cusses 1) the problem of computing lifetime risk of suicide and describes an appropriate methodology, 2) the problem of identifying suicide clusters, 3) statistical approaches that can inform suicide research, and 4) issues in the design of suicide studies. These issues include, case-control studies, risk-based allocation, and sample size and statistical power. Technical details of the statistical models are presented in Appendix A. The methods described here are by no means an exhaustive list of potentially useful approaches in the analysis of suicide data. It is hoped that these examples will provide a perspective on the power of appropriate statistical methodologies in suicide research.
Based on the work of Guze and Robins (1970), much of the psychiatric literature purports that 15 percent of depressed patients will die by suicide. To better understand the foundation of this estimate it is important to understand the various ways in which lifetime risk can be computed. In the case of Guze and Robins (1970), lifetime risk is defined as the proportion of the dead who died by suicide, often termed “proportionate mortality” (see also Goodwin and Jamison, 1990). As pointed out by Bostwick and Pankratz (2000), proportionate mortality is a reasonable estimator of lifetime risk only when the participants are followed until death. In general, however, the studies synthesized in the report by Guze and Robins, typically followed patients for no more than a few years. Furthermore, the participants were hospitalized psychiatric patients, often hospitalized as a precaution for suicide. Both this selection effect and the use of proportionate mortality as an estimator of lifetime risk, lead to an increase in the estimated lifetime prevalence. To obtain a more accurate assessment, Inskip, Harris, Barraclough (1998) calculated percent death by suicide to percent dead overall in a large number of studies. Analyses were stratified by diagnostic group (alcohol dependence, affective disorder, schizophrenia). Unfortunately, the majority of these studies had overall mortality rates of less than 50 percent, so the estimates of lifetime risk (i.e., 100 percent mortality) were extrapolated from the available data. Nevertheless, the lifetime suicide risk estimates were 7 percent for alcohol dependence, 6 percent for affective disorder, and 4 percent for schizophrenia.
In the most statistically rigorous approach to date, Bostwick and Pankratz (2000) compared proportionate mortality to “case fatality prevalence,” which is the number of suicides divided by the total number of patients at risk. Based on a synthesis of 29 studies of hospitalized affective disorder inpatients (19,723 patients), the pooled estimate of proportionate mortality prevalence was 20.0 percent, but only 4.1 percent for case fatal-