COMPUTER-ASSISTED PREDICTION OF OUTCOME OF SEVERELY HEAD-INJURED PATIENTS

Abstract An ability to predict at an early stage the outcome of severely head-injured patients would have several uses. A system has been developed which is based on information obtained by bedside clinical examination. The predictions are based on a data bank collected prospectively since 1968 and which now contains details of over 2000 patients. Although up to 300 items of data were collected on each patient, the use of only eight indicants (the patient's age and features describing the severity of coma) gives predictions that are not improved by the addition of further information. An independence model based on Bayes theorem has been found to be as useful as more complex statistical models. Predictions can be made at different intervals, e.g. at 24 hours, three days, and seven days from the onset of coma. Probabilities are assigned to each of three outcome categories, ‘dead or vegetative’, ‘severe disability’ and ‘moderate disability/good recovery’. There are versions of the program in FORTRAN and PASCAL which run on a North Star Advantage microcomputer under CP M ; the results are printed and provided to the clinician in numerical form or graphically. Surveys of clinician's views about prognosis show a considerable degree of uncertainty. A comparison of the ability of clinicians to predict outcome, using case histories of patients selected randomly from the data bank, showed varying accuracy compared with the true outcome and with computer predictions. Attitudes to computer predictions are being surveyed. In future, real-time clinical predictions will be made, will be compared with simultaneous computer predictions, and computer predictions will be provided as a routine ‘service’ to neurosurgeons.