Computer aided decision making in medicine.

It is now 30 years since Nash's patent Logoscope was introduced as a mechanical aid to diagnosis' and 25 years since mathematical analysis of rational clinical judgment was established as a possibility.2 The aims and principles of computer assisted decision making were laid down by Lee Lusted and Wilfrid Card in the 1960s,34 and over 1000 papers have now been published on this subject, with their scope ranging from single problems such as alcoholism5 and allergy6 to an ambitious attempt to tackle 75% of internal medicine in a unified program.7 Yet, though computers are established in signal processing, data analysis, and physiological modelling and are slowly coming into record systems, they have been used routinely only rarely for explicit help in medical decisions. Recent articles, however, argue that a critical time has arrived.' 9 Four types of programs should be distinguished: algorithmic, automatic interpretation, actuarial, and advice systems. The first of these are simply transpositions to a computer of flow charts for good clinical practice, such as those recently appearing in the BMJr or those for rural health workers in developing countries.'0 The main use of such computer algorithms is in home medical programs such as The Complete Guide to Medicine (for only £6-90), though in the United States they have been adopted for complex protocols for cancer treatment."I The second type of program performs an automatic interpretation of clinical measurements already collected on a computer, such as the recent suggestion for on line detection of rejection of renal transplants.'2 Considerable experience has been gained in automatic interpretation of electrocardiograms, and commercial systems have recently become cheaper and more widespread.'3 The problems of reliable analysis have so far, however, prevented extensive applications in monitoring intensive care,'4 though automatic control of infusion ofdrugs has been reported."' 16 The advantage of this type of system is that it avoids any additional burden on staff and may be attached to standard apparatus, such as the computer interpretation of lung function tests routinely used in at least one clinic.'7 The third, actuarial, type of system provides a measure of the risk of some future event, given certain precisely defined