ity prevalence. As one might expect, in outpatients the rate for case fatality prevalence decreased to 2.0 percent in the analysis of 7 studies of 7,444 affective disorder outpatients. In contrast, the rate actually increased slightly to 24.6 percent for proportionate mortality prevalence.

Bostwick and Pankratz (2000) computed lifetime risk of suicide, using Bayes theorem, as the probability of suicide given death times the probability of death. For example, the overall probability of death in the 29 studies of affective disorder inpatients was 20 percent and of those, 20 percent died by suicide. The product of these two probabilities (i.e., the conditional probability of suicide given death and the prior probability of death) is the Bayes estimate of lifetime risk, which in this case is 4 percent. The Bayes estimate is remarkably close to the case fatality prevalence of 4.1 percent. This finding was consistent for all of the groups examined in their study (affective disorder outpatients = 2.2 percent, affective disorder inpatients = 4.0 percent, Guze and Robins data = 4.8 percent, Goodwin and Jamison data = 3 percent, and the general population = 0.5 percent).

Suicide Clustering

Suicidal behavior in adolescents is a major public health problem (NCHS, 1988). Data suggest that teen suicides often occur in temporal and geographic proximity of one another. This phenomenon is not unlike the concept of an outbreak of a disease in a particular community. Naturally, some clustering of suicides occurs by chance alone even if suicides occur at random. In the study of suicide clusters, the goal is to determine whether or not the outbreaks are occurring to an extent greater than would be expected by chance variation. Past studies have used various populations, such as psychiatric in-patients, high school and college students, marine troops, prison inmates, religious sects etc. (Gould et al., 1990). However, county of residence may be a more sensitive space unit to define a cluster (Gould et al., 1990).

Several statistical methods have been used to detect and statistically assess the time-space clustering of disease (see Gould et al., 1990). The Ederer-Myers-Mantel method (Ederer et al., 1964) is found to be sensitive to temporal clustering as well as time-space clustering. A method proposed by Knox (1964) considers all possible pairs of cases and the time and space distances between them. It establishes clustering by demonstrating a positive relation between the time and space distances of a pair, but required specification of the critical values for time and space to define closeness. This approach was modified by Smith (1982) to define “close in space” as occurring within the same geographic area. Wallenstein and colleagues (1989) provided a formula to assess the practical significance of clusters as well as the statistical significance. Gibbons et al. (1990)



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement