The Prediction of Suicide in Schizophrenia

In view of the fact that the suicide rate among previously diagnosed psychiatric patients is several times that of the general population, the investigation of such groups for factors leading to increased suicidal risk is of great practical and theoretical importance. The present study sought to uncover such factors through the use of case history data collected at the time of admission on 361 schizophrenics hospitalized at the Henry Phipps Psychiatric Clinic. These cases were subsequently followed up for a period of 5 years and the incidence of suicide (12 out of 361) was determined. After deletion of all follow-up information, the case folders of the 12 subsequent suicides were interspersed among those of a random sample of 75 cases drawn from the remaining 349 controls. These 87 cases were then reviewed “blindly” by two experienced psychiatrists who rated each for degree of suicide potential on an 11-point scale. Sociodemographic variables and other diagnostic data were also recorded. Subsequent multivariate analyses were concerned with determining the raters' accuracy as well as with retrieving their strategy; with determining the predictive value of isolated clinical and sociodemographic signs; and with assessing the predictive effectiveness of an optimally weighted composite when applied to the entire series of 361 cases. The results indicated that although none of the signs and ratings, either singly or in combination, yielded individual accuracy of prediction superior to classification by base rate, the clinical judgments of the psychiatrist raters were significantly correlated (p < .01) with outcome. Moreover, retrieval of the raters' strategy via multiple regression resulted in an equation affording accuracy of prediction equal to the original. As expected, the number of previous suicide attempts was the most important single variable associated with outcome as well as the one which most influenced raters' judgments. Previously published "risk" scales were found largely ineffective in this context. Although the substantive results could be considered disappointing in that hardly more than 5 per cent of the variance of the suicide/non suicide dichotomy could be accounted for in the entire series of 361 cases, such an appraisal overlooks the value of identifying high and low risk groups for further study and possible intervention. Moreover, the ready availability of a mathematical substitute for expert clinical judgment renders such pursuits considerably more feasible.