear definitions on what is meant by uptake are necessary and that the usability of the tool should not be compromised by the addition of such information.
Stakeholders were then asked to comment on dermal outputs. They were unanimous in agreeing that outputs on the surface loading on the skin should be provided. They were also in agreement that dermal exposure should consider more than just the hands and forearms and that users should be able to select anatomical areas for inclusion in the assessment which are relevant to the task being considered. When asked to comment on whether skin penetration and dermal uptake should also be included in a model, three stakeholders felt that it would be useful to include these parameters. Three expressed reservations stating that these are very difficult issues and were not sure how they would be dealt with. One reiterated again that he felt that a model should concentrate on assessing exposure and that professional judgement was necessary when dealing with these further issues.
All stakeholders felt that consideration to the ingestion exposure route in a model was unnecessary. This was principally due to the view that ingestion is fairly well controlled by basic hygiene practices and that it would be very complicated and time consuming to assess something that is not very important.
The question of whether a measure of systemic dose would be a useful feature of the revised EASE model was also posed. The general consensus was a very hesitant yes but with many reservations being expressed. For example, given that dermal exposure is not well known in general and users would be required to make a lot of assumptions about many factors this could lead to an unreliable assessment. It was also felt that this was essentially doing the regulator’s job and that professional judgement rather than a model is needed, particularly given that many people place great value on models which may not always be giving reliable outputs. However two stakeholders felt that anything which could help with the process would be useful.
41
3.6 CONCLUSIONS
In conclusion, the interviews were very useful in gaining stakeholders’ views on the EASE model and provided useful indicators on how the model should be developed in the future. However, it was also apparent that some stakeholders had not read or understood the documentation provided with the software or used the on-line help. For example, a comment was made that EASE cannot provide estimates for spray application yet the explanatory text for “wide dispersive use” provides advice on this. This in itself highlights areas that need to be improved and explored further. For example, the accompanying literature and help pages must be clear, unambiguous and presented in a manner that will actually encourage users to read them. Steps should perhaps be taken to determine why users do not use the support tools so that measures can be put in place to improve their effectiveness.
Although stakeholders’ experiences and their use of EASE were very variable, and a broad spectrum of views was obtained, general themes were observed. Overall, stakeholders felt that the model should be updated and exposure ranges “tightened up”. Ideally, they would like the limitations of the model to be addressed. However, it was recognised that this may not be possible given the data available. It is therefore recommended that the first step in improving the model is actually updating the data set on which it is based. Steps should be taken to identify new data sets for inclusion from a variety of reliable sources. Stakeholders were very critical of the dermal model and this also requires priority attention. However it was recognised that further research in this field is necessary before any major changes can be implemented, which obviously has broader implications.
Stakeholders did require the exposure estimates to be more accurate and precise, and the inclusion of other variables in an updated model, such as quantity of substances, duration of exposure and effectiveness of control measures (which stakeholders felt were key determinants of exposure) should help achieve this. However they did also stress that the simplicity of the model should not be compromised to any great extent and that the production of a very complex model may in fact hinder, rather than help, users. This further emphasises the fact that the model, its terminology and accompanying literature must be clear, unambiguous and concise, helping the user generate better exposure predictions. Any future developments of the model should involve extensive testing with end users to ensure, so far as is reasonably possible, that this is achieved.
The subset of stakeholders’ views on the types of outputs they would like to obtain from an ideal model for the purposes of regulatory risk assessment did not provide much clarification, with mixed opinions being obtained. It is important to note that overall the general consensus of opinion appeared that the stakeholders principally want a model that will provide a good reliable estimate of exposure, which can then be taken forward and discussed by professionals. Although the inclusion of information such as uptake and systemic dose was thought useful, stakeholders were cautious about this for many reasons.
Several of the stakeholders reported conducting validation studies of EASE with actual exposure measurements and these helped identify a number of the limitations mentioned. There is however great value in reviewing the validation studies previous ly undertaken to produce a general consensus of opinions and determine how their conclusions have been reached and this is discussed in Section 4.
42
4. REVIEW OF EASE VALIDATION AND CONSISTENCY STUDIES
4.1 INTRODUCTION
Models provide a simplified theoretical representation of reality that may help predict some aspect of a situation where knowledge is incomplete. For a model to be useful it must provide a way of generalising users’ understanding of the world so that predictions can be made across a wide range of situations. Models and measurements are inextricably linked; models must define the types of measurement that contribute to predictions and measurements inform on how well the model works in a specific situation.
It is important to make a clear distinction between models and the software implementation of a predictive tool based on a model. For example, as described in Section 2, the EASE model is defined by the set of questions, the decision logic, and the end-point exposure level ranges. The implementation of this model using the CLIPS software package provides a convenient way of arriving at the model prediction. The focus of this section is on the model rather than the way it has been implemented within the software package.
All human exposure models work by using details of the context within which the exposure takes place to produce an estimate of the exposure in that situation. Models may either be deterministic, based on the underlying physical and/or chemical determinants of exposure, or empirical, based on a statistical analysis of a given data set. Further information on the characteristics of various types of models is described in Section 5.2. Given the need to provide a model that can be generalised across a wide range of situations it is unlikely that a statistical model could be appropriate. However, there is no universally accepted theoretical paradigm for human exposure assessment and so there are a diversity of approaches taken in constructing deterministic exposure models. EASE embodies one such approach to modelling human exposure.
A model’s validity is defined by its ability to provide accurate predictions across the range of situations where it is applicable. The validity of a model can be investigated in many different ways. For example, a simple assessment of validity might review the underlying model structure to ensure that it contains the key determinants of exposure as inputs to the model and that the exposure estimates vary with these determinants in expected ways. This is likely to highlight gross inadequacies in the model but will not provide a great deal of reassurance about the overall accuracy of any predicted exposures. A more fundamental assessment of validity might be obtained by comparing the model predictions with measurements of exposure in a number of different situations. Here model validity might be assessed by the average difference between the predictions and measurements (i.e. bias) and the variation in the difference between predictions and measurements (i.e. variability and uncertainty).
Model consistency is a reflection of the precision of repeated predictions for the same scenario. This therefore reflects uncertainty in the information input to the model. For example, if the general ventilation in a scenario is specified as “fairly good”, then this will most probably need to be interpreted in some way before being incorporated into the model. This would be done either by converting it to a numeric parameter or by aligning it with a categorical classification used in the model. Clearly, assessments of precision must be based on a realistic set of input data otherwise the evaluation may be unreliable.
The purpose of this section is to review the information that may shed light on the consistency and validity of EASE. The model’s structure has been evaluated to ensure that it has validity and information has been obtained that compares EASE predictions with measured exposures
43
from both the peer-reviewed literature and scientific reports. The main part of this assessment focuses on inhalation exposure, because this is the aspect of the model that has received most attention, with a less rigorous assessment of the dermal exposure assessment of EASE.
4.2 EVALUATION OF THE STRUCTURE OF THE EASE MODEL
4.2.1 A Conceptual model of exposure
This section contains the description of a conceptual model that forms the basis of a critique of EASE. Schneider and his co-workers (1999) developed a conceptual model for dermal exposure and in this review extended to include exposure by inhalation. The model is a source-receptor model, where the potential pathways from the source of the contaminant in the workplace are tracked through a number of compartments to the person who is exposed. For in halation the contaminant that enters the nose or mouth is of interest, with the contaminant that lands in the skin contaminant layer being of interest to dermal exposure.
For dermal exposure the conceptual model identifies two environmental compartments (air and contaminated surfaces) plus two compartments associated with the person’s clothing (inner and outer clothing layers). There are four key emission pathways from the source: to surfaces, to the air, to the outer clothing layer or direct emission to the skin. In total there are six further transfer processes linking model compartments: deposition, resuspension or evaporation, transfer, removal, redistribution and decontamination; with in total 26 individual processes. Exposure on the skin is described in terms of mass and concentration of contaminant in the skin contaminant layer, the area of skin contaminated and the duration of exposure.
The conceptual model of inhalation exposure involves a similar series of environmental compartments, although because the air compartment is more important for inhalation this compartment must be elaborated by introducing a second subsidiary air volume around the person’s head (i.e. their breathing zone). The two clothing layers have not been included because this review is less concerned with this level of detail. In addition some of the potential dermal (and inhalation pathways) have been omitted for the sake of clarity.
The model is shown in Figure 8, on the following page. The blue compartments are those for the exposed person, yellow shows those compartments strongly influenced by personal behaviour of the subject, grey the work environment compartments and green the outside air compartment. The pink oval is the source compartment.
44
Figure 8: Simplified conceptual model of inhalation and dermal exposure
Source
Room airBreathing zone
air
Surfaces
Respiratory tract
ClothingSkin
contaminant layer
Outside air
This model may be used to identify possible exposure determinants that should be included in a predictive exposure model. Clearly there are a number of important factors associated with the source and some of these are listed in Table 7. Note, in the following text the word “chemical” has been used to denote any hazardous substance or mixture of hazardous substances.
Table 7 shows ten variables that could affect exposure, with some indication of their relationship with emission. In most cases the same factors can be important in both the inhalation and dermal exposure model, although the impact on emission is not always in the same way. For example, higher volatility for liquids would generally increase emission (to air) leading to subsequently greater inhalation exposure; however, greater volatility will also tend to reduce emission (to surfaces) and consequently reduce subsequent dermal exposure potentia l.
Some of these exposure determinants listed in the table are quite general and, for example, some chemicals such as paints emit volatile substances and as they dry the paint solids form a coating on the surface. This drying process effectively reduces the rate of emission for volatile substances and prolongs the duration of emission. It would be possible to elaborate most of the factors to create a greater range of emission determinants.
45
Table 7: Exposure determinants associated with the source compartment
Inhalation emission factor
Increase in emission potential with…
Dermal emission factor
Increase in emission potential with…
Physical state of the chemical (i.e. solid, liquid, gas or vapour)
Gases and vapours of the chemical (i.e. solid, liquid, gas or vapour)
Liquids
Volatility of liquid chemicals
Higher volatility Volatility of liquid chemicals
Lower volatility
Area of volatile chemicals in contact with air
Larger area Area of volatile
surfaces
Larger area
Air temperature and flow rate over liquid surfaces
Higher temperature or airflow
Air temperature and flow rate over surfaces
Negligible effect
Agitation of liquids or solids
More agitation Agitation of liquids or More agitation (splashing)
Particle size of solid chemicals
Particle size of solid chemicals
Coarser particles
Presence of other substances in a mixture
Unpredictable Presence of other substances in a mixture
Unpredictable
chemical (e.g. spraying, scooping, grinding etc.)
More energetic processes
Process for handling the chemical (e.g. spraying, scooping, grinding etc.)
More energetic processes
Localised ventilation close to the source
Less efficient Localised ventilation close to the source
Small effect
Other control measures such as covers over liquid sources or other containment
Less efficient containment
Other control measures such as covers over liquid sources or other containment
containment
Physical state
chemicals in contact with
solids
Finer particles
Process for handling the
ventilation
Less efficient
Inhalation exposure is exclusively mediated through the air; therefore it is only the upper three compartments in the model that are important in this case. Airflow from the room to the outside air, and vice versa, may occur because of natural ventilation or some form of mechanical general ventilation. Cherrie (1999) and others have demonstrated that the size of the workroom and the quantity of air exchanged between the inside and outside of the building determine the average concentration of the chemicals in the air.
However, there will be a complex series of factors that determine the concentration and mass of contaminant chemicals exchanged between the room air and the breathing zone air. Some of the more important of these factors are listed in Table 8, along with an indication of their influence on the net transfer towards the person.
46
Table 8: Factors affecting transfer from room to breathing zone air
Factor
Location of air inlets and exhaust points for forced ventilation
Presence of air jets (emitting contaminant)
Room shape
Air temperature changes with height or across room
Presence of hot equipment or processes
Movement of people in room
Opening of doors and windows
Duration workers close to sources
Physical work rate of the person
Increase in net transfer with…
Unpredictable – depends on the location of source and worker in relation to the ventilation airflows.
Distance of worker from air jet.
Unpredictable – depends on the location of source and worker in relation to the shape.
Unpredictable
Unpredictable – depends on the location of source in relation to the worker. In a large room, a worker close to hot source will have reduced transfer in relation to a cool source. A worker far from a hot source in a similar room will have higher transfer to their breathing zone than the case with a cool source.
Unpredictable, although more movement will tend to promote more homogeneous conditions in the workroom.
Opening of windows etc should reduce transfer.
Increased duration for someone close to a source will generally increase transfer to the breathing zone.
Increased work rate for someone working close to a source will tend to reduce net transfer to the breathing zone.
The impact of many of the factors associated with the transfer of contaminant chemicals into the breathing zone compartment is not easily predictable. This is because they generally depend on the location of the worker in relation to the source or sources and the airflow patterns in the workroom. Computational fluid dynamics models can deal with this type of problem, but only for specific work situations.
Inhalation exposure is conventionally defined, as the concentration in the breathing zone multiplied by the duration of exposure, although this is often expressed as a time-weighted average concentration averaged over either 8-hours or 15-minutes. In addition, some researchers have suggested that the breathing rate is also an important variable that contributes to the total amount of chemical inhaled and so it is sometimes also considered an important factor in describing exposure for the purposes of assessing risk.
Transfer of chemicals onto the skin is potentially more complex than for inhalation. There are four compartments from which exposure might arise: direct transfer from the source to the skin, contact with contaminated surfaces, contamination of clothing or deposition of contaminants from the breathing zone air. Factors that may influence these transfer routes are shown in Table 9.
47
Table 9: Factors affecting transfer to the skin contamination layer
Factor Increase in net transfer with…
Agitation or pouring or of liquids promoting splashing
More splashing increases transfer
Spills of chemical onto surfaces or clothing
Accidental pouring of chemical onto skin More accidents increases transfer
Less care in working with chemicals will increase transfer
Number of contacts with contaminated surfaces increase transfer
Area of skin in contact with contaminated surfaces clothing worn)
Working in high aerosol concentrations Poor local control measures may increase
Infrequent laundering of work clothing Less frequent laundering of clothing
Many of these factors are related to personal behaviour or are within the control of people at The remaining factors
concern the process and the possibility of increased emission from the source. The effects of most of these factors may in principle be generalised across different workplaces.
Dermal exposure is often expressed as the mass of contaminant on the skin or the mass per unit area multiplied by the area exposed. Cherrie and Robertson (1995) suggested that it would be more appropriate to consider the concentration of the chemical in the skin
This exposure metric is more likely to be related to uptake through the skin than contaminant mass, although the mass is still important if the duration of exposure is sufficiently long.
4.2.2
It must be remembered that EASE was developed more than 10 years ago, at a time when occupational exposure modelling and the realisation of the importance of exposure determinants was only just emerging. reflection of the pioneering work of those who developed EASE, but rather a reflective critique based on current thinking.
EASE clearly recognises what must be one of the key determinants of exposure: the There is a different
logic structure for each of the three physical states, which must be an important part of any exposure model. There is also a division between inhalation and dermal exposure in the
differences in exposure metrics and the differences in the exposure determinants for these different routes. they are “mobile”, is both obscure and unsupported by any research.
The particle size of solid dusts is incorporated into EASE, although again this is basically as a
become airborne (granular). For aerosols the other source terms from the conceptual model are process (dry crushing or grinding, dry manipulation or low dust techniques), presence of
All of these factors have a fairly
More spills will increase transfer
Deliberate immersion of hands in chemicals
Increases in the number of contacts will
Greater skin area contacting surfaces (or less
transfe r rate to the skin
work, e.g. laundering or wearing of appropriate work clothing.
contamination layer multiplied by the area of skin exposed and the duration of exposure.
Evaluation of the structure of EASE
Any criticism in the following text is therefore not a
difference in exposure potential from dusts, liquids, vapours or gasses.
mode l structure, which again must be a necessary prerequisite of any model because of the
However, the criteria for solids contributing to dermal exposure, i.e. that
dichotomy: dusts that may become airborne (respirable/inhalable) or those that cannot
local ventilation and whether the particles readily aggregate.
48
limited scope. For example, “readily aggregate” is defined in terms of whether the substance has a waxy texture or is sticky so it readily aggregates. However the presence of many liquids in dusts can act as an effective dust suppressant and this is not included in this part of the model (although it is included in the “low dust techniques” for handling dusts when liquids are added).
There is a similar range of factors for liquids, although these are often grouped across different areas of the conceptual model. For example, “pattern of use” includes some choices that relate to the control measures (e.g. closed system) and others that refer to the way people interact with the chemicals (e.g. non-dispersive use). For liquids the system calculates the vapour pressure and uses this as part of the judgement of exposure, in line with the approach defined by the conceptual model.
EASE does not include any real consideration of factors other than those related to the source. For inhalation exposure there is no consideration given to the transfer of chemicals from the room air to the breathing zone, although general ventilation is considered as a possible control measure. It is perhaps not unsurprising that EASE does not deal with this type of factor, since as was noted previously in the description of the conceptual model, their effects are not easily generalisable. There is also no consideration in EASE of how workers might interact with the airborne pollutant or even how long they may be exposed to emissions from a process. The latter is a serious omission since exposure will vary pro rata with exposure duration.
For dermal exposure the EASE decision logic is quite crude, relying mainly on the pattern of use (closed system, inclusion into matrix, non-dispersive use or wide dispersive use) and the extent of contact (none, incidental, intermittent or extensive). The same criticisms apply to the pattern of use in the dermal exposure model as in the inhalation model. The extent of contact is only one of the identified transfer factors from the conceptual model.
The authors believe that the exposure metrics used in EASE are poorly defined and this must lead to inconsistent use of the model predictions. For inhalation exposure the output is expressed in terms of concentration (ppm, mg/m3 or fibres/ml), but it is unclear whether this is for a full 8-hour work shift or only for the time while the activity or process operates. Exposure is not defined in the System Properties help section of the model nor in the output screens or files. As discussed in Section 1, the original concept was to estimate 8-hour average exposure, but later the estimates came to be regarded as the average concentration over the duration of a task or activity. For dermal exposure the output from EASE is mass of contaminant chemical per unit area of skin per day. However, these estimates only apply to the hands and forearms, although this is not immediately obvious from the model output. No expla nation is provided about how to interpret these outputs.
4.2.3 Conclusions about the structure of EASE
EASE has a number of characteristics of the conceptual model of exposure described in Section 4.2.1, particularly for the source terms for inhalation exposure. These terms appear to affect predicted exposure in a logical way. However, there are a number of concepts used in EASE that are not easily reconciled with the conceptual model, particularly pattern of use, and the authors are uncertain what, if any, impact this would really have on predicted exposure. There are many factors that will affect exposure that are not included within the EASE model, although it is not clear from the conceptual model how important these are in arriving at an appropriate conclusion about exposure.
The definitions of exposure, both for inhalation and dermal exposures, are imprecise and this imprecision is likely to result in inconsistent application. The model provides insufficient information to fully characterise the exposure. For inhalation exposure, ideally one would like to obtain estimates of the concentration of the contaminant chemicals in the breathing
49
zone, the duration of exposure to this concentration and the persons’ breathing rate while working. For dermal exposure, ideally one should obtain predictions of the concentration and mass of the contaminant chemicals in the skin contamination layer, the area of skin exposed and the duration of exposure.
4.3 IDENTIFICATION OF RELEVANT STUDIES THAT ASSESS THE VALIDITY OR CONSISTENCY OF EASE
The collaborators involved with this study have been involved with the development of EASE, or with its subsequent evaluation, and in some cases with both. They were therefore familiar with the range of organisations that had previously evaluated the model. In addition, all of those contacted during the stakeholder consultation were specifically asked if they were aware of any relevant research (Section 3).
A literature search was undertaken using the bibliographic databases specified in Table 10. In each case the search terms “EASE” and then “EASE” plus “exposure” and “chemical”, were used. It was necessary to add the second two criteria statements to exclude the large number of papers unrelated to this project. The abstracts of all papers meeting the search criteria were scrutinised and the full paper was obtained for any that appeared relevant to this review.
Table 10: Bibliographic databases searched for information on the validity and consistency of EASE
Database information Approximate number of records in database
Search criteria*
Number of records matching search criteria
Number of records about EASE
2002 230,000 EASE 208 3
NIOSHTIC2 2002 0
National Library of Medicine
2002 >12,000,000 EASE + 41 1
Web of Science – ISI inc
2002 17,000,000 EASE + 0
Bibliographic Cut-off date for
HSELINE – HSE
210,000 EASE 606
PubMed –
30
* “EASE+” = Search terms “EASE” plus “exposure” and “chemical”
All of the publications identified in these searches were already available to the researchers, either through personal contacts or from discussions with the stakeholders.
A search was also carried out to identify other relevant information on the Internet. Google (http://www.google.co.uk/) and various other specialist scientific search engines (e.g. Scirus) were used with the same search terms used in the bibliographic database search. Google returned in excess of 700,000 entries, of which a small subset (400) were scrutinised to identify if there was any information relevant to the review of EASE validity and consistency. No relevant entries were identified in this way. Even with much more tightly defined search criteria only about 1% of entries referred to the EASE software package and none of these were of any significance. The key problem for this search strategy was the common use of the word “ease”. It was therefore impracticable to use this strategy to identify research relevant to the validation of EASE.
50
A database of ongoing human exposure research projects collated by the University of Aberdeen and published on the HEROX web site (http://www.HEROX.org/) was also searched. Two projects were listed: one described work already identified from the search of bibliographic databases and the second was this review!
During the stakeholder consultation interviewees were specifically asked whether they were aware of any research to validate EASE, either that they or their organisation had undertaken or work of other scientists. Those who replied positively were asked to provide a reference for a published report describing the work.
The authors were aware of a consistency study undertaken by the HSE alongside other validation work that they commissioned. This had not been published, although an internal report was made available for this review. In addition, HSE had commissioned a study of the validity of the dermal exposure predictions from EASE for people working with lead or tin. This was carried out by HSL and an industry body had commissioned a similar study for inorganic zinc compounds at IOM. Both sets of data were made available for this review.
4.4 SUMMARY OF STUDIES OF EASE VALIDITY AND CONSISTENCY
4.4.1 Inhalation exposure validity studies
Six reports or papers describing studies of the validity of EASE were identified. These are listed in Table 11, below. The earlier studies were mostly either qualitative or provided only limited information about the comparison of EASE predictions with measurements. The final three studies report more rigorous quantitative evaluations of EASE, covering more than 160 EASE end-points with in excess of 17,400 exposure measurements.
Table 11: Summary of studies that have assessed the validity of the inhalation exposure aspects of EASE
Study Number of comparisons
Approximate number of measurements
Type of study
Vincent et al (1996) Nine scenarios Not specified Critique plus
Dervillers et al (1997) Eight scenarios
ECETOC (1997) None Not applicable Qualitative
61 scenarios Not specified Quantitative
Mark (1999) 12,000 Quantitative
Hughson and Cherrie (1999) 4,000 Quantitative
41 scenarios 1,400 Quantitative
case-studies *
Not specified Case-studies
Van Rooij and Jongeneelen (1999)
70 end-points
53 end-points
Bredendiek-Kamper (2001)
* The case studies in the paper by Vincent et al, are the same as those presented by Dervillers et al.
51
Vincent et al (1996) published a report describing a critique of EASE. Some of their criticisms concern technical aspects of the software implementation while others focus on the supporting information and the model structure. They suggested that the documentation of the model is inadequate and the “theoretical foundations” of the program are not explained. In particular, they felt that the basis for deriving the exposure intervals for each end-point should have been provided. In some cases, for example for “closed systems not breached for sampling maintenance” they suggested the exposure range was too low.
They also point out that in EASE there is no allowance made for the proportion of the chemicals in mixtures, i.e. the estimates are the same for mixtures containing 99% of a hazardous substance or 1% of the same substance. However, in the EU Technic al Guidance Document (TGD) that describes the use of EASE for risk assessments it is recommended that such an adjustment be made (EU-DG Environment, Nuclear Safety and Civil Protection, 1995).
They provide a review of the calculation of vapour pressure for volatile substances within EASE. They show that in some cases where vapour pressure is estimated from data at another reference temperature, the error in the EASE algorithm can have a big impact on predicted exposure. They recommend that this aspect of the model should be refined to improve the reliability of the identification of volatility category.
Vincent et al (1996) also criticise the categorisation of “pattern of use”, saying that the available categories are too general to allow a precise estimation of exposure. They recommend expanding the categories on the basis of the industry where the process is carried out, either using the Standard Industrial Classification (SIC) coding scheme or something similar. They note that the model takes no account of personal protective equipment, such as respirators and protective gloves. While this seems to have been a deliberate choice of the developers, based on some philosophical arguments, these researchers felt that the rationale for this decision is not clear. Finally, they criticise the definition of some of the patterns of control, particularly dilution ventilation, which they believe is too vague.
Dervillers et al (1997) and Vincent et al (1996) describe a series of case studies, many of which are common to both papers, and so these two papers are reviewed together. They present nine case studies as shown in Table 12. EASE ranges are not corrected for the proportion of the hazardous substance in the mixture.
52
Table 12: Case studies presented by Vincent et al (1996)
Scenario Substance(s) Range of measured exposure levels
EASE exposure estimates
Units
directly exposed Dichloromethane 24.0 - 340.0 500 - 1,000 ppm
Phenol 0.3 - 1.4 - 1,000
Manufacturing printed circuit
EGME + <0.1 - 18.0 100 - 200 ppm
Manufacture of polyurethane foam
Dichloromethane 15.0 - 140.0 -
Maintenance in bleach plant
Chlorine <0.5 - 8.0 0 - 0.1 ppm
Slate pencil manufacture
Total inhalable dust 0.4 - 11.0 2 - mg/m3
wool MMVF + <0.1 - 0.5 2 - 20 fibres/ml
Production of sodium triborate
Sodium triborate 19.0 * 2 - 5 mg/m3
Charging catalysts Metallic catalyst 4.6 - 775.0 5 - 50 mg/m3
Glazing candy Ethanol 710 - 1,100 500 - 1000
Maintenance in a Lead 0.2 - 2.3 50 - 200 mg/m3
Paint stripping –
500 ppm
boards
10 50 ppm
10
Installing mineral
ppm
petroleum refinery
+ EGME = ethylene glycol monomethyl ether, MMVF = man-made vitreous fibres* Arithmetic mean, range not available
In five of the scenarios the measured exposure levels were less than the EASE predictions, in three cases the measurements and estimates were broadly in agreement and in the remaining two cases the measurements were higher than the EASE estimates. Dervillers et al (1997) and Vincent et al (1996) highlighted the potential problems with categorising a process as “closed” from the data from the maintenance workers exposed to chlorine. Here although the process was closed there were still opportunities for people to become exposed and some relatively high exposure levels were measured.
However, the data used in these studies was generally poorly described and mostly comprised the range of measurements rather than the individual data. In most cases the data represented 8-hour average exposure, but in some cases shorter-term samples were used.
53
Other limited evaluations of EASE
The European Chemical Industry Ecology and Toxicology Centre (ECETOC) have evaluated EASE as part of a more general review of the EU TGD for risk assessment (ECETOC, 1997). They note that the main use of EASE is to provide information for regulatory risk assessments where little or no measured exposure data exists. Some specific criticisms are made, particularly of the definition of terms such as “good” local ventilation and “dry manipulation” used in the model. They also suggest that EASE should address the issue of particle size more fully, although they do not say exactly how this should be done. Their main criticism is the fact that EASE does not include any consideration of the duration of the activity or process and they say that it “may provide strong overestimation if only a few short-term exposure conditions occur during daily work.
Van Rooij and Jongeneelen (1999) describe the outcome of a large number of assessors who attended a short training course on EASE in the Netherlands. In total 117 health and safety practitioners participated in a one-day workshop. During this, evaluations of 61 exposure scenarios were completed, presumably by different course members. There were 35 scenarios with gas or vapour exposures and 26 with dust exposures. In an additional 11 cases it was judged impossible to use EASE to make an exposure assessment.
These authors only had exposure ranges available and they judged the accuracy of the EASE predictions by taking the ratio of the geometric mean of the lower and upper measurements and the corresponding mean value of the lower and upper EASE range. These data are summarised in Table 13, below.
Table 13: Summary of data from Van Rooij and Jongeneelen (1999)
Ratio of EASE estimate to measured exposuresNumber of
scenarios <0.25
(underestimate) (comparable)
>2.5
(overestimate)
Gas/vapour 0
Dust 26 0 50% 50%
exposure 0.25 – 2.5
35 66% 34%
None of the comparisons met the authors’ criteria for underestimation and there was a general tendency for the measurements to overestimate exposure. The average ratio for gases and vapours was 3 and for dusts it was 3.7.
Van Rooij and Jongeneelen (1999) suggested that EASE provided a way of “distinguishing real risk situations from safe situations in a systematic and transparent way”. However, they thought that the user must have adequate training and experience in occupational hygiene to make effective use of the model.
Systematic validation studies
Three recent studies have attempted to provide a systematic evaluation of EASE across a wide range of conditions using existing measurement data. In total over 17,500 individual exposure measurements were available for these evaluations covering 164 different exposure scenarios.
The first two studies were commissioned by the HSE; one at the Institute of Occupational Health (IOH) in Birmingham (Mark, 1999) and the second at the IOM in Edinburgh
54
(Hughson and Cherrie, 1999). Both of these groups summarised their results in slightly different ways, but they both quote the percentage of measurements in the EASE range and whether the EASE estimates were generally in good agreement with the measurements, or too high or low. and all other comparisons in this section, no allowance has been made for the proportion of the hazardous substance in the mixture.
dusts
Workplace/ work activity
Proportion in EASE range
Bias in EASE Number of measurements
Source of data
Mining/quarrying 90% None 636 IOM/IOH
Chemicals production None 12,325
Woodworking 61% None 26 IOH
Various None /
Welding 31% High 98 IOH/IOM
323
Brick manufacture 9% High 1,407 IOM
Metalworking
Oil mist 0% High 34 IOH
It was considered that if more that 50% of the measurements were within the appropriate
there was no bias.
between the predicted EASE ranges and the exposure measurements, while in the remaining In many cases the over prediction was by
about one order of magnitude.
Table 15 shows the corresponding data for gases and vapours.
Their data has been aggregated on this basis and is shown in Table 14. In this
Table 14: Summary of comparisons from IOM and IOH studies - non-fibrous
85% IOH
57% 83 IOM IOH
MMVF processing 12% High IOM
0% High 17 IOH
EASE end-point range then the agreement between the model and the data was good, i.e.
In four out of the nine groupings identified for non-fibrous dusts there was good agreement
five groups the EASE predictions were too high.
55
Table 15: Summary of comparisons from IOM and IOH studies - gases and vapours
Workplace/ work activity
Disinfection
Chemical manufacture
Laying up GRP
Various
Tunnelling
Rubber goods manufacture
Cleaning
Paint spraying
The data from three out of the eight groups for gases and vapours were in agreement with the
measured exposure levels. The differences between the estimates and measurements for
for this category of substance.
Workplace/ work activity
Proportion in EASE range
Number of measurements
Source of data
Manufacture of MMVF 62% None 134 IOM
Asbestos removal None 171
28% Low 79 IOM
Here there was better agreement between the predictions and the measurements, with two of
exposure levels. for some situations EASE predicts zero exposure with no range. aramid fibre exposure was low and if the EASE range had been say <0.1 fibres/ml, the agreement would have been good.
EASE predictions and in the remaining five the EASE predictions were higher than the
gases and vapours were in general greater than for non-fibrous dusts, sometimes by more than two orders of magnitude.
Table 16 shows the data from the IOM study for fibrous dusts – there was no data from IOH
Table 16: Summary of comparisons from IOM study - fibrous dusts
Bias in EASE
51% IOM
Use of para-aramid
the three groups showing good agreement. The data for use of para-aramid fibres was different from most other situations since here EASE generally under-predicted the measured
However, this is partly an artefact of the EASE categories for fibres since In most cases the para-
The data from Bredendiek-Kamper (2001) is summarised below in Table 17.
Proportion in Bias in EASE Number of Source of data EASE range measurements
100% None 18 IOH
67% None 78 IOM
53% None 120 IOM
13% High 197 IOM/IOH
2% High 233 IOM
2% High 126 IOM/IOH
<1% High 331 IOH
0% High 340 IOM/IOH
56
Table 17: Summary of comparisons Bredendiek-Kamper (2001)
Workplace/ work activity
Proportion of scenarios with overlap
Bias in EASE Number of scenarios
Number of measurements
100% None ? 148
Rubber industry 100% None 3
Plastics processing 100% None 2 24
Textile industry None 9 180
Offset printing 40% Low 10 526
Screen printing 605
Various – closed containment scenarios
25
66%
26% High 15
Four of the groups show good agreement with the EASE predictions, in one the EASE predictions were too low and in the other they were too high. It should be noted that the basis for assessing agreement between EASE and the measurements is different from that used with the IOM/IOH studies. Here Bredendiek-Kamper summarised their data with interquartile ranges (25th – 75th percentiles). For the purposes of this review, the data from an exposure scenario is said to agree with EASE if there is any overlap between the interquartile range for the measurements and the EASE end-point range. This is a less stringent criterion than that used with the IOM/IOH data and so better agreement between EASE and the exposure measurements would be expected.
Clearly in this data set there is the suggestion of an inverse association between both the number of measurements in a group and the number of scenarios with the proportion of scenarios with an overlap between the EASE range and the measurements. For example, the three activities with 100% of scenarios overlapping the EASE range had less than 3 scenarios and generally less than 25 measurements. For smaller data sets there is clearly a bigger impact from random processes and it may just be by chance that such good agreement has been achieved for some of the process groups.
The poor agreement and low EASE estimates for offset printing was mainly due to four exposure scenarios where 2-propanol was being used as a solvent. However, the author was unable to say whether this discrepancy was due to differences in the process not accounted for in the EASE estimates or some other reason.
Bredendiek-Kamper looked systematically at whether adjusting for the proportion of solvents in mixtures improved agreement between EASE and the measured exposure levels. In the 16 exposure scenarios where this was done there were 15 where the EASE range originally agreed with the measurements. This was unchanged when the adjustment was made.
The author also looked the level of agreement between low and medium volatility classes. Thirteen out of the 15 (87%) low volatility classes showed agreement between EASE and the measurements, while 9 of the 16 comparisons (56%) for scenarios in the medium volatility class were in agreement. Mark (1999) also reported his comparisons of EASE with exposure measurements for different volatility classes, but he only found good agreement for the very low volatility class.
57
The data from Bredendiek-Kamper are also reported for natural ventilation or technical ventilation (presumably forced general ventilation) in the workplace. There was no indication of any difference between these two types of work environment. The author suggests that this may be because technical ventilation is often installed in workrooms where there are more machines and higher quantities of chemicals are handled. Hughson and Cherrie (1999) similarly found that there was little difference between situations where there was forced general ventilation (53% of EASE predictions were in good agreement with the measurement data) and there was only natural ventilation (40% in agreement).
Other comparisons of the validity of EASE
Esmen (1991) proposed that it was a necessary attribute of any exposure prediction model that it should perform better than a random allocation of exposure from a plausible range of exposures. Therefore if EASE were reliable, it would be expected that a greater proportion of the measured exposures would be within the EASE predicted end-point range than within the randomly assigned end-point range. To test this Hughson and Cherrie (1999) assumed that the plausible range of end-points would comprise the EASE end-points for the state of the substance being studied, i.e. gas or vapour, liquid or aerosol. Each of the EASE end-points was allocated an integer code. A random number was then generated for each set of data, i.e. the groups of measurements for which EASE predictions were made and the appropriate exposure end-point range was selected. The percentage of measurements within the randomly selected range was then calculated.
Figure 9 compares the proportion of exposure measurements inside the randomly selected range with the corresponding value derived from the exposure range allocated by EASE.
Figure 9: Comparison of the proportion of measurements inside the EASE and randoml y allocated EASE ranges (Hughson and Cherrie, 1999)
0.1 1 100 0.1
1
10
100
Per
cen
tag
e in
Ran
do
m E
AS
E r
ang
e Dusts Fibres Vapours
10
Percentage in EASE range
The data in this graph show the fifty-three end-points, with the diagonal line representing those data sets where both EASE and the random exposure range were identical. The bottom right hand area on the graph shows data sets where EASE has been more successful than the random allocation of exposure and the top left area where the random allocation was better than EASE. Ideally, most of the points would have been found in the former area, although in fact there is no indication that EASE is substantially better than a random allocation of
58
exposure range. In 39% of the end-points the EASE prediction was better than the random allocation, whereas for 22% of end-points the random allocation produced a better agreement with the measured exposures. In the remainder of instances the EASE and randomly allocated exposure end-points were equally good (or poor) in the assessment of exposure.
Hughson and Cherrie (1999) also compared EASE with an alternative model involving greater judgement on the part of a human expert. For a selection of the available measurement data in their study the exposure was estimated by an experienced occupational hygienist using a structured subjective assessment method (Cherrie et al; 1996, Cherrie and Schneider; 1999). Only those data sets with 10 or more measurements were selected for this evaluation.
The exposure assessments from this process comprised single values representing mean exposure, rather than the ranges that EASE outputs. To make the data comparable the EASE range closest to the estimated value was selected and allocated to the data set, i.e. the end-point where the difference between the mid-point of the exposure range and the estimated value was smallest. The data from these assessments are shown in Figure 10, along with the corresponding data from the EASE predictions and the random allocation of exposure described above.
Figure 10: Comparison of the reliability of a human expert assessment with EASE and random allocation of exposure (Hughson and Cherrie, 1999)
0.1
1
10
100
Per
cen
tag
e in
Ran
ge
Random Human expert
0.1 1 10 100
Percentage in EASE Range
There were seven surveys where non-fibrous dusts had been measured, four where fibrous dusts had been assessed and seven where vapours had been measured. In about two thirds of these the structured subjective assessment by the human expert contained more than half of the exposure measurements, whereas the corresponding data for the other assessments strategies were 39% for EASE and 22% for the random allocation. From the figure it can be seen that the results from the human expert were mostly above the diagonal line, indicating that they were generally more reliable than a random allocation or EASE at predicting exposure. From this the authors suggest “that there is considerable scope to improve the EASE decision logic to better reflect the important exposure determinants”.
59
4.4.2 Dermal exposure validity studies
Most of those who have commented on EASE have been fairly dismissive of the dermal exposure component. ECETOC (1997) felt that the approach used to estimate dermal exposure was judgmental and hence could be described as “best guesses”.
There are two studies identified that investigate the validity of the dermal exposure assessment section of EASE, one has not been published and the other is only available within a series of scientific reports. These data have been made available for this review. The studies both concern dermal exposure to dusts.
The investigation described by Hughson and Cherrie (2001) involved an assessment of dermal zinc exposure carried out in a variety of factory environments, including galvanising and zinc oxide production. A swab sampling method was used to sample dust on the hands of workers. The work was carried out in three phases and then further work was undertaken to clarify certain aspects of the measurement methodology. The tasks carried out at these factories were categorised in terms of three different EASE endpoints.
Because the authors relied on serial swabs of the skin throughout the working day there was some doubt as to how best to estimate average exposure for the whole day. However, this was resolved in a later series of experiments and the following data presents what the authors consider are the best estimates of average exposure.
Table 18 summarises the data reported in Hughson and Cherrie (2001).
Table 18: Summary of dermal zinc exposures estimates for the hands by EASE category (Hughson and Cherrie, 2001)
Exposure scenario* EASE range (µg/cm2) (µg/cm2)
Measured range (µg/cm2)
Phase 1 survey
A 1,000 6
B 5,000 16
C 15,000
Phase 2 survey
A 1,000 7
B 5,000 53
C 15,000
Geometric mean exposure
100 – 3 – 19
1,000 – 7 – 39
5,000 – 73 49 – 96
100 – 3 – 23
1,000 – 20 – 110
5,000 – 218 97 – 364
* EASE Categories: A – Non dispersive use with intermittent direct handling (galvanising factory) B – Wide dispersive use with intermittent direct handling (zinc oxide –
furnace/warehousing) C – Wide dispersive use with extensive direct handling (zinc oxide – bagging/ re-bagging)
The measured daily exposures to the hands in the three scenarios varied from 3 to 364µg/cm2, which were much lower than the predictions from EASE (100 – 15,000µg/cm2). There is some evidence to suggest that reported exposure was greater during the second series of measurements. Within each phase of the measurement campaign there were clear differences
60
between the three scenarios and as the EASE predictions increased so did the average measured exposure.
Wheeler (personal communication) studied hand exposure for 27 subjects in four factories (three working with lead and the fourth with tin). Dermal exposure was assessed by a hand washing procedure followed by analysis of the rinsing using inductively coupled plasma spectrometry. These data are summarised in Table 19, below.
Table 19: Summary of dermal lead exposures estimates by EASE category (Wheeler, personal communication)
Exposure scenario EASE range (µg/cm2)
Geometric mean exposure (µg/cm2)
Measured range (µg/cm2)
0 or 1 100 0.6
5,000 10.8
0 – 0.03 - 11
2 or 3 100 – 0.5 - 180
Although Wheeler and his colleagues were able to identify four EASE scenarios almost all of the measurements were either in scenario 0 (very low) or 3 (extensive contact, non-dispersive use). All of the data has therefore been combined into two categories “0 or 1”, corresponding to an EASE range 0 to 100µg/cm2, and “2 or 3”, 100 to 5,000µg/cm2.
Again there is a clear indication that higher EASE categories are associated with higher average exposure and the absolute levels of hand exposure are similar to those measured in the IOM studies. However, as before the actual hand exposures to dusts were almost two orders of magnitude lower than predic ted by EASE.
Both studies suggest that for dusts EASE over predicts exposure to the hands, although there is some suggestion that the model frame does capture some aspects of dermal exposure that is related to actual exposure. There are no validation data available to understand how well EASE performs with liquid chemicals or for non-metallic solids.
4.4.3 Consistency studies
Only one study was identified where the consistency of EASE was investigated. This is an unpublished investigation carried out by the HSE, in parallel with the validity studies undertaken by IOM and IOH. The undated draft report was made available for this review and is included in Appendix 2 (Tickner et al, unpublished).
The work describes a comparison of exposure assessments from 15 workplace scenarios undertaken by 34 occupational hygienists. The scenarios covered non-fibrous dusts (6 scenarios), vapours (8 scenarios) and one fibrous dust scenario. These were all briefly described, in some cases accompanied with supporting photographs, with sufficient detail to enable an exposure reconstruction using EASE. Exposure measurement data were not available for any of the scenarios so it was only possible to investigate the consistency of the assessments.
For each scenario there was a clear favoured outcome, although the proportion of assessments in this end-point varied from one scenario to another and according to the physical state of the chemicals. For non-fibrous dusts the proportion of assessments in the most favoured end-point ranged from 50% to just over 90%. In most cases the predictions fell into one of three or four EASE end-point ranges. For the vapour scenarios the proportion of assessments in the most favoured categories ranged from just under 30% to almost 90%. There was a greater
61
spread of outcomes for the vapour scenarios, with between five and sixteen end-points for each scenario. Data for the fibrous dust scenario is not fully described.
The authors note the better agreement between assessors for the dust scenarios compared with the vapour scenarios. They attribute this to the greater number of possible outcomes (end-points) for vapours compared with dusts (59 versus 12).
In both the dust and vapour consistency studies there were a small number of data input errors leading to some of the variation, e.g. choosing the wrong units. However, the main cause of differences between assessors arose for situations where a judgement had to be made; for dusts this was mostly concerned with “pattern of use” and for vapours “aerosol formation”, “pattern of use” or “pattern of control”. Some of this is attributed to misunderstanding the terminology used within EASE.
The authors do not discuss the possibility that the consistency of assessments will depend on the extent of information provided, but this must be the case. For example, if the assessor were provided with a description of a scenario that unambiguously describes what data should be input to EASE then almost all assessors would be expected to concur. However, if the data provided were vague and non-specific then a wide range of outcomes is to be expected. It is believed that the quantity of information is less than could be made available for a single specific workplace, but perhaps typical of that which is available when a regulatory risk assessment is being completed for a generic process.
A number of recommendations were made from this study for improving the software interface for EASE. These include providing better guidance to the user about the procedures for using EASE, clearer help information on terms used in EASE and better training materials, including a number of worked examples. The authors also suggested that the names of some of the terms be changed so that it was clearer that they applied to a wider range of circumstances than might be imputed, e.g. some assessors did not realise that “dry crushing and grinding” should include sanding of wood or other materials.
4.5 OVERVIEW OF THE VALIDITY AND CONSISTENCY OF EASE
The extent of validity studies that are available for EASE is greater than that for most occupational (and non-occupational) exposure models. There are three comprehensive studies available that include more than 17,000 measurements of inhalation exposure covering between 41 and 70 EASE end-points. These produce a fairly clear view that for inhalation exposure EASE tends to either predict close to the measured exposures or to overestimate. There are very few instances where EASE was seen to underestimate inhalation exposure and little evidence that there are identifiable situations where this would generally be expected to occur. The one exception to this, which was seen for para-aramid fibres was an artefact because of the way EASE allocates the exposure in this instance and a more reasonable end-point range would have resulted in good agreement between fibre exposure measurements and predictions.
It might be argued that for a screening model for use in regulatory risk assessments it is an advantage to overestimate exposure. However, in many cases EASE overestimates exposure by two or three orders of magnitude and in other situations it provides a reasonable estimate of actual exposure. Therefore, although it is conservative it is also unpredictable, and this unpredic tability is unacceptable.
There is a consistent difference between the agreement for EASE predictions and exposure measurements for dusts and for vapours, with the latter generally showing poorer agreement. This trend was seen in the validity studies, the consistency study and in the random allocation of end-point ranges. For this reason the difference in agreement has been attributed to
62
another artefact, since there are many more and narrower end-point ranges, making it more difficult for the predicted range to effectively span the range of measurements. The concept of including a range of predicted exposures seems to be a good way of reflecting the uncertainty, although the present studies suggest that these must be selected carefully and should probably more accurately reflect the uncertainty of the exposure predictions.
It was investigated whether the EASE model contains useful predictive power by comparing its predictions with a randomly selected plausible exposure range and with a human expert using an alternative model formulation. In these tests the EASE model was better than random allocation but poorer than the human expert was. This suggests that within the uncertainty of the EASE predictions there are some elements that properly reflect the determinants of exposure, although further improvement should be possible. It is believed that this supports the case for further development of the model. It is worth considering what level of accuracy might be achievable in a model prediction for a single workplace job or task. A dataset of five exposures measurements would allow the median exposure to be confidently predicted to within a factor of 5, whilst ten data points would increase the accuracy to a factor of 31. Given that datasets of this size are fairly common it would be unrealistic to expect a predictive model to necessarily produce predictions better than this.
Perhaps surprisingly, the dermal model also showed some clear associations with exposure to the hands within the two available validation studies. However, as with the inhalation model these were considerable overestimates of actual dermal exposure. It should be noted that there is a much more limited base of information to validate the dermal model and any conclusions about the performance of this aspect of EASE are much less robust than for the inhalation model.
A conceptual model of exposure has been used to try to investigate whether the structure of EASE is appropriate. This analysis showed that while EASE has a number of characteristics described in the conceptual model it is a simplification of what takes place when people are exposed to chemicals. More importantly, it is believed that it does not produce estimates of exposure that are unambiguous or complete. The conceptual model may provide a more rational basis for developing an improved version of EASE. When considering an improved version of EASE it is also important to review other key occupational and non-occupational exposure assessment models available in terms for their underlying structure and usability in order to identify other alternative approaches.
Based on five data points a 95% confidence interval for the geometric mean of a lognormal distribution (geometric standard deviation = 6) would be plus or minus five times.
63
1
64
5. REVIEW OF ALTERNATIVE EXPOSURE MODELS AND ADVANCES IN MODEL IMPLEMENTATION AND COMPUTER
SOFTWARE
5.1 INTRODUCTION
The aim of this section was firstly to identify possible alternative approaches that could be used in the future development of EASE. A large number of models exist and are being developed for the assessment of occupational and non-occupational exposure and a number of these were selected for further investigation.
Secondly, since the original design of EASE there have been a number of advances in artificial intelligence, knowledge management and computer technology, with perhaps one of the most important being those related to the World Wide Web. These advances could potentially offer alternative approaches in terms of developing and implementing a new version of EASE and must therefore be considered accordingly.
5.2 CLASSIFICATION OF MODELS
Models can generally be classified into one of three groups. These are deterministic models, based on a conceptualisation of the indoor environment; probabilistic models and empirical models based on an analysis of available measurements. The characteristics of each of these are discussed in turn.
5.2.1 Deterministic models
Deterministic models vary in complexity and the outputs are presented in the form of point estimates. The most common deterministic models used are box models, which are based on mass balance equations. These equations characterise the rate of release of a substance into a given space and its subsequent behaviour.
At their simplest level, box models determine the concentration of a substance in a “box”. It is assumed that the air in this box is well mixed. The basic form of the mass balance equation is:
dCV = S − CQ
dt
Where V = room volume dC/dt = change in concentration with time S = generation rate C = concentration of contaminant Q = ventilation flow rate
The solution to this equation is:
S S − tC = + (Co − ( exp[ ) Q
] ) Q Q V
Where C0 = concentration at time zero and the following assumptions are met:
1. The contaminant is not present in the infiltration air.
65
2. The rate of extinction of the contaminant (adsorption, absorption, or chemical transformation) resulting from deposition on walls and equipment, condensation of hot vapours and photodegradation of the contaminant is negligible.
3. There is only one generation source within the room. 4. There is instant and perfect mixing.
If the generation rate and ventilation rate is assumed to remain constant over a long time period, such that steady state conditions are reached, the equation reduces to:
SCeq =
Q
It is assumed that the concentration will remain in this steady state until the end of the work shift when it will decrease to zero. This is applicable to continuous processes; however it may not be valid for intermittent work tasks, which finish before a steady state is reached.
In practice, it is highly unlikely that perfect mixing will occur throughout a room, therefore a factor is often introduced to describe the impact of imperfect mixing on the steady-state concentration:
SCeq =
kQ
where k = mixing factor.
The mixing factor is the fraction of ventilation air that is completely mixed with the box air. In many cases the mixing factor is assumed to be 1, which is unlikely to be true even in small rooms. Also, such a simplistic approach may not always be appropriate, for example, for small rooms with stagnant air or high airflow rates, or in large rooms where the source is confined to a small area compared with the overall room volume. Preferably, appropriate ventilation rates should be used, though unfortunately, these are often not known. This approach (i.e. using a mixing factor to modify the equation for a well-mixed room) inadequately describes contaminant concentration and can substantially underestimate exposure and more complex approaches have been suggested (Nicas, 1996). Measurements of mixing factors determined in the absence of occupants may also be misleading as movement and body heat tend to increase mixing. If such factors are ignored then this may lead to the underestimation of mixing and overestimation of concentrations. Very high and very low ventilation rates can cause large deviations of the mixing factor from unity and at moderate ventilation rates, the mixing factor is closer this.
In complex environments, a multiple -box model can be used. This type of model assumes that a room consists of a number of boxes or zones and that the contaminant is perfectly mixed within each of these. For example, the simplest model accounting for imperfect air mixing is a two-zone model which divides a room or space into two zones: a zone representing the space near the source and a second zone representing the rest of the space. It is assumed that the air is perfectly mixed within each zone and that there is airflow between the two. However, it is difficult to define the size and shape of each zone and there isn’t a sudden change in concentration at the zone boundaries. A two-zone model may not adequately describe conditions in a particular room and many zones may be required to fully characterise exposure. Furthermore, throughout the course of a day an individual may move through and work in a number of different zones.
66
The generation or emission rate (S) can be calculated in different ways; one for example, is by using the ideal gas law. However, emission rates can vary substantially from one source to another as a result of design, manufacturing or construction differences. To account for this some models’ use measured emission rates. These are generally limited to a few examples for a given type of source and a narrow range of operating or environmental conditions. It is essential to have knowledge of the variability of emission rates with time, since these can vary by a factor of 100, and to incorporate those into the model.
The simple box model assumes that the contaminant is generated at a constant rate and that the air in the workroom is perfectly mixed, neither of which is true. In addition, the ventilation rate will not be constant due to doors and windows being opened and closed and also changes in environmental conditions. Concentration will also vary throughout the room and over time. Furthermore, the behaviour of the exposed individual will also influence their exposure.
Deposition onto a surface may result in loss of contaminant from the air. Molecules striking a surface can bounce off, be adsorbed or absorbed (and subsequently desorbed) or react to form another species. Aerosols may sediment onto surfaces and very small aerosols may impact onto surfaces by diffusion process. The effect and importance of deposition on concentration depends on the size of the deposition sink relative to other indoor sinks and source terms. These effects can be incorporated into the mass balance equations. Rates of contaminant removal will be important only if they are at least as large as loss rates by air infiltration and ventilation.
When choosing a deterministic model of this type, complexity and usability have to be considered. A simple model may not reflect human, workplace and process parameters and hence accurately predict exposure, it may be difficult to adequately determine the parameters to be used in a more complex model. Although general room ventilation is considered in these models, local ventilation is not taken into account, which means that estimates of exposure are likely to be inflated.
In summary, the advantages of deterministic models are that:
• they are simple to set up and use; • a single point estimate is produced which is easy to understand and interpret and • it is easy to reproduce parameters used for each variable and hence the point estimate
can be consistently reproduced.
However, the disadvantages are that:
• the variability and uncertainty surrounding each parameter is not considered, therefore the potential variation about the point estimate is also not considered and
• a number of assumptions may be made which are not representative of actual conditions, e.g. complete and instant mixing of a substance in air.
5.2.2 Probabilitistic models
Many probabilistic implementations use a deterministic model as their starting point. Exposure varies both between and within individuals carrying out the same job due to differences in working practices and environmental conditions. This variability is accounted for by considering the uncertainty and variability of the input parameters. This is typically done using Monte Carlo simulation, which uses randomly generated numbers drawn from a given distribution to reflect the variability and uncertainty in the deterministic model. The shape of the distribution can be specified for each parameter in the model. The model’s
67
output is in the form of a distribution of exposure rather than a point value and because of this has become increasingly popular. Sensitivity analysis can be carried out to identify the values that have the greatest influence on exposure. This type of approach ideally requires sufficient quality data in order to establish a distribution, given that an inappropriate distribution assigned to a particular parameter will affect the estimated distribution of exposure.
The advantages of probabilistic models are that:
• the predicted exposure can be expressed as a distribution rather than a point estimate, giving the user a better idea of the range of possible exposures for a given situation and
• by using a sensitivity analysis, they allow the most important / influential parameters to be identified.
However, the disadvantages are that:
• these models are more difficult to develop and use; • knowledge of the distribution of parameters is required; • an inappropriately defined parameter distribution may significantly effect the
estimated exposure distribution and • the interpretation of results is more difficult.
Bayesian statistical techniques provide a way of updating probabilistic models as new data becomes available. A distribution of exposures known as a prior distribution is defined which can be based on existing measurements or the output from a probabilistic model. If new measurements become available, the prior distribution can then be updated using Bayes’ rule to produce an updated distribution of exposures (known as a posterior distribution). The use of this technique in occupational hygiene has been recently attracting attention (Ramachandran and Vincent, 1999, Ramachandran, 2001, Wild et al, 2002). The main advantage of Bayesian techniques is that exposure predictions can be updated as new information becomes available.
5.2.3 Empirical models
Empirical models are based on actual exposure measurements from real situations and this class of model may include a range of statistical models. The main advantage of empirical models is that the predicted exposures are based on actual data. However, this in itself presents various disadvantages in that the predicted exposures are dependent on the quality and quantity of the data. For example, if the data does not cover all possible exposure situations, then the model predictions may be based on limited information and be inaccurate.
5.3 IDENTIFYING ALTERNATIVE EXPOSURE MODELS
Approximately 50 to 60 computer exposure assessment programs were identified through professional contacts, scientific literature and Internet search engines and a small number of these were selected for further consideration and review. These were:
1. WPEM v3.2 – Wall Paint Exposure Model. 2. MCCEM – Multi-Chamber Concentration and Exposure Model. 3. CONSEXPO v3.0 – Consumer Exposure Model. 4. BEAT – Bayesian Exposure Assessment Toolkit.
In addition, the exposure reconstruction method devised by Cherrie (1999) was also considered.
68
WPEM is based on a deterministic mass balance model. MCCEM and CONSEXPO use a combination of deterministic and probabilistic techniques, though the deterministic components alone can be used if required. Both WPEM and MCCEM are part of a suite of exposure assessment tools and models developed by the Office of Pollution Prevention and Toxics (OPPT), a division of the US EPA (http://www.epa.gov/opptintr/exposure/) and were downloaded from the Internet. The exposure reconstruction method is based on a deterministic model with a probabilistic add on component. BEAT uses a Bayesian approach.
These five models were selected partly because they covered the three categories of models described in Section 5.2 and also because they appeared to be both widely used and readily available. With the exception of the exposure reconstruction method, these models are all implemented using a computer program and a graphical user interface. A number of other spreadsheet models were briefly considered but were rejected for further review on the basis that they could only be used by experts in modelling.
A researcher experienced in exposure assessment and modelling reviewed the models. When reviewing the model implementations, a number of factors were considered. These included usability, routes of exposure, appropriate situations for use, substances covered, amount of knowledge required by the user, amount of input information required, strengths and weaknesses and whether any reviews or validation studies had been carried out. Each of the five models is considered in turn.
Reviews of some of the computer exposure programs available have previously been undertaken (Price, 2001). A searchable database which contains information on models used by OECD member governments and industry to predict health or environmental effects, exposure potential and possible risks has been developed by OECD (http://webdomino1.oecd.org/comnet/env/models.nsf). Information on use, limitations and validation is included, although the information may not be up-to-date.
5.4 WPEM v. 3.2
The Wall Paint Exposure Model (WPEM) was developed by the US EPA’s Office of Pollution Prevention and Toxics (OPPT). It estimates an individual’s inhalation exposure to airborne concentrations of a chemical released from latex or alkyd wall paint during or after the painting of a residence or office, for brush or roller application. It was developed to allow industry developers and health and safety offic ials to both easily and accurately identify chemicals in paint formulations, which pose potential exposure problems. Emission and sink models developed from small chamber data are incorporated into the mass balance equations.
Indoor air concentrations in one or two zones are calculated using a mass balance model. A constant application rate over time and an exponentially declining emission rate are assumed. For alkyd primers and paints there is a choice of either an empirical single -exponential or semi-empirical emission model. For latex paints either a single -exponential or a double -exponential model can be used to calculate emission. These models were derived through applying non-linear regression techniques to chamber test data.
There are two types of sink models. The first is a one-way sink where the contaminant can enter but not leave the sink. In the second, reversible sink, the contaminant can both enter and leave the sink.
Some values for adsorption and desorption have been established from a limited number of chamber tests. However, no default values are supplied, as the developers believed there was apparently not enough information to set them. It is suggested that unless there is good evidence to the contrary, users assume that there are no indoor sinks. This assumption is likely to result in higher estimates of exposure being made, particularly with respect to the
69
one-way sink model. Although exposures made using the reversible sink model could also be higher, this will be influenced by activity patterns.
The computer interface comprises a number of screens labelled: “Introduction”, “Painting Scenario”, “Paint and Chemical”, “Occupancy and Exposure” and “Execution”. It is easy for the user to proceed through these screens sequentia lly and the following information is required.
Painting Scenario screen
1. Building volume and airflow rates. 2. Percent of building painted. 3. Whether walls, ceilings or both are painted. 4. Amount of paint used, painting rate and painting duration.
Paint and Che mical screen
1. Type of paint and primer/paint density. 2. Properties of the chemical to be modelled, weight fraction in the primer/paint. 3. Chemical emissions model for primer and paint. 4. Indoor sink model (optional).
Occupancy and Exposure screen
1. Type/gender of exposed individual. 2. Individual’s location and breathing rate during the painting event. 3. Weekday and weekend activity patterns (locations, breathing rates). 4. Number of painting events in lifetime. 5. Length of lifetime and body weight.
Execution screen
1. Title of run and length of model run. 2. Results (exposure estimates) after execution. 3. Option to view/print a report summarising inputs and outputs.
On each screen, areas where user inputs are required are shown in white. Optional inputs are shown in grey and the user can override these defaults, which are entered for most parameters. Once the type of building is chosen, the building volume is automatically set. The air exchange rate is automatically set depending on the type of residence or building chosen. Information on 24 chemicals is included although the user may add additional chemicals if they wish. The molecular weight, vapour pressure and weight fraction of the chemical in the primer and paint are required.
An example of the exposure estimates calculated is shown in Figure 11.
70
Figure 11 – Typical output from WPEM
Four exposure estimates based on the inhaled dose are calculated and reported. These are:
1. Lifetime average daily dose. 2. Average daily dose. 3. Acute potential dose rate. 4. Single event dose.
These estimates require information on the inhalation rate, frequency of events, duration of an event, years of exposure, body weight and average time.
Additional output containing information on time-varying concentrations within the building and concentrations to which the individual is exposed is automatically saved in a comma separated value (csv) file, which can be viewed in Microsoft Excel or any other spreadsheet program.
There is a comprehensive manual available from the OPPT web site, which includes both details on input and interpretation of the output produced (GEOMET Technologies, Inc, 2001). There are six default scenarios, complete with description, which can be used. These scenarios include exposure for both amateur and professional painters, a child or an adult located in the non-painted part of the house and also an office worker at work during the week after the building has been painted at the weekend.
On the Execute screen there is a “Model Limitations” button which provides information on the limitations of the model. The emission algorithms used in the model are based on chamber tests for interior application of alkyd or latex primer/paint. Other limitations are detailed overleaf.
71
WPEM v3.2 was produced following a review of an earlier version (cited in http://www.epa.gov/opptintr/exposure/docs/wpemqa.htm). A number of small-chamber tests were carried out by ARCADIS Geraghty & Miller Inc (reported in GEOMET Technologies, Inc, 2001) for the EPA in order to characterise concentrations of various volatile organic compounds (VOCs) emitted from different formulations of both alkyd and latex primer and paint. The model was evaluated by carrying out two experiments in a test house owned by the EPA. In one of the experiments, alkyd primer and paint were used and in the other, latex primer and paint. For the alkyd primer and paint undecane and p-xylene were selected. Generally, there was good agreement between measured and estimated values for both chemicals, though predicted concentrations were usually higher.
The advantages of WPEM v3.2 are that it is easy to use, a high level of expertise is not required and the limitations of the model are clearly listed both online and in the manual. However, the model is limited in that:
• the way in which the task is performed is not considered; • the results are presented as point estimates and uncertainty is not considered; • the emission and sink models used are derived from a small number of small-chamber
tests with fixed air exchange rate, fixed loading of wallboard, fixed production application rate and for one type of application;
• uniform mixing is assumed and • the indoor-outdoor air exchange rate is considered to be constant over time.
5.5 MCCEM 1.2 (BETA VERSION)
The Multi-Chamber Concentration and Exposure Model (MCCEM) is an interactive computer program also developed by the US EPA Office of Pollution Prevention and Toxics (OPPT). This model can be used to assess inhalation exposure to chemicals released from products in domestic premises. Each residence can be divided into up to four zones or chambers and exposure to a particular chemical is modelled. MCCEM uses a system of simultaneous mass balance equations to estimate exposure concentrations in indoor environments and is much more sophisticated than WPEM.
There are a number of input screens labelled: “House”, “Run Time”, “Emissions”, “Sinks”, “Activities”, “Dose”, “Monte Carlo”, “Options” and “Report”. These screens are easily followed through sequentially.
On the House screen there is a database of residence data from which the user can select the most appropriate situation for the exposure scenario being considered. The database contains information on different types of residences for different seasons of the year. Information on the air exchange rate, house and room volumes and inter-room airflow is available. Information on chemical emission rate from the substance being used, and activity pattern has to be defined by the user. Providing rate constants are supplied, the model also takes account of the impact of sinks on exposure concentrations.
On the Run Time screen, the user supplies the start time for the run, the length of the run and the reporting interval for indoor concentrations.
On the Emissions screen, the emission rate is set for each of the selected zones. The user can choose from constant, single exponential or double exponential source models. Alternatively, there is an option for the user to enter emission rates. Using emission rate data, however, presents problems since measurements will typically be made at one point in a particular room and may not be representative of conditions throughout. The Sinks screen is optional and allows the user to specify reversible or non-reversible sinks in each zone.
72
On the Activities screen the user specifies time-varying locations and the breathing rate of the exposed individual. On the Dose screen information on the number of times each year the scenario occurs, the number of years of exposure, body weight and number of years of life is specified.
The Monte Carlo screen is optional and if selected, allows the user to specify distributions for infiltration, emission rate and for parameters relating to sinks. The user can select from four distributions: triangular, normal, uniform or log normal. However, the way in which the parameters of these distributions are set is not intuitive. For the triangular distribution, values for the lower and upper bounds have to be specified as a ratio of the mode. For the normal and uniform distribution the lower bound is specified as the ratio of the mean and the upper bound is automatically calculated since the distribution is symmetric. For the log normal distribution the coefficient of variation should be specified.
On the Options screen the user can specify a title, record notes, specify initial concentrations, choose to run a single chamber or multi-chamber model and choose to apply a saturation-concentration constraint.
Although the sensitivity analyses option can apparently be applied to the indoor volume, infiltration rate, emission rate, decay rate or outdoor concentration by applying a multiplicative value between 0.001 and 1000, efforts to locate and execute this option were unsuccessful. Attempts to obtain clarification of this problem proved unsuccessful during the time constraints of this review.
The model is executed by pressing the “Execution” button and for the deterministic model, the following results are presented:
1. Lifetime average daily dose. 2. Average daily dose. 3. Average daily concentration. 4. Single event dose. 5. Peak concentration. 6. Acute potential dose rate. 7. Time when the acute potential does occurred. 8. Average inhalation rate.
Where the Monte Carlo analysis option is selected, the mean, standard deviation and maximum concentration are presented for lifetime average daily concentration, average daily concentration, lifetime average daily dose, average daily dose, peak concentration, single event dose and acute potential dose rate. In addition, the average inhalation rate is also given.
An example of the output screen is shown in Figure 12.
73
Figure 12 – Typical output from MCCEM
Additional output containing information on time-varying concentrations within each zone and concentrations to which the individual is exposed is automatically saved in a csv file. Again, these can be viewed in Microsoft Excel.
Although there is comprehensive online help, which includes a worked example, there is no user guide provided with this version.
In the online help there is a section on the model’s evaluation. The current program is based on an earlier DOS version which had been peer reviewed (Koontz and Nagda, 1991). The reported evaluation consisted of simply comparing the output from the current version with the DOS version to ensure that the algorithms used were working correctly and to evaluate accuracy. Agreement with the DOS version was established. Koontz and Nagda (1991) also report on previous research where the output from MCCEM was compared with other indoor air models including COMTAM (developed by the National Institute for Standards and Technology) and INDOOR (developed by the USEPA Office of Research and Development) (GEOMET, 1989). MCCEM, CONTAM and INDOOR produced virtually identical results when provided with the same inputs for the same scenarios. However, these evaluations cannot be considered to constitute a proper validation of this exposure model.
A study by Nagda et al (1995) compared indoor air concentrations of toluene from an adhesive used for installing floor tiles to those estimated using MCCEM. Emission data inputs for the model implementation was determined from small chamber experiments. Air exchange rates and interzonal airflow rates were measured in the research house. Three zones were defined: the bedroom where the adhesive was applied, the remaining upstairs area and the downstairs area. Two runs were carried out. The results for only two zones were presented. There was relatively good agreement between predicted and measured
74
concentration for the bedroom in which the adhesive was applied. Agreement was not so good for the remaining upstairs area.
A further reference was identified as possibly containing information on a validation of MCCEM (GEOMET, 1995), however, this document could not be obtained.
Overall, the advantages of MCCEM are that:
• a Monte Carlo option is available; • the user can select different physical and environmental conditions for the given
scenario from a database of residences; • time varying emission rates can be specified; • the impact of indoor sinks can be accounted for; • the indoor space is divided into more than one chamber, rather than assuming a
single, well, mixed zone and • activity patterns and other human factors are taken into account.
However, as with all other models, MCCEM has a number of disadvantages. For example;
• some knowledge is required to select appropriate parameter values; • knowledge of different distributions and distribution of parameters is required when
usin g the Monte Carlo option; • the choice of values to define distributions is not intuitive; • data for emissions and sinks is required and • information on activity patterns is also required.
Fehrenbacher et al. (1996) evaluated the mass balance models used by the EPA to estimate inhalation exposure to new substances. Sensitivity analysis and Monte Carlo simulations were carried out to determine what model-input parameters most affected the estimated exposure. The Monte Carlo analysis showed that input values must be carefully selected and that generation rate, ventilation flow rate and vapour pressure were particularly important. The model was evaluated by comparing the estimations with monitoring results for six different published studies. This model does not account for the use of control measures such as local exhaust ventilation and it might be expected that exposure would be overestimated. This was evident in the comparisons, with the model predictions generally overestimating actual measured exposure. Predictions were within one order of magnitude of the measured values for all six studies. Fehrenbacher et al. (1996) concluded that such models are appropriate for screening purposes and that where they are used for other purposes, all assumptions should be carefully considered and improved upon if possible.
5.6 CONSEXPO 3.0
The CONSumer EXPOsure Model (CONSEXPO) has been developed by RIVM (the Dutch National Institute of Public Health and the Environment) and allows users to estimate exposure to chemicals contained in consumer products. It is designed to assess exposure to non-professional indoor users of consumer products. Inhalation, dermal and ingestion routes of exposure are considered. Total exposure is determined by combining contact, exposure and update scenarios for each route of exposure. There is a default database containing values for models and parameters linked to types of products. This was not available for evaluation.
Alternatively, users can specify contact, exposure and uptake by selecting the appropriate scenarios and models from predefined lists. For the contact component the frequency of use, total duration of contact, duration of actual use and start of contact (set to zero if exposure is
75
constant) have to be specified. Exposure by inhalation, dermal and ingestion routes can be considered and the total exposure determined. For each of these routes a number of exposure models of varying complexity are available for selection. For inhalation exposure these are typically mass balance models.
5.6.1 Inhalation exposure and uptake
Inhalation exposure
The models vary in complexity and are presented to the user in the form of scenarios. For inhalation there are 6 scenarios, based on 6 different models. These are:
1. Constant concentration. 2. Source and ventilation. 3. Evaporation from pure substance. 4. Evaporation from a mixture. 5. Indoor gas. 6. Paint, the spray; well mixed and the spray cloud.
The user is presented with a choice of scenarios rather than models and the user may therefore be unaware of the limitations of the models behind their selection.
Constant concentration scenario
In this scenario it is assumed that the concentration in a single room is constant. It is suggested that this may be used in the following situations:
• for screening purposes; • when information is limited and • for suddenly released gases or highly volatile compounds.
Information is required on the amount of product released into the room, room volume and weight fraction of the chemical compound in the product.
Source and ventilation scenario
In this scenario a room has a source that emits a chemical compound into the air. This room is ventilated with ambient air. A mass balance equation is used to describe exposure and this scenario should be used when:
• a steady release rate can be defined and • the maximum concentration is below the saturation in air.
Information on the generation rate, ventilation rate, ambient concentration, breakdown rate and room volume is required. Perfect mixing is assumed and sinks are not taken into account.
Evaporation from pure substance scenario
It is assumed in this scenario that a pure substance evaporates into a room. The evaporation rate depends on the difference in vapour pressure between the saturated air at the product/air boundary layer and the air in the room. The room is ventilated with ambient air and it is assumed that equilibrium will be reached between the vapour pressure of the substance and the concentration of the substance in the air of the room. This scenario should be used when a pure substance is being assessed and the dynamics of evaporation are important.
76
This model requires the properties of the compound to be known. Information is also required on the surface area of the product in contact with air, room volume and the effective ventilation rate. Again a well-mixed atmosphere is assumed.
Evaporation from mixture scenario
In this scenario chemicals evaporate from a product consisting of a mixture of compounds. The evaporation rate depends on the difference in vapour pressure between the compound in the mixture and in the air. It is assumed that the model consists of two parts: the chemical of interest and all the other chemicals together.
This scenario is only valid for estimating exposure over short periods of time since the concentration of the compound in the mixture decreases due to evaporation. It is assumed that the room is ventilated with clean ambient air and so the concentration of the compound in air will reach equilibrium. It is intended to extend this scenario to a more general one in the future.
This scenario should be used when:
• a compound evaporates from a matrix; • the concentration of the compound in the matrix is nearly constant during evaporation
and • the dynamics of evaporation are important.
Information is required on the surface area of the product in contact with air, temperature in the room, room volume, effective ventilation rate, total amount of product and weight fraction of the compound in the product.
Indoor gas scenario
This model predicts carbon monoxide concentrations from burner and room characteristics. It assumes a single room with a natural gas burner emitting exhaust gas containing carbon monoxide and carbon dioxide into the room. It has been shown that hot exhaust gas separates the room air into a warm upper and a cold lower layer. The model distinguishes between three air layers: the air directly above the burner, the air in the warm upper layer and the air in the cold lower layer. Exchange between the layers decreases with increasing temperature gradient.
Mass balance equations govern the behaviour of carbon monoxide and carbon dioxide and this model should be used to predict exposure to carbon monoxide emitted by a burner in a single room.
Information on carbon monoxide production, the fraction of exhaust gas (both carbon monoxide and carbon dioxide) directly lost after emission, room volume, surface of the room, and outlet height is required. A mixing factor describing how well both upper layers mix and the upper and lower compartments mix by exchange of air, the kilowattage of the burner, exhaust ventilation and outward directed ventilation in the upper air compartment next to the burner is also required. Carbon monoxide production is burner specific therefore generalisations cannot be made.
Validation experiments by Dijkhof et al (1999) (cited in van Veen, 2001) showed that the model performed well for ventilation volumes below 0.5 air changes per hour. It over predicted for higher ventilation volumes, though this can be corrected for by including a parameter describing the amount of exhaust gas directly removed.
77
Painting scenarios
Painting
This model predicts exposure to compounds evaporating from paint applied to a surface in a single room. The painted surface is divided into two layers: an upper one in direct contact with the air and a lower one that acts as a store. A fixed ventilation rate is assumed. This model should be used when:
• there is a single room; • the concentration of the chemical in the matrix decreases significantly or is fully lost
and • the dynamics of evaporation are important.
Information is required on the painted area from which chemicals may evaporate, the amount of product used, the weight fraction, density of the paint, layer exchange rate, fraction of paint applied to the upper layer during painting, and the mean molecular weight of the other compounds in the paint. Information on room volume, effective ventilation and room temperature is also necessary.
Spray: well-mixed room model
This model assumes that a spray generates droplets that are well mixed in the room and that no evaporation of the chemical occurs. If the volatility of the chemical is high, the source and ventilation model is more appropriate.
This model should be used when:
• the chemical does not evaporate; • a spray is used to apply the paint; • indoors and • the air in the room is assumed to be well mixed.
Information is required on the average amount of formulation sprayed per unit time, the weight fraction of the chemical, the airborne fraction, density, room volume, effective ventilation, droplet size and release height.
Spray: cloud model
This model describes exposure to an aerosol emitted by a spray can. It is assumed that a cloud of spray droplets is formed which the ventilation does not remove. This cloud divides the room into two compartments: one containing the cloud of droplets and the other containing the rest of the room. It is also assumed that the user is exposed to the cloud compartment and that the non-user is exposed to the remaining room compartment. If the cloud is mixed in the room then the well mixed spraying model is more appropriate.
The spray cloud model should be used when:
• a spray is used as application; • indoor use; • the aerosol consists of a defined cloud; • the user is assumed to have direct contact with the cloud and • the non-user is assumed to have contact with the evaporated chemical in the room and
no contact with the cloud.
78
Information is required on the average amount of formulation sprayed per unit of time, density, weight fraction, airborne fraction, droplet size, release height, radius cloud, target area, room volume, effective ventilation and solvent evaporation.
Inhalation uptake
Three models are available to calc ulate uptake in the lung:
Fraction model
Uptake is defined as the product of the fraction taken up, the inhalation rate, the respirable fraction and the exposure. Total uptake is defined by integrating this product over the duration of exposure.
Diffusion model
This is based on the concentration difference between the lung air and lung blood
Equilibrium flow model
This model is based on equilibrium exchange between a compound in the lung air and the lung blood.
5.6.2 Other routes of exposure and uptake
Exposure and uptake from dermal and oral routes will only be considered briefly in this review since the main focus is on exposure and uptake by the inhalation route.
Dermal exposure and uptake
For dermal exposure there are five different scenarios which the user can choose from. These are:
1. Fixed volume. 2. Diffusion in product. 3. Migration to skin. 4. Transfer Coefficient. 5. Contact rate.
In addition, dermal exposure will also occur concurrently with inhalation exposure and so setting inhalation exposure automatically sets a dermal exposure.
The fixed volume scenario assumes that the product is well mixed with no diffusion gradients and hence that the volume of product in contact with the skin is constant over time. This fixed volume can be either a small amount spilt on the skin or a large volume in contact with the skin, e.g. during dish washing.
The diffusion in product scenario assumes that the product is not well mixed and that a diffusion gradient governs transport of the chemical. It also accounts for evaporation of the chemical.
The migration to skin scenario is relevant when a chemical migrates from clothing onto the skin. The transfer coefficient scenario can be used when the skin comes into contact with a contaminated surface. Information on the area of a surface wiped by the skin per unit of time
79
is required. For the contact scenario dermal exposure is calculated from a contact rate, with the contact rate defining how much of the product deposits on the skin per unit of time. The transfer coefficient and contact scenarios are particularly relevant to biocide risk assessment.
One of the drawbacks of the above scenarios is that they assume that only one scenario is relevant when this may not always be the case. For example, an individual using a disinfectant solution may have exposure both by direct contact with the solution and by indirect contact with the solution due to contact of contaminated clothing with the skin.
For dermal uptake there are three possible models to select from. The first is the fraction model, which can be used when little information about the product is available. The diffusion model is more complex and uses the concentration difference between the product and the blood to calculate the amount taken up per unit time. The final model is the SKINPERM model that can be used to estimate the skin permeation coefficient from aqueous solutions using properties such as molecular weight and the octanol/water partition coefficient. Again, uptake is proportional to the concentration difference and skin permeability. This model is based on the work of Wilschut et al. (1995).
Oral exposure and uptake
There are four exposure scenarios to select from for oral exposure:
1. Single ingestion. 2. Hand to mouth contact. 3. Leaching from product. 4. Article migration from product to food.
In addition, a fraction of the oral exposure will be derived from the ingestion of aerosols or particles inhaled and deposited in the respiratory tract.
The single ingestion scenario describes uptake from products that are swallowed. The hand to mouth contact scenario estimates exposure as a result of dermal exposure to the hands and subsequent hand-to-mouth contact. The leaching from product scenario assumes leaching from objects placed in the mouth. The migration scenario is relevant to situations where a chemical migrates from packaging to food, which is then consumed. As for the dermal scenarios, it is assumed that only one scenario is relevant, which may not always be true.
There are two models, the fraction and the diffusion model (of which there are two types), which the user can choose from to describe oral uptake. The fraction model is based on an absorption factor. For the diffusion models, only passive diffusion is considered, since there are no estimates for active diffusion parameters and the intestine is described as a long tube. There are two types of diffusion model: the complete radial mixing tank model which assumes that product travelling through the intestine releases the compound radially into the wall of the gut and the mixed tank model which assumes that the intestine is one well mixed compartment from which uptake occurs. The complete radial mixing model is considered to provide a better representation of the gut and results in a larger uptake estimate.
5.6.3 Outputs and usability
The results can be displayed either as point estimates or in graphical form. The Monte Carlo method is only used when there is more than one parameter with variation. A typical output screen is shown in Figure 13.
80
Figure 13: Typical output from CONSEXPO
The program allows each contact, exposure and uptake parameter to assume a distribution. There is a choice of three distributions – normal, log normal or uniform. Measured data can also be used to establish an empirical distribution. For the normal distribution the mean and standard deviation are required, for the log normal distribution the median and coefficient of variation are necessary and for the uniform distribution, the upper and lower bounds. Where measured data is used, the program constructs a cumulative distribution from the data and samples from it.
Sensitivity analysis can be carried out for any parameters for which distributions have been defined. At present this can only be done for a particular parameter on a local basis where all other parameters are at their best estimate. A global sensitivity analysis cannot be carried out with the present program.
The program is complicated to use and requires information on a considerable number of parameters to be input. Depending on the scenario chosen, between three (e.g. constant concentration scenario) and 11 parameters (e.g. indoor exhaust gas scenario), need to be set. Although the user chooses a scenario rather than a model, knowledge of the underlying models is required in order to select the required parameters properly.
It was evident that not every parameter needs to be filled in for the model to run, for example, not all inhalation uptake parameters in the manual’s tutorial example were required. It is unclear to what extent this affects the answers obtained. There is a comprehensive manual (van Veen, 2001) to accompany the software, complete with tutorial, however, the example does not cover the Monte Carlo option.
81
A series of fact sheets have been developed which describe main categories of consumer products such as paint, cosmetics, and pest control products and these provide guidance on models and exposure parameters to use. In addition, Weegels and van Veen (2001) have studied consumer behaviour and it is intended to use this information to refine the default parameters set for the various models used.
5.6.4 Advantages, disadvantages and review of CONSEXPO
The advantage of this very detailed model is that all exposure routes can be considered and total exposure can therefore be calculated. It is also thought that the defaults database should be useful although this was not available for evaluation.
However the disadvantages of using CONSEXPO are that:
• it is difficult and complicated to use;• only one scenario at a time can be considered;• due to its complexity and number of input parameters, different users may obtain
different answers for the same scenario; • considerable knowledge is required to select appropr iate models; • information on a large number of parameters is generally required; • knowledge of different distributions and distribution of parameters is required and • the choice of values to define distributions is not intuitive.
Two papers have been published which compare inhalation exposure estimates to those obtained during activities associated with painting (van Veen et al. 1999, van Veen et al. 2002). These experiments were intended to validate the paint exposure model. In the earlier paper, van Veen et al. (1999) reported good agreement between estimated values and experimental values for n-alkanes associated with painting. In their most recent paper van Veen et al. (2002) estimated exposures to dichloromethane, where estimates were compared with experimental values obtained during paint stripping. The model accurately predicted exposure at the upper range of experimental values, but over predicted in four out of six experiments.
The exhaust gas model (http://webdomino1.oecd.org/comnet/env/models.nsf), used in inhalation exposure estimation, has also apparently been validated, however this information was not available to the researchers at the time of reporting.
Work on CONSEXPO is still continuing and the developers hope that a version containing the complete defaults database will be distributed via the web site. The user interface is also being updated and it is anticipated that this will be implemented by end of 2003.
5.7 BEAT 1.13
The Bayesian Exposure Assessment Toolkit (BEAT) model is being developed by the HSE and is intended to be used for estimating dermal exposure. It is based on the ideas of Phillips and Garrod (2001) and comprises a collection of software models built around a database containing more than 500 exposure measurements. It is still in the process of being developed and is not generally available.
The user defines an exposure scenario, inputting information on tasks performed (for example, immersing, mechanical treatment, manual dispersion, hand tool dispersion incidental exposure), the product (including its physical state, environment, distance from source etc) and technique (including application rate and pressure, frequency and extent of contact etc). In most cases, drop-down menus are available for the user to select from. The
82
program then compares this information with existing scenarios in the database and selects similar ones. This comparison is carried out on a number of levels.
Firstly, the physical state of the products being used is compared and scenarios with different states are disregarded. Secondly, the tasks making up each job are compared. Tasks are grouped into Dermal Exposure Operation (DEO) units. These units were devised as part of the RISKOFDEM project which is a multi-centre Pan-European project aimed at improving understanding of the nature and range of dermal exposures to hazardous substances throughout the EU (RISKOFDERM, 2001). These are general categories of tasks involving the potential for exposure. They are compared on the basis of time spent on each task. Thirdly, similarity is considered at the level of each DEO unit. For example, the modifiers for manual dispersion are application rate and orientation. Each modifier is assigned a ‘similarity score’ based on its similarity with an existing job. These similarity scores are multiplied within each DEO unit to provide a similarity score for that unit. A total similarity score can then be calculated using the formula on p17 of the manual (Warren, 2002). This can then be translated into beliefs about the median rate of exposure. Table 3 on p19 of the manual illustrates how similarity scores relate to beliefs about the rates of exposure. For example, a similarity score of 5 corresponds to a belief that it is 95% likely that the median rates of exposure are within a factor of 10.
Once the related scenarios have been selected, the Bayesian posterior beliefs can be calculated by clicking the ‘Predict’ button.
To obtain the full output including the predicted dermal exposure, the results first have to be printed out, since only the indicative distribution matrix is shown on the screen. This is a 4 by 3 matrix representing the 12 combinations of four median rates of contamination and three levels of exposure variability. The values given are estimated probabilities that the exposure distribution for the scenario of interest corresponds to each of the 12 indicative distributions relative to each other. Therefore, in the example in Figure 14, the medium/wide log normal distribution is around three times more likely than the medium/intermediate distribution, which relates to a most appropriate median exposure of 20 mg/min.
Positioning the cursor over a particular cell gives the median and 75th and 95th percentiles for that distribution. The printed output provides information on the scenario of interest including the performed tasks, product information and technique. This is followed by a description of the jobs that were considered to be closely related to the scenario of interest. Finally, the indicative distribution matrix with the most plausible cell and predicted dermal exposure beneath are printed.
The contribution of each of the chosen similar scenarios can be investigated by de-selecting other scenarios and re-running the model. Alternatively, the effect of omitting one of the scenarios can be investigated by de-selecting that particular scenario and re-running the model.
There is a comprehensive manual which explains how the model works (Warren, 2002). An example, with interpretation of the output is provided and on-line help is also available.
83
Figure 14: Typical output from BEAT
The advantages of BEAT are that:
• in depth knowledge about Bayesian techniques is not required; • more than a point estimate is provided; • more than one task can be considered simultaneously and • it is easy to use, despite the complexity of the underlying model.
However, the disadvantages of the model are that the:
• like any model that incorporates an ‘expert system’, the predicted estimates will depend upon the quality of the encapsulated knowledge;
• some mathematical knowledge is required to interpret results; • predicted exposure estimates will depend on the quality and quantity of data available
and • there is no provision for using actual exposure data from the proposed scenario to
update the posterior distribution. (Since the review was carried out, a new version of BEAT has been developed which allows actual data to be used).
As this model is still under development no validation has been carried out.
84
5.8 EXPOSURE RECONSTRUCTION MODEL
Cherrie et al. (1996) and Cherrie and Schneider (1999) developed a structured exposure assessment method for use in occupational epidemiological studies. This model is intended to predict inha lation exposure and uses information about the physical state of the substance, together with other environmental and workplace related variables.
The model comprises a source term, which is dependent on the intrinsic properties of the contaminant e.g., the dustiness of a solid or the vapour pressure of a liquid, the way the contaminant was handled, e.g. careful lifting and stacking or dropping and finally, the effectiveness of any local controls. These three parameters are multiplied together to provide the active emission of the source. Three further parameters are incorporated into the basic model. These are the passive or fugitive emission, the fractional time the source of the contaminant was active and the efficiency of any respiratory protection.
The model simplifies the dispersion of contaminants away from sources using two notional spatial regions: the near-field, which is a volume around the worker whose exposure is being investigated and the far-field, which comprises the remainder of the work environment. General ventilation in a workroom will have an impact on the contaminant concentration in both the far- and the near-field. In this scheme the intrinsic and passive emissions nominally have concentration units (e.g. mg/m3). This would correspond to the airborne concentration generated with a certain “standardised” handling and no local control measures.
The exposure estimate is developed on an arbitrary scale and does not have any units. This value must be multiplied by a conversion factor (based on the appropriate exposure limit) in order to give an exposure level. For example, if the predicted total exposure is 0.61 for respirable dust, the calculated exposure will be 2.44 mg/m3 (the occupational exposure limit (OEL) is 4 mg/m3).
The reliability of the method has been assessed (Cherrie and Schneider, 1999, Semple et al. 2001). In the first case, exposures for 63 jobs involving man-made mineral fibres, asbestos, styrene, toluene and mixed respirable dust were reconstructed from detailed descriptions of the jobs and tasks. Most of these reconstructions were carried out by two experienced hygienists. In general, there were statistically significant associations between estimated and measured level, though there was also a tendency to overestimate exposure. Correlation, however, was poor for styrene. In the second study five occupational hygienists estimated exposure for 40 tasks covering asbestos, man-made-vitreous fibres, dust and polycyclic aromatic hydrocarbons. On this occasion guidance and training were provided prior to the estimations being made. Agreement between the assessors’ estimates and the measured data was good to excellent though again there was a tendency for assessors to overestimate exposure levels by, on average, two to fourfold. Estimates were improved when the estimates from two or three assessors were aggregated. However, little benefit was obtained from using additional assessors.
Estimates made using this model were compared with those made using EASE and measured data and were generally found to be better (Hughson and Cherrie, 1999), with two thirds of the estimates being in good agreement with the exposure measurements compared with only 39% for EASE. These are described more fully in Section 4.
The advantages of this method are that:
• it provides a structured and logical approach to exposure estimation; • both the immediate area (near-field) around worker and the wider environment (far-
field) are considered;
85
• more than one task can be considered in one exposure assessment and • it allows the main causes of exposure to be identified.
The disadvantages are as follows:
• the predicted exposure estimates depend on the quality of information available; • some mathematical knowledge is required to interpret results; • the estimates will vary depending on the “expert” carrying out assessment i.e. a high
level of subjectivity required; • good guidance and training is required and • it is not yet available as a stand-alone computer program for general use.
This model is currently set up on an Excel spreadsheet. Recently, Monte Carlo methods have been incorporated into the spreadsheet using the @RISKTM software package. This has allowed the influence of uncertainty in parameters to be investigated (Semple et al, in press). Triangular distributions were assumed for emission, handling and general ventilation and uniform distributions for the time the source is active and the percentage of time spent on the task. For triangular distributions the minimum, midpoint and maximum values have to be specified. For uniform distributions minimum and maximum values have to be specified. In this example, local ventilation and respirator protection, were not incorporated into the Monte Carlo simulation since neither were relevant to the particular jobs considered. Sensitivity analysis can also be carried out using @RISK. The advantages of incorporating the Monte Carlo methods are that parameter distributions are set, rather than the user having to choose them and the choice of defin ing characteristics for distributions is more intuitive. The disadvantage is that training is required in order for the user to use the model correctly.
5.9 DISCUSSION ON ALTERNATIVE EXPOSURE MODELS
The exposure assessment model implementations considered in this section encompass deterministic, probabilistic and empirical models. Although only a few model implementations were studied in depth, they are considered to be representative of the types of implementations currently available. All of the types of models used have their advantages and disadvantages. For example, it is difficult for a deterministic model to adequately represent a particular situation, although more complex deterministic models should produce more realistic exposure models. The exposure reconstruction method, which uses a simple mathematical expression, combines both work and environmental factors and more realistically represents exposure than the mass balance equation approach. In addition, properties of the substance of interest, handling, room size, general ventilation and the effectiveness of local exhaust ventilation and respiratory protection, are all considered. Probabilistic techniques depend on the distribution of the model parameters being adequately characterised. Expert knowledge is required to determine the prior distribution required as a starting point to apply Bayesian techniques. Empirical models are dependent on the quality and quantity of data on which they are based. MCCEM, CONSEXPO and the exposure reconstruction method combine both deterministic and probabilistic techniques. As further measurements become available the output from such models could be further updated using Bayesian techniques. Output from EASE is currently in the form of a range based on actual measurements. It should be possible to use the measurements to produce a distribution of exposure estimates, which could also be updated by the actual user when new measurements become available.
In order to use a particular model properly, the user should fully understand the model and also its strengths and weaknesses. The assumptions and appropriateness of the input parameters should be considered fully before use. However, it is unlikely that users will properly understand or evaluate the appropriateness of the models used or the default values
86
provided. The user is more likely to be concerned with how easy a program is to use than understanding its input values and uncertainty associated with its estimates. Indeed this was evident in some of the responses obtained from the stakeholder interviews in Section 3 where the usability of the model was emphasised.
The user friendliness of a package is also important and the implementations considered here vary in this respect. WPEM is the most straightforward to use, leading the user through step-by-step. Although MCCEM also guides the user, it is more complex than WPEM and a greater depth of knowledge is required by the user to define parameters and select options. However, with practice it is relatively simple to use. Despite the complexity of its underlying principles, BEAT is easy to use though as yet only dermal exposure is considered. Unsurprisingly, CONSEXPO, which considers all routes of exposure and uptake, requires considerably more knowledge to use it properly. With the exception of the exposure reconstruction method, all are menu driven. Currently training is required before the exposure reconstruction method can be used; however, the implementation of a structured computer program is planned which should reduce this need.
Finally, it is important that the exposure estimates obtained are reproducible for a particular scenario when different assessors use the model. With the exception of the exposure reconstruction method, no attempt has been made to address this issue in any of the other model implementations considered. Given its complexity, it is anticipated that more problems in this respect will be experienced with CONSEXPO, than with the simpler and more transparent models such as WPEM.
The exposure reconstruction model has been evaluated for a number of different scenarios. In addition, estimates using this model have been shown to compare more favourably with measured values than those obtained using EASE. BEAT, which is still being developed has not reached the validation stage. The remaining models have been evaluated to a much more limited extent and for a limited number of scenarios.
The ideal computer exposure program should be user friendly and guide the user through it in a logical manner. Any specialist mathematical or statistical knowledge required should be limited. Different users should be able to obtain comparable answers for the same scenarios. On-line help and guidance should be available and guidance on the interpretation of the results provided. A comprehensive manual should be available for further reference. Full examples should be provided for the user to reproduce. All assumptions and limitations must be adequately described. Most of the programs considered here have not been properly validated, which limits their use. Evaluation and validation of any model should be ongoing and default values updated if necessary.
5.10 REVIEW OF ADVANCES IN MODEL IMPLEMENTATION AND COMPUTER SOFTWARE
Over the last few decades there have been a number of advances in artificial intelligence, knowledge management and computer technology, with overviews of the most important to this review being discussed in turn.
5.11 ADVANCES IN ARTIFICIAL INTELLIGENCE
5.11.1 Inexact Reasoning
Much of the attention of the Artificial Intelligence community has been devoted to the study of how to reason in the absence of full information. This is very relevant to EASE given that reasoning about new substances, for which little data is available, is necessary. At the time of the initial development of the model, an expert systems approach using a knowledgeable
87
occupational hygienist's experience was implemented given that no suitable alternatives were available. However, other technologies, developed over the last twenty years, might be more appropriate under certain circumstances.
5.11.2 Fuzzy Logic
Fuzzy logic was originally investigated at the University of California, Berkeley in the 1960s (Zadeh, 1965) and a complete calculus has been constructed. Instead of the standard binary truth-values of Boolean logic, where 0 represents false and 1 represents true, fuzzy logic deals with truth values that are real numbers. These are in the range 0 to 1, where 0 now represents absolutely false, 1 represents absolutely true, and other values represent values somewhere in between. This is intended to deal with common sense concepts, for example, height and temperature, for which acceptable values can be short and tall, or hot and cold respectively, and for which strict numerical equivalents are often not appropriate.
Fuzzy logic values must be distinguished from probabilities. For example, if a temperature is quoted with fuzzy logic as being 0.6 hot, this means that the temperature is fairly hot, whereas a probability value of 0.6 would be interpreted as meaning that there is a 60% chance that the temperature is hot as opposed to not hot. However, in probability theory and in this example, the temperature must be either hot or not hot and it may not be possible to determine which it is.
Fuzzy logic has been applied very successfully, largely by Japanese companies, to control circuits for devices such as washing machines and camera auto-focus systems.
5.11.3 Case Based Reasoning
Another approach is Case Based Reasoning, which was first investigated by Schank (1982) at Yale in the 1970s. Case based reasoning is based on the theory that new problems may be similar to those encountered previously and that the past solutions applied may also be useful in the current situation. Typically, case based reasoning systems maintain a “case base” of previously encountered problem-solution pairs. A new case is matched against the case base and all matching cases with their paired solutions can be retrieved. For more complex problems, the retrieved solutions may need to be adapted to suit them better. Case based reasoning can be adaptive or learning systems providing feedback on the effectiveness of retrieved solutions can be used to modify the matching algorithm.
Case based reasoning has been used very successfully in the detection of credit card fraud and the provision of diagnostic support and helpdesk facilities.
5.12 ADVANCES IN KNOWLEDGE MANAGEMENT
As manufacturing and tangible assets have become less important than services and specialist knowledge, organisations such as BP (Collison, 1999) and Hewlett-Packard (Sieloff, 1999) have realised the importance of “knowing what they know” and how to utilise this information effectively. This has resulted in considerable activity and growth in the field of Knowledge Management. Knowledge management is concerned; not only with managing “knowledge” itself, but also with the management of processes that it acts upon and how the knowledge is developed, preserved, used, and shared.
As Artificial Intelligence is principally concerned with understanding what knowledge is and how it can be represented so that it can be used, the tools and techniques developed over the last thirty years for studying and building knowledge-based systems are immediately applicable to the general concerns of knowledge management.
88
5.12.1 Knowledge Acquisition
Knowledge acquisition addresses the extraction, initial analysis, and interpretation of knowledge, so that it can be represented and encoded for use in a knowledge-based system. Many techniques have been developed to elicit different types of knowledge from different types of experts during various stages of the acquisition process. Examples include structured and unstructured interviews, protocol analysis, concept sorting, and “20 questions”. The issues addressed at this stage are mainly concerned with the knowledge itself and the tasks to be performed.
Very similar techniques are immediately applicable to extracting knowledge from documents. Indeed, as a document has been created deliberately, extracting knowledge from these can be much simpler than extracting knowledge from people. Also, if the document is available in an electronic form, the process can usually be automated. To a certain extent, the more structured a document is, the easier it is to understand and extract the knowledge contained within it. Most technical documentation, such as regulatory and statutory publications, is highly structured, thus providing a good basis for knowledge extraction and management.
As well as statistical methods such as principal factor analysis, machine learning techniques can be used to perform (semi-) automatic knowledge acquisition. These include case based reasoning (previously discussed in Section 5.12.3), rule induction, which can be used to pick out key features of examples for generalising as formal rules, and artificial neural networks. Artificial neural networks can be used to either create a generalised non-linear predictive model of the system described by the data or to identify significant groupings in data sets. All machine learning techniques require a reasonably large set of historical data.
Once the information has been captured, it must be analysed and organised before it can be represented usefully. Knowledge analysis has been the focus for most European work in this area since the early 1980s and is centred on the KADS project at the University of Amsterdam (Schreiber et al, 1993). The KADS approach, which is based on a set of interacting models, each of which is used for a particular aspect of the knowledge, provides standard libraries of models of expertise and problem solving methods.
5.12.2 Knowledge Representation and Inference
Knowledge representation is concerned with producing descriptions of the world that can be communicated between people, organisations and computers, and which can be used by them to understand and make inferences about the world.
There are many different kinds of knowledge, such as the concepts involved, how they are classified and related to another, how uncertainty and imprecision are dealt with and how tasks such as classification, design and diagnosis should be carried out. For each of these, there are various possible approaches and representations. Knowledge representation is also concerned with issues including the absence of knowledge, defaults, inheritance, and conflicting knowledge.
In principle, techniques for representing knowledge are distinct from what and how inferences can be drawn from it however, in practice, the two are usually inter-linked. How knowledge is represented has a strong influence on how easy it is for a particular system to analyse or use this. Many domains such as mathematics, linguistics and psychology have mature, well-understood representations of their own fields of knowledge. Many of the different techniques and formalisms developed in these fields have been used extensively by knowledge-based systems dealing wit h other domains. These techniques and formalisms include predicate logic, semantic networks, production systems, frames, causal networks and
89
objects. More recently, representations and languages for ontologies have been investigated and are discussed in Section 5.12.3.
Techniques for assisting decision-makers in reasoning with uncertain and conflicting information have been developed independently in the field of management science. Now that scientists in both fields are aware of each other's research there is a great deal of scope for merging of the two approaches.
5.12.3 Ontologies
A major requirement for managing knowledge is an agreed set of terms and their meanings to remove ambiguity and ensure accurate communication and understanding by all parties involved.
An ontology is “an explicit representation of a particular area of expertise expressed in terms of the concepts involved, their definitions and their inter-relationships” (Uschold and Tate, 1998). Multi-disciplinary research in the development and use of ontologies has grown rapidly since the early 1990s, with major international organisations such as Boeing (Uschold and Tate, 1998) and Unilever (Uschold et al, 1998) being heavily involved. Ontologies provide essential communication mechanisms between people and organisations and for inter-operability between systems.
Ontologies are currently being investigated and used in many different domains. Within the natural language community they are being used for characterising meaning and sense, within the database community, they facilitate the inter-operability of heterogeneous databases and within artificial intelligence, they provide a focus for the capture and representation of domain knowledge.
5.13 ADVANCES IN COMPUTER TECHNOLOGY
As well as advances in artificial intelligence technologies, there have been relevant advances in general computer technology, particularly those related to the World Wide Web. Compared to the situation ten years ago, the information world has been transformed.
5.13.1 World Wide Web Technology
The near universal acceptance of networked Personal Computers (PCs) for providing support tools, along with advances in software concerning the World Wide Web (the Web), has made it practical to distribute and share information electronically, both internally within an organisation or department and externally to others outwith the organisation.
It must be emphasised the use of Web technology does not necessarily imply any commitment to distributing documents over the Internet. Although the original rationale of the Web was an interactive display of hypertext documents transferred over international networks using HTTP (the Hypertext Transfer Protocol) using a suitable browser, for example, Netscape Navigator or Internet Explorer, this technology is just as effective on individual desktop workstations with no network access. Web technology provides publishers with the option of allowing users to either access publications on-line or obtain the same material directly from, for example, a CD-ROM using exactly the same browser software.
Through the use of Uniform Resource Locators (URLs), Web technology allows references to be incorporated into a document. By changing what the URL points to, a reference can be changed from a traditional citation of where the information can be found, to the actual information itself providing it is available in a suitable online format. As well as this ability
90
to refer to supporting information uniformly and flexibly, it is also possib le to embed code within a document. This means that rather than simply providing a formula which must be applied to set of data, an embedded program can request this, carry out the calculation and present the result, thus significantly reducing the risk of error.
The publication of the Standard Generalised Mark-up Language (SGML) standard (ISO 8879) in 1989, addressed the problems of maintaining documents that are expected to have a life expectancy greater than that of any specific word processor or text storage format presently used to hold or manipulate them. The normal problems inherent in any tool-specific approach include factors such as the continued evolution of the tool and concern about the long-term existence of the company responsible for marketing and developing the tool. To address these issues, SGML allows a document to be described by separating out its structure and content. The basic technology used to represent a document on the Web, the Hypertext Mark-up Language (HTML), was derived directly from the experience of SGML, though was initially considerably simplified.
Over the last decade, the Web’s capabilities have been greatly expanded in this area, particularly through the definition of the Extensible Mark-up Language (XML) (Bray et al, 1998). XML supports a much greater degree of customisation of the structure for different classes of document, and associated “style sheets” allow the presentation of a document to be much better separated out from its content.
5.13.2 Platform Independence
One of the major benefits of the Web is platform independence, a desirable property that was lost in EASE 2.0. Through defining and adopting standards such as HTML and XML, it is now possible to represent information in a manner that does not depend on a particular company's products, be it IBM, Microsoft, Apple etc. This has been greatly facilitated through the development of Web browsers that allow documents to be viewed on almost any computer platform, providing they have been properly structured according to these standards.
There have also been major developments in the ability to “animate” these documents; in other words, to have code associated with them which is also platform independent. This is a very significant advance as it was generally accepted that programs would be compiled into the native instruction set of the machines they were intended for, thus resulting in a different executable program being required for each platform. In particular, the Java and JavaScript languages have become widely used for Web based purposes as both are more closely linked to Web browsers, such as Internet Explorer and Netscape Navigator, than the underlying platforms.
Platform independence available now means that documents can be both held and operated in a platform independent manner. In principle, documents can be viewed using generally available tools such as Web browsers, with any additional mechanisms required to support specialised facilities being provided by an integrated, platform independent, knowledge-based system.
5.14 ADVANCES IN COMPUTER HARDWARE
In 1994, an average professional workstation used by the academic research community was driven by a processor running at around 100MHz, with tens of megabytes of main memory and a hard disc with a capacity of a few hundred megabytes. Today, an entry level PC will be based on a processor running at around 2GHz, with a minimum of 256Mb of main memory and 40Gb of hard disc. This would cost, on average, under £600.
91
Driven by the multimedia explosion of the Web, commodity computing today offers capabilities far in excess of those of research computers ten years ago. In particular, excellent graphics are provided on even the cheapest and least capable configurations, providing support for viewing electronic documents. Printing capabilities, driven by digital photography, are also an order of magnitude better and again cheaper, than they were a decade ago; non-specialist colour ink jet and laser printers can produce output that rivals that of specialist printing companies. This enormous increase in capability has fuelled users' expectations therefore software systems nowadays must be presented attractively or they will simply not be used as was reflected in the stakeholder interviews.
5.15 DISCUSSION ON IMPLEMENTATION AND SOFTWARE CONSIDERATIONS
The position is very different now to what it was when EASE was first proposed. The EASE task itself is better understood, due largely in part to having built the system and making it widely available . This has also helped validate the use of computer-based systems for providing support in this area.
As discussed there have been significant advances in technology since 1993. Advances in Artificial Intelligence technology, particularly in case based reasoning, could offer alternative approaches when formulating a new EASE model. However, the advances in web-based technology provide a radically different environment within which any future EASE system will need to be deployed.
It is important to remember that the EASE model is only a part of the EASE system. When a new system is built, much of the effort will need to be devoted to providing the structure through which the model will be accessed. As users' expectations of systems have dramatically increased due to the much-increased power of their computers, the EASE system will have to provide significantly enhanced communication and user support facilities. Much of this can be simplified by using an existing framework; in particular by taking advantage of the standard facilities provided by Web browsers. This would also have the additional benefit of encouraging platform independence.
The Web is able to provide an eminently suitable context for presenting the EASE model for two further reasons. Firstly, an essential feature of EASE is that the model is provided in a natural, easy to use manner. The Web is based around the concept of a navigable document, possibly containing embedded programs, and its navigation model is immediately suitable for guiding a user intelligently through the sequence of steps needed to use the model correctly. Secondly, the Web provides simple, standardised ways of providing supporting information either directly as special purpose Web pages or as links to authoritative sources of further information.
As noted in Section 5.13.1, the use of Web technology does not necessarily imply any commitment to using the Internet, although that does become a possibility. If the system were provided over the Internet, system maintenance, development and distribution become simpler as a change made to the master system becomes immediately available to all of the system's users. Provision over the Internet means that availability enquires can be dealt with by an informative Web page and there is no need for any mechanism for logging and despatching copies of the system to individuals. It is also easier to monitor system use.
The EASE system developed so far has been presented as a stand-alone tool that is available for use completely independently of any other tool. Given the development in Web technologies, this is no longer necessary and there may well be benefits from presenting related tools through a common framework that gives them a similar appearance, provides consistent predictable behaviour, and allows them to exchange information. It also
92
encourages collaboration between different teams of developers. It is noted that COSHH Essentials documentation is now available through the Web (http://www.coshh-essentials.org). The COSHH Essentials Web page’s start by determining what sort of task is being considered and this is also needed for EASE. Ensuring that the same task classification is used by both systems (if appropriate) could be an immediate potential benefit.
93
94
6. A DISCUSSION OF THE WAY FORWARD
6.1 INTRODUCTION
We have set out to scrutinise the underlying structure and philosophy of the EASE model and provide a critical appraisal of its utility and performance. In doing this we have consulted a variety of stakeholders and reviewed the available scientific literature. This work has enabled us to set out a vision for the future development of EASE so that it can continue to be a useful tool for regulatory risk assessment.
For almost ten years EASE has provided a way for regulators to estimate occupational chemical exposure. It has served well and there is no evidence that it has resulted in unsafe regulatory decisions. However, there is both documentary evidence and the strong opinion amongst many stakeholders that the predictions from EASE are not sufficiently reliable for future needs. The information used to define the exposure ranges in EASE is likely to be more than 20 years old and there have been many changes in European workplaces during this time. There is also increasing concern about the highly conservative nature of the EASE predictions and the potential impact of these on the competitiveness of European businesses. However, one of the strengths of EASE is that it maps the situation for which an exposure estimate is required on to a situation for which you have some information. The weakness is that this is not always done well.
6.2 WHAT SHOULD AN IMPROVED EASE DELIVER?
What do stakeholders want from EASE? There are currently many characteristics of EASE that are valued, but often there are associated limitations. These include:
• transparency of the model and the underlying assumptions, although there is concern than the model does not encompass all of the key exposure determinants;
• linkage to actual workplace measurements, although as we have seen these are increasingly becoming archaic;
• precautionary predictions, although perhaps too precautionary;
• provision of predicted exposure ranges, although their meaning is perhaps not sufficiently explicit and
• EASE is perceived as being easy and quick to use.
At present EASE provides exposure estimates for the inhalation and dermal exposure routes. Stakeholders felt that both of these routes were important for regulators to consider. There was less enthusiasm for the inclusion of accidental ingestion into the model framework, although other exposure models we reviewed have this possibility. We believe that for some types of substance, for example metals and biocides, inadvertent ingestion may be important and so we consider it should be included in any further development of EASE.
Stakeholders had diverse opinions about the choice of exposure metrics for the model output. For all routes of exposure the model should produce estimates of daily (8-hour) exposure and separate exposure estimates for tasks that may be carried out within a working day. In this way regardless of how the user prefers the data to be presented they can access the information they wish. We believe that for inhalation exposure the model should produce estimates of the concentration of the chemical in the worker’s breathing zone and the estimated average breathing rate. It may also be useful to have predictions of the likely particle size distribution for aerosols, particularly whether the aerosol is mostly inhalable or
95
respirable. For dermal exposure the main exposure parameters should be the mass of chemical on the skin surface, the concentration of the chemical on the skin and the area of skin exposed. For accidental ingestion the exposure metric should be the estimated mass of chemical taken into the mouth. The model should also explicitly contain information about the duration of the activity or activities that are being assessed, particularly if the aim is to produce a time weighted average exposure estimate based on information about subsidiary tasks or activities.
However, the most important question to answer before any further development of EASE can be undertaken is what is the key purpose of the model? EASE was originally developed for assessment of exposure for substances new to the European marketplace where it was envisaged that there would be very little information available about the uses of the product. Today EASE is used to fill in gaps in exposure measurements in the regulatory risk assessment of existing substances, to provide data to support the setting of occupational exposure limits and to predict exposure levels in a specific workplace. It can be argued that the requirements for accuracy and precision differ between each of these uses. For example, in the case of risk assessment of new substances it may be particularly advantageous to be precautionary and so a model with fairly poor accuracy that overestimates may be acceptable. Using a model to provide exposure estimates for setting exposure limits or to estimate the risk in a specific work environment requires greater accuracy and precision if the output is to be useful.
The four uses of models outlined above usually differ in the quality of information available for the exposure assessment. In the case of new substances there may perhaps only be general information about the possible exposure scenarios. In the case of existing substances there should be better information available about the actual patterns of use and so it should be possible to obtain a more reliable estimate of exposure. In addition, there may also be exposure data available that could be used to augment the model estimates. Finally, for a specific workplace prediction there could be detailed information available about the type and effectiveness of local ventilation systems, quantities of substance used, room temperatures and so on. All of this information could be used to support more detailed and hopefully more accurate exposure modelling.
6.3 THE WAY FORWARD
The validation studies that we have reviewed make a clear case for improved accuracy in the exposure estimates, particularly for gases and vapours. We believe the extent of the overestimation in the current version of EASE is unsustainable; regulators are being too precautionary and, because of this, industry is being forced to control hazards more carefully than may be necessary. The question is whether it is possible to develop the present version of EASE to give the desired characteristics or whether a more radical solution is needed.
It would be possible to continue developing the existing version of EASE to address the limitations we have identified and to improve the accuracy of the predictions. For example, some authors have made suggestions to improve the calculation of vapour pressure (Vincent et al, 1996) and others have provided suggestions for improving the estimation of vapour concentrations arising from complex liquid mixtures (Urbanus, personal communication). These changes are likely to improve the reliability of the model predictions for specific substances or in specific use scenarios. However, we do not believe these types of changes would deliver the improvements that the stakeholders wished for. Perhaps this type of approach would be most suitable if EASE were just to be used for new substances, as it had originally been envisaged.
96
Others have suggested that improved reliability could be achieved by carefully defining a subset of substances and uses for which EASE assessments are valid. To some extent this could already be done using the information available from the studies reviewed in Chapter 4. This approach runs counter to the underlying philosophy of EASE that it should be applicable to any and all substances and patterns of use. However, again this type of strategy could be useful if EASE was to be restricted to a screening model for new substances, and it could usefully augment the evolutionary development of EASE described above. Overall we have concluded that if we are to achieve the stakeholders aspirations for EASE then we need to adopt a more revolutionary approach to the model, although clearly the actual approach will depend on clear agreement of the regulators and other stakeholders on the purpose of EASE. We have reviewed a range of consumer and workplace exposure models and concluded that none of them provides a clear, better alternative to EASE. Indeed, EASE currently has many characteristics that we believe are superior to some of the models reviewed. An alternative approach is clearly merited.
Figure 15 shows the general scheme that we suggest for further development of EASE. It comprises both improved model software and the ability to combine this with available data.
Figure 15: Proposed structure for the development of EASE
Deterministic model inc. Monte
Carlo module
Bayesian process to combine data
and model output
Exposure estimates for risk
assessment
Similarity module to select data for risk
assessment
Exposure database, with contextual
information
We consider that a deterministic model is both appropriate for the further development of EASE and could provide more accurate exposure estimates. From our earlier discussion we feel that the deterministic part of the model should be developed with the target of predicting the median exposure level in a single work situation to within a factor of three of the true value. The approach developed by Cherrie, Schneider and others for inhalation exposure (Cherrie et al, 1996; Cherrie and Schneider, 1999) produced better agreement with measured exposures than did EASE, which suggests that there is further scope to improve the reliability of exposure assessments for regulatory risk assessments. Much of the impetus for a deterministic model has come from the development of a clearer theoretical understanding of how people become exposed, as for example in the conceptual model of dermal exposure developed by Schneider and colleagues (Schneider et al, 1999).
The BEAT model for dermal exposure to biocides has demonstrated that it is possible to develop an empirical model that can adapt to take account of new data. While we feel that this
97
may be an appropriate way forward in a situation with a relatively limited range of scenarios it is unlikely to be sufficient on its own for the huge diversity of situations that may be encountered with general chemical use with only limited available measurement data. Nevertheless, BEAT has shown how the combination of model predictions and measurement data can provide a powerful framework for exposure assessment. We believe that this approach should be used in any further development of EASE. It will ensure that stakeholders are reassured that the exposure estimates arising from this process are linked to workplace measurements, with the added advantage that these links can easily be kept up to date. Currently exposure assessment is usually based upon either models or empirical data. Whilst the estimates from these contrasting approaches might be compared, unlike the proposed scheme, there is no single coherent exposure estimate. By combining these two information sources the proposed re-development of EASE would represent a major advance in the harmonisation of exposure assessment.
Models and measurements are complementary: models can provide exposure estimates for all possible situations where a chemical is used using information about the circumstances surrounding exposure, while measurements can provide accurate exposure estimates for a limited set of exposure circumstances. We believe that combining both sets of data will maximise the possible accuracy and precision of any regulatory exposure assessment. It is foolish, as some have suggested, to dispense with model estimates when measurements are available since there is no real guarantee that a limited measurement data set reflects the range of situations covered by the risk assessment.
In our opinion the goal of the exposure estimation should be to produce the most accurate and precise estimates of exposure possible. While a conservative assessment is clearly valued by stakeholders rather than having it implicit in the assessment we feel it is more appropriate to explicitly add in this conservatism afterwards through a safety factor based upon the model variability and uncertainty. The ideal therefore, is a model that provides an accompanying measure of the uncertainty in the predicted exposures.
It is envisaged that the available exposure measurements to be used in a risk assessment will be held in a database along with appropriate contextual information. The relevant information in the database and the associated model predictions could be combined using a Bayesian statistical framework, similar to that in the BEAT model. The University of Aberdeen, IOM and the University of Utrecht are currently collaborating on the development of a database for use within the European chemical industry and in other organisations. This database, known as the CEFIC Exposure Management System (CEMAS), will enable inhalation and dermal exposure measurements to be stored with a wide range of contextual information. We believe that this could provide the basis for the type of approach we have outlined here. Data could be collected routinely by industry or government, or could be specifically collected to address the needs of the regulatory risk assessment.
In many cases there may be data in the database that is relevant to the particular scenario being evaluated, but not an exact match. For example, the substance to be assessed may be the same, but the formulation of the chemical product may be different, or the use pattern may be the same but the substance different. It would be desirable to use this data in updating the model predictions, but to do this we would need to have some criteria for judging when a scenario was “similar” enough to another to allow for this. We believe that there should some further research to investigate the use of similarity algor ithms for judging when data from one situation may be used to help predict exposure in another situation. This may be possible using our theoretical understanding of exposure, statistical analysis of existing data sets, along with the knowledge acquisition techniques that have been developed within the field of artificial intelligence. This work would involve the use of structured questionnaires, to assess the expert opinion of occupational hygienists, human exposure researchers and regulatory risk assessors concerning the basis for their current choice of data for use in risk assessment.
98
To facilitate the combination of model predictions and data using Bayesian techniques we must first produce a model that, as well as the average exposure, also predicts both the variability in the exposure and the uncertainty. The variability arises from differences between workplaces, between different workers in the same work scenario and differences for the same worker from day to day. The variability of exposure is important because it is the basis for evaluating what might be considered as a “reasonable worst case”, which is an integral part of the subsequent risk characterisation stage of the risk assessment and defining an appropriate exposure range. We envisage that rather than having predefined ranges as at present an updated version of EASE would present the user with a confidence interval for the mean predicted exposure. Uncertainty is the additional variation in the model predictions because of lack of information or understanding of how a particular factor may impact on exposure. This is also important because it indicates the likely level of confidence that may be placed in the estimated worst case exposure. Estimation of variability and uncertainty in the mode l may easily be carried out using Monte Carlo simulation techniques.
6.4 UPDATING EASE
We are not able to specify exactly what form the model may take, but we feel that the conceptual model we have sketched earlier could form a reasonable basis for future development. This has clear advantages in making the basis for the exposure assessment transparent to stakeholders. Figure 16 reproduces this conceptual model.
Figure 16: Conceptual model of inhalation and dermal exposure
Source
Room airBreathing zone air
Surfaces
Respiratory tract
Clothing Skin
contaminant layer
Outside air
The most appropriate way forward is to identify the range of transfer factors and compartment descriptors that are linked to the model and develop a simplified deterministic model of exposure from these. The inhalation model of Cherrie and his co-workers (Cherrie, 1999) is compatible with this approach and we believe that it is a reasonable starting point. In this scheme there are two equations that describe the exposure for a single task: one for material emitted from the source that goes directly to the breathing zone compartment (also known as
99
the near-field) and one for material emitted to the room air and then to the breathing zone (known as the far-field). It is assumed that there is negligible contaminant in outside air, a reasonable assumption for occupational exposure scenarios, and so exchange of air between the room and the outside serves to reduce the contaminant concentration in the room air.
It is possible to develop a similar assessment scheme for dermal exposure, although this is likely to be more complex given the greater number of potential pathways for contaminant to transfer from the source to the skin. However, it is encouraging that the existing version of EASE, which is very limited in only really taking account of contact with surfaces, has some predictive power. Similarly, a conceptual model could be constructed for ingestion exposure and this could be used as the basis for a predictive model.
However, one of the strengths of EASE is that it maps the situation for which an exposure estimate is required on to a situation for which you have some information. The weakness is that this is not always done well. BEAT uses a similar, but more sophisticated, approach, and case-based reasoning also uses this approach.
The process that EASE uses is particularly relevant to regulatory risk assessment because the regulator is unlikely to have detailed knowledge of the way in which the chemical is used. EASE is less useful for predicting exposures in individual situations where more detailed information is available.
Another strength of EASE is that it can be applied to any industrial situation, although finding the appropriate combination of pattern of use and pattern of control sometimes requires a great deal of creative thinking. A deterministic model may not be capable of total coverage.
6.5 CONCLUDING REMARKS
In our review we observed that there are considerable differences between the approach taken for exposure assessment for consumer exposure to chemicals and that used for occupational exposure. The existing EASE and CONSEXPO models are fundamentally different and this is likely to result in completely different estimates of exposure even when the underlying scenarios are similar, e.g. brush painting in an enclosed space. The conceptual model that we have outlined is sufficiently general that it would be applicable in both the workplace and in consumer scenarios. It seems inappropriate that consumer and occupational exposure assessment techniques are so discordant and efforts should be made to try to harmonise, as far as possible, these two regulatory processes.
The overall complexity of the proposed model for occupational exposure makes it an absolute necessity to develop a completely self-contained computational package that does not require the user to have any mathematical expertise. Assuming a suitable exposure database is available, several additional steps can be identified: 1. development of appropriate deterministic models for inhalation, dermal and ingestion
exposure; 2. develop new EASE software based on these models and ensure that the output includes a
distribution of estimated exposures and the associated uncertainty in these predictions (known as the “prior” in Bayesian terminology);
3. similarity algorithms must be constructed into the database search facilities to allow the automatic extraction of relevant exposure data to update the model predictions;
4. integrating the Bayesian updating software to combine the prior distribution of exposures with the available “similar” measurements and
5. develop integrated software for presenting the estimated exposure parameters and displaying the uncertainty distributions for model outputs.
100
The final integrated model should be disseminated to end-users through the Internet either as a collection of downloadable software modules or, more ideally, through a web-enabled version that can be used online. The model must contain its own self-contained help system.
We believe that the further development of EASE must be seen as a priority for regulatory risk assessment of chemicals and that this initiative should be taken forward within a European context. Acceptance of any new scheme will be dependent on all relevant stakeholders being committed to the proposed approach. Before a decision can be made on the best way to model and estimate exposure, the model developers, the regulators and industry need to discuss the purpose and intended use of a successor to EASE; both of these will have a strong bearing on the form of the model.
101
102
ACKNOWLEDGEMENTS
The authors would like to thank all those individuals who took part in the stakeholder interviews and Christiaan Delmaar and Jacqueline van Engelen. RIVM, The Netherlands, for their advice on the CONSEXPO model. Our thanks are also extended to those past and present HSE and HSL personnel who assisted with this research.
The UK Health and Safety Executive, the European Chemical Industry Council (CEFIC) and the American Chemistry Council jointly funded this research.
103
104
REFERENCES
Bray T, Paoli J, Sperberg-McQueen CM. (1998) Extensible Mark-up Language (XML) 1.0. http://www.w3.org/TR/REC-xml.
Bredendick-Kämper S. (2001) Do EASE scenarios fit workplace reality? A validation study of the EASE model. Appl Occ Environ Hyg; 16: 182-187.
BSI (2001) The TickIT Guide. Issue 5.0. British Standards Institute.
Burns DK, Beaumont PL. (1989) The HSE National Exposure Database – (NEDB) Ann Occup Hyg; 33(1): 1-14.
Cherrie JW. (1999) The effect of room size and general ventilation on the relationship between near and far field concentrations. Appl Occup Environ Hyg; 14: 539-546.
Cherrie JW, Robertson A. (1995) Biologically relevant assessment of dermal exposure. Ann Occup Hyg; 39: 387-392.
Cherrie JW, Schneider T. (1999) Validation of a new method for structured subjective assessment of past concentrations. Ann Occup Hyg; 43: 235-246.
Cherrie JW, Schneider T, Spankie S, Quinn M (1996). A new method for structured, subjective assessments of past concentrations. Occupational Hygiene; 3: 75-83.
Collison C (1999) BP-Amoco's knowledge repository - connecting the new organisation. Knowledge Management Review, 7, March/April.
Dervillers J, Domine D, Bintein S, Karcher W. (1997) Occupational modelling with EASE, SAR and QSAR in Environmental Research; 6: 121-134.
Devine JM (1993) The estimation of occupational exposure to chemicals. In OECD (1993) Environment Monograph No. 70 – Occupational and Consumer Assessments (OCDE/GD(93)128) OECD Paris Annex II – The Estimation of Occupational Exposure to Chemicals. pp. 35-50. Available by e-mail from [email protected]
Dijkhof W, Bakker S, Meuleman J (1999). Maximale belasting innovatieve gastoestellen. Fase B Validatio rekenmodel en voorstel regelgeving. Report HGT/99115/DIJ/031, GasTec, Apeldoorn. Cited in van Veen MP. CONSEXPO 3.0 Consumer exposure and uptake models. RIVM report 612810 011 2001.
Esmen NA. (1991) Analysis of strategies for reconstructing exposure. Appl Occup Environ Hyg; 6: 488-494.
ECETOC (1997) Exposure Assessment in the Context of the EU Technical Guidance Documents on Risk Assessment of Substances. ECETOC Report No. 35. Brussels: ECETOC.
EU-DG Environment, Nuclear Safety and Civil Protection (1995) Technical Guidance Document on Risk Assessment for Human Health of New and Existing Substances.
European Commission (1993) Directive 93/67/EEC on Risk Assessment for New Notified Substances European Commission, Brussels.
105
European Commission (1993b) Council Regulation (EEC) No 793/93 of 23 March 1993 on the evaluation and control of the risks of existing substances [Official Journal L 84, 05.04.1993].
European Commission (1994) Regulation No 1488/94 on Risk Assessment for Existing Substances European Commission, Brussels.
European Commission (1996) Technical Guidance Document in support of Commission Directive 93/67/EEC on Risk Assessment for New Notified Substances and Commission Regulation (EC) No 1488/94 on Risk Assessment for Existing Substances European Commission, Brussels ISBN 92-827-8011-2.
European Commission (1996b) Technical Guidance Document in support of Commission Directive 93/67/EEC on Risk Assessment for New Notified Substances and Commission Regulation (EC) No 1488/94 on Risk Assessment for Existing Substances Part III. European Commission, Brussels ISBN 92-827-8011-2.
Fehrenbacher MC, Macek GM. Estimating Dermal Exposure to Workers in the U.S. EPA’s New Chemicals Program International Occupational Hygiene Association. Paper presented at the First International Scientific Conference on Occupational Hygiene Brussels 1992.
Fehrenbacher MC, Hummel AA. (1996) Evaluation of the mass balance model used by the Environmental Protection Agency for estimating inhalation exposure to new chemical substances. Am Ind Hyg Assoc J; 57: 526-536.
GEOMET Technologies, Inc. (1989) MCCEM User’s Guide: Version 2.1, Germantown, Maryland (Report No. IE-2130). Cited in Koontz MD, Nagda NL. (1991) A multi-chamber model for assessing consumer inhalation exposure. Indoor Air; 4: 593-605.
GEOMET Technologies, Inc. (a division of Versar, Inc.) for USEPA Office of Pollution Prevention and Toxics Washington, DC and National Paint and Coatings Association Washington, DC. Wall paint exposure model (WPEM) version 3.2 User’s guide 2001.
GEOMET Technologies, Inc. MCCEM: Multi-Chamber Concentration and Exposure Model and user’s guide, developed for US EPA Office of Pollution Prevention and Toxics, Washington DC, Nov 1995.
HSE (1992a) Internal paper – Estimation of Occupational Exposure to Chemicals: Development of an Expert System
HSE (1999) The Assessment of Workplace Exposure to Substances Hazardous to Health – The EASE Model Version 2 for Windows
Hughson GW, Cherrie JW. (2001) Validation of the EASE model in relation to dermal zinc exposures. Edinburgh: Institute of Occupational Medicine (IOM Report TM/01/01).
Hughson GW, Cherrie JW (1999) External validation test of EASE for Windows version 2. Contract research report 3846/R51.172, 1999.
ISO (2000) ISO 9001:2000 Quality Management Systems - Requirements. International Organisation for Standardisation.
Koontz MD, Nagda NL (1991). A multi-chamber model for assessing consumer inhalation exposure. Indoor Air; 4: 593-605.
106
Mark D. (1999) Validation of the EASE model. HSE contract report no. R51.172,
Nagda NL, Koontz MD, Kennedy PW. (1995) Small-chamber and research-house testing of tile adhesive emissions. Indoor Air; 5: 189-195.
Nicas M. (1996) Estimating exposure intensity in an imperfectly mixed room. Am Ind Hyg Assoc J; 57: 542-550.
OECD. http://webdomino1.oecd.org/comnet/env/models.nsf.
OECD (1993) Environment Monograph No. 70 – Occupational and Consumer Assessments (OCDE/GD(93)128) OECD Paris Annex II – The Estimation of Occupational Exposure to Chemicals Available by e-mail from [email protected]
Phillips A, Garrod A. (2001) Assessment of dermal exposure – empirical models and indicative distributions. Appl Occup Environ Hyg; 16: 323-328.
Price PS. (2001) An evaluation of the potential for use of existing exposure software (or software currently under development) in a tiered approach to the assessment of exposures and risks to children. LiFELINETM, prepared for the American Chemistry Council.
Pryde E. (2003) EASE past and present. Unpublished communication.
Ramachandran G. (2001) Retrospective exposure assessment using bayesian methods. Ann Occup Hyg; 45: 651-667.
Ramachandran G, Vincent JH. (1999) A bayesian approach to retrospective exposure assessment. Appl Occup Environ Hyg; 14: 547-557.
RISKOFDERM (2001) Risk assessment for occupational dermal exposure to chemicals. First year report. TNO, Zeist, The Netherlands, March 2001.
5thRISKOFDERM (1999) Contract number QLK4-CT-1999-001107. Frame work programme of the European Community, European Commission, Luxemburg.
Risk Assessment and Toxicology Steering Group (1999) Exposure assessment in the evaluation of risk to human health. Leicester: MRC Institute for Environment and Health.
Royal Society of Chemistry (1992) Risk Management of Chemicals, Chapter 11 ISBN 0-85186-4657-8
US Environmental Protection Agency. Exposure assessment tools and models. http://www.epa.gov/opptintr/exposure/
Schank R. (1982) Dynamic Memory: A Theory of Learning in Computers and People. Cambridge University Press.
Schneider T, Vermeulen R, Brouwer DH, Cherrie JW, Kromhout H, Fogh CL. (1999) A conceptual model for assessment of dermal exposure. Occup Environ Med; 56: 765-773.
Semple SE, Proud LA, Tannahill SN, Tindall ME, Cherrie JW. (2001) A training exercise in subjectively estimating inhalation exposures. Scand J Work Environ Health; 27: 395-401.
Semple S, Proud L, Cherrie JW. (in press) Use of Monte Carlo simulation to investigate uncertainty in exposure modelling. Submitted to Scand J Work Environ Health.
107
Sieloff CG. (1999) If only HP knew what HP knows: the roots of knowledge management at Hewlett-Packard. Journal of Knowledge Management; 3(1): 47-53.
Schreiber AT, Weilinga BJ, Breuker JA. (1993) editors. KADS: A Principled Approach to Knowledge-Based System Development, volume 11 of Knowledge-Based Systems. Academic Press, London.
Tickner J (unpublished) EASE consistency study (see Appendix 2).
Uschold M, Jasper R. (1999) A framework for understanding and classifying ontology applications. In Benjamins VR, Chandrasekaran B, Gomez-Perez A, Guarino N, Uschold M, editors, IJCAI-99 Workshop on Ontologies and Problem-Solving Methods (KRR5)
Uschold M, Tate A. (1998) Putting ontologies to use. The Knowledge Engineering Review; 13 (1): 1-3.
Uschold M, King M, Moralee S, Zorgios Y. (1998) The enterprise ontology. The Knowledge Engineering Review; 13. Special Issue on Putting Ontologies to Use, eds. Uschold M and Tate A.
Van Rooij JGM, Jongeneelen FJ (1999) An assessment of the quality of estimation of the concentration of chemicals with the EASE expert system; testing of measurements. HSE report 99/17.
van Veen MP. (2001) CONSEXPO 3.0 Consumer exposure and uptake models. RIVM report 612810 011.
van Veen MP, Fortezza F, Bloemen HJTh, Kliest JJ. (1999) Indoor air exposure to volatile compounds emitted by paints: experiment and model. J Exp Anal Environ Epidemol; 9: 569-574.
van Veen MP, Fortezza F, Spaans E, Mensinga TT. (2002) Non-professional paint stripping, model prediction and experimental validation of in door dichloromethane levels. Indoor Air; 12: 92-97.
Vincent R, Dervillers J, Bintein S, Sandino JP, Karcher W (1996) Evaluation of EASE 2.0. Assessment of occupational exposures to chemicals. INRS/CTIS, ECB.
Warren N. (2002) User manual for Bayesian Exposure Assessment Toolkit version 1.13.
Weegels MF, van Veen MP. (2001) Variation of consumer contact with household products: a preliminary investigation. Risk Analysis; 21: 499-511.
Wild P, Sauleau EA, Bourgkard E, Moulin J-J. (2002) Combining expert ratings and exposure measurements: a random effect paradigm. Ann Occup Hyg; 46: 479-487.
Wilschut A, ten Berge WF, Robinson PJ, McKone TE. (1995) Estimating skin permeation. The validation of five mathematical skin permeation models. Chemosphere; 30: 1275-1296.
Zadeh LA. (1965) Fuzzy sets. Information and Control; 8: 338-353.
108
Appendix 1: Personnel involved in the development of EASE
Paul Beaumont Managed and developed NEDB from the late 1980s to early 1990s. Paul undertook the fundamental work on the database content to organise the exposure information into coherent groups that could then be analysed to generate and validate the exposure ranges for different categories of substance use and control. With others, including Eric Pryde, Paul produced a prototype KBS for determining occupational exposure levels. This was undertaken in 1992 alongside a major programme of other work.
Murray Devine Head of Health Policy Directorate’s Chemicals Policy Division from 1993 to 1998. Murray was responsible for HSE’s strategy on substances hazardous to health and created the conceptual model that was used to develop EASE.
Jeff Friar Head of the Chemical Agents Group in Technology Division’s Occupational Hygiene Unit from 1991 to 1997. Jeff managed the project that developed EASE, including the research project with HSL and AIAI that transformed the logic structure into a workable software model.
Diane Llewellyn Diane worked on EASE with John Tickner from 1997 to 2000 and with Andy Phillips from 2001-2002. Diane undertook the analytical work to create the revised logic charts for low dust techniques for version 3 of EASE and revised the user manual to include examples of the use of the model.
Len Morris Len was responsible for dermal exposure for the Occupational Hygiene Unit from 1993-4. Len was involved in the early development of the dermal exposure prediction part of EASE along with Jeff Friar.
Eric Pryde Eric was the main contact between the Occupational Hygiene Unit and HSL at Sheffield. He developed the prototype version of EASE that was accepted by the European Commission, and managed the contract with AIAI to produce the CLIPS version of the model. He was extensively involved in producing each version of EASE, and in 1999-2000 he developed the database of substance properties for version 3 of EASE and the implementation of the empirical formulae that are used in that version to calculate vapour pressure for water-soluble gases.
John Tickner Managed NEDB and EASE from 1997 to 2000. He instigated the work to develop version 3 of EASE.
Acknowledgement Dr Ian Guest (formerly of GlaxoWellcome) provided data and expertise to help create the lowest ranges of the revised logic chart for dusts.
109
110
Appendix 2: EASE Consistency study
1. Background
The assessment of occupational exposure in risk assessments cannot always be based on exposure measurement. Constraints (e.g. time and funds) on carrying out a survey to gather enough data to make reliable judgements, often make this an impractical option. An alternative is to use modelling techniques to predict or estimate contaminant concentrations in the workplace. The EASE (Estimation and Assessment of Substance Exposure) model was developed for this purpose.
EASE is a knowledge-based ele ctronic expert system. It is essentially a series of decision trees. It asks a set of questions about the numerous factors that influence exposure, e.g. about the properties of the substance, its pattern of use and measures used to control exposure. The exposure prediction is determined by the choice of answers made.
For any one question (apart from those asking for physical data), the EASE user must select from a number of representative categories, hence the model relies on a certain amount of professional experience and judgement. The main purpose of this study is to examine how consistent EASE users are in making those decisions, and to subsequently develop improvements in the model where these are found necessary.
2. Details of tests
Using information in reports and files, together with personal experience, 15 workplace scenarios were developed. Each scenario was designed to contain sufficient information to enable the test subjects to use the EASE model to predict exposures. Substances covered included fibrous and non-fibrous dusts, as well as vapours from liquids. Scenarios came from a range of industries. The scenarios given to the test subjects can be found in Section 7 of this report, they are written in italics.
Test subjects (34 in all) were asked to read through each scenario and then input the information at the appropriate stage in the EASE model. The EASE model logs each decision a user makes in an electronic file. Each file was given a unique name. All of the information recorded was subsequently entered into a database to aid data analysis.
3. Details of subjects
The test subjects were all occupational hygienists, with a range of backgrounds and years of experience in the field. For each scenario the subject was asked if they were familiar with the situation described (this was recorded on a worksheet), this allowed for any differences between the experienced and inexperienced subjects to be gauged.
4. General overview
To gauge consistency of use of the EASE model, for each scenario the percentage of subjects reaching the same EASE outcome was calculated. For each scenario there was a clearly favoured outcome, demonstrating a degree of consistency. The percentage of subjects in the favoured outcome group, varied with each scenario. The range was from just under 30% for scenario 7 to just over 90% for scenario 3. The average figure over all the scenarios was 62%. The spread of outcomes for each scenario ranged from 3 to 16.
111
Agreement was much better in the dust scenarios (scenarios 1 to 6 and 15) than in the gas/vapour/liquid aerosol scenarios (7 to 14).
Most inconsistencies occurred when EASE asked the test subjects questions, which required judgmental decisions, although some operator errors occurred when inputting factual information (e.g. entering incorrect units). The results of the study suggest that some of the inconsistencies arose because of misunderstandings in the terminology used in the EASE model. It is anticipated that improvements in the model will remedy this and lead to greater consistency of use. The suggested improvements are summarised in Section 8 of this report.
5. Overview of dust scenarios
Chart 1: Distribution of EASE outcomes for the 6 dust scenarios
% o
f co
un
t
100
90
80
70
60
50
40
30
20
10
0
Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6
Typically the subject’s predictions fell into 3 or 4 of the 10 possible outcomes for non-fibrous dusts. For each scenario, this graph shows the percentage split across those 3 or 4 outcomes.
The graph shows that EASE use was fairly consistent. On average just over 75% of the test subjects reached the same conclusion for a scenario. Subjects were consistent in entering the more factual information for a scenario, for example, the physical properties of the dust (with the exception of deciding if the dust aggregates) and details of controls. Inconsistencies mostly arose from the judgmental decision on pattern of use. To increase consistency, it is suggested that definitions and help boxes are improved. Further details are given in Section 7.
6. Overview of gas / vapour / aerosol scenarios
The number of possible outcomes for gas and vapour scenarios are far greater than for dust scenarios, they total 59. Charts 2 and 3 show the spread of results for each of the scenarios. It is perhaps unsurprising that test subjects were less consistent with these scenarios, compared with the dust scenarios, on average 50% of the test subjects reached the same conclusion. As in the dust scenarios, inputting of factual data caused a few problems, although some operator errors were made in these scenarios (e.g. entering incorrect units). For these scenarios inconsistencies arose in making judgements on aerosol formation, pattern of use and pattern of control. As in the dust scenarios, some of the difficulties appeared to arise from misunderstanding the terminology used to describe EASE categories. Section 7 gives the details, together with suggestions for improving consistency of use.
112
Chart 2: Distribution of EASE outcomes for the vapour scenarios
% o
f co
un
t
90
80
70
60
50
40
30
20
10
0
Scenario 7 Scenario 8 Scenario 9 Scenario 10
Typically the subject’s predictions fell into several of the 59 possible outcomes for vapours. For each scenario, this graph shows the percentage split across those outcomes.
Chart 3: Distribution of EASE outcomes for the vapour scenarios
% o
f co
un
t
100
90
80
70
60
50
40
30
20
10
0 Scenario 11 Scenario 12 Scenario 13 Scenario 14
Typically the subject’s predictions fell into several of the 59 scenarios possible outcomes for vapours. For each scenario, this graph shows the percentage split across those outcomes.
7. Details of each scenario
7.1 Scenario 1 – the Long Thin Brick Company
The company makes bricks from clay. A survey concentrated on the soft mud process of the brick making procedure. This involves extruding clay to shape it and, cutting the bricks to size before they are sent for firing. Dust can arise from the surfaces of the bricks as they dry, and from spills of clay drying on the floor. Personal samples were taken of respirable silica
113
levels. What would you predict the results to be? (68% of subjects felt experienced in predicting exposure for this scenario.)
The subject’s predictions fell into 4 of 10 possible exposure ranges for a non-fibrous dust. The critical decisions for predicting exposure to a non-fibrous dust are entering the pattern of use and deciding if the dust aggregates.
Aggregates 88% N 22% Y Use pattern 62% DM 35% LDT 3% DC&G
LEV 100% N
Comments
The areas where decisions were inconsistent were choice of pattern of use and choice of whether the particles aggregate. The choice for pattern of use was quite difficult. The process being described was wet (soft mud), however, the scenario states that dust can arise from surfaces and floors as they dry out.
For each pattern of use EASE defines the categories in a help box. Low dust techniques include wet processing and any other technique where sufficient care is exercised to substantially reduce potential exposures. It is also used if the substance is handled in a booth, ventilated cabinet or clean room. 35% of subjects selected low dust techniques for the pattern of use. Selecting low dust techniques with no LEV predicts an exposure range of 0 to 5 mgm-
3 dust. For this scenario the only low dust technique that does not include LEV is wet processing. If wetting the substance were 100% effective then there would be no dust exposure. This begs the question, where does up to 5 mgm-3 of dust come from? Assuming that this level allows for less than 100% effective wetting and that it allows a degree of ‘drying out’, then this needs to be made clear in the EASE definitions. Perhaps then, the 62% of subjects who selected dry manipulation, despite the described process being wet, would have opted for low dust techniques. Dry manipulation is defined as any manipulation of the dry material. The scenario did not mention any manipulation of the dry clay, this has been assumed by the subjects. The person who selected dry crushing and grinding fell within the experienced group.
If the definitions of low dust techniques and dry manipulation were to be improved, it is likely that the choice between the two would have been easier to make.
The only other inconsistency for this scenario was in deciding if the dust aggregates. Of the 4 subjects (22%) who selected yes the particles aggregate, 3 were experienced and 1 inexperienced. EASE defines readily aggregating solids as substances which are waxy in texture or which are in some other way sticky so that particles of the solid readily aggregate and will give rise to less dust than those solids with particles which don’t readily aggregate. Assuming that this applies to dry particles and not stickiness due to wetting (a different issue altogether) then this should be made clear in the EASE definitions. Some examples of both types of material would be helpful here.
7.2 Scenario 2 – Falldown Furniture Ltd.
A factory makes flat-pack self -assembly furniture. One of the processes is to sand the component parts of the furniture. This is done in an open factory using either a belt sander, drum sander or bobbin sander. All of the machines are supplied with LEV(a typical machine is shown in a photograph). What levels of inhalable wood dust might be anticipated? (76% of subjects felt experienced in predicting exposure for this scenario.)
114
The subject’s predictions fell into 4 of 10 possible exposure ranges for a non-fibrous dust. Predictions were more consistent than in scenario 1, with 82% coming up with an exposure range of 2 to 10 mgm-3. Subjects consistently enter details of physical properties and LEV. As in scenario 1, inconsistency arises in decisions on pattern of use and aggregation of the dust.
Aggregates 94% N 6% Y Use pattern 85% DC&G 9% DM 6% LDT
LEV 100% Y
Comments
When the EASE screen for selecting the pattern of use appears, the first option of dry crushing and grinding is highlighted. This means that the definition of this use pattern is automatically displayed in the right hand side help box. The definition specifically mentions sanding by hand and sanding by machine as examples of dry crushing and grinding. Even so, 15% of subjects did not select this option. If these subjects did not read the definition before making their choice, this could explain their decision, inferences can be made from the names of the categories. Sanding does not immediately come to mind as a crushing and grinding operation, dry manipulation would seem more appropriate. Similarly, if the definitions have not been read, subjects might select low dust techniques because LEV is present, not realising that decisions on LEV will be dealt with at a later stage. When ‘select type of process used’ (i.e. pattern of use) comes up on the screen, then EASE should state here that LEV will be dealt with at a later stage. At present this is only mentioned in the definition of dry crushing and grinding. This might prevent subjects selecting low dust techniques as the pattern of use where LEV is present.
Better category names should be chosen so that less inferences are drawn from the name itself.
For example;
* Tasks where dust is generated at a fairly high velocity (e.g. grinding, fettling, tumbling, sanding, crushing, blowing down)
* Tasks where dust is generated at low to moderate velocity ( e.g. debagging, weighing, brushing, sieving)
* Tasks where dust is suppressed or contained (e.g. wetting, enclosure of process)
No explanation is offered as a reason why 2 subjects considered that wood dust aggregates.
7.3 Scenario 3 - Complex Precision Casting Ltd.
A small foundry uses sand as the moulding medium for casting metal articles. Silica dust will therefore be encountered at different processes in the foundry. Of partic ular concern are the 6 men working in the fettling shop. They are using hand-held tools to finish small to medium sized workpieces in a booth with LEV (picture provided). The question is, should this be backed up with the use of respirators? Using EASE, give an estimate of the levels of exposure encountered in this situation. (71% of subjects felt experienced in predicting exposures for this scenario.)
The subject's predictions fell into 3 of 10 possible exposure ranges for a non-fibrous dust. 91% of the subjects arrived at the same outcome. As before, the critical decision was the pattern of use and aggregation of the dust
115
Aggregates 3% Y 97% B Use pattern 91% DC&G 6% DM 3% LDT
LEV 100%
Comments
It is interesting to note that the subject selecting low dust techniques in this scenario, does so for every dust scenario where LEV is used. This would seem to give credence to the explanation given in the last scenario, the subject equates low dust techniques with the use of LEV, not realising LEV will be dealt with in the next stage.
The two subjects selecting dry manipulation were both unfamiliar with this workplace scenario (i.e. inexperienced), they may not have known that 'finishing work pieces with hand tools' was a grinding operation.
7.4 Scenario 4 - Oventuf Glass plc
Selenium metal powder is used as a de-colouriser in the production of glass for ovenware. Selenium removes the natural green colour of the glass. Dry selenium powder is weighed out in the laboratory on the open bench, and is then tipped into a mixer with other materials. In the laboratory, a technician carefully weighs out 22 kg of the powder into a plastic container, which is then covered and transported to the mixer. Quite often, a small amount of powder is spilt on the bench and is cleaned up using a brush. After each batch is weighted out, the bench is cleaned with a wet cloth. For the purposes of this exercise, we are interested in the inhalation exposure of the technician while weighing out the powder. The laboratory is about 10m square, and 2.5 m high. There is an extract fan on one wall, which is not always switched on, and there is no LEV on the bench. What is the likely concentration of selenium in the technician's breathing zone? (29% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 3 of 10 possible exposure ranges for a non-fibrous dust. The only inconsistencies arose from selection of pattern of use.
Use pattern 88% DM 9% LDT 3% DC&G
Comment
Of the 4 subjects who did not select dry manipulation for this scenario, 3 described themselves as inexperienced. There is no obvious reason for the inconsistent choice of use pattern for this scenario.
7.5 Scenario 5 - XYZ Electronic Materials Ltd
Thin layers of gallium arsenide are machined from large crystals using diamond-tipped disc cutters. The cutting operations take place in a small workshop containing 6 machines. Each machine is connected to a common cutting fluid system which provides a continuous drench of a water/oil mixture at the cutting tools. The machines are fitted with LEV consisting of 450 x 60 mm slots positioned close to the cutting tool and extracted via flexible ducting. Use the EASE model to estimate the possible range of exposure to dust at the machining process. (29% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 4 of 10 possible exposure ranges for a non-fibrous dust. 85% of subjects had an outcome of 0 to 1 mgm-3. Again, there were no inconsistencies in entering physical properties. Differences occurred in the choice of pattern of use, aggregation of dust and, despite the scenario mentioning LEV, selection of LEV present or absent.
116
Aggregates 18% Y 82% N Use pattern 91% LDT 6% DM 3% DC&G
LEV 94% Y 6% N
Comments
It is difficult to make any suggestions to improve consistency of use for this particular scenario, other than to reiterate what was said in scenario 1, i.e. improvements need to be made to the definitions of use patterns and aggregation. Also, several examples need to be included with each definition.
7.6 Scenario 6 - Sweeney Todd's Pie Emporium Ltd.
A meat pie bakery employs about 20 staff who work together in the main bakery area (30m x 40m x 5m high). The room has extract fans in the walls and all the handling and preparation of the pastry is done in the room. This includes dough mixing, kneading, cutting, and pie making. The finished pies then go straight to the oven for baking. As well as the bulk use of flour for pastry-making, flour is liberally spread on to dough and work surfaces to prevent sticking. Use the EASE model to assess the exposure of the bakery employees to flour dust. (53% of subjects felt experienced in predicting exposure for this scenario.)
Aggregates 29% Y 71% N Use pattern 6% DC&G 91% DM 3% LDT
LEV 100% N
Comments
Another quite difficult scenario. Dry flour is liberally used for a number of tasks and most subjects opted for dry manipulation. 29% of these subjects opted for dust aggregates, presumably because they felt that it would be incorporated into the sticky pastry. EASE needs make clear that aggregation as described in the model, is an inherent property of the dust and is not due to wetting of the dust.
The two subjects who selected dry crushing and grinding were inexperienced, so may not have been familiar with tasks undertaken in a bakery. It is interesting to note that even though a number of different tasks were described in the scenario (whic h may in reality give rise to different levels of exposure), none of the subjects thought to produce different EASE predictions for the different tasks. The EASE manual needs to give more practical help in highlighting the ways in which EASE can be used. An automatically displayed on-screen introduction summarising some of the practical issues of EASE use might be helpful (perhaps a window which can be switched off in the future when the user is more experienced).
7.7 Scenario 7 - Skimfast Boats Ltd
Quite a large factory manufactures a range of pleasure craft (speedboats and larger motor boats). The process under investigation is laminating, i.e. applying styrene resin and sheets of glass fibre to a mould using brushes and hand rollers. LEV was in use for some of the tasks sampled. However, due to the nature of the work it was not always possible to apply extraction near to the source of emission. In fact, there was little difference in the results of those that used LEV and those that didn't. During the work large doors, through which the boats were moved in and out, were often open. This was said to be common practice. Using EASE, what sort of exposure ranges would you predict for this scenario? The vapour pressure for styrene is 5 mmHg at 20° C. (58% of subjects felt experienced in predicting exposure for this scenario.)
117
The subject's predictions fell into 9 of 59 possible exposure ranges for a gas/vapour aerosol scenario. Inconsistencies arose in determining the pattern of use, the control methods and aerosol formation.
Aerosol 35% Y 65% N Use pattern 65% WDU 29% NDU 6% M
Controls 79% DHV 15% DH 3% S 3% LEV
Comment
For this scenario the choices are far from clear. Opinion between even experienced subjects was divided on whether using brushes and rollers to apply the styrene resin generated an aerosol.
The majority of subjects opted for a wide dispersive use pattern but, 29% opted for non-dispersive use. The EASE definitions for the two categories are somewhat vague, and it might have been anticipated that subjects with knowledge of the laminating process in the boat building industry would have selected wide dispersive use, however, experience was not a factor in the choice of use pattern. More precise definitions with examples may aid consistency here.
Most subjects noted the open doors in the factory and chose direct handling with dilution ventilation as the pattern of control. EASE is clear that the situation described in this scenario (i.e. large spaces provided with suitably located openings for ingress and egress of air) are included in the dilution ventilation category. One subject selected LEV as a pattern of control - it is mentioned in the scenario, although it is described as ineffective. In the EASE model LEV is assumed to be effective and the definition states that to qualify as being present it must be effective.
7.8 Scenario 8 - Gypsy Rose Furniture
A small factory providing a paint stripping service are concerned about the levels of dichloromethane their workers are exposed to. Items to be stripped (e.g. doors, dressers, tables and smaller wooden items) are lowered into a large tank of DCM, either by hand or hoist. Once in the tank as the paint begins to soften, the operator scrapes or brushes the paint off. The work takes place in an open workshop and the tank has lip extraction. The vapour pressure of DCM is 350 mmHg at room temperature. (46% of subjects felt experienced in predicting exposure for this scenario).
The subject's predictions fell into 5 of 59 possible exposure ranges for a gas/vapour/aerosol scenario. Inconsistencies occurred in the same 3 areas as scenario 7.
Aerosol 32% Y 68% N Use pattern 82% NDU 18% WDU
Controls 79% LEV 9% DHV 6% DH
Comment
If “aerosol formed” is selected, the EASE model automatically defaults to a high tendency to become airborne. In this scenario, the physical properties of DCM mean that the tendency to become airborne is high anyway, so the question of aerosol formation has no influence on the outcome. The activity which might trigger a subject to select aerosol created for this scenario
118
is brushing. It is interesting to note that a number of the subjects who selected “aerosol formed” for this scenario, did not select it for the previous scenario where both brushes and rollers were used.
19% of subjects selected a wide dispersive use pattern for this scenario - these subjects may be taking into account the hoisting activities undertaken. This scenario does not strictly conform to the wide dispersive use definition. It is possible that the subjects are making inferences from the category name. As stated in the comments section of the last scenario, the definitions need to be more precise.
Lip extraction is mentioned in this scenario and most subjects opted for LEV as the method of control. However, when the stripped items are hoisted out of the tank they would be beyond the influence of the LEV, perhaps the reason why 21% of subjects did not select this control method. This scenario is another example of where the work being carried out could have been broken down into two tasks and EASE used to model each, i.e. make one prediction for the part of the job where the LEV would reduce exposure and one prediction for the removal of items from the tank when exposure is likely to be higher. None of the model users did this. EASE users need more information on how the model may be applied.
7.9 Scenario 9 - Sky-high Engineering Services Ltd
The company undertake contract maintenance of hydraulic equipment and assemblies for commercial aircraft. The equipment is disassembled, cleaned, repaired and reassembled on laboratory benches in a large well-lit workroom. The workroom is well-ventilated by a large extract fan in the wall, but there is no LEV. Parts are cleaned by brushing them with solvent (neat Genklene), and drying them off with tissues. The solvent is kept in steel pots on the bench, the pots have inward-draining funnel-shaped tops, in which the brushes sit when not in use. As this is skilled work, some care is taken to prevent overuse of solvent, there is little or no evidence of pools of solvent on the benches, and used tissues replaced in covered waste bins. Degreasing takes up only about 10% of the total time spent on maintenance work. In order to estimate the overall exposure of technicians, use EASE to estimate their exposure to the solvent during degreasing activities. The vapour pressure of Genklene is 16.5 Kpa at 25°C.
The subject's predictions fell into 14 of 59 possible exposure ranges for a gas/vapour/aerosol scenario. The inconsistencies occurred in the same categories as previous scenarios.
Aerosol 28% Y 72% N Use pattern 70% NDU 27% WDU 3% M
Controls 67% DHV 24% S 6% DH 3% LEV
Comment
The results from this scenario provide definite evidence that a degree of operator error is occurring when EASE is being used. Two subjects entered the incorrect units for the vapour pressure of the solvent, and one subject said that LEV was in use when the scenario stated that no LEV was used. It seems likely that some of the inconsistencies that occurred in the other scenarios (some of the 'less readily explained' choices) may simply be down to errors during inputting of information, rather than differences of opinion on the part of the EASE users.
7.10 Scenario 10 - Grotgone Degreasers Ltd
Large metal castings are degreased by spraying with 1,1,1-trichloroethane in a large ventilated booth, prior to painting. This work is undertaken by 1 man for about 2 hours a
119
1
day. He wears an air-fed visor while inside the spray booth. The booth is ventilated by a downdraft arrangement and the mean downward air velocity is in the region of 0.1 to 0.2 ms-
. Return air grilles are at the sides of the booth, at floor level. The castings are quite complex, and solvent runs freely off them during spray application, falling to the floor and running off in gutters for recovery. Estimate the concentration of the solvent in the spraybooth atmosphere during degreasing activities, using EASE. The vapour pressure of 1,1,1-trichloroethane is 16.5 Kpa at 25°C. (47% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 9 of 59 possible exposure ranges for a gas/vapour/aerosol scenario.
Aerosol 94% Y 6% N Use pattern 82% NDU 12% WDU 6% M
Controls 73% LEV 18% DHV 6% S 3% DH
Comment
Even though this scenario states that the solvent is sprayed, two subjects (6%) selected “no aerosol formed”.
The definition of wide dispersive use gives painting as an example of this category, so it is understandable that some subjects selected this use pattern for scenario 10. But the definition states that such a use may deliver uncontrolled exposure to not only the operator but to others in the vicinity. In this scenario the operator was working in a downdraft booth, so does not fit the definition. Inclusion onto a matrix was an unusual choice and this scenario does not fit this category. Having said that, the definition of inclusion onto a matrix is not helpful. Inclusion onto a matrix only occurs as an option in gas/vapour/liquid scenarios, yet, the examples described in the definition all relate to dusts. Clearer definitions and examples are needed.
As mentioned in scenario 7, to qualify as being present LEV must be effective. Since the capture velocity in the booth was pretty low to effectively capture a spray, it would not have been surprising for the experienced subjects to have opted for dilution ventilation. 18% of the subjects did select this category but all bar one were inexperienced.
6% of subjects selected segregation as the method of control. The operator is segregated from the rest of the workforce but is not himself segregated from the source of emission. EASE needs to make clear that segregation refers to the separation of the operator from the source of emission.
7.11 Scenario 11 - Wypekleen Sponges Ltd
The company manufactures synthetic sponges from viscose material. The base material for sponge formation is produced in a large reaction vessel where batches of cellulose-based feedstock (e.g. waste cotton, flax or pulp) are mixed with carbon disulphide, As the plant is largely enclosed, exposures are generally low, but there are problems during charging of the reactor. When a new batch is to be loaded, the operator stands by a stainless steel chute located above the feed pipe to the reactor. A hatch in the feed pipe is opened, and the operator drops the contents of an overhead monorail tub onto the chute and pushes it into the hatch with a rod. The hatch is closed and 30 litres of carbon disulphide is metered into the vessel via a sealed pipe. The charging operation takes 3-5 minutes. The hatch does not make a gas-tight seal with the feed pipe, as indicated by smoke tube observation during the loading operations, and during charging the operator is exposed to carbon disulphide from the previous batch when the hatch is opened. There is no LEV at the hatch. Use EASE to assess
120
the exposure of the operator to carbon disulphide during the charging operation. The saturated vapour pressure of carbon disulphide is 297 mmHg at 20° C. (26% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 8 of 59 possible exposure ranges for a gas/vapour/aerosol scenario.
Aerosol 18% Y 82% N Use pattern 79% NDU 9% WDU 9% M 3% CS
Controls 68% S 15% DHV 3% FC 3% LEV
Comment
The comments made before apply equally here. This scenario clearly demonstrates the misunderstanding of the segregation category. It appears that segregation has been confused with use of the substance in an enclosed plant.
7.12 Scenario 12 - Reelykomfy Shoes Ltd
The company manufactures ladies shoes, producing about 8000m2 per week of synthetic material for the uppers. This material is rolled from a synthetic resin-based feedstock. Drums of Vithane resin, which contains a high proportion of dimethylformamide (DMF) are placed under a mixing machine, and polyurethane adhesive and colouring material are added. The drums are covered with a loose fitting lid during mixing, and left unattended until mixing is complete. The workshop has a good level of general ventilation, but no LEV is provided at the mixers. Use EASE to estimate the exposure of the operator to DMF while the resins is being prepared for mixing. The vapour pressure of DMF is 3 mmHg at 20°C. (21% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 16 of 59 possible exposure ranges for a gas/vapour/aerosol scenario.
Aerosol 35% Y 65% N Use pattern 24% M 3% CS 61% NDU 12% WDU
Controls 54% S 36% DHV 3% FC 3% LEV
Comment
Segregation has again been confused with plant enclosure in this scenario. EASE should be amended so that as well as giving a definition of each category, some examples of what isn't covered by the category are included.
7.13 Scenario 13 - Coloursupps Ltd
This is a large printing company, producing weekend colour supplements for national newspapers. The presses are housed in a large workshop about the size of an out-of-town DIY warehouse. Press operation is continuous for several days at a time, and the presses are shut down, usually at weekends, for cleaning, maintenance and preparation for the next run. 8 men are involved in the cleaning-down operation which takes several hours. The work consists of pouring toluene liberally onto the dirty ink rollers, using watering cans with sprinkler nozzles. The inking blades are removed and washed by hand in a trough full of toluene. The large rollers are removed from the machine and cleaned by hand using toluene
121
soaked swabs and scrapers. The printing shop is ventilated by wall-mounted fans, but there is no LEV at the presses themselves. Estimate the exposure of the cleaners to toluene using EASE. The vapour pressure of toluene is 21 mmHg at 20° C. (65% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 7 of 59 possible exposure ranges for a gas/vapour/aerosol scenario.
Aerosol 50% Y 50% N Use pattern 26% NDU 74% WDU
Controls 74% DHV 26% DH
Comment
The EASE definition does state that to qualify for “aerosol is formed”, large quantities of aerosol are produced, what it should also make clear is that the potential exists for the droplets to be inhaled. It is unlikely that the watering can used in this scenario would generate an aerosol in the breathing zone of the operators.
7.14 Scenario 14 - Highpower Engines Ltd
Engine blocks for large automotive diesel engines are painted in open-fronted water-backed spray booths. The face velocity at the booths is 1-2 ms-1. The blocks are placed on rotating turntables by a mobile hoist and sprayed with a paint containing 30% xylene. The sprayers stand at the front of the booths and rotate the turntable to reach all parts of the block. Each block takes about 10 minutes to paint. Estimate the exposure of the sprayers during the painting operation using EASE. The saturated vapour pressure of xylene is 9 mmHg at 20° C. (64% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 5 of 59 possible exposure ranges for a gas/vapour/aerosol scenario.
Aerosol 97% Y 3% N Use pattern 3% M 3% WDU 94% NDU
Controls 3% S 3% DH 94% LEV
Comment
For the gas/vapour/aerosol scenarios, this scenario produced the greatest consistency amongst the test subjects - 88% reached the same conclusion.
7.15 Scenario 15 - High-tech Renofibres Ltd
Refractory ceramic fibres are often used to lag furnaces and kilns. Unfortunately, because they become brittle with heat over time, the need replacing every 2 years or so. What sort of levels of refractory ceramic fibres in air might be predicted during the replacement process, assuming that whilst the area is separated from the main workroom, no engineering controls such as LEV are applied and the workers will be relying on RPE to prevent exposure? (44% of subjects felt experienced in predicting exposure for this scenario.)
The subject's predictions fell into 4 of 16 possible exposure ranges for a fibrous dust scenario.
122
TBA 12% Hi 88% Me Use pattern 24% DC&G 76% DM
Comments
All the subjects correctly recognised this scenario as a fibrous dust problem and all correctly entered that LEV was not in use. The critical decisions in a fibrous dust situation are in determining the tendency of fibres to become airborne and in determining the pattern of use.
The EASE model names the most common fibres likely to be encountered in the workplace and categorises them into 1 of 3 groups, high, medium and low. EASE states that ceramic fibres fall in the “medium” tendency to become airborne group. Even so, 12% of the test subjects selected a high tendency to become airborne, perhaps on their own perceptions of the behaviour of ceramic fibres rather than reading the EASE definitions.
24% of the subjects (mostly experienced) selected dry crushing and grinding as the pattern of use. The description of the scenario does not readily fit in with the definition of this use pattern in EASE. It may be that the test subjects have assumed that some activities not mentioned in the scenario take place, and that these fall into the dry crushing and grinding category.
8. Conclusions and Recommendations
The study has demonstrated that although there is a degree of consistency amongst EASE users, there is room for improvement. The study indicates that the problems are mostly arising because of misconceptions about the terminology used in EASE. The suggestions for amending the model to improve consistency are summarised in the list below;
1. On the opening screen of EASE, put up a warning that for each screen, the definitions of all the categories should all be read before a selection is made.
2. On the opening screen of EASE, warn the user to read through the EASE log file when finished, to check for inputting errors.
3. The screen after the opening screen should contain brief information on how EASE can be applied e.g. scenarios should if needed be broken down into discrete tasks/parts of a procedure and EASE predictions made for each task rather than make one prediction for the composite scenario as suggested in scenarios 6 and 8. The screen should also show what modifications can be made to the EASE outputs e.g. time-weighting, mixtures (albeit with a warning on the limitations). This screen might be in the form of a window which can be switched off once the user becomes familiar with EASE.
4. The following definitions should be clearer and more precise • all the pattern of use categories • aggregation • aerosol formation • all the methods of control categories
5. More and better examples should be given in the definitions of the above categories.
6. Consider the existing category names with a view to renaming them, so that inferences cannot be drawn from the names alone. See comments in Section 7.2.
123
7. Provide a larger area of screen for the definitions of categories, preferably so the whole definition can be seen at once. At present, sometimes only a paragraph of a larger explanation is shown, and users may not always realise that they need to scroll down to see the remainder of the text.
8. Consider putting together a short software training pack with worked examples.
9. As a follow up to this study undertake modifications and rerun the tests to see if improvements have been made.
J Tickner D Llewellyn L Dooley
KEY TO ABBREVIATIONS
Y YES N NO
Me MEDIUM Hi HIGH CS CLOSED SYSTEM M INCLUSION INTO/ONTO A MATRIX
NDU NON DISPERSIVE USE WDU WIDE DISPERSIVE USE DM DRY MANIPULATION
DC&G DRY CRUSHING AND GRINDING LDT LOW DUST TECHNIQUES FC FULL CONTAINED
LEV LOCAL EXHAUST VENTILATION S SEGREGATION
DH DIRECT HANDLING DHV DIRECT HANDLING WITH DILUTION VENTILATION
124
Appendix 3: Stakeholder interviews questions and prompt
1. Background information on stakeholder
Name:
Company name and address:
Job title:
Brief job description:
(Which of the following categories does the respondent fall into?• Practising industrial/occupational hygienists • Regulators/consultants working on chemical risk assessments (specify which) • Occupational Epidemiologists • Occupational toxicologists • Other (specify)…………………………….
2. Extent of users experience with EASE
What do you think EASE is used for?
How long have you been using EASE?
How often do you use EASE?
Did you experience any installation problems / software conflicts with EASE?
Did you find the computer interface clear and easy to use?
Did you find the explanatory text provided to help users categorise exposure scenarios clearenough?
Have you used EASE to predict dermal exposure or inhalation exposures (or both)?
What substances have you used EASE for?
Ask for examples of specific scenarios in which it was used.
How do you use the exposure ranges that are produced on each occasion EASE was used?Do you use the highest/lowest/average value? Were you under the impression that the model outputs were precise indicators of exposure? Did you want to be able to make an exact prediction of exposure using these ranges?
Did they require task specific exposure levels to help decide on appropriate controls, did they require ballpark estimates of average exposure over longer time periods or did they require worst-case estimates?
Do you feel that the estimates were accurate and precise for the situations you have explored? How concerned are you about the precision and accuracy of the predicted exposures?
Did EASE fulfil your requirements?If not, why not?
125
Have you attempted using EASE for any of the following substances and what were your experiences?
• Dusts containing more than one substance of interest. • Liquids with very low vapour pressures (e.g. isocyanate) • Mercury (or similar) • Oil mist • Mixtures of solvents • Welding fume
How easy/difficult is it to place real-life work situatio ns in terms of the EASE categories?
In general, do you have enough information about the workforce, tasks etc. to feed into EASE or do you have to make broad assumptions about the working practices?
What other applications might you use EASE for in the future?
3. Validation of EASE against measured exposures
Did you validate the results obtained from EASE against measured exposures?
If yes, why was this undertaken? What did the validation consist of? How did the measured exposures compare with predic tions? What was the outcome?
If no, why was no validation undertaken? Were you confident in the results obtained from EASE?
If using EASE again, would you wish to validate it by some means?
Are you aware of any work done by others to validate EASE predictions?
4. Strengths, limitations and suggestions for improvement
In general, what do you think the are the strengths and limitations of EASE?
How do you think it can be improved?
Prompts:
Do you think the ranges are too wide or too narrow?Would a point estimate or a distribution of exposures be more useful?Would it be better to structure EASE around more specific tasks or processes?
Given the specific purposes that you have used EASE for, what do you think were its limitations? Again, how do you think it can be improved?
If you were to specify a new version of EASE how accurate and precise do you think the exposure estimates should be?
Prompts:Accurate i.e. over a range of substances and scenarios the predicted average exposure to within plus/minus 3%, 10%, 30%, 100%, 300% of the true average?
126
Precise i.e. on any individual scenario the predicted average exposure within plus/minus 10%, 30%, 100%. 300% of the true average and the predicted upper percentile (say 95%ile or upper range bound) against the same categories?
5. Future Developments
Do you think that an improved version of EASE is needed for regulatory risk assessment?
Prompts: More accurate / precise estimates Better inhalation or dermal predictions Inclusion of ingestion and / or injection exposures Easier to use software
6. Info and comment on other exposure models
Have you used any other exposure models?
If yes, describe what was used? Why were these used? Did these alternative methods fulfil the desired purpose? Did they validate results from these models with actual exposure measurements? What was the outcome? What were the limitations of the alternative models?
Overall, what were their thoughts on these as possible alternatives to EASE?
If no, why have you not used any alternative exposure models?
End of interview:
Thank participant and ask if they have any questions.
Would you like to receive a copy of our report on this work? If yes, check name, address
127
Printed and published by the Health and Safety ExecutiveC30 1/98
Printed and published by the Health and Safety Executive C1.25 08/03
ISBN 0-7176-2714-4
RR 136
9 78071 7 6271 41£20.00