Computer-Aided Diagnosis and Negligence

Abstract

The phrase ‘expert system’ has been widely used to describe computer systems which are capable of performing at or near to the level of an expert. Expert systems may be recognized by their construction or by their performance. Structurally, expert systems usually comprise a knowledge-base, an inference engine and an interface with the user as shown in Figure 1. The knowledge-base contains the raw material of the expert system; the rules and facts representing the expertise. An important part of that knowledge-base usually will be heuristic in nature and this is particularly so with respect to expert systems in medicine. A large amount of a specialist's knowledge is informal and experiential in nature and this heuristic knowledge is often what sets the specialist apart from the general practitioner or indeed sets the latter apart from the medical student. The inference engine is a computer program which attempts to resolve the user's enquiries by operating on and interacting with the knowledge-base. Finally, the interface with the user serves two purposes: first, to make the system relatively easy to use and second, and very importantly, to provide an explanation and justification for the results, advice and suggestions obtained from using the system (Winfield, 1982). There will be other parts to an expert system which are used to refine and modify the knowledge-base. The performance of an expert system is of utmost importance and it is the most fundamental test for whether a computer system falls into this classification. d'Agapeyeff (1984) defines expert systems as being: ‘programmed to a significant extent, from an explicit representation of empirical human knowledge; readable by those who provided the knowledge and, potentially, by similarly knowledgeable users and managers; able to provide explanations of their reasoning on demand; quickly alterable with (comparatively) low risk of unwanted side effects.’ d'Agapeyeff describes the knowledge-base as being empirical, suggesting that it is informal or heuristic knowledge rather than comprising clear and formal rules as written in textbooks. This does not preclude the inclusion of formal knowledge in the expert system but it is clear that heuristic knowledge occupies a key role in the development of expert systems. Much medical knowledge is heuristic in nature; for example, an experienced doctor might use a rule in diagnosis such as: ‘If symptoms A and B are observed then C is plausible but certainly not D’. In the field of medicine, expert systems will enable specialist expertise (a rare commodity) to be available to non-specialists such as general practitioners. Potentially, a general practitioner can have the skill and experience of specialist consultants at his fingertips. The most famous early medical expert system is MYCIN (Buchanan and Shortliffe, 1984) which is used to assist in the diagnosis of blood infections and to suggest treatments. Of course, many expert systems in medicine and in other fields have been developed as research or training tools and are neither intended nor designed to be used in a real decision-making capacity. However, expert systems technology has developed to such a state that it is increasingly likely that they will be used in the future by a general medical practitioner when advising patients or diagnosing a patient's condition. This in itself may be a very good thing and may lead to the earlier diagnosis of ailments bringing the benefits to the patient in terms of quicker appropriate treatment or earlier referral to a consultant or the selection of the most effective subsequent tests. There are, however, some serious drawbacks which may have important legal consequences if the advice or diagnosis obtained by a general practitioner using an expert system is defective and the patient suffers injury or pain as a result. The defect may relate to a fault in the expert system itself or it may be a result of misinterpretation by the general practitioner; for example, he may have misunderstood the scope or limitations of the system he is using. Following a description of the basic principles of medical negligence which are relevant to expert systems and the typical relationships involved in the use of an expert system, the liabilities arising from defective advice obtained from a faulty expert system are discussed together with the general practitioner's position; the latter may still find himself liable if the fault lies within the expert system itself. Then, the legal position in circumstances where the system itself is operating satisfactorily is explained. For example, the defective advice may have resulted from the inappropriate choice of or use of the expert system. Finally, looking somewhat into the future, the potential liability deriving from a failure to use a proven and available expert system will be discussed.