Several of our findings deserve to be highlighted: (1) In adjusted analyses, almost all temporally prior DSM-IV mental disorders were positively associated with the subsequent first SA. Highest elevated risks were found for PTSD (> 15-fold), dysthymia and panic disorder (> sevenfold), and GAD as well as OCD (each > fourfold). (2) Among the comorbidity categories of no prior mental disorder, 1, 2, and 3 or more prior disorders, on average each increasing unit was positively associated with the subsequent first SA (> eightfold). Risk elevations were threefold for 1 disorder, sevenfold for 2 disorders, and 19-fold for 3 or more disorders. (3) At the minimum, more than 50% of first SAs were preventable in the exposed group if the disorder had been causal and prevented. At the maximum, over 90% of first SAs were preventable in the PTSD group. (4) Half of all DSM-IV disorders each accounted for more than 11% of first SAs in the total sample. Total incidence reductions of first SA by more than 20% and even over 30% were possible, if nicotine dependence or specific phobia had been prevented, respectively.
Our results are best compared to those of the NCS-A study by Nock et al.  and the MAMHS study by Borges et al. . They also used a representative sample of adolescents and young adults, included a wide range of specific DSM-IV mental disorders as risk factors for the subsequent first SA, and analyzed data with a discrete-time survival model based on retrospective AOO information. Our 5.5% overall cumulative lifetime incidence of SA is somewhat higher than the prevalence of 4.1% found by Nock et al.  and the prevalence of 3.1% found by Borges et al. . A possible explanation for our higher estimate might be that our sample included older individuals up to age 34 and that we used prospective-longitudinal data for the estimation of the prevalence . In good agreement with Nock et al.  and Borges et al.  as well as with our earlier analyses [10, 18], we observed a higher cumulative incidence of SA in women than in men.
In our study, PTSD strongly elevated the risk for a subsequent first SA. In the NCS-A data , PTSD was the only anxiety disorder with an elevated risk, whereas in the MAMHS data  PTSD was not predictive. When turning to representative community studies of adults, PTSD is also inconsistently reported to elevate the risk for (first onset) SA [for a review see 35]. In a recent community study of adults in South Africa, prior PTSD was found to be the strongest predictor for the subsequent first SA (adjusted for other DSM-IV mental disorders) . In a cross-national analysis of representative adult samples from 21 countries around the globe, prior PTSD strongly elevated the risk for the subsequent first SA in both developed and developing countries , thus suggesting that ours was not a chance finding. Panic disorder was predictive in our study but not in either the NCS-A study  or the MAMHS study . Among the anxiety disorders evaluated in the CHDS study , panic disorder elevated the risk for an SA in subjects 16–25 years old. Of the three anxiety disorders, it was the only one that remained significant across several sets of confounders, for example, other disorders and life stress. In a review on anxiety disorders and risk for suicide, Sareen  concluded that especially PTSD and panic disorder have often been found to be independent risk factors for SA, which is supported by our results for adolescents and young adults.
Among the affective disorders, the risk estimate of dysthymia was relatively large, compared to MDD and any bipolar disorder. In the MAMHS study , dysthymia had the highest risk estimate of all disorders analyzed. In the NCS-A data , the combined group of MDD/dysthymia elevated the risk for the first SA, yet the compound diagnosis prevents direct comparison to our results. In the CHDS data , affective disorders were reported as a compound diagnosis elevating the risk for an SA in 15–21 year olds, also preventing further comparisons to our results.
DSM-IV substance disorders performed relatively weakly in predicting the subsequent first onset of SA. In our study, drug abuse/dependence was predictive, yet alcohol abuse/dependence was not. In the NCS-A study , no substance disorder reached the significance level and in the MAMHS study  only drug abuse was predictive. However, nicotine dependence yielded a considerable risk estimate in our study, requiring an explanation. Aside from our own previous results where both non-dependent regular smokers and dependent smokers at baseline were at higher risk for their first future SA , we found one other study of young people that took the temporal sequence of tobacco use and SA into account. Using the MAMHS data of 12- to 17-year-old Mexicans, Miller et al.  reported elevated risks for the first SA along all different tobacco-related habits [i.e., (irregular) use, weekly use, daily use, dependence] and along all different sets of confounders. In the CHDS study, Boden et al.  reported associations between the number of cigarettes per day before the ages 16, 18, and 21 years and the subsequent (not first) SA between these ages. However, the temporal associations pooled over the 3 assessment periods attenuated to non-significance when controlled for non-observed fixed factors, suggesting that even more than 20 cigarettes per day was not an independent risk factor for subsequent SA. This contradiction may be explained by the smaller sample size (between 935 and 983) and the differences in the two data analyses.
Discussing our findings in light of results from adult population samples, all our significant associations have also been observed in adult samples [e.g., 37, 38, 41]. Our results extend these findings insofar as they impressively show that even in a young community sample, empirical evidence of temporal associations exists, adjusting for various confounders.
Several studies  have suggested that risk for subsequent SAs seems to be comparable across different mental disorders. Hoertel et al. , therefore, investigated in a large population sample whether this observed elevated risk for SA could be due to a general underlying psychopathology dimension. Using structural equation modeling, this group showed that effects of mental disorders on risk of SAs seem to occur through an underlying common general psychopathology dimension. On the other hand, Nock et al. , who also found associations between virtually all included mental disorders and subsequent SA in the World Health Organization World Mental Health Survey, demonstrated that the point risk estimates, vary remarkably by type of disorder. Given this variation in estimates, Nock et al.  interpreted their findings as arguing against an underlying common psychopathology dimension. Our results are in accordance with Nock et al.’s , since our risk estimates also differ remarkably across different disorders. Whether the associations between a variety of prior mental disorders and subsequent SAs are a result of a common psychopathological dimension is beyond the scope of our paper. This surely interesting question should be addressed in future analyses.
We also found that the number of DSM-IV mental disorders was a risk factor for the first SA in our study. Having 1, 2, or 3 or more mental disorders all elevated the risk for a subsequent first SA. This finding is in relatively good agreement with other community surveys [e.g., 43, 44]. Comparable to earlier EDSP analyses based on cross-sectional baseline data , our data even support a clear dose–response relationship between number of comorbid mental disorders and the first SA.
AFs have not been estimated so far in studies evaluating the risk of mental disorders for the first SA. It certainly seems impressive that any of the mental disorders that elevated the risk for the subsequent first SA bear the potential of reducing this tragic outcome by at least 57% (drug abuse/dependence) and up to 93% (PTSD). Moreover, anxiety disorders seem to be a very important intervention target in the subpopulation of adolescents and young adults, because not only is the AOO of anxiety disorders earlier compared to other groups of mental disorders but also SA might be prevented to a considerable extent.
PAFs in conjunction with our study characteristics were reported in the CHDS study . Anxiety disorders accounted for 7.5% of SAs in the total birth cohort sample of individuals aged 16–25 years. In our study, PAFs for single anxiety disorders ranged from 4.7 (OCD) to over 30% (specific phobia). Unfortunately, the results of the CHDS study  and the results of our study cannot be compared directly due to the use of the disorder group (anxiety) instead of single anxiety disorders in the former. When turning to adult community samples that reported associations between prior mental disorders and the subsequent first SA we found only one study that reported PAFs. Bernal et al.  reported that MDD accounted for 28% of first SAs in a sample of more than 21,000 adults (aged 18 years and over) across 6 European countries. GAD accounted for 4% of first SAs. The PAF of 28% is almost 3 times as much as our result of 11.5% for MDD. Most likely this is due to differences in age and other methodological characteristics. In other studies with adult community samples, PAFs were reported not for single disorders but for disorder groups only (e.g., mood disorders, any mental disorder; [46,47,48]), were based on results that did not take into account the temporal order of risk factor and outcome [49,50,51], or did not point to the subsequent first SA .
Our study points out how essential it is to consider the context of SAs, in terms of age, sex, and psychopathology. When considering SAs or suicidal behavior in general, heuristics do more harm than good; that is, inquiring about suicidal phenomena only if the (young) patient reports depressive symptoms is important, yet not inquiring otherwise might often be fatal. Furthermore, a relatively rare phenomenon such as suicide and all of its cognitive and behavioral derivatives should be approached on the population level with much more determination, as discussed by Knox et al. . The determination to prevent SAs on the population level of course is synonymous with major efforts professionally, politically, and monetarily. Our study might serve to guide the selection of promising prevention targets for the general population of adolescents and young adults. After all, “it seems likely that earlier identification and earlier symptomatic relief is an important component of the prevention and treatment of youth suicidal behavior” .
The present study has several methodological strengths. First, we used a representative community sample of adolescents and young adults with an observation period that fully included the high-risk period of both the first onset of mental disorders and the first SA. Second, our analyses are strengthened by the inclusion of all individuals who reported an SA rather than individuals who received medical attention, and by the inclusion of a comprehensive set of predictor diagnoses that allowed us to evaluate a broad range of mental disorders. Third, we used an interviewer-administered standardized interview.
Our study also has several limitations. First, our analyses were based on self-reported data, which are always prone to recall bias. However, this bias may have been lessened by our longitudinal study design of 3 follow-up assessments over a 5- to 10-year period, which decreased the time frame for retrospective assessments. Second, several disorders known to be associated with suicidal behavior were not analyzed in our study (e.g., schizophrenia, personality disorders), because they were not assessed. Third, the assessment of SA at T0 and T1 was limited to individuals who reported depressive symptoms over at least 2 weeks. Although this might have led to a more conservative prevalence estimate initially, the later follow-up waves assessed lifetime SA in all subjects. Fourth, the AFs and PAFs are preliminary quantitative appraisals of the impact of mental disorders on the risk for onset of a first SA. The preliminary status and the inherent assumption of causality warrant caution in interpreting the results.
These limitations notwithstanding, our study provides valuable new information about mental disorders and SA in adolescence and young adulthood. From a public health perspective, our AFs suggest that the prevention of mental disorders among adolescents and young adults could substantially reduce the incidence of SAs. Our results also clearly demonstrate the importance of considering not only depression but also the full range of mental disorders when evaluating the risk for suicidal behavior. Given the strong associations between comorbidity and SA, clinicians should always conduct a suicide risk assessment among patients presenting with multiple mental disorders. Finally, our results point to the need for future work to increase our understanding of the increased risk for suicidal behaviors during adolescence and of the causal pathways linking mental disorders to suicidal behaviors.