Suicide prediction revisited

Data from a 1983 prospective study of suicide in a cohort of 4800 psychiatric inpatients were reanalyzed using logistic regression, which is more appropriate for a binary outcome. The results were the same as in the previous study: too few of the subsequent suicides were identified and there were too many false positives to make this procedure useful. Several additional "artificial" logistic regression analyses were done: one series randomly removed increasing numbers of nonsuicide cases to increase the base rates; another series added an increasingly powerful hypothetical "test." Both of these maneuvers helped, but fell well short of perfection.