A Compositional Approach to Nursing Terminology - CiteSeerX

TERMINOLOGY AND CLASSIFICATION

A Compositional Approach to Nursing Terminology Hardiker N and Kirby J Medical Informatics Group, Department of Computer Science, University of Manchester, Manchester M13 9PL, UK The development of standardised vocabularies within nursing has been an important research activity for a number of years. Current representations generally take the form of taxonomic vocabularies. These are seen as important as they provide a structure for retrieving and analysing data from automated systems. However, there is increasing evidence to show that traditional taxonomic vocabularies are unsuitable for capturing detailed clinical data. This paper describes how GRAIL (GALEN Representation and Integration Language) is being used within the TELENURSE project to develop a representation of nursing terminology which is sufficiently expressive for documenting detailed clinical data while retaining the benefits of traditional taxonomic vocabularies.

Introduction The development of standardised vocabularies to represent nursing terminology and to describe nursing practice has been an important research activity for many years. The development and increasing use of computer-based nursing information systems have further emphasised the importance of this activity. The result has been the emergence of a number of representations. Problems with traditional taxonomic vocabularies The majority of the commonly reported standardised nursing vocabularies take the form of taxonomic vocabularies. Taxonomic vocabularies are terminological systems in which concepts are related by hierarchical relations i.e. generic ‘is-a’ relation and partitive ‘part-of’ relation, and other associative and pragmatic relations1. Examples within nursing include the North American Nursing Diagnosis Association Taxonomy I (NANDA), the Nursing Interventions Classification (NIC), the Home Health Care Classification (HHCC) and the Omaha Community System (Omaha). These representations are seen as important because they provide a structure for retrieving and using nursing data from automated systems2. Other reasons cited for organising nursing concepts into taxonomies include: to formalise and expand knowledge about nursing practice, to assist in determining the cost of nursing services, to help to target resources more effectively and to make explicit the role played by nurses in health care 3. Monohierarchical taxonomic vocabularies that are exhaustive and that guarantee disjunction are seen as useful for statistical evaluation1. Thus it could be argued that taxonomic vocabularies have a useful role to play in activities such as data retrieval and data analysis. However there is increasing evidence to show that taxonomic nursing vocabularies are not able to represent the detailed clinical data within patient records4. As such they are poorly suited for representing day-to-day nursing care. One study was carried out to test the feasibility of using the third version of the Systematized Nomenclature of Human and Veterinary Medicine (SNOMED) to represent the terms used by nurses to describe patient problems5. It should be noted that SNOMED III contains all of the nursing diagnoses from NANDA. This study found that NANDA terms alone provided matches for only 30% of the patient problems described by nurses in the study. It is clear from these results that NANDA alone does not provides the coverage necessary for nurses to record

1

TERMINOLOGY AND CLASSIFICATION patient problems (interestingly the inclusion of SNOMED III terms and combinations of SNOMED III terms increased the proportion of matches to 69%). Another study was carried out to compare the ability of terms from NIC and from the medically-oriented Current Procedural Terminology to represent the clinical terms used by nurses and patients to describe nursing interventions6. The results of matching NIC terms to clinical terms used by nurses and patients to describe nursing interventions are given as ‘encouraging’. However the examples cited of selected clinical terms and their matching NIC interventions show that comprehensiveness of scope is at the expense of clinical detail. For example, the relatively abstract NIC term ‘Hypoglycaemia management’ is given as a match for the relatively detailed clinical term ‘Fingersticks for blood sugar’. NIC has been criticised previously for being insufficiently fine-grained for capturing differences in practice4. One reason concerns the fact that traditional taxonomic vocabularies are constructed by enumerating all of the possible terms that are to be represented; and organising these terms within a hierarchy. In constructing any enumerative scheme, developers must limit the number of concepts to include as the total number of concepts would be unmanageable, both in terms of development and in terms of practical application. As such, enumerative representations tend to be tuned to a single purpose or to a group of closely-related purpose; re-use for other purposes proves very difficult. Indeed HHCC and Omaha have been criticised for lacking the specific vocabulary of acute care and NANDA Taxonomy I has been criticised for not covering all fields of specialty practice4. Solutions to problems concerning expressiveness Linear lists An alternative approach to the traditional taxonomic vocabulary is the linear list. A linear list is simply a collection of terms relevant to a domain1. One study claims that it may be possible to develop a list of standard terms that is capable of representing the universe of terms actually used to record data elements in a patient record4. However, there are many outstanding issues arising from this study: •

The study was confined to two specialist fields, Orthopaedics and Thoracic/ Cardiovascular surgery; there is no indication as to how the list of standard terms employed within the study might scale up to include other areas of practice. • 11 auditors were involved in the study, matching statements from patient records to standard terms in a code book. While the reliability of the auditors, that is the accuracy of term matching, was a consideration within the study, the results given omit any discussion on the degree of detail captured by the standard terms and on the exactness of term matching. • Within the study term matching was performed manually and it is not clear how this process may be automated, nor how the standard terms might be re-used for other purposes. Until these issues are resolved, the usefulness in practice of such an approach is questionable. A compositional approach Within the GALEN project (Generalised Architecture for Languages, Encyclopedias and Nomenclatures in Medicine) a new approach to representing clinical terminology has been developed in the form of the GALEN Representation And Integration Language (GRAIL) 7.

2

TERMINOLOGY AND CLASSIFICATION GRAIL is a terminological language that provides a means of capturing the knowledge that underpins clinical terminology in a formal compositional model from which all and only sensible clinical concepts can be generated8. By decomposing complex concepts into primitive concepts, other schemes such as SNOMED attempt to address the problems associated with enumerative representations. However, although SNOMED provides a framework or meta-model for constructing complex clinical concepts, it is impaired by the lack of specific rules for determining which combinations are clinically sensible. Thus it cannot prevent the creation of clinically meaningless concepts; nor is it able to control combinatorial explosion. A GRAIL model consists of a taxonomy of concepts and a set of rules or ‘sanctions’ to dictate how these concepts may sensibly be combined. For example, it might be sensible to combine the elementary concepts Mobilising and Ability to create a composite concept which defines ‘Mobility’: (Ability which refersTo Mobilising). As they are created, composite concepts are classified automatically in the taxonomy. For example, if the concept Mobilising subsumes Walking, then the composite concept which defines ImpairedWalkingAbility will be subsumed by ImpairedMobility.

Mobilising

Walking

Ability

Ability which < refersTo Mobilising hasState Impaired>

Ability which < refersTo Walking hasState Impaired>

The result is a multiple hierarchy of clinically sensible concepts which are defined to an arbitrary level of detail. Practical application of GRAIL The TELENURSE project The TELENURSE project is an accompanying measure in the European Telematics Application Programme. Its primary aim is to promote consensus among nurses across Europe around the use of the International Classification of Nursing Practice (ICNP) which is being developed by the International Council of Nurses. Each concept within ICNP is explicitly defined and classified in terms of the generic relation. The alpha version of ICNP 3

TERMINOLOGY AND CLASSIFICATION has two dimensions: nursing phenomena and nursing interventions. At the time of writing, the ICNP classification of nursing interventions was undergoing change. The remainder of this discussion is therefore restricted to the ICNP classification of nursing phenomena. As the ICNP classification of nursing phenomena is monohierarchical, it may be well-suited for statistical evaluation. However, as an example of a traditional taxonomic vocabulary it is not well-suited to the task of recording day-to-day nursing care. In contrast, the development of GRAIL has been driven in part by the data entry requirements of users of clinical applications. The use of GRAIL within TELENURSE The GALEN approach has been applied successfully within other areas9; nursing presents new challenges. There is within nursing a resistance to recording meaningful, concise information concerning the nursing care of patients10. This is compounded by a general confusion about the nature of nursing information. In the context of nursing interventions four main origins for this confusion have been identified3: 1. The contextual nature of nursing information leading to confusion between intervention (nurse behaviour) and assessment and evaluation (patient behaviour); 2. The use of synonyms e.g. action, activity, treatment, order; 3. The lack of conceptualisation of how a number of actions might fit together, resulting in long verbose care plans; 4. Inadequate decision making among nurses in selecting and prioritising interventions. Within TELENURSE the GALEN approach is being applied in an attempt to overcome the first three of these problems; the final problem requires a change in nursing practice. An existing GRAIL medical foundation model has been extended to incorporate nursing concepts. This has involved the development of GRAIL definitions for ICNP concepts. For example, the ICNP concept ‘Nursing Phenomena’ has been explicitly defined in GRAIL as: Phenomenon which hasRelevantDomain NursingDomain. Such definitions restrict the possibility of ambiguity and make explicit any contextual influences. In GRAIL, any number of unique names for individual concepts is permitted, thus facilitating the controlled use of synonyms: (Phenomenon which hasRelevantDomain NursingDomain) name NursingPhenomenon. As specific detail is added, GRAIL concepts are classified automatically. The resulting subsumption hierarchy provides a ‘bridge’ for different levels of abstraction and represents a conceptualisation of how concepts interrelate. Transforming hope into working achievement A significant problem with enumerative representations is the fact that any reasoning behind the decisions made during the construction of the scheme is locked inside terms or concept 4

TERMINOLOGY AND CLASSIFICATION definitions. For example, a nurse may have a clear understanding of the enumerated term ‘Able to walk a short distance’. However a computer can have no such understanding and thus cannot utilise the underlying clinical concepts in managing the scheme. A major motivation for the modelling activity within TELENURSE has been the need to ensure that computers will be able to manage the ultimate structure and content of ICNP. As part of TELENURSE two prototype nursing care management systems are being developed. GALEN technology will make ICNP, in the form of a GRAIL model, available to users of these systems in order to transform the potential benefits of using standardised vocabularies into working achievements. Summary A number of well-founded standardised nursing vocabularies have been developed over recent years. The majority of these take the form of taxonomic vocabularies. While such representations may be appropriate for statistical analysis of relatively abstract data, they are unable to capture the detail of day-to-day nursing care. GRAIL provides a mechanism for representing clinical data at any level of detail. Within the TELENURSE project a model built in GRAIL will be used to make ICNP available to users of clinical applications without compromising the operational needs of those users. The result will be a representation of nursing terminology which is sufficiently expressive for documenting highly detailed clinical data; and one which retains the benefits of traditional taxonomic vocabularies. References 1. Ingenerf J Taxonomic Vocabularies in Medicine: The Intention of Usage Determines Different Established Structures. In: Greenes RA et al. eds. Proceedings of the Eighth World Congress on Medical Informatics., Amsterdam North-Holland: 1995:136-139. 2. Grobe. Nursing Intervention lexicon and taxonomy study: Language and classification methods. Adv Nurs Sci, 1990;13(2):22-33. 3. McCloskey JC, Bulechek GM, Cohen MZ, Craft MJ, Crossley JD, Denehy JA, Glick OJ, Kruckeberg T, Maas M, Prophet CM, Tripp-Reimer T. Classification of Nursing Interventions. J Prof Nurs 1990;6(3): pp151-157. 4. Ozbolt JG, Russo, M, Stultz MP. Validity and Reliability of Standard Terms and Codes for Patient Care Data. In: Gardner RM (Ed.), Proceedings of the 19th Annual Symposium on Computer Applications in Medical Care. Hanley & Belfus Inc., Philadelphia, 1995:37-41. 5. Henry SB, Holzemer WL, Reilly CA, Campbell KE. Terms Used by Nurses to Describe Patient Problems: Can SNOMED III Represent Nursing Concepts in the Patient Record? J Am Med Informatics Assoc 1994161-74. 6. Henry SB, Holzemer WL, Reilly CA, Miller TJ, Randall C. A Comparison of Nursing Intervention Classification and Current Procedural Terminology Codes for Representing Nursing Interventions in HIV Disease. In: Greenes RA et al. eds. Proceedings of the Eighth World Congress on Medical Informatics. Amsterdam:North-Holland, 1995:131-135. 7. GALEN Implementation and Modelling Teams. GALEN Documentation Volume H: A Guide to GALEN's Concept Services. Available from: Project Coordinator, Medical Informatics Group, Department of Computer Science, University of Manchester, M13 9PL, 1994. 8. Rector AL, Bechofer S, Goble CA, Horrocks I, Nowlan WA, Solomon WD (1995). Why GRAIL. Paper in preparation. Medical Informatics Group, University of Manchester. 9. Kirby J et al. (1996), A System for Entering Detailed Clinical Data. To be presented at: Healthcare Computing, 18-20 March 1996, Harrogate, UK. 10. Howse E, Bailey J (1992). Resistance to documentation - a research issue. Int J Nurs Stud, 29 (4): pp 371380.

5

TERMINOLOGY AND CLASSIFICATION

Nursing Intervention Intensity and Focus: Indicators of Process for Outcomes Studies Grobe SJa, Hughes LCb, Robinson Lc, Adler DCc, Nuamah Ic and McCorkle Rc a

School of Nursing, University of Texas at Austin, Austin, TX 78701-1499. bSchool of Nursing, Wichita State University, Wichita, KS, 67260-0041. cSchool of Nursing, University of Pennsylvania, Philadelphia, PA, 191046096. Outcomes research has become increasingly important in the current health care environment and for informatics research efforts. Recent efforts in automating clinical data for use in outcomes studies has focused attention on the need to represent the processes of care in the classic structure-process-outcome models of care. This paper reports on use of the Nursing Intervention Lexicon and Taxonomy for classifying interventions to characterize two process of care variables: intervention intensity and intervention focus. Study results demonstrate that these variables are descriptive and provide promise for describing processes of nursing care for describing clinical care.

Introduction Outcomes research and benchmarking studies are increasingly important in today's healthcare arena wherein cost and quality studies are being conducted to examine efficiency, effectiveness and quality of care. And, although much attention has been focused on outcomes of care, much less attention has been directed toward describing the processes of care. In this current environment of capitated and managed care it becomes increasingly critical to find ways to describe and measure these processes of care. This paper describes a preliminary way for describing and measuring care process in a study of the home care of cancer patients. Models used to guide outcomes research are those of Donabedian and Holzemer (Donabedian, 1982; Holzemer, 1995). In Donabedian's structure-process-outcome model, structure includes the resources used in providing care, process includes the activities that comprise care, and outcomes are the consequences of both of these on health. Holzemer's model extends Donabedian's; it has a horizontal axis of inputs, processes and outcomes and a vertical axis that includes the three constituents generally involved in health care encounters: the client, the provider and the setting. While much research in nursing and health care has been focused on organizational structure (i.e., Holzemer's inputs) and outcomes, little attention has been focused on describing care processes. This study reflects one of the dimensions of Holzemer's model: Process (from the horizontal axis) as reflected by nurses' care interventions as documented during their home care of breast and prostate cancer patients. Importantly, because the main study reflected a controlled clinical trial, this nursing documentation was not encumbered by external Medicare reimbursement considerations. Although the study uses manual methods to test the feasibility of using classified nursing interventions to describe the processes of care, it represents an important preliminary informatics effort to demonstrate "proof of concept," prior to investigating automated methods for both extracting and classifying similar interventions from narrative nursing recording. Study purpose The premise of this paper is that, it is possible to describe care processes, using two patterns of nursing care variables: nursing intervention intensity and intervention focus, as demonstrated in a controlled clinical trial of home care of cancer patients (McCorkle, PI: NIH, NINR R01 NR03229, 1992-96). Intervention intensity is defined as the frequency of interventions; and, intervention focus is defined as the categories of interventions used most

6

TERMINOLOGY AND CLASSIFICATION frequently during care. These two variables, intensity and focus, once measured reliably and validly, can then be used for examining relationships among the care processes and specific care outcomes. This paper describes these two patterns of care variables for a subset of McCorkle's postoperative cancer patients receiving the experimental home nursing care (the standard nursing intervention protocol (SNIP), specifically the breast (N=22) and prostate (N=33) cancer patients. McCorkle's standard intervention protocol consisted of 8 contacts including 3 home visits and 5 telephone calls over 4 weeks by a clinical nurse specialist (masters prepared nurse), begun within a week after discharge from the hospital. Clinical nurse specialists documented their home care in paper records that served as the source of study data. Their documentation was free of any constraints imposed by reliance on reimbursement. In fact, these nurses were encouraged to document completely all the care they provided for each patient and their caregiver(s) using a modified SOAP format. Study procedures Once the cancer patients (breast and prostate) had received the standard nursing interventive protocol, interventions were manually extracted and transcribed, by a single member of the research team, from nurses' narrative SOAP recording (modified with an I category added to include interventions, making it a SOAIP format). Then, three trained individuals, independently classified each of the interventions using a 7 category scheme, i.e., the Nursing Intervention Lexicon & Taxonomy (NILT) (Grobe, 1996). The category names and their short descriptive concepts include: Care Information Provision (CIP), teaching; Therapeutic Care Psychosocial (TCP), supporting; Care Vigilance (CV), monitoring status; Care Need Determination (CND), assessing need for care; Care Environment Management (CEM), obtaining resources for care; Therapeutic Care General (TCG), performing care procedures; and Therapeutic Care Cognitive Understanding and Control (TCCU&C), encouraging selfcare. A source definition, typical of those used by categorizers is provided for CIP along with prototypical intervention examples. Brief definitions for the remaining categories are provided in Figure 1. Care Information Provision (CIP): the deliberative, cognitive, physical or verbal activities of informing or teaching that assist individuals (may be client, family, significant other(s) or caregivers(s), who are the focus of care to acquire or use care information intended to maintain or improve the existing state or general condition and maximize the response to therapy. Prototypical examples include: Advise patient to seek information from ET nurse; Explain how to splint an incision; Review medications with patient; Teach patient to perform dressing change; and Inform wife of possible side effects of medications. Figure 1 NILT Category Definitions and Examples CND: Care Need Determination: •assessment of need for care including: past health, baseline health state, role management, health beliefs and values. Examples: Determined her perception of the cause of fatigue; Assessed patient’s knowledge of her disease; Explored patient’s expectations for recovery. CV: Care Vigilance: •assessment of physical status, physiological, mental or emotional status, monitoring of status or devices. Examples: Assessed pain; Monitored incontinence; Assessed mastectomy site. CEM: Care Environment Management:

7

TERMINOLOGY AND CLASSIFICATION •evaluation of the environmental or familial context for care, referring or influencing the use of resources. Examples: Referred patient to Reach for Recovery; Offered numerous resources for depression; Suggested use of Wellness Center; Encourage to talk with someone in physician’s office. TCG: Therapeutic Care-General: •performance of procedures, and physically based activities. Examples: Encouraged physical activity, i.e., short daily walks; Reapplied ace bandage to right chest area; Encouraged kegal exercises; Performed dressing change on incision line. TCP: Therapeutic Care-Psychosocial: •performance of psychologically based therapies. Examples: Assured patient that feelings were normal; Reassured her that improvement is not unidirectional; Listened actively; Supported patient emotionally; Encouraged open dialogue with husband. TCCU&C: Therapeutic Care Cognitive Understanding and Control •activities to enhance self care, control and independence. Examples: Supported patient’s need to decide when she is ready; Supported patient’s decision making ability; Provided positive feedback about her problem solving skills; Supported patient in his self care efforts. CIP: Care Information Provision: •informing about care and care procedures and therapies. Examples as above.

Once all the interventions were categorized, those interventions [for which there was not total agreement] were then discussed by the 3 categorizers for placement into a single NILT category. This agreed-upon final category was the ‘category of record’ used for this paper. All interrater agreement scores (Cohen's kappa) were calculated prior to this discussion, which was simultaneously used to maintain coder training. Interrater agreements (Cohen's kappas) for the breast and prostate charts were .7492 and .7944 respectively. Study results The patterns of care variables include intervention intensity and focus. Each is described next using the intervention data from the records of 22 breast patients and 33 prostate cancer patients. Table 1 Intervention intensity by contact* for breast and cancer patients Contact 1 2 3 4 5 6 7 8 9

Breast N % 257 20.9 134 10.9 138 11.2 158 12.9 122 9.9 103 8.4 116 9.4 137 11.1 44 3.6 20 1.6 1229

N = 22 Breast and 33 Prostate Patients

Prostate N % 242 15.0 247 15.4 199 12.4 201 12.5 160 9.9 197 12.2 144 8.9 133 8.3 63 3.9 23 1.4 1609

Total N % 499 17.6 381 13.4 337 11.9 359 12.6 282 9.9 300 10.6 260 9.2 270 9.5 107 3.8 43 1.5 2838

* Contact = either home visit or phone call

Intervention intensity for each patient contact is illustrated first in Table 1. Each contact represents either a home visit or a telephone call. For the breast cancer patients, the first 8

TERMINOLOGY AND CLASSIFICATION contact is more intensive with one-fifth of all interventions occurring during this first contact. By the fourth contact, both breast and prostate patients had received 55% of all interventions for their course of care. It is noteworthy that a few patients in both cancer groups received more than eight contacts, with these additional contacts representing about 5% of all interventions. Two patients received 10 contacts. Intervention Intensity by type of contact (either home visit or phone call) illustrates that in general, almost two-thirds of all interventions occurred during home visits (Table 2). Statistically significant differences exist between breast and prostate patients (p=0.046) on intervention intensity by type of contact, although this difference appears marginal. Table 2 Intervention intensity by type of contact* (home visit or phone call) for breast and prostate patients1

Breast (n = 1229) Prostate (n = 1609) Total (n = 2838) * p = 0.046

1

Home Ints 792

visits % 64.4

Phone Ints 437

calls % 35.6

978

60.8

631

39.2

1770

62.4

1068

37.6

N = 22 Breast and 33 Prostate Patients

Intervention focus, i.e., what categories predominated during care, are illustrated in Table 3. Prostate patients received more teaching (CIP) (p