A Dynamic Classification Approach for Nursing

Abstract

Nursing has a long tradition of classification, stretching back at least 150 years. The introduction of computers into health care towards the end of the 20^th Century helped to focus efforts, culminating in the development of a range of standardized classifications. Many of these classifications are still in use today and, while content is periodically updated, the underlying classification structures remain relatively static. In this paper an approach to classification that is relatively new to nursing is presented; an approach that uses formal Web Ontology Language definitions for classes, and computer-based reasoning on those classes, to determine automatically classification structures that more flexibly meet the needs of users. A new proposed classification structure for the International Classification for Nursing Practice is derived under the new approach to provide a new view on the next release of the classification and to contribute to broader quality improvement processes.

Introduction

Classification describes both the act of classifying (organizing by class according to common features), and the result of classifying or being classified. The act of classifying is as old as health care itself; classes of disease, such as gout, have steered treatments for generations of health care practitioners over thousands of years. Structures resulting from the act of classifying are more recent although they could hardly be considered new phenomena. For example, Nosologia Methodica, Sistens Morborum Classes, Genera Et Species by Francois Boissier De Sauvages, now widely regarded as a foundation for the International Classification of Diseases, was published in 1763, nearly one quarter of a millennium ago.

Nursing has a long history of classification. In Notes on Hospitals¹, published one hundred years after Nosologia Methodica, Florence Nightingale argued the need for more comprehensive hospital records. In the same book, she also presented a set of tables to provide ‘an uniform record of facts’ (p 162), for example:

• Total sick population

• Number of cases

• Annual proportion of recoveries.

In order to populate these tables, it was obviously necessary to classify individuals in terms of what constitutes sickness, what is meant by a case, and what should be regarded as recovery.

The introduction of computers into health care, a further one hundred years later, led to a renewed interest in classification. In the US, the Nursing Minimum Data Set (NMDS)² provided a framework for the consistent collection of nursing-oriented data. Standardized classifications were developed to provide the classes of nursing diagnoses, interventions and outcomes that would serve to populate the NMDS.

Examples of classifications recognized by the American Nurses Association (ANA) as being suitable for use in computer-based systems³ include the Clinical Care Classification (formerly Home Health Care Classification)⁴, NANDA International⁵, Nursing Intervention Classification⁶, Nursing Outcome Classification⁷, the Omaha System⁸, and the Perioperative Nursing Data Set⁹. These classifications have been developed manually, with individuals or groups of individuals identifying, defining and organizing classes by consensus. While the content of the classifications is periodically updated, the underlying classification structures are relatively static, possibly due to the difficulties associated with manual processes. This paper describes a new more dynamic approach to classification that uses the formal definitions of classes and automated reasoning to determine how they are organized (and re-organized).

Background

The International Classification for Nursing Practice (ICNP^®)¹⁰, developed by the International Council of Nurses, is also recognized by the ANA. In common with other nursing terminologies, early versions of ICNP were developed and maintained manually. However, recognizing the increasing sophistication of ICNP users and the resulting increasing complexity of the classification, the ICNP Strategic Advisory Group endorsed the use of automated approaches to development and maintenance, commencing with ICNP Version 1. All subsequent versions of ICNP have been based on the same technology, the main differences between releases being the addition of new classes and the retirement and replacement of classes that are no longer required in practice

At the heart of ICNP is an ontology – a formal representation of ‘concepts’ or classes and their interrelationships. The ICNP ontology is represented in the Ontology Web Language (OWL)¹¹; the development team uses the Protégé ontology editor¹² to manipulate the ontology.

In an OWL ontology, each class is formally defined. A class may be related to other classes through a class hierarchy i.e. a ‘kind of’ hierarchy based on the generic relation. For example, Analgesic is represented within ICNP as a kind of Drug as shown in Figure 1. Classes may be related to other classes via properties. For example, for AdministeringDrug, AdministeringAct is associated with Drug via a property called hasInterventionalTarget as shown also in Figure 1.

Figure 1.

Asserted relationships within the OWL representation of ICNP

It is possible to use software tools, such as FaCT++¹³, to perform automated reasoning on OWL ontologies to determine inconsistencies (mistakes in definitions), redundancies (classes that may be equivalent) and subsumption relationships (classification). The latter is particularly important in the context of this paper as it is this feature that supports the development of the ICNP classification and takes it from a manual to an automated process. Building on the previous example, as Analgesic is a kind of Drug, a reasoner can infer that AdministeringAnalgesic would also be a kind of AdminsteringDrug, as shown in Figure 2.

Figure 2.

A relationship, inferred through automated reasoning, within the OWL representation of ICNP

The ICNP OWL ontology therefore comprises both elementary classes, such as Drug and AdministeringAct, and composed classes such as AdministeringDrug. The elementary classes provide the building blocks for the composed classes.

ICNP may be delivered in a number of different formats – this flexibility is one of the significant advantages of a heavily-automated classification such as ICNP. Some users choose to engage with ICNP via a 7-axis model in which the elementary classes only are organized into 7 different axes such as Action, Target, Means, etc. These axes explicitly mirror and therefore comply with the structures within the terminology standard ISO 18104^14,15. Other users choose to engage with ICNP via composed (or pre-coordinated classes only). Previously, these composed classes were delivered in the form of simple non-hierarchical lists pertaining to diagnostic classes and interventional classes.

The remainder of this paper describes the approach taken by the ICNP development team in deriving a hierarchical representation of these composed classes, with a focus on diagnostic classes. The approach forms part of a much broader quality improvement process which has been reported elsewhere¹⁶.

Method

The inferred diagnostic class hierarchy of ICNP Version 2 (i.e. the classification resulting from automated reasoning) is organized at the upper level around two axes. First, there is a distinction between whether a diagnostic class is considered as an undesirable or negative phenomenon e.g. Dyspnoea, or as a desirable or positive phenomenon e.g. Hope. At the next level in the hierarchy, there is a distinction between whether a diagnostic class, either positive or negative, is actual i.e. it is actually occurring, or whether it is potential i.e. there is a risk or chance of it occurring. Classes corresponding to potential negative diagnostic phenomenon are considered to be a Risk. A graphical partial representation of the Version 2 diagnostic inferred hierarchy is given in Table 1. Note that in Version 2, there are no classes that would be considered potential positive diagnostic phenomenon.

Table 1:

A tabular partial representation of the ICNP Version 2 diagnostic inferred hierarchy

DiagnosticPhenomenon
	PositivePhenomenon
		ActualPositivePhenomenon
			Hope
		PotentialPositivePhenomenon
	NegativePhenomenon
		ActualNegativePhenomenon
			Dyspnoea
		PotentialNegativePhenomenon
			RiskForConstipation

While this organization of classes is useful in terms of developing and maintaining the classification itself, it represents a view of the world that is unfamiliar to nurses in practice. A different view might bring a number of benefits:

• facilitating navigation of the classification through a more familiar structure

• providing a framework to support data entry by partitioning the world in a more useful way

• facilitating aggregation of data through the identification of more meaningful classes.

The ICNP development team (n=5) collectively agreed on a new structure for the upper levels of the ICNP class hierarchy; one that would a) provide a more familiar structure for the users of ICNP, b) test the process of deriving automatically a new classification structure, and c) contribute to the wider quality improvement processes. As agreed among the team, the new structure would distinguish between diagnoses that involve body processes and those that involve psychological processes. OWL classes were formally defined to represent these two diagnoses:

• Any body process that is either positive or negative and either actual or potential

• Any psychological process that is either positive or negative and either actual or potential

A third class of diagnosis would ensure coverage by capturing diagnoses that involve neither body processes nor psychological processes:

• Any diagnostic process that is neither a body process nor a psychological process that that is either positive or negative and either actual or potential.

Results

After automated reasoning, five hundred and ninety six diagnostic concepts were organized into the new classification structure. In the new hierarchy, 3 classes had 5 immediate parents, 134 had 4 immediate parents, 138 had 3 immediate parents and the remainder (321) had 2 immediate parents. This demonstrates the richness of the inferred hierarchy and the benefit of an automated approach to classification. As an example of this richness, NegativeMedicationSideEffect has just one parent in the asserted hierarchy – a generic MedicationSideEffect (neither negative nor positive). In the new inferred hierarchy, the same class, appearing as a BodyProcessPhenomenon, has 2 additional immediate parents - thanks to its formal definition it is also classified as a NegativeResponseToTreatment and a Complication.

A tabular partial representation of the new inferred hierarchy is given in Table 2.

Table 2.

A tabular partial representation of the new inferred ICNP hierarchy

DiagnosticConcept
	BodyProcessPhenomenon
		Allergy
		EffectiveTissuePerfusion
		RiskForConstipation
	PsychologicalProcessPhenomenon
		Hope
		Confusion
		RiskForLoneliness
	NonBodyNonPsychologicalPhenomenon
		ElderAbuse
		EffectiveRolePerformance
		RiskForFall

This new classification structure provides an alternative view on ICNP, with diagnostic classes grouped according to the requirements, in this case, of the ICNP development team. It also contributes to the wider quality improvement process of ICNP. Indeed the new hierarchy revealed on inspection several issues that needed to be addressed prior to the 2011 release of ICNP. These issues included: errors or missing properties in definitions for classes, lack of a closure axiom to bound NonBodyNonPsychologicalPhenomenon, transitivity propagating erroneously across properties, and lack of intermediate parent classes.

Conclusion

Prior to this study, the decision to underpin ICNP with an OWL ontology had already delivered positive benefits in terms of developing and maintaining the classification. The present study has demonstrated that the same heavily-automated approach can support a relatively dynamic reorganization of the classification in order to match a range of changing requirements. The particular classification structure (reflecting to some extent a Cartesian dualistic view of the world) developed largely automatically within this study was agreed in advance among the ICNP project team. This new classification structure forms the basis for the display of diagnostic classes in the 2011 release of ICNP. However, the inherent flexibility of the ontological approach means that even this structure need not be set in stone. It is important that individual classes remain stable – and this is assured through the formal computable definitions for classes. However, it is desirable and inevitable that future releases of ICNP will use additional classification structures to match the changing needs of nursing practice.