development of a compositional terminology model for nursing orders

6
International Journal of Medical Informatics (2004) 73, 625—630 Development of a compositional terminology model for nursing orders Susan Matney a, * , Catherine Dent b , Roberto A. Rocha c a Intermountain Health Care and University of Phoenix, Salt Lake City, UT, USA b University of Utah, Salt Lake City, UT, USA c Intermountain Health Care and University of Utah, Salt Lake City, UT, USA KEYWORDS Controlled medical terminologies; Nursing terminologies; Provider order entry (POE); Nursing orders; Healthcare standards Summary Aim: Develop a compositional terminology model for nursing orders that would conform to the existing standard health level seven (HL7) messaging stan- dard for clinical orders. Develop and evaluate the set of attributes needed for a pre-coordinated concept for a single nursing order, using a replicable three-step mod- eling process. Results: A terminology model for nursing orders was developed using empirical data. The model was validated against nursing research and standards lit- erature, and evaluated using 609 nursing orders that were successfully mapped to the structure. The representative services came from 20 Intermountain Health Care (IHC) hospitals, demonstrating the generalizability of the model and its attributes across many care settings. © 2004 Elsevier Ireland Ltd. All rights reserved. 1. Background Intermountain Health Care (IHC) has a history of processing and managing large subsets of coded clinical data captured in electronic form. These data are generated and managed in a collection of disparate computer systems. Each one of these systems is interfaced to a centralized ‘‘Clinical Data Repository’’ (CDR), where the clinical data is systematically encoded and stored [1]. The CDR was designed to provide a truly longitudinal view of the clinical data collected from each patient. Information is collected from multiple inpatient and outpatient sites, providing a complete view of the course of care for each patient over time. The CDR is currently the primary electronic data source for outpatient records. * Corresponding author. Tel.: +1 801 442 4488; fax: +1 801 442 6996. E-mail address: [email protected] (S. Matney). The collection and integration of clinical data from multiple disparate systems requires the exis- tence of a terminology server [2]. The terminology server is a database capable of storing the clinical concepts used by these systems, along with their associated terms and relationships. In the case of the CDR, this essential component is called the ‘‘Health Data Dictionary’’ (HDD) [1]. The HDD is a terminology server that is continuously being up- dated and developed by both 3M and its customers, including IHC. The HDD is not a process or an ap- plication that manipulates data. Rather, the HDD serves as a translation engine for the various inter- faces and application programs that retrieve and store data in the CDR. IHC is currently in the process of developing a new provider order entry (POE) system. The clinical concepts used by POE, including a wide variety of nursing orders, will be defined within the HDD and ultimately stored in the CDR. As a general rule, or- ders will be sent to the CDR using standard health level seven (HL7) messages [3]. HL7 defines a ser- 1386-5056/$ — see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ijmedinf.2004.04.006

Upload: susan-matney

Post on 05-Sep-2016

219 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Development of a compositional terminology model for nursing orders

International Journal of Medical Informatics (2004) 73, 625—630

Development of a compositional terminologymodel for nursing orders

Susan Matneya,*, Catherine Dentb, Roberto A. Rochac

a Intermountain Health Care and University of Phoenix, Salt Lake City, UT, USAb University of Utah, Salt Lake City, UT, USAc Intermountain Health Care and University of Utah, Salt Lake City, UT, USA

KEYWORDSControlled medicalterminologies;Nursing terminologies;Provider order entry(POE);Nursing orders;Healthcare standards

Summary Aim: Develop a compositional terminology model for nursing orders thatwould conform to the existing standard health level seven (HL7) messaging stan-dard for clinical orders. Develop and evaluate the set of attributes needed for apre-coordinated concept for a single nursing order, using a replicable three-step mod-eling process. Results: A terminology model for nursing orders was developed usingempirical data. The model was validated against nursing research and standards lit-erature, and evaluated using 609 nursing orders that were successfully mapped to thestructure. The representative services came from 20 Intermountain Health Care (IHC)hospitals, demonstrating the generalizability of the model and its attributes acrossmany care settings.© 2004 Elsevier Ireland Ltd. All rights reserved.

1. Background

Intermountain Health Care (IHC) has a history ofprocessing and managing large subsets of codedclinical data captured in electronic form. Thesedata are generated and managed in a collectionof disparate computer systems. Each one of thesesystems is interfaced to a centralized ‘‘ClinicalData Repository’’ (CDR), where the clinical datais systematically encoded and stored [1]. The CDRwas designed to provide a truly longitudinal viewof the clinical data collected from each patient.Information is collected from multiple inpatientand outpatient sites, providing a complete view ofthe course of care for each patient over time. TheCDR is currently the primary electronic data sourcefor outpatient records.

* Corresponding author. Tel.: +1 801 442 4488;fax: +1 801 442 6996.

E-mail address: [email protected] (S. Matney).

The collection and integration of clinical datafrom multiple disparate systems requires the exis-tence of a terminology server [2]. The terminologyserver is a database capable of storing the clinicalconcepts used by these systems, along with theirassociated terms and relationships. In the case ofthe CDR, this essential component is called the‘‘Health Data Dictionary’’ (HDD) [1]. The HDD is aterminology server that is continuously being up-dated and developed by both 3M and its customers,including IHC. The HDD is not a process or an ap-plication that manipulates data. Rather, the HDDserves as a translation engine for the various inter-faces and application programs that retrieve andstore data in the CDR.IHC is currently in the process of developing a

new provider order entry (POE) system. The clinicalconcepts used by POE, including a wide variety ofnursing orders, will be defined within the HDD andultimately stored in the CDR. As a general rule, or-ders will be sent to the CDR using standard healthlevel seven (HL7) messages [3]. HL7 defines a ser-

1386-5056/$ — see front matter © 2004 Elsevier Ireland Ltd. All rights reserved.doi:10.1016/j.ijmedinf.2004.04.006

Page 2: Development of a compositional terminology model for nursing orders

626 S. Matney et al.

vice as ‘‘any activity that must be scheduled priorto its performance,’’ and an order as ‘‘a requestfor a service.’’ Nursing orders represent a type oforder used in POE systems. There is currently nocontent for nursing orders in IHC’s HDD, but therewere nursing orders data available in a legacy sys-tem, called HELP [4], and also hundreds of paperorder sets, that included nursing orders, within the20 IHC hospitals.

2. Objectives

The purpose of this project was to develop and val-idate a detailed compositional terminology modelthat could be used to generate pre-coordinatedconcepts for nursing orders. The international stan-dards organization defines a concept as ‘‘unit ofknowledge created by a unique combination ofcharacteristics’’ and pre-coordination as ‘‘a com-positional concept representation within a termi-nology system, mapped to a single identifier’’ [5].The compositional model was used to assist withthe categorization and organization of the orderconcepts, allowing the detection of patterns andtrends, and the identification of synonyms. Theresearch questions that guided this study were: (1)which attributes should be included in completenursing order concepts? (2) how do the attributes ofa nursing order concept compare to those definedfor nursing actions?There were at least four reasons why we chose

to create pre-coordinated concepts for nursing or-ders based on a detailed compositionalmodel. First,there needed to be sufficient granularity in the or-der concepts to enable appropriate categorizationand organization. No public formal terminology sys-tem existed with the appropriate degree of granu-larity.Second, there is a specific health level seven

(HL7) message type for patient services, the‘‘Observation Request Message’’ (ORM). The‘‘Observation Request Segment’’ (OBR) is the seg-ment in an ORM that contains the item of service.HL7 specifies the type of data that can be usedin each field of the OBR segment. Field OBR-4,the ‘‘Universal Service Identifier,’’ identifies theindividual order [2,3,6]. Since only one field isused for the orderable service identifier, a singlepre-coordinated concept is the best option. Wechose to create an aggregate concept becausewe concluded that more than one attribute wasneeded to fully define a nursing order Fig. 1.Third, well-structured models, such as the Euro-

pean Committee for Standardization (CEN) modelfor concepts to support nursing [7] and the Interna-

Fig. 1 The third field of an observation request (OBR)segment is the filler order number, a permanent identifierfor an order and its associated observations. The fourthfield is the universal service identifier and contains theidentifier code for the requested order. The fifth field ispriority.

tional Classification for Nursing Practice (ICNP) [8],already exist for nursing actions or interventions,but they are not specifically tailored for nursing or-ders. A nursing intervention is the result of an order.Fourth, while no starter set of nursing order at-

tributes existed, sensible starting lists of nursing or-ders were available at IHC in both electronic and pa-per form. The availability of starter lists made thedevelopment and validation of the compositionalmodel possible.

3. Methods

This research project was performed over a3-month period. The group responsible for the de-velopment of the model included medical vocab-ulary engineers, in-house standards writers, andnursing analysts working on the new POE system.The developers were also responsible for the vali-dation of the model. At that time, all participantswere employed by IHC. An initial analysis was con-ducted to explore possible methods of represent-ing nursing orders. A multi-axial representationalapproach, similar to SNOMED [9] and Logical Obser-vations Identifiers Names and Codes (LOINC) [10],was chosen.The data used for model development were

electronic nursing orders collected from a legacysystem as well as paper ‘‘standing orders’’ usedwithin the 20 hospitals of IHC. During the initialstages, a ‘‘top—down’’ approach was used fordefining the order concept components, resultingon a few high-level attributes. Next, the high-levelattributes were partitioned into more detailedones, until sufficient granularity was obtained,enabling the unambiguous representation of thenursing orders [11].Step one was to identify the required attributes

of the compositional model. The common attributeswere identified by decomposition of a subset of indi-vidual nursing orders. After the common attributeswere identified, additional nursing orders were usedto test and validate the attributes.

Page 3: Development of a compositional terminology model for nursing orders

Development of a compositional terminology model for nursing orders 627

The second step was to determine if any existingterminology systems, described in the research orstandards literature, had already defined similar at-tributes. The literature review was focused on nurs-ing terminology systems that were used for eithernursing services or nursing actions/interventions.Three core nursing terminology systems matched

these review criteria. The first one was the‘‘Patient Care Data Set’’ (PCDS) developed byOzbolt [12]. PCDS has been recognized by the Amer-ican Nurses Association (ANA) as one of the vocab-ularies to be used by nurses. PCDS includes nursingproblems, goals, and orders, and it is the only ANAapproved terminology system that includes nursingorders.The second terminology system was the

‘‘International Classification of Nursing Practice’’(ICNP), created by the International Council ofNurses. ICNP is another ANA approved terminologysystem and it has been established as a commonlanguage for describing nursing practice. ICNP in-cludes a nursing action classification that is dividedinto eight axes [8].The third system was the terminology model for

nursing actions created by the European Commit-tee for Standardization (CEN), Technical Committee251 (TC251) [7,13].The last step was to test the compositional termi-

nology model using empirical data that are repre-sentative of the orders that would be needed by thenew POE system. The validation process was doneby taking all the gathered orders, from the HELPsystem and paper, and representing them using thenew model. A spreadsheet containing the compo-sitional model was distributed to members of theteam. There were a total of four people decompos-ing nursing orders using the model. Synonym listswere created, helping with the identification of du-plicates. We also concluded that a common list of‘‘Action Types’’ was needed, because there weremany synonyms identified for nursing actions. Forexample: ‘‘promote’’ was considered synonymousto ‘‘assist,’’ ‘‘facilitate,’’ and ‘‘progress.’’ Actionsfrom Grobe’s ‘‘Nursing Intervention Lexicon andTaxonomy’’ (NILT) were used to help develop thelist of actions [14]. Disagreements on how to modela given order were handled by discussions leading toconsensus. The participants ultimately determinedthe adequacy of the model.

4. Results

A list of attributes was identified for the compo-sitional terminology model for nursing orders (seeTable 1). At first, the compositional model had all

the attributes of a fully specified order, includingfrequency, priority, route, etc. However, the intentwas to use the resulting pre-coordinated order con-cepts only for the ‘‘Universal Service ID’’ of the or-der segment (OBR-4), so the attributes represent-ing concepts expressed in other HL7 message fieldswere deleted. A specific attribute for representingtemporal details of events related to orders, suchas ‘‘preoperative’’ and ‘‘at discharge’’ was addedduring this initial phase.Examples of the resulting pre-coordinated nurs-

ing order concepts (represented in XML) are listedbelow.Example 1: ‘‘Assess skin areas in contact with

oxygen delivery device’’

Example 2: ‘‘Get patient up to chair withtwo-person assistance’’

Example 3: ‘‘Apply betadine to ulcers on left footwith 2 × 2 gauze’’

Example 4: ‘‘Notify physician if systolic bloodpressure greater than 160mm/Hg’’

The second objective was to determine if theseattributes were substantiated in the research andstandards literature. The attributes of the modelwe developed were compared to the three modelsidentified in the literature (see Table 2).The attributes of the PCDS orders axis were eval-

uated. The attributes included from the PCDS were‘‘Subject,’’ ‘‘Object,’’ and ‘‘Action.’’The axes we matched from ICNP were ‘‘Action

Type,’’ ‘‘Target,’’ ‘‘Means,’’ ‘‘Time,’’ ‘‘Topology,’’‘‘Location’’ (divided into two–—‘‘body site’’ and

Page 4: Development of a compositional terminology model for nursing orders

628 S. Matney et al.

Table 1 Nursing Order Attributes

Order attributes Definition Examples

Action typea Delivery mode Educate, assess, measure, feedActivity focusa The subject of the action Pain, bleeding, patient, etc.Body location The body location Arm, leg, nosebody side The side of the object/subject Left, rightMeans The device, equipment, or mechanism

used to perform the action or processDevice (wheelchair, walker, prosthesis. . . )

Method The method to be used for the order Auscultation, palpation, observationBeneficiary The person, persons, or group of

people who would benefit from theaction or process

Client (patient), family, group, community. . .

Timing The timing of the action or process Pre-op, during delivery, at dischargeCondition A condition that was required–—either

before process was carried out, or forcompletion of the process

If O2Sat < 90, until systolic blood pressure ≥90

Location The location action would take place Hall, patient room, department

a Denotes mandatory fields.

‘‘spatial’’), and ‘‘Beneficiary.’’ There was a sepa-rate field in the HL7 message for ‘‘Route,’’ thus,this attribute was not used.When we compared the attributes of the pro-

posed compositional terminology model to the Eu-ropean Committee for Standardization (CEN) Tech-nical Committee 251 terminology model for nursingactions [7,13], the only attribute in our model thatwas not in the CEN model was ‘‘Location.’’ The CENmodel provided the best match to our model.Following the model development and validation

against published models, the developers gener-ated 609 pre-coordinated nursing order conceptsusing the proposed compositional terminologymodel. All attributes were used; none were lackingand no new attributes were needed.

Table 2 Nursing order attribute comparison

Attribute PCDS ICNP CEN

Action type* Action Action type ActionActivity focus* Object Target TargetBody location Location:

body siteSite

Body side Topology TopologyMeans Means MeansMethod DeviceBeneficiary Subject Beneficiary BeneficiaryTiming Time TimingCondition Is

associatedwith

Location Location:spatial

5. Discussion and implications

The first question that needs to be addressed iswhy not just use the CEN or ICNP models, sincethey match so closely to our proposed model? Asstated before, our intent was to create a termi-nology model specific for nursing orders, and notnursing actions. We also wanted to have a composi-tional representation that could be used to consis-tently generate aggregated (pre-coordinated) con-cepts for use with HL7 order messages. The CENmodel had almost all the atomic parts but not theactual pre-coordinated instances. Another relatedquestion is how are nursing orders different fromnursing actions or interventions? In reality, orderscan encapsulate one or multiple actions. An orderfor a protocol is similar to a laboratory battery or-der. It spawns multiple interventions and can possi-bly be completed by more than one person or dis-cipline. Also, interventions have to be modeled ina different way, using items such as units, codeddata, and/or text fields. An example is an order forvital signs. There is only one order but it results inmultiple observations: blood pressure, heart rate,respiratory rate, and temperature. Also, the POEapplication IHC is currently developing requires thespecification of the person who can issue the or-der, and the person that can perform the action. Aphysician usually initiates patient orders, but nursesor therapists usually carry out the interventions.Therefore, the ordering roles are attached to theorder, while the execution roles are properties ofthe interventions and actions.The PCDS was the only model we found for nurs-

ing orders. Unfortunately, we could only find a de-

Page 5: Development of a compositional terminology model for nursing orders

Development of a compositional terminology model for nursing orders 629

scription of the attributes and not the actual ordersthemselves, since it is not in the public domain. ThePCDS aggregate orders would be a good starter setfor our analysis, if they were available.We intend to contribute our data set to open stan-

dards organizations such as LOINC or HL7. We alsoplan to share our pre-coordinated concepts withwhoever may be interested in using them as startersets for their own POE systems.An intriguing question that also needs to be

answered is why has not anyone thought of thisbefore, despite the fact that there are many POEsystems today? Ozbolt was the only researcherwe found who had identified specific attributes oforders. One possible explanation is that existingPOE systems are using proprietary (locally devel-oped) nursing concepts, without the preoccupationof adopting a terminology model to try to ensureoptimal consistency and maintainability of theseconcepts.There are many benefits in creating composi-

tional orders. The first reason is in terminologycreation and maintenance. The ability to check forsynonyms by each specific attribute will decreaseredundant concept creation. The second benefitis the ability to attain domain completion. If wewant to find every order that has been created for‘‘pain,’’ we can query for ‘‘pain’’ in the activityfocus. If we want to query for all the action typesthat are assessments, we can query for ‘‘Assess’’in the action type column.In addition to the questions raised by compar-

isons with other efforts, during the prospectivemapping of the nursing orders, one issue was how tomap a complicated order such as ‘‘Apply and main-tain a cervical collar?’’ It was determined that thisneeded to be two orders. Another issue that wasdiscussed was the creation of separate concepts fordifferent conditional orders such as ‘‘temperatureabove a certain value,’’ where temperature couldbe expressed as a range of numbers. The final de-cision on how to model the conditional orders hasnot been made. However, it was apparent that theuser should not have to wade through many dif-ferent orders with the same meaning, with onlydifferent conditional attributes because it wouldbe very time consuming. In these cases, the con-dition should probably contain a field for ‘‘value’’that the user will complete at ordering time (seeExample 4).

6. Conclusion

A compositional terminology model for nursingorders was created, guiding the development of

pre-coordinated order concepts that can be usedin the OBR-4 field of the HL7 order message. Nurs-ing orders were gathered from many IHC facilities,on both paper and electronic form. The modelwas compared to three similar models identifiedin the literature. Finally, successful mapping of609 orders confirmed the validity of the resultingmodel.The proposed model requires additional eval-

uation. Evaluation criteria could include domaincompleteness, structural and semantic relation-ships among terms, degree of ambiguity of thepost-coordinated concepts, and overall utility [15].Further, evaluation is necessary to determine howthe new POE system will use the orders and therelationships between the ordered service and theprovided care.

Acknowledgements

We would like to thank Stanley M. Huff, M.D.–—Mentor; Susan J. Grobe, R.N., Ph.D., F.A.A.N., fortechnical advice; and Susanne Miller, R.N., M.S., fordomain expertise

References

[1] 3M Health Information Systems, Making sense of the data:using a medical data dictionary to integrate, share, andunderstand clinical data, 3M Health Information Systems,2002.

[2] R.A. Rocha, S.M. Huff, P.J. Haug, H.R. Warner, Designing acontrolled medical vocabulary server: the VOSER project,Comput. Biomed. Res. 27 (6) (1994) 472—507.

[3] Health Level Seven, Health level seven standard version2.4: an application protocol for electronic data exchangein healthcare environments, Health Level Seven Inc., AnnArbor, MI, 2001.

[4] G.J. Kuperman, R.M. Gardner, T.A. Pryor, HELP: a dynamichospital information system, Springer-Verlag, New York,NY, 1991.

[5] ISO/IEC 17115, Health informatics–—vocabulary for termi-nological systems, International Organization for Standard-ization, Geneva, Switzerland, 2002.

[6] HL7, Glossary of terms, Health Level Seven, Ann Arbor,Michigan, 2001.

[7] CEN ENV 14032, Health informatics–—system of conceptsto support nursing, CEN, Brussels, 2000.

[8] J.J. Warren, A. Coenen, International classification fornursing practice (ICNP): most-frequently asked questions,J. Am. Med. Inform. Assoc. 5 (4) (1998) 335—336.

[9] K.E. Campbell, A.D. Das, M.A. Musen, A logical foundationfor representation of clinical data, JAMIA 1 (3) (1994) 218—232.

[10] S.M. Huff, R.A. Rocha, C.J. McDonald, et al., Develop-ment of the LOINC (logical observation identifier namesand codes) vocabulary, JAMIA 5 (3) (1998) 276—292.

[11] C.E. Barr, H.J. Komorowski, E. Pattison-Gordon, R.A.Greenes, Conceptual modeling for the unified medical lan-

Page 6: Development of a compositional terminology model for nursing orders

630 S. Matney et al.

guage system. In: 12th SCAMC, 1988, IEEE Computer Soci-ety Press, 1988, p. 158-1.

[12] J.G. Ozbolt, The patient care data set: profile,http://ncvhs.hhs.gov/990518t3.pdf, Vanderbilt University,1999, pp. 1—16.

[13] CEN Pre-standard, Health informatics categorial structuresfor systems of concepts nursing phenomena and actions,CEN, Brussels, 2000.

[14] S.J. Grobe, The nursing intervention lexicon and taxon-omy: implications for representing nursing care data in au-tomated patient records, Holistic Nurs. Pract. 11 (1) (1996)48—63.

[15] S.B. Henry, C.N. Mead, Nursing classification systems: nec-essary but not sufficient for representing ‘‘what nursesdo’’ for inclusion in computer-based patient record sys-tems, JAMIA 4 (3) (1997) 222—232.