The aim of this study was to evaluate the prevalence of death by suicide and suicide attempts in antidepressant and placebo treatment groups of antidepressant clinical trials over the past 17 years and to compare these rates to those from programs conducted prior to 2000. Deaths by suicide in antidepressant-treated patients and suicide attempt rates in both antidepressant and placebo-treated patients decreased significantly in more recent programs. Overall, these data support our hypothesis that patients in investigational antidepressant programs after 2000 had a lowered rate of death by suicide and suicide attempts.
Specifically in the antidepressant group, the number of suicide deaths per 100,000 patient exposure years after 2000 has dropped by 96% of the rate pre-2000 indicating a major shift in the prevalence of depression trial suicide deaths in the past decade and a half. Mirroring this decline, the rate of suicide attempts per 100,000 PEY has also decreased by 84% in programs after 2000. While the decline over time in the suicide death rate of placebo-treated patients was not statistically significant, the significant decrease in the rate of suicide attempts in the placebo group mirrored that of the antidepressant group (an 85% decrease). These decreases were verified with significant findings from regression analyses weighted by program PEY.
These observations are intriguing since there has been a dramatic increase in suicide rates in the general US population in the years following 2000 (Curtin et al. 2016). There are several possible explanations for this decrease in suicide and suicide attempt rates in antidepressant clinical trials in the face of a rising suicide rate in the general population. The first possible explanation that we explored was the potential for enhanced treatment efficacy for suicide prevention among more recent antidepressants. However, support for this theory was not found. The decrease in suicide and suicide attempt rates observed in the antidepressant treatment group were also seen in the placebo group, indicating that medication treatment effects are not likely responsible for the overall observed decline. Additionally, our exploration of drug-placebo differences in suicide and suicide attempt rates revealed no significant differences. We would expect that if medication effects were at play in these data, they would occur exclusively in the antidepressant-treated patients and therefore we would assume that drug-placebo differences in suicide and suicide attempt rates might be more readily observed.
Another potential explanation that we explored was the possibility that key demographic features, such as age and sex, of the patient populations may have changed between pre- and post-2000 programs. In our analysis of these 14 antidepressant programs, we found that the mean age stayed within the range of 40–50 years and did not vary significantly between pre- and post-2000 programs. With regard to percent female, programs consistently enrolled around 60% female patients, with this proportion staying consistent over time. Due to a lack of reporting, analysis of other details of patient demographics (including race, diagnostic history, and familial mental health history) was not possible with these data. However, age and sex are well-established modifiers of suicidality and our findings indicate that these factors have not differentially influenced the suicide and suicide attempt rates between pre- and post-2000 programs.
Given these findings, it is not likely that this phenomenon of decreasing suicide and suicide attempt rates in antidepressant clinical trials reflects changes in antidepressant efficacy or in the demographics of the population sample. A final possible explanation that we considered involves the changes that have occurred in the design and conduct of depression trials. It is possible that this notable decrease in deaths by suicide in more recent antidepressant clinical trials may be partially due to implementation of stringent entry criteria for participation in these trials, including higher awareness of suicide risk by clinical investigators screening patients and pharmaceutical companies’ physicians and scientists. This may additionally be due in part to the increased use of operationalized and systematic suicide risk assessment tools such as the Columbia-Suicide Severity Rating Scale (C-SSRS) (Posner et al. 2011) and the Sheehan Suicidality Tracking Scale (S-STS) (Coric et al. 2009). Due to the proprietary nature of protocol details, reporting of screening procedures is limited within these data and therefore this explanation remains an untested theory. Additionally, selection biases resulting in the exclusion of more severe depression presentations (such as mixed depression or depression with psychotic features that have established increased risk of suicide) may have further suppressed suicidal patients from enrolling in trials of antidepressants.
If screening procedures have become more effective at excluding suicidal patients from clinical trials of antidepressants, then there may be an undesired result—the effects of antidepressant treatment on depressed patients with moderate or high suicidality may not have been able to be evaluated in more recent trials because these patients may have been systematically excluded. On the other hand, the ability to exclude patients with moderate or high suicidality may have a benefit in the reverse. Such effective suicidality screening procedures could be used to intentionally include patients with higher risk for suicidality. Used in this way, clinical trials treating populations at higher risk for suicidality could evaluate the short-term efficacy and the long-term prophylaxis of a candidate anti-suicidality medication. Inclusion of acutely suicidal patients in trials designed to evaluate antidepressant efficacy for suicidality reduction has been successfully modeled in a trial of citalopram combined with lithium (Khan et al. 2011).
There are significant statistical limitations of these data primarily because suicides are rare events and therefore statistical analysis on their occurrence can be severely limited by the amount of exposure. In these data, the deaths by suicide in the placebo group occurred very infrequently overall and PEY for placebo exposure was less than 15% of what it was for antidepressant exposure (see Table 1). Floor effects (in that the number of suicides cannot drop below zero) and reduced statistical power from such inadequate exposure in the placebo group may have limited the ability to find statistical significance for the decrease in the number of deaths by suicide. Because suicide attempts are much more frequent, this measure may have a better sensitivity for detecting changes in the suicidality of patients (both antidepressant and placebo treatment groups showed significant findings in suicide attempts over time).
It is important to note that these data are limited to approved antidepressants only and while suicide data from trials of antidepressants that never gained approval may exist, such data are not publicly accessible and therefore were not available to us for this analysis. Additionally, suicide data may be influenced by the duration of a study—for example in longer term studies, survivor bias may be at play. However, in the case of rare events such as suicide, more events are more likely to be observed over a longer follow-up. Conversely, shorter duration of follow-up may underestimate the number of suicide events. We would have liked to examine any possible effects of the duration of the studies included in the ISS reports; however, the respective contribution of short-term and long-term studies is not given and cannot be accurately estimated from these summary data. Because the placebo data comes primarily from acute, Phase III trials and placebo exposure is generally kept to a minimum, the average follow-up per patient is certain to be less than the antidepressant treatment group which can include open-label, long-term follow-up studies and additional exposure from acute active-controlled trials. Differences between study designs regarding randomization ratios and other pertinent trial design factors also could not be examined because these summary data were pooled. Although we would have liked to use the Poisson regression as a statistical method to examine the continuous decline in suicide rates over time, we were unable to due to the fact that we did not have access to patient-level data. The statistical limitations of rare events and low exposure, particularly in the placebo-exposed group, must be kept in mind when examining these data, but we do not believe that such limitations preclude examining these data with the methods that are available to us.
Finally, because we only have access to data in the public domain (the ISS reports as presented via the Freedom of Information Act on the FDA Access Data website), we are only able to accurately examine suicidal behaviors that are recorded and reported consistently over time. While completed deaths by suicide and suicide attempts have always been considered by the FDA to be adverse events, which are required to be reported in the safety summaries, other measures of suicidality and suicidal behavior have not been consistently measured and reported in these trials. In this context, it is important to note that while measures of suicidality (such as MADRS item-10 or the C-SSRS) have been frequently used in recent years in the selection of patients for trials of antidepressants, these measures are not presented independently as outcome variables in the New Drug Approval packets on the FDA website. Similarly, completed suicides and suicide attempts as this study examined are not presented as outcome measures in the review of efficacy either alternatively or in addition to being reported as adverse events in the safety summary.
What can be concluded from these data is that suicide rates in antidepressant clinical trials are not static—changes in design and procedures have the potential to dramatically alter the rates of suicide and suicide attempts observed within the context of these trials. For this reason, caution must be exercised in extrapolating suicide data from antidepressant trials and generalizing these rates to the greater population.
In conclusion, deaths by suicide and suicide attempts have decreased significantly in antidepressant clinical trials following 2000 compared to the decade before 2000. Basic demographic features of the patients have remained consistent and medication treatment effects on suicidality were not apparent in these data. These findings may reflect enhanced screening procedures and the effective exclusion of suicidal patients in clinical trials for depression.