The information value of clinical data

Abstract

Decisional aids are dependent on the accuracy, comprehensiveness and decisional value of the observations in the database and the test subject. Diagnostic data stem from the observations of the patient, the doctor and third party tests. An information-utility index was used to compare the diagnostic contribution of these sources. In a diagnostic database of 314 cases of jaundice, twenty-two findings of high diagnostic usefulness were discovered among 6 diseases. Sixteen of these (73%) were patient's observations, 4 were doctor's (18%), one the patient's age and one a ward test. Five laboratory tests had lesser average scores. In three databases--on jaundice, abdominal pain and low back pain the 7 commonest diseases were examined, totalling 1018 patients. In a simple Bayesian procedure, the patient's observations alone correctly classified 50%, the doctor's observations a further 16%, the patient's age and sex another 4% and the prior probability of the diseases, an additional 4%. These figures varied considerably between the diseases. When all 21 diseases with more than 10 cases in the database were considered, the respective figures were 53, 15, 3 and 3%. A European Community medical research initiative is collecting data on 10,000 cases of jaundice and of acute abdominal pain to establish an adequate diagnostic database for the development of decision support systems.