mental disorders overlap to a large degree (Regier, Farmer, Rae, Locke, Keith, Judd, and Goodwin, 1990; Boyd et al., 1984). The study of epidemiological data on age of onset for co-morbid disorders (to determine which of two disorders came first) has begun only recently (e.g., Merikangas, Eaton, Angst, Kraemer, Canino, Rubio-Stipec et al., 1993). The relative risks for symptoms of one disorder, or for the disorder itself, in predicting first onset of a second disorder, are probably not as high as for the precursor signs and symptoms that specifically belong within the diagnostic cluster of the second disorder; but these relative risks may be higher than other risk factors, and sufficiently high to warrant use in screening in some situations. For example, the relative risk for first onset of major depressive disorder for those with a panic attack is 3.4, as estimated in a time-dependent proportional hazards model (Andrade, Chilcoat, and Eaton, 1993). The prevalence of panic attack is about 10 percent (Eaton, Dryman, and Weissman, 1991). Applying the formula for attributable risk yields an estimate of 19 percent. The distinction between indicated preventive interventions and treatment interventions can be further clarified as the precursor symptoms of coexisting disorders become more understood.
The transition to adulthood is poorly understood, in spite of the fact that it is probably the age period when most adult disorders have their peak rates of incidence. There are prevalence surveys of mental disorders in adults, such as the ECA study, and in children, such as those reviewed in Table 5.8. The ECA study is the only one in the United States that estimates incidence of specific disorders in adults, and there is only one study estimating incidence rates for a DSM-III disorder in a population under age 18 (Lewinsohn et al., 1993). There are no studies that estimate incidence of specific disorders during the age period of the transition to adulthood, that is, from about age 15 to about age 25. Studies under way using synthetic cohort designs (in which the effects of aging can be studied by using a cross-sectional design that includes a range of ages but requires strong assumptions about the effects of birth cohort and historical period [Beltes, Cornelius, and Nesselroade, 1979]) have the capacity to yield estimates of incidence, and to study the effects of risk factors that are relatively close to the time of the beginning of the study. But synthetic cohort designs will not be able to study the effects of combinations of these risk factors and ones that occur later. Because these combinations may be very important in the transition to adult