Assessing the clinical impact of a predictive system in severe head injury

Many computer-assisted diagnostic or prognostic systems have been described in the literature, but as yet such systems have had little impact upon routine clinical practice. There are many possible explanations for the limited success which these systems have enjoyed, but one of the most significant must be the dearth of well-designed and carefully executed trials to assess their performance. Many evaluations only address the scientific question of whether the diagnoses or prognoses are accurate; they stop short of answering the crucial pragmatic question of whether the system benefits patients, i.e. is clinical care improved by the use of such a system? This brief paper describes a predictive system in severe head injury which has been developed by an international group based in the Institute of Neurological Sciences at the Southern General Hospital, Glasgow. The various steps involved were first the establishment of a clinical databank according to a strict protocol and with carefully designed scales to quantify the impairment of consciousness and outcome, the development of prognostic models relating outcome to initial severity, the comparison of the performance of this model with predictions made by clinicians, and finally a multi-centre prospective trial assessing the clinical impact of the system. The results of this trial are not yet available, but a consideration of the design of the trial raises important points which are relevant to many other clinical fields where 'decision aids' need to be evaluated.