Antiepileptic Drugs and Suicide Attempts in Patients With Bipolar Disorder—Reply

In reply

Mentari et al¹ from the US Food and Drug Administration (FDA) raise questions and concerns regarding the conclusions of our recent article² on the relationship between antiepileptic drugs (AEDs) and suicide attempts. In contrast to the FDA meta-analysis of randomized controlled trials of AEDs, which showed significantly increased risk of suicidal events (ideation and behavior) relative to placebo, we found no evidence of increased risk of AED treatment and suicide attempts in an observational cohort study of 47 918 patients with bipolar disorder. Our analysis compared periods of risk before and after treatment initiation (following diagnosis) and between individuals who were and were not treated with an AED. Our analysis adjusted for prior suicide attempts (in the year prior to diagnosis), concomitant treatment with other AEDs, lithium, antidepressants, and antipsychotics and age, sex, and year (2000-2006). All patients in our cohort had at least 1 year of continuous observation before and after their index bipolar disorder diagnosis. Results of our analysis indicated that rates of suicide attempts significantly decreased following treatment initiation and that overall there were no significant differences between patients who did or did not receive AED treatment. In a subgroup of patients who did not receive concomitant other AED, lithium, antidepressant, or antipsychotic treatment, AED treatment was associated with a significant decrease in suicide attempt rates (3 in 1000 patient-years vs 15 in 1000 patient-years; event rate ratio, 0.19; 95% confidence interval [CI], 0.08-0.47; P < .001). A series of sensitivity analyses that used advanced matching strategies and person-time monthly exposure analyses confirmed the overall results.

Mentari et al raise 3 major criticisms of our study. First, they suggest that the lower rate of suicide attempts following treatment initiation may be due to increased clinical monitoring and support following treatment initiation and that high levels of disease impairment, including suicidal behavior, may lead to treatment initiation.

We agree that as one gets further from the index episode the rate of suicide attempts will “regress toward the mean,” leading to potential bias of simple pretreatment vs posttreatment comparisons. It is precisely for this reason that we conducted a sensitivity analysis using a person-time analysis that adjusts for the overall decay in suicide attempt rate over time and defines exposure on a month-by-month basis. In this way, we were able to compare rates of suicide attempts within individuals during periods in which they took an AED vs times when they did not, regardless of when those periods occurred. The results of this analysis revealed that rates of suicide attempts were significantly higher during months when patients were not taking an AED relative to months that they were (odds ratio [OR], 0.59; 95% CI, 0.47-0.75; P < .001). This finding was further replicated in the subsample (n = 662) of highest-risk patients who had made a suicide attempt in the year prior to their index diagnosis (OR, 0.35; 95% CI, 0.17-0.74; P < .005). These findings reveal no evidence of increased risk of suicidal attempts associated with AED treatment.

Second, they suggest that the reason that we did not see a difference between treated and untreated patients is because of confounding by indication, ie, differences in disease severity between the 2 groups. This is a very important point because we would expect patients who received treatment to be more severely ill and have higher unadjusted suicide attempt rates. In fact, we found no difference in overall suicide attempt rates after adjusting for prior suicide attempts, concomitant treatments, and demographic factors and in patients who were treated with an AED only, significantly lower suicide attempt rates. This is just the opposite of what would be expected based on confounding by indication.

Third, they criticize our comparison of the entire observation period for untreated patients with the postdrug-only period for AED-treated patients. While it makes no sense to attribute risk of a drug to adverse events that occurred prior to taking the drug, we did explore this question of the time of drug initiation in a sensitivity analysis in which we matched treated and untreated patient pairs using propensity score matching and then further matched each pair of treated and untreated patients in terms of the time at which the treated patient took the AED. This directly addresses the concern regarding differential periods of observation over the course of illness between treated and untreated patients, and again, no association between AED treatment and suicide attempt rate was found.

Finally, Mentari and colleagues computed the raw unadjusted suicide attempt rate combining the predrug and postdrug initiation period for AED-treated patients and showed that it was higher than for untreated patients (OR, 1.57; 95% CI, 1.34-1.84). We reperformed their analysis adjusting for age, sex, year, previous suicide attempts, and concomitant treatment and found that differences in these variables accounted for much of the effect (OR, 1.33; 95% CI, 1.12-1.58). What this analysis shows is that, overall, patients who were treated with an AED were more severely ill than the untreated patients, making the reduction of suicide attempt rate following treatment even more impressive. This is consistent with Table 2 of our article, which showed that the pretreatment suicide attempt rate in patients who were ultimately treated with an AED was significantly higher than the rate for untreated patients, again showing that if anything, treated patients should be at higher risk for suicide attempt. If AED treatment significantly decreases this elevated risk, it is hard to imagine that AED treatment is responsible for creating the risk.

A fundamental difference between the FDA study and our study is that our study was restricted to suicide attempts (n = 446 attempts) whereas the FDA data consisted of 4 completed suicides, 38 attempts, 4 preparatory acts, and 96 ideations. Our study had more than an order of magnitude more cases of suicidal behavior. Conversely, the FDA study benefits greatly from randomization of the individual studies whereas our study is observational and one can never be fully confident that all potential sources of bias have been removed. However, meta-analysis is itself an observational study of studies and heterogeneity introduced by combining multiple drugs used for multiple indications can lead to bias in conclusions drawn about the class of drugs as a whole. Finally, we are surprised that the FDA takes exception to the results of our study given that it replicates the lack of a significant effect of AED treatment on suicidality in psychiatric patients found in their own study (OR, 1.51; 95% CI, 0.95-2.45).

Correspondence: Dr Gibbons, Center for Health Statistics, University of Illinois at Chicago, 1601 W Taylor, Chicago, IL 60614 (rdgib@uic.edu).

Author Contributions: Dr Gibbons had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Financial Disclosure: Dr Gibbons has served or is currently serving as an expert witness for the US Department of Justice, Wyeth, and Pfizer Pharmaceuticals in cases related to antidepressants and anticonvulsants and suicide. Dr Mann has received research support from GlaxoSmithKline and Novartis. Dr Brown directed a suicide prevention program at the University of South Florida that received funding from JDS Pharmaceuticals.

Funding/Support: This work was supported by National Institute of Mental Health grants MH062185 (Dr Mann), R56 MH078580 and MH8012201 (Drs Gibbons and Brown), and MH040859 (Dr Brown).

This article was corrected for errors on December 16, 2010.