Antiepileptic Drugs and Suicide: A Systematic Review of Adverse Effects

Antiepileptic drugs (AEDs) have been described as potential risk factors for suicidal behavior [1]. In 2008, the Food and Drug Administration (FDA) in the USA reported a 2-fold increased risk of suicidal ideation or behavior for 11 AEDs (odds ratio, OR, 1.80, 95% confidence interval, CI, 1.24-2.66) [2].

Suicide is a serious public health concern around the world. In 2011, the prevalence rate in Europe was 13.9/100,000 persons per year [3], with the highest rates among men and older individuals, although the prevalence among adolescents and young adults has recently increased [4,5]. These figures are even higher among people with epilepsy; the reported lifetime prevalence rate of suicide and suicide attempts are between 5 and 14.3% [6]. Psychological autopsy studies have demonstrated that 90% of suicide victims suffer from a psychiatric condition at the time of death [7].

The following trends and observations led to this systematic review (SR): (1) the high and widespread use of AEDs to treat epilepsy and other conditions including off-label indications [8,9,10]; (2) the use of new AEDs among the elderly [11,12]; (3) the fact that AED users have the highest risk of suicide within their population age group [3], and (4) the potential public health implications of the association between AEDs and suicide. The aim of this review is to systematically summarize the available evidence from randomized clinical trials (RCTs) and observational studies on the association between AEDs and suicide.

We searched PubMed (1980-April 2012), the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, the Database of Abstracts of Reviews of Effects (1980-February 2012), PsycINFO (1980-February 2012), and ClinicalTrials.gov (April 5, 2012). The search strategy included MeSH terms and free-text words. We combined the terms ‘antiepileptic drugs' and each of the Anatomical Therapeutic Chemical Classification level 4 and 5 drugs for N03A (version 2012) with the terms ‘suicide', ‘intentional self-murder', ‘killing oneself', ‘hastened death', ‘subintended death', ‘miscalculation', ‘fatal self-harm', ‘suicide intent', ‘suicide attempt' and ‘suicide behavior', using the Boolean operator AND. We included studies published in any language mastered by the authors P.F., E.B., M.S. and L.I. - English, any Romance language, and Swedish. We also electronically searched the following peer-reviewed journals: Neuropsychopharmacology, CNS Drugs, Neurology, all of the Lancet group publications, Science Direct, and the Wiley Online Library. In addition, we reviewed the reference lists from the included studies. Two of the three authors we contacted provided additional information.

We aimed to identify RCTs and observational studies related to any type of AED that included suicidality (defined as completed or attempted suicides) as a primary or secondary outcome. Studies that included suicidal ideation or deliberate self-harm with no intention to die or unknown intention as outcomes were only included if the authors also collected information on completed or attempted suicides, as defined by Posner et al. [13].

P.F. screened the titles and abstracts yielded by the bibliographic search, and E.B. screened the above-mentioned journals electronically to assess whether they met the eligibility criteria. We retrieved the full-text article if an abstract was selected or if there was uncertainty regarding the eligibility of a study. The inclusion criteria were RCTs or observational studies, and a population aged 18 years or older. The reference group was placebo or no treatment. We excluded case reports and studies conducted with data from spontaneous report databases.

The search on ClinicalTrials.gov was limited to the trials that used a random allocation method, had a safety-related outcome and were limited by age (≥18 years). We excluded ongoing trials.

P.F. and E.B. extracted the data using a predetermined form. To better assess the heterogeneity between the studies, we developed a checklist specifically for this SR of adverse effects (see online supplementary material; for all online suppl. material, see www.karger.com/doi/10.1159/000356807). Any discrepancies were resolved by discussion.

Because we expected clinical (e.g. different drug exposures, different participants, different indications for use, different definitions of suicide) and methodological (inclusion of randomized and observational studies, different confounders included in the adjusted models) heterogeneity, we decided to conduct a narrative synthesis [14].

We initially included 10 publications. See figure 1 for the numbers of studies that were screened, assessed for eligibility and included in the final analyses, as well as the reasons for exclusion; 1 prospective cohort study [15] had flaws in its statistical results and was therefore excluded. Since this search was conducted, a new SR [16] and a new-user design study [17] have been published and included in this SR. Ultimately, we included 11 published studies. For a description of the 11 included studies and the cohort study excluded, see table 1.

Gibbons et al. [18 ]conducted 2 population-based prospective cohort studies between 2000 and 2006 -1 study involving patients with bipolar disorder reported a risk of suicide for any AED of 0.88 (95% CI 0.72-1.08), and another study [19] which included patients who had filled one prescription of gabapentin showed an effect of this AED on suicide of 0.93 (95% CI 0.76-1.14). Gabapentin on patients with bipolar disorder in this study showed a protective effect (OR 0.62, 95% CI 0.41-0.94). In both studies, the measurement of the exposure did not account for compliance, dosage and duration of treatment, and it focused on any use over a time span of 1 year. This methodology may have led to nondifferential exposure misclassification. Suicide may have been underreported, resulting in nondifferential outcome misclassification. Not all potential confounders were considered, although in the study of the bipolar patients, the authors adjusted for suicide attempts in the year before the diagnosis.

Sondergård et al. [20] found an increased risk of suicide associated with one prescription of any AED among patients with a severe bipolar disorder (OR 3.30, 95% CI 2.17-4.99); however, they did not consider previous suicide attempts or other cotreatments in the model. In an attempt to control for confounding by indication they adjusted for readmission and the use of lithium as a marker of bipolar disorder severity. The number of prescriptions was a proxy measure for the duration of treatment. They did not consider dosage or compliance.

In the case-crossover analysis of Olesen et al. [21], the authors found a 2-fold increased risk of suicide with the use of AEDs (OR 1.84, 95% CI 1.36-2.49).There was a potential for nondifferential exposure misclassification (as an average daily dose was calculated instead of use of prescribed daily dose) and for confounding by changes in indication or the severity of the risk of suicide over time. The authors did not state the reason for the time window they selected and failed to include sensitivity analyses for different time exposure windows. There is also a potential for reverse causation bias. The cohort analysis was not considered in this review as the reference drug was carbamazepine.

There were 2 community-based case-control studies. On the one hand, Arana et al. [22] reported an association between suicide and the use and nonuse of AEDs in 11 cohorts and concluded that the increased risk was dependent on the underlying disease. On the other hand, they also reported an increased risk of suicide in patients on AEDs without epilepsy, depression or bipolar disorder (OR 2.57, 95% CI 1.78-3.71). The authors reported that 17% of these patients had a pain-related diagnosis. In the subgroup analyses, comparing current users and nonusers of AEDs, the current use of AEDs provided a protective effect for patients with epilepsy alone (OR 0.59, 95% CI 0.35-0.98); however, patients with depression alone had an increased risk of suicide (OR 1.65, 95% CI 1.24-2.19). Conversely, in patients with bipolar disorder, the use of AEDs seemed uneventful (OR 1.13, 95% CI 0.35-3.61). Andersohn et al. [23] showed that only AEDs with high risk of depression were associated with an increased risk of suicide in epilepsy patients (OR 3.08, 95% CI 1.22-7.77). In both studies [22,23], the results could be biased by confounding by indication. Furthermore, there was a possibility of nondifferential outcome misclassification because suicides may have been underestimated. Residual confounding from the type of epilepsy was also a possible explanation for the results in the study of Andersohn et al. [23] because the code-defined diagnoses for type of epilepsy were not validated. The small numbers of cases and controls could have also decreased the power of the study to detect an effect, particularly for individual drugs. In the study of Arana et al. [22], the authors excluded all individuals with a familial or personal history of suicide, which could have led to an overall underestimation of the risk of suicide.

Pugh et al. [17] conducted a study with in- and outpatients aged 65 years and older. Demographic, psychiatric comorbidity and chronic pain variables were used to construct a propensity score to control for confounding in the model. The authors reported a hazard ratio of 3.90 (95% CI 2.93-5.19) for any AED use. The authors did not assess the effect of prevalent users on the results. Except for levetiracetam, marketed in 2004, all the other AEDs have long been on the market. Thus, some of the patients included may not have been naive to AEDs. The authors developed a propensity score model and gave a c-statistic value of 0.66. The c-statistic tells us how well the model with the observed covariates predicted the probability of receiving an AED. The c-statistic usually ranges between 0.5 (chance classification) and 1 (perfect classification) [24]. However, Pugh et al. [17] did not provide the distribution of the propensity scores among the AED users and nonusers, which would have been more informative of the distribution of the confounders between groups [24]. Thus, it was difficult to assess whether there was any error in their estimation. In addition, there was the potential for nondifferential misclassification of suicide.

Three SRs [2,25,26] included randomized trials of efficacy; suicide was collected as a potential adverse event in these trials. The FDA's meta-analysis [2] reported an increased risk of suicide for all AEDs (OR 1.80, 95% CI 1.24-2.66). Van Lieshout and MacQueen [25] and Premkumar and Pick [26] reported on the effects of lamotrigine on suicide in patients with bipolar disorder and depression (OR 0.26, 95% CI 0.02-2.77) and with schizophrenia (OR 2.97, 95% CI 0.12-72.18). The results of all these 3 SRs could have suffered from an ascertainment bias of adverse events, even though the FDA's meta-analyses used a standardized protocol to detect and assess potential suicide events [2]. In the FDA's meta-analysis, suicide was described post hoc by the sponsor of the drug. In addition, there were differences in the duration of treatment and treatment discontinuation between the placebo and treatment arms, with higher rates of discontinuation in the drug arm and a longer duration in the placebo arm.

The review by Fountoulakis et al. [16] concluded that there were not enough data to confirm the association between an increased risk of suicide and AEDs as a group. With regard to individual drugs, they concluded that lamotrigine and topiramate may increase the risk of suicide, whereas carbamazepine and valproic acid were protective. However, this review searched a single bibliographic database. It included all population subgroups except patients with mood disorders, any type of comparator except studies that only compared AEDs and lithium, and any type of suicidal behavior. The authors restricted the type of studies to those with a naturalistic design, database studies and prospective trials, without further specifications. They did not assess the quality of the included studies.

Concerning individual drugs, studies have shown an increased risk of suicide for patients taking phenobarbital (1 study [21]), phenytoin (1 study [17]), carbamazepine (1 study [22]), valproic acid (3 studies [17,21,22]), lamotrigine (3 studies [2,17,21]), topiramate (3 studies [2,17,18]), gabapentin (1 study [17]) and levetiracetam (2 studies [17,23]). However, a similar number of studies demonstrated a protective effect - valproic acid (1 study [23]) and gabapentin (1 study [19]) - or no effect - phenobarbital (2 studies [17,23]), phenytoin (2 studies [21,23]), valproic acid (3 studies [2,18,22]), lamotrigine (5 studies [18,22,23,25,26]), topiramate (2 studies [22,23]), gabapentin (5 studies [2,18,21,22,23]), levetiracetam (3 studies [2,22]) and oxcarbazepine (3 studies [2,18,21]) - of these AEDs on suicide.

Table 2 shows the effect measures and the corresponding 95% CIs grouped by population subgroup and/or indication for use and AEDs.

Since the publication of the FDA alert on the risk of suicide associated with 11 AEDs, several observational studies [27,28], as well as reviews and expert opinions [29,30,31,32], have attempted to disentangle this controversial relationship. We ultimately reviewed 11 studies that reported findings on the effects of AEDs on suicide compared to a placebo or no treatment; 3 studies [2,17,21] reported a 2- to 4-fold overall increased risk of suicide with AEDs as a group. For epilepsy, the FDA meta-analysis [2] showed an increased risk of suicide, whereas Arana et al. [22] did not find an increased risk of suicide for patients with epilepsy, but did find an increased risk of suicide in depressive patients. However, the close relationship between epilepsy and other neurological and psychiatric disorders has determined the use of AEDs for different indications [33]. Andersohn et al. [23] showed that the risk of suicide in epilepsy patients depended on whether the AEDs had a high or low risk of depression. Concerning mental illness, Arana et al. [22] showed an increased risk of suicide among patients with depression, but not among patients with bipolar disorder. Sondergård et al. [20] showed that any AED could increase the risk of suicide in patients with bipolar disorder under specialty out- and inpatient care. Gibbons et al. [18] showed a protective effect of AEDs for the same type of patients, although this study included the outpatient healthcare sector and as it required a 1-year follow-up of patients to be included, patients who committed suicide during this time period might have been excluded. In the study of Arana et al. [22] it seemed that the effect of AEDs on bipolar patients was uneventful. For indications other than epilepsy, depression and bipolar disorder, Arana et al. [22] showed a 3-fold increased risk of suicide.

For Pugh et al. [17] and Sondergård et al. [20] the use of AEDs may be a marker of disease severity, as patients in hospital have an increased risk of suicide attempts. Moreover, in Pugh et al. [17], depression and previous suicidality were more prevalent among patients exposed to AEDs. Confounding by indication could partly explain their results. In the study of Olesen et al. [21], during the case time period an increased risk of suicide could precede the prescription of an AED. Several studies presented in this review [18,20,21] did not adjust for previous suicidal behavior, which has been reported as the most significant risk factor for attempted and completed suicides [34]. Gibbons et al. [18] adjusted for suicide attempts in the year before the diagnosis of bipolar disorder. Andersohn et al. [23] reported that they adjusted for prior suicidal behavior, but the extent of the adjustment was unclear. The same issue applied in the adjustment for the type of epilepsy. Residual confounding could have masked the effect of an association in all of the observational studies we included. In the FDA's meta-analysis [2], the epilepsy patients took the AED under study as an adjuvant to other AEDs, which may indicate more severe disease. Suicide in this group of patients could be related to the ‘forced normalization' phenomenon [35].

Another potential flaw in all the studies presented in this review is the nondifferential misclassification of suicide which was assumed to be a limitation in all healthcare database studies. Theoretically, nondifferential disease misclassification would bias the effect estimate towards the null, which may cause a problem for interpreting the results of studies reporting a null or very weak association between AEDs and suicide [36]. There was no description of how adverse events were collected or how causality of an adverse event to the drug was assessed in the clinical trials reviewed in the included SRs. In the FDA's meta-analysis [2], adverse events were measured retrospectively based on the patient's spontaneous reports.

The dosage of the treatment and the compliance could not be estimated in several observational studies [17,18,19,20,21,22,23]. This fact could introduce nondifferential exposure misclassification. Andersohn et al. [23] did calculate the duration of the prescription minimizing this type of bias. The nondifferential misclassification of AEDs could bias the effect estimate towards the null. Furthermore, many studies [17,18,19,21] focused on ‘ever use' of an AED over a long time span, which may also have biased the effect estimate towards the null.

This review showed that the effect estimates were heterogeneous across populations or by indication. An explanation for these contradictory findings could be, for example, that the patients on AEDs represented a subgroup of patients at particularly high risk of suicide, and AEDs would serve as a marker for the severity of disease.

The main strength of this SR is that we included studies that had a placebo or nontreatment group as the reference group. The inclusion of randomized and observational studies resulted in a broad overview of the relationship between AEDs and suicide. Another strength of this study is the standardized narrative synthesis. In an attempt to assess clinical heterogeneity we grouped the study results by population subgroup or by indication. Because of the few numbers of studies we could not group by type of study. Comparing the results between studies is difficult because of the differences in the definition of the exposure or outcome as well as the confounders adjusted for in the statistical model. In addition, we developed a checklist to assess the studies we included in this SR of adverse effects. Although this checklist needs to be validated, it is a step toward assessing the quality of any type of study included in such SRs.

Even though we searched in several bibliographic databases, we did not search EMBASE, which is a complementary bibliographic database to the PubMed database. We did not attempt to search gray literature. Hence, there is potential for publication bias. Moreover, we did not identify any clinical trials, although they were in our inclusion criteria. However, we assessed the effect of AEDs on suicide using SRs that included randomized trials of AED efficacy. This problem was the result of the difficulty in searching the bibliographic databases for adverse events in RCTs on efficacy. It is unlikely that clinical trials include any terms related to assessment of adverse events in their titles, abstracts or keywords.

In an attempt to overcome some of the flaws detected in the published studies we believe that a new study should be conducted retrieving the information from a large healthcare database because suicide is a rare event. The exposure to AEDs should be as detailed as possible to account for the duration of the exposure. In addition, to avoid outcome misclassification, the study should try to link the database with a mortality register to better capture the causes of death, or search for cases in the database using free words. Alternatively, a review of a sample of medical charts to validate the diagnoses could be conducted to ensure the detection of all potential cases. Due to the close relationship between several neuropsychiatric disorders and suicide, and the perils of confounding by indication, all of the diagnostic codes for the covariates relevant to the study should also be validated. We propose a set of confounders: epilepsy (type, age at onset, mono- or polytherapy), mood disorders, schizophrenia, neuropathic pain or other indications including off-label use, antipsychotics, other central nervous system drugs, personal history of previous suicide attempts, and socioeconomic status. Age and gender should be considered potential effect modifiers. Ideally, the severity of the diseases included as confounders should be thoroughly assessed, directly or through a proxy variable such as referrals to the hospital or to specialty care. In addition, any new RCT conducted to determine the efficacy of a new AED or to determine a new AED indication should include suicidality as a specific safety-related outcome.

There is no clear evidence to confirm or rule out an association between the use of AEDs and suicide because of the heterogeneity at the clinical and the methodological level.

Most of the studies considered an overall effect of AEDs as a single class effect. This assumption may need to be revisited in light of the different effects of the different AEDs on mood.

The research leading to these results was conducted as part of the PROTECT consortium (Pharmacoepidemiological Research on Outcomes of Therapeutics by a European Consortium (www.imi-protect.eu), which is a public-private partnership coordinated by the European Medicines Agency. The PROTECT project has received support from the Innovative Medicine Initiative Joint Undertaking (IMIJU) (www.imi.europa.eu) under grant agreement No. 115004.

P.F., E.B., M.S., X.V., M.R., S.S., J.H. and L.I. do not have any conflict of interest. J.A. belongs to the European Federation of Pharmaceutical Industries and Associations (EFPIA) member companies in the IMIJU, and costs related to their part in the research were carried by the respective company as in-kind contribution under the IMIJU scheme. The PROTECT project has received resources which are composed of financial contribution from the European Union's Seventh Framework Programme (FP7/2007-2013) and in-kind contribution from EFPIA companies. The views expressed are those of the authors only.