An assessment of the utility of suicide prediction.

Efforts to predict a future event assume varying levels of confidence depending on its base rate and the error rate of the prediction instrument. Most researchers working with suicide prediction instruments seem tacitly to assume they will be able to predict a future suicide most of the time. Applying basic decision theory on a neuropsychiatric hospital population indicates that researchers using a prediction schedule will be unlikely to predict a future suicide beyond a 20% level of efficiency. Contrary to the general clinical view, eliminating false negatives was shown to be more practical than eliminating false positives in increasing the efficiency of a predictive schedule.