Introduction

Antidepressants are widely used to treat diseases such as depression, anxiety, obsessive compulsive disorder, neuralgia, and restless legs syndrome [1]. Their use has increased five-fold over the past 20 years in the United States and approximately three-fold over the past 15 years across 25 European countries [2]. Although antidepressants are effective and generally well-tolerated, there are increasing reports of potential adverse effects of antidepressants, including bone fractures [3], metabolic syndrome [4], bleeding [5], and a possible increased risk of seizures.

Epilepsy, a neurological disorder characterized by random recurring seizures of varying intensity, affects approximately 1% of the global population [6]. It most commonly involves intense shaking or twitching and loss of consciousness; mild seizures can produce a brief moment of blank staring or temporary changes in behavior. Seizure is caused by many factors, among which drug use is a common external cause [7]. Several cases of antidepressant-associated seizure have been identified in patients on high oral doses [8,9,10]. However, the risk of seizure at therapeutic doses remains unclear. In a cohort study, Hill et al. [11] found that all classes of antidepressant exposure were associated with an increased risk of seizures in people aged 20 to 64 years. However, Coupland et al. [12] reported that this risk varied according to the type or time of antidepressant exposure; they subsequently showed that antidepressant use increases the risk of seizures, but the results were inconsistent in a subgroup analysis based on individual antidepressants. Because various factors associated with antidepressant exposure (i.e., exposure time, antidepressant type, and individual antidepressant) may differentially affect the risk of fracture, these factors must be analyzed separately.

Considering the widespread use of antidepressants, it is important to determine whether there is a relationship between antidepressant exposure and the risk of seizures. In this study, we conducted a systematic literature review and meta-analysis to assess this association.

Methods

Data sources and search strategy

This work followed the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines [13]. We comprehensively searched the PubMed and EMBASE for relevant English-language publications from the dates of inception to May 2022. The search strategy, performed using Medical Subject Heading terms and the free keywords, comprised 3 groups of keywords: 1 group was about intervention (e.g., “antidepressant OR tricyclic OR TCA OR monoamine oxidase inhibitor OR MAOI OR selective serotonin reuptake inhibitor OR SSRI OR selective noradrenalin reuptake inhibitor OR SNRI”), 1 about outcome of seizure (eg, “epilepsy OR epilepsia OR epileptic OR epilepticus OR seizure OR epilepsy OR aura”), and 1 about study design (eg, “risk OR ratio OR epidemiologic OR observational OR observation OR case-control OR cohort”). The reference lists of retrieved articles were manually examined to detect reports not found in the database search.

Study selection

Studies were initially identified based on titles and abstracts, then subjected to full-text evaluation. Studies were considered eligible if they met the following criteria: (1) population: adults receiving antidepressant, with or without a formal psychiatric disorder diagnosis; (2) exposure: antidepressant use; (3) comparison: no treatment or low dose antidepressant treatment; (4) primary outcome: incidence of seizures; and (5) study design: randomized-controlled trials, case-crossover or self-controlled case series, and cohort, nested case-control, and case-control studies.

The exclusion criteria were (1) reviews, letters to the editor, or conference abstracts; (2) basic or animal experiments; (3) case reports and series; (4) lack of a control group; and (5) inability to isolate seizure-related data.

Data extraction and quality assessment

The following data were extracted: first author, publication year, study design, study location, ascertainment of antidepressant exposure, assessment of seizures, number of participants, follow-up year, statistical adjustments, and study quality. The most adjusted effect-size estimate was included in the analysis if more than one estimate was listed. The risk of bias in an individual study was evaluated by two independent reviewers; areas of disagreement were resolved by consensus. The methodological quality of the included studies was assessed using the Newcastle–Ottawa Scale (NOS) [14], as recommended by the Cochrane Collaboration for evaluation of observational study quality. A score of > 7 points was considered an indicator of a high-quality study.

Outcome assessment

The primary outcome assessed was the risk of seizures during antidepressant use compared with no treatment. To test the robustness of the association and the influence of each study on our estimates, a sensitivity analysis was performed by omitting one study at a time. In an attempt to explain possible heterogeneity between studies, we performed subgroup analyses based on the study design (case-crossover, cohort, nested case-control, or case-control), antidepressant type (selective serotonin reuptake inhibitors [SSRI], serotonin and norepinephrine reuptake inhibitors [SNRI], and others), individual antidepressants, and the definition of the exposure window (within 30 days or after antidepressant use). Following the results of the subgroup analysis, we conducted meta-regression analyses.

Data synthesis and analysis

Data were analyzed using STATA ver. 13 (Stata Corp., College Station, TX, USA). Quantitative meta-analyses were conducted for outcomes reported in at least two eligible studies. Heterogeneity among studies was assessed using the χ2 test and I2 statistic; an I2 value of 0–25% represents insignificant heterogeneity, 26–50% represents low heterogeneity, 51–75% represents moderate heterogeneity, and > 75% represents severe heterogeneity [15]. The DeSimonian and Laird random-effects model was used when studies were heterogeneous; otherwise, the Mantel–Haenszel fixed-effects model was used. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for the associations between antidepressant exposure and subsequent risk of seizures [16]. Splitting one study into several estimates leads to substantial more weight is assigned to this study in the meta-analysis, especially in random effects model. Thus, we explored fix effects model to generate a pooled OR if more than 3 estimates from one study are included, and then this pooled OR was used in the meta-analysis. Potential publication bias was assessed using funnel plots and quantified with Egger’s regression test [17, 18]. P-values < 0.05 were assumed to indicate statistical significance.

Results

Search results

The initial systematic literature search of two databases identified 3,704 articles. After the exclusion of 1,321 duplicates, the titles and abstracts of 2,383 studies were screened. 63 potentially eligible studies were retrieved for full-text review. 55 articles were excluded after the full-text review because they did not meet the inclusion criteria. Two British studies extracted data from the QResearch primary care database, with distinct age ranges for the individuals included. Furthermore, the remaining two British studies obtained data from two different data sources. Consequently, all four British studies were included in the current review. Two studies conducted in Taiwan were also included in this study because they had different study designs, although there was some overlap in terms of study period and participants. We performed a sensitivity analysis to assess the influence of each study on heterogeneity, and a summary estimate by excluding each of these studies from the same database one at a time. Ultimately, eight studies [11, 12, 19,20,21,22,23,24] satisfied the predetermined eligibility criteria. No related randomized controlled trials were identified. The selection process and exclusion justification are presented in Fig. 1.

Fig. 1
figure 1

Flow chart of the search process and study selection

Full size image

Characteristics of the included studies

Table 1 presents the main characteristics of the studies included in the systematic review. The eight studies were published from 2011 to 2023, with sample sizes ranging from 4,325 to 1,317,532 individuals, for a total of 1,709,878 patients. Among the eight studies, four were conducted in the UK [11, 12, 19, 20], two were conducted in Taiwan [21, 23], one was conducted in Canada [22], and one was conducted in Sweden [24]. The studies included two case-control studies [22, 23], two nested case-control studies [19, 20], one cohort study [11], and one case-crossover study [21]. The remaining two studies [12, 24] used case-crossover and cohort study designs. Three studies [11, 12, 19] selected exclusively depressive patients without antidepressant exposure as a control, one [22] enrolled patients with depression who were exposed to bupropion as a control, and one [23] enrolled patients with depression who were exposed to low-dose antidepressant as control. The overall NOS score was 51 of 63 (81%), which is considered representative of overall high quality; the details of the critical appraisal are provided in Tables S2 and S3.

Table 1 Characteristics of the included studies
Full size table

Meta-analysis

Overall-analysis

A meta-analysis of all studies with 1,709,878 individuals revealed that antidepressant use significantly increased the risk of subsequent epilepsy (OR 1.59, 95% CI 1.39–1.83; P < 0.001), with high heterogeneity (I2 = 89.2%) (Fig. 2). As shown in Figure S1, there was no evidence of publication bias. Sensitivity analysis showed that the significance of the comprehensive effect did not significantly change after studies using the same data source were eliminated one by one.

Fig. 2
figure 2

Forest plot of the risk of seizure in relation to antidepressant exposure

Full size image

Subgroup analysis

Table 2 presents the results of the subgroup analyses. The subgroup analyses according to study design showed significant associations for the case-crossover studies (OR 2.35, 95% CI 1.7–3.24; P < 0.001), cohort studies (OR 2.03, 95% CI 1.57–2.6; P < 0.001), nested case-control studies (OR 1.74, 95% CI 1.41–2.15; P < 0.001), and case-control studies (OR 1.32, 95% CI 1.15–1.51; P < 0.001) (Figure S2).

Table 2 meta-analysis for studies included in this analysis
Full size table

In the analysis limited to five studies that enrolled controls, including depressive patients with or without exposure to low-dose antidepressants, who had an indication for antidepressant use; notably, a significant association was observed among patients using antidepressants (OR 1.53, 95% CI 1.34–1.74; P < 0.001) (Figure S3).

In the analysis limited to three studies that exclusively enrolled depressive patients, a positive relationship was observed among patients using antidepressants (OR 1.83, 95% CI 1.54–2.17; P < 0.001; Figure S3).

In subgroup analyses according to antidepressant type, significant associations were observed for the use of an SSRI (OR 1.48, 95% CI 1.32–1.66; P < 0.001, SNRI (OR 1.65, 95% CI 1.24–2.19; P = 0.001), or other antidepressant (OR 1.77, 95% CI 1.4–2.25; P < 0.001). However, the use of a TCA (OR 1.27, 95% CI 0.84–1.92; P = 0.249) (Figure S5) was not significantly associated with an increased risk of seizures.

When studies were grouped according to individual antidepressants, positive relationships were detected for the use of citalopram (OR 1.62, 95% CI 1.35–1.95; P < 0.001), escitalopram (OR 1.62, 95% CI 1.35–1.95; P < 0.001), fluoxetine (OR 1.44, 95% CI 1.2–1.73; P < 0.001), paroxetine (OR 1.56, 95% CI 1.3–1.87; P < 0.001), sertraline (OR 1.76, 95% CI 1.35–2.29; P < 0.001), mirtazapine (OR 1.48, 95% CI 1.24–1.78; P < 0.001), venlafaxine (OR 1.88, 95% CI 1.36–2.61; P < 0.001), and fluvoxamine (OR 1.38, 95% CI 1.03–1.83; P = 0.029). No significant association was observed among patients using amitriptyline (OR 1.22, 95% CI 0.75–1.98; P = 0.427), dosulepin (OR 0.86, 95% CI 0.27–2.72; P = 0.797), lofepramine (OR 2.12, 95% CI 0.96–4.69; P = 0.065), trazodone (OR 2.03, 95% CI 0.67–3.16; P = 0.21), or duloxetine (OR 0.97, 95% CI 0.75–1.25; P = 0.818) (Figure S6).

In subgroup analyses based on the definition of the exposure window, a significant association was observed between studies that evaluated seizure development within 30 days after antidepressant use (OR 2.35, 95% CI 1.7–3.24; P < 0.001) and those that evaluated seizure development after antidepressant use (OR 1.49, 95% CI 1.26–1.76; P < 0.001; Figure S7).

Meta-regression analysis

Considering the higher combined OR associated with shorter-term antidepressant exposure, a univariate meta-regression analysis was conducted to investigate the potential sources of variation in the exposure window. This analysis revealed no significant differences in seizure risk (standard error (SE) = 0.132, P = 0.104; Figure S8).

Discussion

By combining data from available studies, this study demonstrated that antidepressant exposure was associated with an increased risk of seizures. Subgroup analyses indicated that this risk varied according to antidepressant type and individual antidepressant.

An investigation of the association between antidepressant exposure and seizures should consider antidepressant indications, which have been associated with an increased risk of seizure [25]. Thus, the ideal control for confounding according to indication is a reference group of patients with psychiatric disorders who did not take antidepressants. Five of the included studies limited their cohorts to patients diagnosed with psychiatric disorders or exposed to low-dose antidepressants to reduce potential indication bias. The significant association persisted when data from these five studies were pooled; the pooling partially controlled for indication confounding. The pooled analysis of the four case-crossover studies [11, 12, 21, 24] also supported a positive relationship. Case-crossover is a novel strategy that controls for between-person confounders by comparing the risk and reference periods in each patient. Risk periods are defined as the time during or after an exposure. Thus, although a psychiatric disorder itself is associated with a risk of seizures, our findings suggest that confounding according to indication could not explain the observed association.

Our further analyses demonstrated that only new-generation antidepressants were associated with an increased risk of seizures. The precise biological mechanisms underlying the relationship between SSRIs/SNRIs and seizures are unclear. Several SSRIs inhibit G-protein-activated inwardly rectifying potassium (GIRK) channels. It has been suggested that the activation of GIRK channels induces neuron membrane hyperpolarization via potassium ion efflux, ultimately reducing neural excitability [26]. Thus, the inhibition of these channels by SSRI exposure may increase neural excitability and produce a lower seizure threshold. Additionally, antidepressant use has been linked to hyponatremia [27], which may represent an additional potential mechanism underlying the association between antidepressants and seizures [28]. Furthermore, previous epidemiological investigations demonstrated that the risk of seizures was higher for SSRIs than for TCAs [27], which may partially explain the findings of subgroup analyses based on antidepressant type and individual antidepressant.

Another important issue is the definition of the antidepressant exposure window, which varied among the included studies. In our study, we used a 30-day period as the exposure time window among case-crossover studies. After data from the case-crossover studies had been pooled, a more pronounced risk of seizures was observed, suggesting an acute effect of antidepressants. This finding was reinforced by the relatively lower, but significant, risk observed in the overall analysis. The lower risk present in more long-term antipsychotic users might be related to the effects of tolerance and cross-tolerance to antidepressant drugs.

The main strength of our study was that we performed detailed subgroup analyses according to study design, antidepressant type, and individual antidepressant. We conducted additional analyses to control for confounding according to indication. Nevertheless, this work had several limitations. First, the robustness of our findings was hindered by the small number of included studies. Thus, the subgroup analysis was limited by the sample size, and further studies are needed. Second, similar to other meta-analyses of observational studies, uncontrolled confounding variables may have affected the results. Previous studies have shown that antidepressant use increases the risk of falls [29], and the associated traumatic brain injury is a strong risk factor for seizures. However, this potential confounder was considered in a single included study. Future studies should fully adjust for potential confounders to rule out alternative explanations. Third, there was evidence of high statistical heterogeneity for the pooled results. Subgroup analyses were performed to examine the source of the heterogeneity, but the variables (study design, type of antidepressant, individual antidepressant, or exposure window) evaluated did not thoroughly explain the source of heterogeneity. Thus, we can be less certain about these results. Finally, the definition of antidepressant dose in the included studies was inconsistent; therefore, we could not determine whether this parameter was associated with the risk of seizures.

In conclusion, the findings of this meta-analysis suggest that SSRI/SNRI, but not TCA, use is significantly associated with risk of seizures, especially among short-term users. Clinicians should ensure that antidepressants are prescribed only for patients with a clear indication and be cautious when prescribing antidepressants to patients who have an underlying increased risk of seizures.