Representing Clinical Knowledge in the PEN&PAD Clinical Data Entry System
Helen Fowler, Richard Gain and John Kirby
University of Manchester / VAMP Health Limited
Teaching Company Scheme Project
Introduction
The PEN&PAD - Practitioners Entering Notes and Practitioners Accessing Data - approach to clinical data entry developed by the University of Manchester Medical Informatics Group will be available to GPs this year1. A collaborative project with VAMP Health Limited has reimplemented and extended the PEN&PAD data entry user interfaces and the underlying terminology representation software in a Microsoft Windows™ environment. The result is the PEN&PAD Clinical Data Entry System (CDES) which has been designed as an extension to existing GP computer systems - in the first instance, VAMP's ViSion system.
The PEN&PAD CDES user interfaces address well-known problems of current approaches to data entry: finding the desired clinical topic; entering appropriate levels of additional detail on the topic; and adding further information on associated symptoms, examination findings and diagnosis.
In the Topic Selector, a symptom (or disease) is selected by selecting a body location, optionally a body system, and a symptom (or disease) which is either located in the body location - symptom in arm - or associated with the body location - cough (in chest).
Having selected a topic, it may be entered directly into the patient's record. Alternatively further details may be added by means of a data entry form as shown in Figure 2. Data entry forms allow further details on the presenting complaint to be added in the Descriptors (qualifiers) section as well as entering additional information on Associated Symptoms, Examination findings, and Diagnosis.
Both of these user interfaces are driven by the clinical knowledge base which is at the heart of the PEN&PAD CDES. This paper will focus on the approach to the representation of clinical knowledge including: discussion of taxonomies of concepts; presentation of medical knowledge to the user; and the use of pragmatic knowledge in the construction of data entry forms.
Taxonomies of concepts
The clinical knowledge is represented using the GALEN approach to clinical terminology2. A version of the GALEN Terminology Server, known as the TSM, has been developed, which allows a model of terminological and pragmatic clinical knowledge to be entered and then used to drive the user interfaces. The starting point for the development of the model used in the PEN&PAD CDES is the Concept Reference (CORE) model developed in the GALEN project. The CORE model contains the high-level conceptual framework as well as most of the anatomical concepts used in the application. This model has been extended to include a taxonomy of a wide range of symptoms and signs as well as some initial diseases with more currently being added.
Body locations
The model used in the PEN&PAD CDES contains a taxonomy - classification structure - of body locations which includes both internal and surface anatomy, for example, stomach and epigastrium. Locations are initially placed in taxonomies according to what they are, rather than their position in the body or in a body system. For example, the veins, lymph vessels and airways are modelled as kinds of tubular hollow body structures which transport substances: blood, lymph and air. The fact that they are components of different body systems can also be superimposed on the concepts so that they are both tubular hollow structures and parts of the cardiovascular system, lymphatic system and respiratory system. As well as defining that specific components can be part of the major body systems, it is also possible to define the sub-locations of single structure, for example, that a long bone has a proximal end, distal end, and shaft.
Pathological phenomena
Symptoms, signs and diseases are concepts, and are entered into the model at positions in a taxonomy which is automatically maintained by the TSM. However, their main initial axes of classification are not as symptoms, signs and diseases but as processes and lesions. That is, some symptoms, signs and diseases can be considered to be processes (for example, cough, tenderness and atrial fibrillation), whereas others are lesions (for example, lump, palpable abdominal mass and carcinoma).
The distinction between symptoms, signs and diagnoses is based on the role of the concept in the consultation: a symptom is usually reported by the patient; signs are observed by the doctor; and a diagnosis represents some form of conclusion drawn by the doctor. It is possible for some things, such as lump, to be signs and symptoms, and it is even possible for some things, such as abscess, to be diagnoses also. In this way, a further taxonomy is superimposed on the original, giving rise to a multiple hierarchy where most concepts have more than one parent. This means that a lump reported by the patient (lump as a symptom) is classified as both a kind of lesion and a kind of symptom.
Presenting terminological medical knowledge to the user
The terminological model of clinical concepts drives the user interfaces in the PEN&PAD CDES. The three main manifestations of this are:
Combining pathological phenomena and location
Much of medicine is concerned with pathological phenomena in locations of the body. A key aspect of the model is the use of rules - known as sanctions - which define what pathological phenomenon may occur in particular body locations. Thus a fracture may occur in a bone but not in the liver. These sanctions define what new concepts may be created when the user selects a symptom (or disease) and a location when using the Topic Selector. Using sanctions, the Topic Selector ensures, therefore, that only terminologically permissible combinations can be selected and entered into the patient record.
The location selected in the Topic Selector may need to be further refined on the data entry form by selecting a sub-location and/or a laterality. The model contains sanctions which define the sub-locations of body locations (if any) and the laterality (if applicable).
Descriptors
Sanctions are also the basis for modifiers - or qualifiers - which may be attached to pathological phenomena and locations. Modifiers are important clinically, and include general things like severity, onset, progress, duration, aggravating factors, relieving factors, etc. Other modifiers are more specific to the particular phenomenon, for example, the consistency and mobility of a lump, the colour and texture of a mole, etc. Modifier sanctions drive the data entry forms producing most of the Descriptors section of form topic, the associated symptom pop-up forms and the signs details in the Examinations section.
In addition to modifiers, the other important group of descriptors represents the cases where there are causal relationships between the topic and other pathological phenomena. For example, if the form topic is RASH then the descriptors section includes PAIN and ITCHING meaning that the rash is causing pain and/or itching. In more natural language, this would translate into "Painful rash", "Itchy rash" or "Painful itchy rash".
Natural language
Because the modifiers and other descriptors have a regular structure, rules about word order and use of the descriptors can be used to produce a more readable natural language expression. The sanctions are also the focus of these linguistic rules, which will produce expressions such as:
"Severe productive cough for 3 days, getting worse"
rather than:
cough severity: severe
productive productive
character:
duration: 3 days
progress: worse
which appears on the data entry forms.
Pragmatic knowledge in data entry forms
As well as the descriptors of the topic, data entry forms contain a collection of other concepts relevant to the topic: associated symptoms, examinations and some common diagnoses. The relationship between the form topic is not terminological, but is based on pragmatic knowledge about what might usually be relevant to the topic.
Associated Symptoms
For a form topic of COUGH, it may be relevant to record the presence or absence of associated symptoms: wheezing, haemoptysis, chest tightness, and dyspnoea. If noting only the presence or absence any of these symptoms is not sufficient, then the 'more' buttons provide access to pop-up forms on which greater detail may be entered. Of course, many other associated symptoms are possible and these may be selected using a graphical symptom selector based on the Topic Selector.
Examinations
Examinations are more complex because they consist of collections of signs. Broadly there are two kinds: examinations of body systems, and examinations of body locations. Examinations are designed with a nested structure allowing the entry of summary statements of in some cases NORMAL and in others ABSENT or PRESENT while at the same time allowing greater detail to be added where necessary. For example, in the respiratory examination:
the presence or absence of Sputum, Wheeze and Crackle, and that Breathing and Percussion are normal can be entered. The 'more' buttons provide access to pop-up forms on which greater detail may be entered when required. Again, it is possible that other examinations would be carried out, and these may be accessed from the main form.
Diagnoses
A list of common diagnoses are included on the form in order to simplify data entry. If the desired diagnosis is not in the list, then it can be selected using a graphical user interface based on the Topic Selector. As with the presenting symptom, the selected topic can be accepted on its own or further details added. For example, if fracture of femur was selected then the form would contain descriptors for the sub-location, laterality, severity, the form of the fracture, open/closed, etc.
Conclusion
The user interfaces of the PEN&PAD CDES are based on models of terminological and pragmatic clinical knowledge. The terminological knowledge supports the graphical Topic Selector which allows the selection of combinations of symptoms (or diseases) and a body location, and the descriptors on the data entry forms which allow greater detail to be entered on symptoms, signs and diseases. The pragmatic knowledge collects together on data entry forms associated symptoms, examinations and diseases which may be relevant to the form topic. All individual concepts - symptoms, signs and diseases - entered by the user are also classified automatically in relation to the existing model of medical knowledge. This classification provides the basis for advanced search and manipulation facilities.
References
1 Nowlan WA. Clinical workstation: Identifying clinical requirements and understanding clinical information. International Journal of Bio-Medical Computing 1994; 34:85-94
2 Rector A, Zanstra P, Solomon D, and The GALEN Consortium. GALEN: Terminology Services for Clinical Information Systems. In Health in the New Communications Age. Laires M, Ladeira M, Christensen J. IOS Press, Amsterdam, 1995:90-100