A decade of declining adolescent suicide rates came to an end in 2004 with an 18% increase in adolescent suicides from the previous year (1). This disturbing increase in the adolescent suicide rate coincided with publicity about the relationship between antidepressant treatment and suicide risk in children and adolescents and a subsequent decline in antidepressant prescriptions. These concerns eventually resulted in a black box warning from the Food and Drug Administration (FDA) that was based upon a carefully conducted meta-analysis of pediatric antidepressant trials that showed a higher rate of spontaneously reported suicide-related events in subjects assigned to antidepressants versus placebo (2). Three articles in this issue of the Journal directly grapple with these two apparently contradictory sets of observations: the risk for suicidal behavior conveyed by antidepressant treatment versus the potentially protective effects of antidepressant treatment against suicidal behavior.
The FDA recognized that the classification of suicidal events in extant clinical trials was haphazard and potentially misleading. Therefore, in order to most accurately assess the risk for suicidal ideation and behavior associated with antidepressant treatment, the FDA commissioned Posner and colleagues, who developed a clinically meaningful and reliable classification system for suicidal events that served as the basis for the FDA meta-analysis. One difficulty that cannot be overcome with this classification system is that only spontaneously reported adverse events were available for classification. It is unclear if what is being measured is the impact of medication on the threshold for spontaneous reporting between treatments or on actual suicidal ideation and behavior. In fact, the same FDA analyses that showed an increase in the risk for spontaneously reported suicidal events associated with antidepressants showed a tendency toward a protective effect against new-onset and worsening suicidal ideation in the subset of trials for which suicidal ideation was assessed systematically. The consistent classification system described in the Journal can truly quantify the impact of treatment on suicidal risk only when it is consistently and prospectively applied.
Observational studies of treatment can provide additional clues about the relationship between treatment and suicidal behavior, although the results have been contradictory (3, 4). Simon and Savarino provide an elegant illustration of the need to consider the previous history of suicidal behavior and its role in referral when evaluating the relationship between treatment and subsequent suicidal behavior. They examined a large database of outpatient claims from a prepaid health plan and compared the rate of suicide attempts in the 90 days before and the 180 days after initiation of antidepressant or psychotherapy treatment. The peak period of attempts was in the month before the initiation of treatment, regardless of the setting or type of treatment. The authors interpret these findings to mean that a suicide attempt was a common precipitant for referral to treatment and that the decline in suicide attempts after treatment initiation was due to nonspecific therapeutic factors.
Gibbons and colleagues take a similar approach to that of Simon and Savarino by examining the rate of suicide attempts before and after the initiation of treatment in the Veterans Administration (VA) health care system. In contrast to the study by Simon and Savarino, Gibbons and colleagues have a comparison group with a diagnosis of depression that did not receive antidepressant treatment, although it could not be determined with the extant database whether the patients in any of the groups received psychotherapy. All of the treatment groups, regardless of the type of antidepressant prescribed, showed large decreases in the rates of suicide attempts after initiation of treatment. In addition, Gibbons and colleagues also showed that the rates of suicide attempts in patients treated with an antidepressant were roughly one-third of those observed for patients who were not treated with an antidepressant. It is possible that, as posited by Simon and Savarino, these findings are at least in part due to a high rate of referral for treatment of depression shortly after a suicide attempt. However, the decline in risk for suicide attempt in the treated compared to the untreated group makes a stronger case that such changes are due to the protective effects of treatment per se. Another possible explanation for the higher rate of suicidal behavior in the untreated group is that lack of treatment may be due to lack of adherence and concomitant psychosocial factors (e.g., personality disorder, alcohol abuse) that may increase suicidal risk. Although the FDA’s analyses of randomized trials found an elevated rate of suicide attempts in young adults ages 18–24 treated with drug versus placebo, in this study, within the same age stratum, the risk of suicide attempt in the treated group was one-third that of the untreated group (4). It is possible that the divergence between the FDA’s analysis and the report by Gibbon et al. is accounted for by a difference in the baseline characteristics of the two samples: patients at high risk for suicidal behavior are excluded from clinical trials, whereas such patients were highly represented in this VA-based observational study.
Observational studies can never definitively demonstrate causality. However, the results of both studies by Gibbons et al. and Simon and Savarino are consistent with a protective role of treatment against emergent suicidal behavior, and perhaps just as important, these data show a pattern that is exactly the opposite of what one might expect if antidepressants were associated with increased suicidal risk.
One of the primary mandates for physicians is primum no nocere—first do no harm. This collection of articles in the Journal advance that mandate. Harm can be caused inadvertently by the iatrogenic effects of treatments as well as by clinicians’ reluctance to use effective treatments. Clearer classification of suicidal behavior leads to more consistent assessment and communication about risk so that the prevalence and severity of “harm” can be articulated. However, the assessment of risk attributable to antidepressants requires a systematic application of this classification approach and cannot rely solely upon the retrospective recall of spontaneously reported events. The two articles examining the relationship between treatment and suicidal behavior, while not proving that antidepressants and psychotherapy reduce suicide risk, do strongly support the converse conclusion: it is much more likely that suicidal behavior leads to treatment than that treatment leads to suicidal behavior. With more specific and systematically assessed information about suicidal behavior, clinical trials can be more informative about suicide risk. Moreover, as both articles by Simon and Savarino and Gibbons et al. demonstrate, the rate of suicidal behavior in depressed patients initiating treatment is quite high before any exposure to antidepressant or psychotherapeutic treatment. In order to advance our treatment of real patients in real world settings, we need to conduct clinical trials that include, rather than exclude, these high-risk patients. Only by empanelling depressed patients at significant suicide risk into randomized trials and then systematically assessing the impact of treatment on suicidality and depression will we be able to delineate the effects of antidepressants and psychotherapy on depression and suicide risk. In the meantime, the disturbing increase in the suicide rate in adolescents at a time when antidepressant treatment is becoming less frequent in this population should serve as a warning that the consequences of not receiving treatment for depression may be fatal.