Why clinical information standards matter

Most clinicians probably know clinical information standards as the coding systems used for representing clinical concepts. They include longstanding systems such as the International Classification of Diseases and the Diagnostic and Statistical Manual of Mental Disorders as well as more recent systems such as the Read Clinical Classification and Snomed, though coded vocabularies themselves are only the visible tip of the standards iceberg. Work on clinical information standards is not glamorous, but such standards are fundamental for progress in health care. They are essential resources for future clinical decision support, audit, governance, research, education, and training. In a very real sense, these standards define what can be said about patient care.

In this issue Brown et al describe a crossover trial comparing the accuracy and usability of two clinical terminology standards in a setting designed to reflect tasks and circumstances commonly encountered in contemporary British general practice (p 1127).¹ The paper may seem of limited relevance, in that it is focused on Britain, on primary care, and on two specific clinical standards, both of which are probably obsolescent. Also in this issue is a report describing the practical use of clinical information coding for managing diabetes in south London primary care (p 1130).² Again, this is a paper whose main focus is on British primary care.

However, these papers do have wider relevance, as much for the general questions they raise as for the particular questions they answer. For example they raise questions about what clinical information standards doctors need, who should develop them, and how they should be evaluated.

What standards do we need?

The answer to the first general question—what standards do we need—is easy (though hard to execute). We need three types of clinical information standards: document structuring standards; term lexicons; and ontologies.

Document structuring standards specify what information items a particular clinical document must, may, or cannot contain; how the items are represented; and where in the document the items are to be found.^3–5. For example, a structuring standard for an operating theatre list might specify that for each patient there must be precisely one field called “date of birth,” containing a value of the form “1970-02-17” and appearing immediately after a field called “sex.”

Term lexicons are sets of coded terms. Their primary purpose is to resolve synonymy. Thus a term lexicon may say whether “myocardial infarction” should be taken to mean the same as “coronary thrombosis.”

Ontologies represent domain knowledge in a form that can be exploited by computers. For example, an ontology covering surgical procedures might allow a computer program to deduce that a “flexible bronchoscopy” is a type of investigation of the lungs that can be done as a day case.

The answer to the second general question—on who should develop standards—is difficult. Organisations that might develop clinical information standards include national or international standards organisations^6–8 (as with the International Classification of Diseases, for example); health related government agencies (as with Read codes); academic research groups⁸; professional colleges and societies; commercial companies or consortia⁹; charities; health trusts; or combinations of these, as with Snomed Clinical Terms.¹⁰

Large organisations have large resources but may be slow, bureaucratic, and unfocused, whereas small organisations are often dynamic and focused but may not have the resources to distribute and maintain standards. Commercial companies have extensive experience of implementing standards in clinical information systems—for example, by incorporating facilities within their systems for browsing and selecting clinical terms from large coded vocabularies—but they cannot be expected to fund efforts which are not commercially beneficial. Clinicians understand the domain, but may have limited theoretical or practical knowledge of information representation technologies, whereas computer scientists cannot be expected to have insight into clinical working practices.

The answer to the third general question, about evaluation, is also difficult. Standards can be judged only if there is a precise statement of their intended purposes, and such purposes may be inherently imprecise and evolutionary. Different types of standards are interdependent in practice, though standards are usually developed and evaluated in isolation. Evaluation in realistic clinical circumstances is the ideal, as in this week's comparison by Brown et al, but by this stage the cost of development will already have been incurred.

Since the three types of clinical information standard are highly interdependent, there should be greater coordination in their development. Although the number of individuals active in clinical standards work is small, it is important that those who evaluate and assess standards should be independent of those who develop standards.

Building ontologies is complex and challenging, but they are crucial for next generation clinical information systems. Lexicons such as Read Codes version 5, Clinical Terms version 3, and Snomed Clinical Terms incorporate some ontological content, but their primary purpose is to enumerate terms, and their ontological content is simply not sufficient to support sophisticated reasoning in future computerised clinical information systems, which must provide much greater clinical decision support than is found in current systems. More work is required on rich ontologies.

It is tempting for policymakers to believe that clinical information standards (and clinical information technologies in general) are now nearing maturity, and therefore that an appropriate strategy is to identify the best current systems and promote their adoption.¹¹ However, there is little evidence to suggest that we are approaching a plateau of maturity in health information standards or health information technologies generally. On the contrary, the next decade will be characterised by accelerating and unpredictable innovation. In such circumstances we should accept the need for large scale experimentation in health information systems. Brown et al have shown that even small scale studies can show significant benefits of one standard versus another. In itself, their study perhaps does not justify wholesale abandonment of Read Codes version 5 in favour of Clinical Terms version 3, but clearly there is a need for more widespread independent clinical evaluation of standards.