thesis proposal - information...
TRANSCRIPT
TECHNISCHE UNIVERSITEIT EINDHOVEN
Thesis proposal Process model transformations in health care
E.L.M. van den Broek BSc
7/4/2010
Deriving a way to translate actual (low‐level) process models from the health care field into High‐Level process models that are understandable to specialists in the Health Care field, in order to increase their usability.
1
1. Delineation of the thesis project
1.1. What precisely is being researched, and why?
1.1.1. State the goal of the thesis
Deriving a way to translate actual (low‐level) process models from the health care field into High‐Level process models that are understandable to specialists in the Health Care field, in order to increase their usability. These specialists are involved in reorganizations in the information systems field (for example: selecting a new hospital information system), but do not have a background in either modeling, information systems or computer technology. Models involved need to be abstracted, but must keep a certain level of detail to be of use.
1.1.2. State the benefit of the research
By aiming to increase the understandability of process models, the goal will be to increase the usability of process models. By being more understandable, the usage of process models in the HC field can increase, allowing for quality improvements of decisions in the HC field. It will provide software developers with a way of presenting their products in a detailed, structured and yet understandable way to HC professionals. Using a single language will also help HC professionals with comparing offers by different software providers, to see which product matches the current processes in the organization best.
1.1.3. State the known knowledge in this area
Research shows that usage of process models in HC is low. Literature suggests that both from a consumer perspective [1] and from a provider perspective, these models seem to be lacking. This does not necessarily mean they do not exist, but they are not provided to other parties. Meanwhile, research confirms the importance of process models, since it is stated that knowledge about the business processes is needed in order to decide which information systems an organization needs to evolve in a better organization [2]. Lacking knowledge about business processes leads to the acquisition of suboptimal products. Even if these products fit current processes and improve efficiency, the business process does not evolve with it [2]. Both from the consumer perspective and from the provider perspective, literature describes why the presence and usage of process models is desirable.
When process models are available (mostly the case for the provider side), these models tend to use different modeling languages. This makes comparison between models difficult and also the integration with the consumers’ own models [3]. While reducing the number of modeling languages would be one solution, a more practical solution lies in model transformations. Research on this has mainly focused on transforming a model from a specified language into another specified language [3][4][5][6][7]. In many cases, the ‘transform to’ language is considered to be an industry standard [6][7]. A potential problem with (automatic) transformations is the subtle violation of a language’s standard that input models may pose [5].
Research by Mendling et al. [8] was conducted on understandability of models.
2
A company in this field is iCON health care in the Netherlands. The company is active in advising hospitals (and other companies) about different software solutions, where simulation is used for demonstrating how such a solution is being (or could be) used at a certain hospital. These simulations use high‐level models that are created by iCON, of which iCON claims that they are understandable for HC professionals.
1.2. What are the conclusions and what is the value of the conclusions?
This research should lead to conclusions that provide clarity about how understandable, high‐level process models should look like. More important will be, how such models can be constructed by transforming existing low(er) level models, using existing tools. This concerns both the abstraction process, as well as the visual/graphical language to use to present these process models. Based on these concerns, strengths and weaknesses of existing tools will be discussed. This in result will lead to suggestions for improvements for one or multiple existing tools. The focus will be on their usage in the HC field; results should be generalized across the health care field for information systems.
1.3. What should be done as a result of the conclusion?
Based on the results, it should be possible to build a tool to automatically translate low level process models into the high level process models that will be defined in the research. The actual programming of such a tool lies outside the scope of the research, but the research should provide a programmer with all necessary information concerning the transformation between models, to develop such a tool.
3
2. Research design
2.1. Methodology
2.1.1. Are you reusing some existing solution or developing something from scratch?
The focus will be on the evaluation of existing tools for model transformations. Using available low‐levels from within the field, these existing model transformation tools will be evaluated. A framework should be used for evaluating the tools. Rad et al. [3] have already developed a framework for evaluating modeling languages in the HC field. The framework can help in deciding what a desirable language is for higher‐level models. Since it focuses on HC‐specific issues, this framework will also be a good starting point for developing a framework for evaluating transformation tools. If possible, the research will include the actual adjustment/extension of an evaluated tool.
The models used will originate from the HC field. Either hand‐crafted models can be used, or models might be generated from logs, by using a tool such as ProM. The reason for using either method, will be discussed below.
Hand‐crafted models
These models are constructed mainly by software developers and are constructed for internal usage (for workflow enactment, for process simulation, or for demonstration purposes). They can be readily used, if developers are willing to supply these models. Workflow enactment models could be made available by IT organizations such as ChipSoft or Siemens (in the context of the EZIS and SOARIAN suites respectively.) Models can also be developed by third parties such as iCON, who model the processes at a hospital (for simulation and demonstration purposes).
The quality of and the language used for these models is likely to vary among developers and organizations [3]. This makes it hard to develop a universal solution that is applicable to all low‐level models available from software developers and organizations. The main advantage of using pre‐developed low‐level models by software developers and organizations is that these models focus solely on either the software product offered or on the hospital process. There is no ‘interference’ of the other processes (i.e. the influence of an information system on the process of a hospital), thereby providing a much better basis for comparing the own hospital process with the process of the offered product(s).
Generating models from logs (ProM tool)
ProM is a very versatile tool for applying process mining on log files of a wide variety of types. The result is typically a low‐level model of the process. The acceptance of different types of log files makes ProM usable in many situations, and thus provides similar source models for different hospitals. The main disadvantage of basing the model on log files is the required implementation of an information system, in order to provide logs that ProM can handle. Though other (non‐electronic) log files can be used for manual construction of process models, this procedure is too time consuming in order to be considered for this research. Requiring an information system will naturally exclude scenario’s in which hospitals are about to purchase an information system. From a systems perspective, the acquiring of an information
4
system has a very large impact on the hospital, thus eliminating this scenario from the research would be a limitation.
The problem with pre‐developed models is a quality aspect, while the problem with models generated from log files is based on a time aspect, since the duration of the thesis project is predefined and therefore limited. Valuing the quality aspect, the usage of models that are generated using ProM is more desired in this situation. As a result of this, the research will focus on post‐HIS situations in hospitals, however it should be possible to replicate the research for pre‐HIS situations by using process models from pre‐HIS situations. The option to use pre‐developed models will be kept as a “fallback”, in the case that process mining fails on log traces. Since these models are already available, no delay would be caused by switching to these hand‐crafted models.
Figure 1 ‐ schematic overview of the goal
Figure 1 shows more clearly how the generated models will be used, and how iCON can contribute. The upper process shows the way of working at iCON, as told by the employees: at a health care organization, there will be a certain process. The professionals at iCON try to understand this process by interviewing those involved in the process. The result is an abstract of the actual (highly detailed) process, which is very useful for supporting decision making in software implementations. The lower process shows process mining. Log traces from the actual process are used to ‘mine’ the process. The result is a highly detailed model, that corresponds very accurately to the actual process. However, it is far to detailed to be used by HC professionals. Thus the aim of the research will be to transform these highly detailed process models, to more abstract process models that are similar to those offered by iCON.
2.1.2. How will you get your ideas for your solution development?
Using the tools on the available low‐level models will lead to problems, since research has shown that development on these tools is not finished. Due to the differences between tools, it is assumed that problems caused by one tool, may not be present in other tools. The development of the solution will focus on these differences. While comparing advantages and disadvantages of different tools, the goal is to remove disadvantages from tools by incorporating advantages from other tools. It is also expected
5
that iCON can contribute to the solution development, since this company has experience with the development of understandable models, aimed at the HC field.
2.1.3. What criteria (test, etc.) shall be used to test the validity of the answer?
In order to assess the validity of the answer, the solution should be used on actual low‐level models. The transformed models should then be reviewed by HC professionals for understandability. Since iCON has experience with developing models for the health care industry, they should be able to perform this review. However, in order not to exclude actual HC professionals, the understandability of the high‐level models should also be verified by HC professionals. iCON is interested in what HC professionals think of the models that they develop themselves, thus reviewing the models developed for this research should be done by both HC professionals and iCON. This way validity is increased and iCON benefits from feedback from the HC field.
2.1.4. Why was the chosen method selected?
The method described above was selected for a number of reasons:
‐There has been a vast amount of research on model transformations so far. This research has focused on multiple tools and multiple languages, which has resulted in different functionality. By comparing multiple tools and multiple languages, indirectly this research is used. Building a new tool from scratch would ignore all this previous research. Part of the research on tools has been goal‐driven, where tools have been tested to perform a specific task (e.i. refactoring, translation). With the fourth iteration of the International workshop on Graph‐Based tools (GraBaTs) in 2008, this workshop is converted to a contest (now known as Transformation Tool Contest) where different modeling tools are put to the test, performing different kind of tasks [9]. Results of this contest may aid in the selection of tools to review.
‐Since the HC field proves to be different from mainstream business in many aspects, it is important to make sure that the solution is applicable to the HC field. The best way to achieve this, is by basing the research on models that are found in the HC field. Meanwhile, since iCON puts effort in this research, it is important that the research has also practical relevance. By using their models for input and by having them review the developed models, this relevance can be achieved.
2.1.5. Can you identify some intermediate steps in your solution development?
The research can largely be broken into two parts. The first part will focus on the selection of a suitable modeling language for presenting higher‐level models to HC professionals. The second part will focus on the evaluation of transformation tools and the subsequent improvement of (a) tool(s). Some intermediate steps that can be identified, as shown in Figure 2. A more detailed overview can be found further below.
6
Part IV: Transformation
‐Derive transformation rules‐Test/review rules
Part III: Evaluation of transformation tools‐investigate different transformation tools
‐development of evaluation framework
‐evaluation of transformation tools (using actual low‐level models from the HC field)
‐improvement of transformation tool(s) [optional]
Part II: Modeling languageSelect model language to use by:
‐Investigating different modeling languages (possibly with suggestions by iCON)
‐Use of framework by Rad et al. (6)
Part I: Models
generate low‐level models from log traces
(using ProM)
Investigate desired format of high‐level models
(based on iCON models)
Figure 2 ‐ Intermediate steps
2.1.6. To what extent will the findings be capable of generalization?
Findings should be capable of being generalized across the entire HC field of process management. Within the HC field, process management is often better known as ``care path documentation’’ or ``procedure, policy and guideline development’’. Within the HC field, the results may be specific to large organizations or clusters (hospitals, rehabilitation centra, collaborations between general practitioners and insurers, …) It may be possible that the results are generalizable towards any process‐oriented large organization or cluster (also from other industry segments, such as the automotive industry, the logistics industry, …) However, this would be an unexpected positive development, but this is not part of the scope of this research.
2.1.7. How great can the reliability of the results be expected to be? And the validity?
Due to the reliance on existing tools (which are constructed, based on earlier research), it is expected that the reliability of the results will be high. Applying the results to field‐models, further increases the validity. The use of process mining will also have a positive effect on the validity, since the risk of using models that do not fully correspond to the real situation is much lower. Attributing to this is the fact that there is no reliance on developers concerning the release of their models and the quality and
7
8
language of these models. Comparing with reliability, it is expected that validity is much harder to achieve. Using multiple models from as many different aspects within the HC field helps in improving validity.
2.1.8. What is the main field of research where you plan on searching for relevant
sources?
Most research on this topic has been carried out in the information systems and process management field and the cross‐over field between HC and information systems. This last field is actually more a sub‐field of the information systems field. The HC field will provide research focusing on the way of doing business in HC.
Below is a short list of publication outlets that have been used in preparation to the thesis. It is expected that these outlets will continue to be relevant in the thesis. This list is non‐exhaustive.
International Journal of Medical Informatics
Journal of Medical Systems
Journal of Theoretical and Applied Electronic Commerce Research
Workshops and conferences, such as the MHPW2010 workshop, contributed valuable knowledge about information technology in health care, and will continue to do so during this research.
Since the HC field of information systems is frequently visited in publications for information technology, it is expected that a lot of information will be supplied by different publication outlets. Of course, the literature study conducted earlier will be an important source of both information and references for the thesis.
2.1.9. What are the main publication outlets that are relevant for your thesis?
Some of the consulted publications should also be interested in the current research project. The International Journal of Medical Informatics and the Journal of Medical Systems seem to share their scope with the topic of the research being conducted for this thesis project. Because this research aims at increasing usability of process models to HC professionals, outlets that are read by these professionals are also of interest. Quality Management in Healthcare and Health Business are examples of non‐academic publications that reach the target audience.
9
3. Project plan & time schedule Figure 3 (previous page) shows a detailed planning for this thesis project. Below, this planning will be described in more detail: both risks and input from iCON will be discussed.
Part I: Models
Investigate process mining
A mainly theoretical research on the usage of process mining. This research is an extension to the research that has been conducted for the literature study. The focus will be on which parameters and plug‐ins to use when mining. Included in this, is a short research on the differences between heuristics mining and fuzzy mining.
Research on process mining is strongly supported by the Information Systems department at the faculty of Industrial Engineering & Innovation Sciences (IE&IS) at the TU/e. A lot of knowledge and literature on process mining has originated at this department, thus it should be no problem to acquire required information.
Risk impact: non‐significant
Generate low‐level models from log traces
Using process mining, log traces are translated to process models. Knowledge acquired in the previous phase is used for deciding on configuration for mining. If available, logtraces provided by iCON will be used. Because of the direct relation between these logtraces and models developed by iCON, this is a very valuable process. In case iCON is not able to supply logtraces before the deadline, it should be evaluated if it would still be valuable to do this process at a later point in time.
Availability of log traces may prove to be difficult, but some log traces are already available. If no other traces are available, this won’t cause delays but will influence the completeness of the research.
Risk: low
Investigate process models from different sources
There is an abundance of different types of process models, using different languages and different formats. The models will be analyzed on their strengths and weaknesses. The goal is to find desirable characteristics that will make up the process model format specified later. This process is mainly explorative.
10
Risk impact: Moderate
Due to the differences between the languages, it is hard to compare them. Some of the languages used are either new or still in development. Development on these languages during research may influence the outcome of observations done earlier.
Risk aversion: upon starting the process, the current format of models and the current syntax and semantics of languages will be used. Changes in models, syntax and/or semantics will not be considered unless these changes will have a large impact and these changes can be applied within time constraints.
Process model format [MILESTONE]
Finishing part I of the research should lead to a desired format for process models, that will be used in later parts as the ‘transform to’ format.
Basing on the results of the model investigation, it should be possible to develop a preferred format. The experts at iCON will be consulted for decision support.
Part II: Modeling language
Part II and Part III take place in a loop: modeling languages are reviewed one at a time and subsequently matching transformation tools are reviewed. If a modeling language proves to be unsuitable after both the language itself and the available transformation tools have been reviewed, a new modeling language is selected and both processes take place again. This approach was selected because it may be unnecessary to evaluate all pre‐selected modeling languages: once a suitable modeling language has been found, it is possible to proceed.
Investigate different modeling languages
This process is a continuation of the “investigation”‐phase conducted in part I, but more in‐depht. The aim here is to investigate to which extent a language is able to conform to the desired format, as constructed at the Process model format milestone.
This process takes place with one model at a time. Duration of the process will depend on the number of iterations needed before a suitable language is found. Therefore, the risk associated with this process is mainly time‐oriented. When more iterations of this process are needed, this will result in a longer total process.
Risk impact: high
Risk aversion: see ‘investigate different transformation tools’
11
Construct modeling language evaluation guidelines
Based on the process model format constructed, guidelines will be proposed to aid in the evaluation of a modeling language. A starting point for constructing these guidelines will be the framework developed by Rad et al. [3]. Further input for guidelines can be provided by iCON. Because the number of iterations for evaluating a modeling language is unknown, it is considered to be unnecessary to develop a full framework for evaluating languages.
Due to the mainly theoretical nature of this process, the associated risk is expected to be non‐significant.
Risk impact: non‐significant
Apply guidelines
As a part of the evaluation of the modeling language, the constructed guidelines are applied to language under research. There is no risk associated with this process.
Investigate different transformation tools
Similar to investigating different modeling tools, also different transformation tools will be investigated. This investigation will be related to the modeling language that has been investigated, as transformation tools are typically able to only transform from one input language to one output language. This investigation should lead to a taxonomy that clarifies the differences and similarities between tools.
As with the modeling language investigation process, this process might be executed multiple times: for every modeling language that is research, also transformation tools will be investigated. Added difficulty originates from the sometimes poor documentation and construction of transformation tools, as encountered in preparation for this research. Development on many tools has not finished, resulting in the need to use Work‐in‐progress builds or beta‐versions of tools.
Risk impact: high
Risk aversion: Both the investigation of modeling languages and transformation tools can potentially delay the research, since multiple iterations of both processes are possible. In order to prevent this from happening, a modeling language and an accompanying transformation tool should be selected before the end of may. In case the modeling languages and transformation tools investigated up till that point all prove to be unable to fully achieve the goal stated for this research, the best language and tool should be selected. Shortcoming of both the language and the tool must be documented, as should the impact of these imperfections on the final result.
The incompleteness of both documentation and development on transformation tools is unavoidable. Fortunately, most transformation tools are either developed at the TU/e or are developed with support
12
from researchers at TU/e. Contacting these people may decrease a delay caused by incompleteness of documentation and/or software.
Construct transformation evaluation guidelines
Based on the process model format constructed and the modeling language selected, guidelines will be proposed to aid in the evaluation of transformation tools. The taxonomy develop in the previous process, can be used as a starting point here, since it provides different characteristics of transformation tools. The constructing of guidelines focuses on deciding which behavior and which characteristics are desired.
Due to the mainly theoretical nature of this process, the associated risk is expected to be non‐significant.
Risk impact: non‐significant
Apply guidelines
As a part of the evaluation of the modeling language, the constructed guidelines are applied to language under research. There is no risk associated with this process.
Test understandability of format
In order to make sure that the research achieves its goal, it is important to test the understandability of the format proposed within the HC field. HC professionals will be asked about their opinion on the proposed format, using the modeling language selected in part II and the transformation tool in part III. It is assumed that iCON can arrange contact with some HC professionals. Also, iCON should review the understandability of the proposed language/format for process models. It is best to meet with iCON before the HC professionals, so the feedback from iCON can already be processed, thus making it more likely that the HC professionals will find the models understandable. Based on the feedback received, it may be necessary to make changes to the previous sections.
In the worst case scenario, the selected modeling language and transformation tool(s) need to be discarded, thereby forcing another entry in iteration loop that consists of part II and part III. Afterwards, another test for understandability should be conducted. This would result in a serious delay.
Risk impact: very high
Risk aversion: since the impact of this risk cannot be changed, it is better to minimize the likelihood of this risk. By including people from the HC field in an earlier phase, it will be much more clear what is expected from the HC field.
13
Part IV: Transformation
Derive transformation rules & Test/review rules
Having chosen a modeling language and a transformation tool, rules need to be defined to achieve the desired format. These rules need to be well documented, as they should provide input to a programmer who can implement these rules into (an) existing transformation tool(s). Another import aspect is the testing of these rules, for which these rules should be applied to multiple models. The best test would be to apply these transformation rules on logtraces from iCON customers. This allows to compare the result of automatic transformation with the result iCON achieves by their approach.
At the moment, it is impossible to state how many rules are required for successfully transforming models to the desired format. As a result of this, it is hard to plan the duration of this activity. The same goes for the test/review process: the duration of this process is unknown in advance.
Risk impact: moderate / high
Corrective actions: since the risks described above are unavoidable, it is important to have a strategy to cope with these risks. The worst case scenario would be that there are no transformation rules to be found. In that case it should clearly be documented why this is the case. Proposed strategies that did not result in transformation rules should be discussed and it should be analyzed why this did not work.
Quality of rules is more important than quantity of rules. In other words: it is better to have a few transformations rules that are unable to transform an entire model but work on different models, than having a lot of rules that can only transform one specific model correctly and fully.
3.1. Time schedule Due to the involvement of iCON, not all deadlines are equally strict. Because of dependence on them, certain activities might be postponed for a week. When this is the case, it is important that resources are shifted to the next activity, in order to not delay the entire project. iCON has seen and agreed upon this planning.
deadline Process Involvement iCON March 30th 2010 Investigate PM
Generate low‐level models Logtraces of iCON customers Investigate process models iCON process models
April 2nd 2010 Process model format Discuss findings / initiate “investigate modeling language”
April 16th 2010 [hand in ‐ milestone] process model + pre‐investigation of models
Arrange meeting(s) with client(s) for pre‐face of ‘understandability test’
14
15
April 23rd 2010 Deadline for considering new modeling languages
April 30th 2010 Investigate modeling language Investigate transformation tool
May 14th 2010 Test understandability Test understandability: HC professionals & iCON
[hand in – milestone] Modeling language [hand in – milestone] Transformation tool
June 25th 2010 Derive transformation rules Test/Review rules [hand in – milestone] Transformation rules
Discuss results & progress
July 30th 2010 Delays Write report [hand in] Final report & presentation Final presentation
Table 1 ‐ time planning, deliverables in bold, meetings in italic
[1] ChipSoft, ChipSoft productportfolio, 2009. [2] G. Vassilacopoulus en E. Paraskevopoulou, “A Process Model Basis for Evolving Hospital Information
Systems,” Journal of Medical Systems, vol. 21, 1997, pp. 141‐153. [3] A.A. Rad, M. Benyoucef, en C.E. Kuziemsky, “An Evolution Framework for Business Process Modeling
Languages in Healthcare,” Journal of Theoretical and Applied Electronic Commerce Research, vol. 4, 2009, pp. 1‐19.
[4] F. Gottschalk, W. van der Aalst, M. Jansen‐Vullers, en H. Verbeek, “Protos2CPN: Using colored Petri nets for configuring and testing business processes,” International Journal on Software Tools for Technology Transfer, vol. 10, 2008, pp. 95‐109.
[5] O. Muliawan, P. van Gorp, en D. Janssens, “Executing a Standard Compliant Transformation Model on a Non‐standard Platform,” Lillehammer: 2008.
[6] P. Jiang, Q. Mair, en J. Newman, “Using UML to Design Distributed Collaborative Workflows: from UML to XPDL,” WETICE '03: Proceedings of the Twelfth International Workshop on Enabling Technologies, 2003.
[7] H. Zha, Y. Yang, J. Wang, en L. Wen, “Transforming XPDL to Petri Nets,” Business Process Mangement Workshops, Heidelberg: Springer Berlin / Heidelberg, 2008, pp. 197‐207.
[8] J. Mendling, H. Reijers, en J. Cardoso, “What Makes Process Models Understandable?,” Business Process Management, Springer Berlin / Heidelberg, 2007, pp. 48‐63.
[9] P. van Gorp, “TTC 2010: Transformation Tool Contest Website,” http://is.ieis.tue.nl/staff/pvgorp/events/TTC2010/, 2010.