A nationwide study of initiation of antidepressant pharmacotherapy and the risk of seizures

Abstract

Objective:

The present study aimed to examine whether antidepressant initiation increases the risk of hospitalizations and unplanned outpatient visits for seizures. Research has provided conflicting evidence as to whether antidepressant initiation causes seizures. Because epilepsy and depression are comorbid, this remains an important question, particularly in the care of those already at-risk for seizures.

Methods:

We used Swedish-register data, including 658,766 antidepressant initiators and 1:1 age-, region-, and sex-matched non-initiators, ages 12–65. We used filled prescriptions to identify any antidepressant and serotonergic antidepressant and inpatient hospitalizations and unplanned outpatient (to avoid coding routine epilepsy maintenance as a seizure) visits to identify seizures, respectively. We first compared seizure visit incidence between antidepressant-initiators and matched non-users in the year following initiation from 2006–2013. To examine seizure risk over months pre- and post-initiation, within-individual analyses compared risk during the month one year prior to initiation with all subsequent months. We examined associations for any antidepressant and serotonergic antidepressants, as well as for any initiator and initiators with a history of seizures.

Results:

Our matched-cohort results showed higher incidence of seizure visits among antidepressant users compared with non-users (e.g., adjusted incidence rate ratio [IRR]=3.14, 95% confidence interval [CI]=2.83–3.49). In within-individual analyses, the months after initiation were associated with higher incidence of seizure visits when compared with the month one year prior to initiation (e.g., one month after initiation IRR=1.96, 95%CI=1.64–2.34), but in individuals with a seizure history we observed weaker or no associations in the months after initiation (e.g., two months after initiation IRR=1.12, 95%CI=0.87–1.45). Notably, irrespective of potential seizure history, the months preceding initiation were associated with the greatest risk (e.g., one month before initiation IRR=2.86, 95% CI=2.42–3.38).

Conclusions:

Our findings suggest that there may be an elevated risk of seizures during antidepressant treatment, though the period of highest risk was before the initiation of antidepressants. Risk for seizure visits was lower among individuals with a history of prior seizures, which may be reassuring for the clinical care of these patients or indicate lack of treatment seeking following seizures. This study highlights the need to consider seizure risk across time; the failure to account for these dynamics may help account for discrepant findings in previous studies.

INTRODUCTION

Individuals with epilepsy often have comorbid depression and anxiety.^1–5 One explanation for this is that antidepressant use may increase risk for seizures.^4,6–19 Several observational studies have reported elevated risk of seizures at therapeutic doses.^{4,7–10,17,19} However, some human and non-human animal research has suggested that antidepressants often have an anti-seizure effect.^6,20–23

However, antidepressant use may just a reflect the fact that epilepsy is more common in people with depression/anxiety than it is in the general population,^1–5 as untreated depression may itself be a risk factor for seizures.¹ Previous studies have documented a bidirectional relationship between symptoms of depression and epilepsy has that is likely to be at least partially genetic in origin,^1,5,24,25 and research has also indicated a socio-economic link between these conditions.²⁶ Thus, depression (and/or other indications) may confound associations between antidepressant use and seizures.

The extant research has investigated the safety of antidepressant use in the general population and among those with previous seizure history.^4,6–19 Most studies on the safety of antidepressant use among patients with epilepsy report no increase in seizure frequency^16,18 or a decrease in seizure frequency related to medication initiation.^6,22 However, this research has been underpowered^{6,12,16,18,22} and has not ruled out confounding factors.^6,12,18,22 Studies in the general population have used larger samples, yet findings have been inconsistent.^{4,7–10,17,19} Research investigating this question has also typically not examined risk of seizures over time, despite the time-varying nature of seizure risk⁵ and depression,¹ as well as the growing body of evidence that the time before and after a seizure is related to higher risk for depression.^1,5,24,25 One seminal paper in the field examining clinical trial data is often cited as sufficient evidence that antidepressants do not cause seizures.¹¹ However, there was a low base rate of seizures in this study (n=60), even among a large sample of 4,419 antidepressant users Further, almost half antidepressant users who had seizures (n=26) were treated with bupropion.¹¹ This suggests that even larger samples are required to fully examine such a rare, but important adverse event like seizures. As such, evidence to date regarding the risk of seizures related to antidepressant use remains inconclusive.

More research is needed to clarify if antidepressant initiation increases the risk for seizures or if associations are better explained by alternative explanations, including confounding by indication. The current study took a comprehensive approach to reconcile discrepancies in the previous literature. We compared incidence of seizure-related medical visits 1) in a matched cohort (see below) of antidepressant initiators and non-initiators, and 2) prior to and after antidepressant initiation within individuals. In addition to adjusting for confounding that was stable over time (e.g., history of traumatic brain injury [TBI], stroke, central nervous system [CNS] infection, brain tumors) using a within individual approach, we adjusted for additional medications (including anti-seizure medications [ASMs]) and examined the pattern of seizure risk across medicated and unmedicated periods.

METHODS

Standard protocol approvals, registrations, and patient consents

The institutional review board at Indiana University and the regional ethical review board in Stockholm, Sweden, approved this study. According to Swedish law, informed consent was not necessary because the study used data available from national Swedish registries.

Data Availability Statement

The data used in this study are from national register information. The authors had no special privileges in accessing the data. Dissemination of personal information is regulated by the Swedish Secrecy Act. In accordance with Swedish law, researchers seeking access to individual-level data must apply for permission through a Research Ethics Board (etikprovningsmyndigheten.se) and from the primary owners, Statistics Sweden (https://www.scb.se/en/services/guidance-for-researchers-and-universities), and the National Board of Health and Welfare (https://www.socialstyrelsen.se/en/statistics-and-data/statistics/).

Data Source

Individuals in Sweden are provided unique registration numbers, allowing the tracking of information over time in linked national data registers for research purposes. The Prescribed Drug Register (PDR) provided information on prescriptions filled starting in July, 2005.²⁷ The National Patient Register (NPR) provided information on inpatient diagnoses since 1987 and specialist outpatient diagnoses since 2001, using the International Classification of Diseases (ICD) versions 9 or 10.^28–30 We used the Medical Birth Register for information on sex and country of birth.^31,32 We used the Integrated Database for Labor Market Research and the Education Register for information on socioeconomic factors.³³

Sample

We included individuals, ages 12–65, in Sweden who initiated antidepressant use between July 1^st 2006 and December 31^st 2013 and a set of comparison individuals who did not use antidepressants during the study period. Comparisons were matched 1:1 to users on year of birth, sex and county of residence in the year of initiation. Antidepressant initiation was defined as the first antidepressant prescription dispensed after a minimum of 365 days washout. We defined treatment periods with two dispensed prescriptions falling within 120 days of each other, as prescriptions for oral medications are typically not filled for more than 90 days at a time in Swedish psychiatric care.³⁴ We added 30 days to allow for non-adherence. Non-initiators were assigned the initiation date of their matched user. Follow-up ended one year after the initiation date; matched pairs were censored by whichever of the following came first: emigration, death, or discontinuation (30 days after the last filled prescription in a treatment period) of the initiating antidepressant in the user. We allowed for repeated seizure events for individuals in our analyses.

We present descriptive information on our sample of 1,317,532 (60.8% female) individuals, half of whom had an initiating antidepressant, in Table 1.

Table 1.

Descriptive information on n=1,317,532 individuals included ages 12–65, followed from one year prior to initiation through 1 year post-initiation

	Users	Non-users
N	658,766 (100.0%)	658,766 (100.0%)
Female	400,283 (60.8%)	400,283 (60.8%)
Age category
12–25 years	135,388 (20.6%)	135,388 (20.6%)
26–35 years	125,513 (19.1%)	125,513 (19.1%)
36–45 years	144,040 (21.9%)	144,040 (21.9%)
46–55 years	131,850 (21.0%)	131,850 (21.0%)
56–65 years	121,975 (18.5%)	121,975 (18.5%)
No. events	5,276	1,562
No. individuals with event	2,571 (0.4%)	888 (0.1%)
No. individuals with a history of seizures
At 1 year before prescription	6,558 (1.0%)	4,326 (0.7%)
At 1 month before prescription	7,362 (1.1%)	4,608 (0.7%)
At prescription	7,534 (1.1%)	4,630 (0.7%)
Mean follow-up (days)	510.4	510.4
Initiating antidepressant group
SSRI/SNRI	470,814 (71.5%)	NA
TCA	86,630 (13.2%)	NA
Other	106,032 (16.1%)	NA

SSRI=selective-serotonin reuptake inhibitor. SNRI=selective serotonin-norepinephrine reuptake inhibitor. TCA=tricyclic antidepressant.

Measures

Exposure Definitions:

We identified antidepressant exposure using anatomical therapeutic chemical (ATC) codes for filled prescriptions beginning with N06A, which are codes for antidepressants. We also used an alternate definition that included selective-serotonin reuptake inhibitors and selective-norepinephrine reuptake inhibitors (SSRIs/SNRIs) only (ATC codes beginning with N06AB, N06AX16, N06AX17, N06AX21, N06AX23) because 1) the majority of individuals in our sample initiated a serotonergic antidepressant (n=470,814, 71.5%) and 2) there are discrepant findings in the literature according to subclass of antidepressant medication. We did not examine other antidepressant subclasses because of the rarity of use (Table 1).

Seizure Visit Definitions:

We defined seizures based on primary ICD-10 diagnosis of epilepsy and seizure (G40 and G41) from any inpatient hospital visit or unplanned outpatient visit recorded in the NPR. We included only unplanned (i.e., scheduled without prior planning with provider) outpatient visits in order to avoid inclusion of outpatient visits for routine epilepsy management that are often scheduled in advance (in which case a seizure may not have occurred recently). Although we included diagnoses to capture seizures that may have occurred prior to an epilepsy diagnosis, we did not have comprehensive coverage of all seizure diagnoses in our linkage of registers. We recognize our outcome definition is imperfect in a number of ways. For instance, misclassification may occur as codes reflect provider judgment about what was primary or most relevant during a particular visit or hospitalization. Additionally, less severe seizures may be missed, as it is common among those with established epilepsy to only seek treatment if medications need adjustment and/or if a serious injury has occurred. We defined previous history of seizures by any inpatient or outpatient ICD 9 (345, 780.39) or ICD-10 (G40 and G41) diagnosis of epilepsy or seizures (excluding infantile spasms recorded in ICD-9: 345.6) prior to the follow-up.

Covariates:

We considered known or hypothesized confounding factors for inclusion in our matched-cohort models. Sociodemographic characteristics included birth country, highest level of education between parents and the index individual, and family income. Missing values (for education and income) were included as a separate category in analyses. Sociodemographic variables were measured one year prior to initiation. Our matched-cohort analyses also adjusted for any history of psychiatric diagnoses (using ICD codes; Table S1) one year prior to initiation, including attention-deficit hyperactivity/disorder (ADHD), autism spectrum disorder, alcohol use disorder, substance use disorder excluding alcohol, anxiety disorder, compulsive disorders, bipolar disorder, depression, suicide attempt, personality disorders, and schizophrenia.

Finally, within-individual analyses (see below) also adjusted for concurrent psychotropic (i.e., antipsychotic medication, anxiolytics, opioid analgesics, hypnotics/sedatives, and ADHD medication) and ASM use identified using ATC codes (Table S1) for filled prescriptions. A sensitivity analysis also added adjustment for any concurrent diagnosis of a psychiatric disorder.

Data analytic plan

The cohort extraction was conducted using SAS 9.4; all analyses were carried out using R 3.5.0. We estimated associations between antidepressants and seizures using Poisson regression (R-package gnm). First, we examined whether those who initiated antidepressant medication were at greater risk of seizure visits compared with matched individuals who did not initiate an antidepressant medication during the entire medicated period (for initiators). We estimated models before and after statistically controlling for the measured covariates. This approach mirrors much of the previous work attempting to understand if antidepressants cause seizures, which has adjusted for covariates, including mood and anxiety diagnoses. This approach also mirrors prior work, which has often do not consider the interdependent fluctuation of seizure risk with depression (and thus any depression treatment)¹. As such, our inclusion of this analysis was meant to compare previous approaches to our other, more robust analyses, which do consider such fluctuations in addition to providing adjustment for stable confounding by design. Thus, we do not see these model estimates as providing a conclusive answer to the aims of the current study but rather important context for our more rigorous analyses. Models, using both exposure indices (i.e., all antidepressants and serotonergic antidepressants). We also estimated associations in the entire cohort and specifically among those with a previous seizure history, as previous research suggests that risk may differ in these groups.^4,6–19

Second, we plotted the seizure visit rate per 1000 person-years in each month of our study period—that is, each month one year prior to initiation through one year after initiation. We did this to explore periods of higher and lower absolute risk, which advances prior work that treats all time the same across medicated and unmedicated periods. Thus, we also used this approach to inform selection of a no-treatment referent period in our within-individual analysis (see below).

Finally, we used conditional Poisson regression to conduct within-individual comparison of the incidence of seizure visits in months prior to and following initiation. Any seizure visits that occurred on the day of initiation were considered to be prior to initiation. This approach uses each person as their own comparison (i.e., compares risk for seizures in the same individual before and after medication initiation), which adjusts for unmeasured time-stable confounding (e.g., stable psychiatric symptoms; genetic factors; history of TBI, stroke, CNS infection, brain tumors) by design.^35,36 We used the month one year prior to initiation as the reference category. We compared seizure visit incidence in this month to all subsequent months during the study period, both before and after initiation. Models were estimated for both exposure indices, as well as among all initiators and those with a previous seizure history. We estimated associations both with and without adjustment for for psychotropic medications and ASMs.

We conducted sensitivity analyses to examine possible sources of bias in our main analyses. First, it is possible that our inclusion of outpatient visits captured some routine maintenance of epilepsy and/or follow up several days or weeks following a seizure event, despite being unplanned. This could lead to misclassification and/or a lack of specificity in time to the any relevant seizure event.To address this, we re-estimated our main analyses using inpatient diagnoses for seizures only. Second, our decision to censor individuals who discontinued their medication during follow-up could have introduced bias if discontinuation was related to subsequent seizure risk (informative censoring). This was important since discontinuation was quite common (i.e., 80.9% and 75.5% of individuals that initiated any antidepressant and SSRIs/SNRIs discontinued within a year, respectively). To understand the potential influence on our main findings we re-estimated models without censoring these individuals. Third, we sought to rule out the influence of psychiatric conditions during follow-up in addition to psychotropic medications and ASMs. Fourth and fifth, we examined whether associations differ by sex and age.

RESULTS

Matched-cohort analyses

In our unadjusted models, we observed higher incidence of hospitalizations and unplanned outpatient visits for seizures among those who initiated any antidepressant (incidence rate ratio [IRR]=3.74, 95% confidence interval [CI]=3.37–4.14) and among those that initiated SSRIs or SNRIs specifically (IRR=3.55, 95% CI=3.16–3.99), compared with matched comparisons. After adjustment for covariates, we observed some attenuation of these associations, but higher seizure visit rates remained among initiators of any antidepressant (IRR=3.14, 95% CI=2.83–3.49) and SSRIs/SNRIs specifically (IRR=3.03, 95% CI=2.69–3.41). Incidence was also higher in initiators with a previous seizure history, although the difference from comparisons was less pronounced (IRR _any=1.60, 95% CI _any=1.42–1.80; IRR_SSRI/SNRI=1.46, 95% CI_SSRI/SNRI=1.27–1.67; Table 2). Although these findings provide important context given much of the prior literature has used a similar approach, these results contrasted our more time sensitive approaches that also accounted for more confounding factors (see below). .

Table 2.

Matched Cohort, Between Individual Models Comparing Seizure Visit Incidence in the Year Following Antidepressant Initiation

	IRR (95% CI)	IRR (95% CI)
	Any antidepressant	SSRI/SNRI
General Population
Unadjusted	3.74 (3.37–4.14)	3.55 (3.16–3.99)
Adjusted	3.14 (2.83–3.49)	3.03 (2.69–3.41)
Among Individuals with a Seizure History
Adjusted	1.60 (1.42–1.80)	1.46 (1.27–1.67)

IRR=incidence rate ratio. CI=confidence interval. SSRI=selective-serotonin reuptake inhibitor.

SNRI=selective serotonin-norepinephrine reuptake inhibitor.

Absolute Rates of Seizures

Among individuals who initiated an antidepressant, 1,882 (0.3%) had one or more seizure visits in the year prior to initiation, and 2,187 (0.3%) had one or more seizure visits in the year following initiation (Table 3). Although seizure visit incidence was low overall, it was higher among those who initiated antidepressants compared with those who did not initiate an antidepressant, even before initiation occurred. In fact, seizure visit incidence was at its highest in the months immediately preceding initiation. After initiation, incidence decreased and remained stable but did not return to the lowest risk point, which occurred before initiation (Figures 1, S1, and S2).

Table 3.

The number of seizure visits among initiators, regardless of if they discontinued their antidepressant

	Year before initiation	Year after initiation
N events	N (%) individuals	N (%) individuals
0	656,884 (99.71)	656,579 (99.67)
1	1,120 (0.17)	1,351 (0.21)
2	404 (0.06)	420 (0.06)
3	160 (0.02)	182 (0.03)
4	84 (0.01)	105 (0.02)
5+	114 (0.02)	129 (0.02)

Figure 1.

Absolute rate of seizure visits in months among users of any antidepressant

Within-individual analyses

Comparisons with 12 months prior to initiation as the referent

We observed higher incidence of visits for seizures in the months after initiation compared with 12 months prior to initiation (e.g., one month after initiation [IRR_any=1.96 95% CI=1.60–2.39; IRR_SSRI/SNRI=2.09, 95%CI=1.68–2.60]) consistent with our pattern of findings for absolute risk. Seizure visit incidence was highest in the months immediately prior to initiating the medication (e.g., one month prior to initiation [IRR_any=2.86, 95% CI=2.42–3.38; IRR_SSRI/SNRI=3.17, 95%CI=2.58–3.88]; Figure 2 and Table S2). We observed a slightly different pattern among those with a previous seizure history, however. Whereas the month prior to initiation similarly showed highest incidence of visits for seizures relative to the month one year prior to initiation, most associations with the months following initiation were either weak or not elevated (e.g., two months after initiation [IRR_any=1.12 95% CI=0.87–1.45; IRR_SSRI/SNRI=1.16, 95%CI=0.85–1.59]; Figure 3 and Table S2). Adjustment for psychotropic medications and ASMs among all antidepressant users produced commensurate results (Figure 4).

Figure 2. Unadjusted within-individual incidence rate ratio (95% CIs) of seizure visits in users in months relative to initiation (baseline: 12 months before initiation).

CI=confidence interval. SSRI=selective-serotonin reuptake inhibitor. SNRI=selective serotonin-norepinephrine reuptake inhibitor.

Figure 3. Unadjusted within-individual rate ratio (95% CIs) of seizure visits in users in months relative to initiation (baseline: 12 months before initiation), among those with a previous seizure history.

CI=confidence interval. SSRI=selective-serotonin reuptake inhibitor. SNRI=selective serotonin-norepinephrine reuptake inhibitor.

Figure 4. Within-individual incidence rate ratio (95% CIs) of seizure visits in users in months relative to initiation for users of any antidepressant (baseline: 12 months before initiation), adjusted for concurrent psychotropic and anti-seizure medications.

CI=confidence interval.

Sensitivity Analyses

Findings from sensitivity analyses suggest that our conclusions were not influenced by our index of seizure visits, censoring, or associations with psychiatric diagnoses. We also observed similar patterns in males and females, as well as across the age range (Figures S3–S7).

DISCUSSION

The present findings replicate previous research documenting an association between antidepressant use and seizures,^4,6–19 as we observed higher incidence of hospitalizations and unplanned outpatient visits for seizures among those that initiated antidepressants compared with matched comparisons. This approach mirrors what much of the current literature has done to address this question; although associations remained after adjustment for several indications of use, including depression and anxiety disorders, residual confounding likely still impacts these results. This is because this approach does not account for several unmeasured factors that are also important to consider (e.g., TBI, stroke). Additionally, it does not account for the interdependent and time-varying relationship between depression, its treatments, and seizures.^1,5,24,25

In our within-individual analyses, seizure visit risk was higher in months following medication initiation, compared with the month one year prior to initiation. The risks following initiation remained fairly stable up to one year of treatment. However, associations were weaker and, in some cases, non-existent among individuals with a seizure history. This could suggest that antidepressant initiation does not influence seizure risk in individuals with epilepsy.

Alternatively, these findings could also be due to different selection patterns into antidepressant treatment among people with more chronic epilepsy, or a lack of seeking medical care following seizures among this population. This latter interpretation may indicate that any seizure increase or exacerbation not captured by our data may reflect a lack of perceived need for care. This may indicate that any worsening of seizures is either minimal or at least perceived as such.

Notably, the months just preceding medication initiation were associated with the highest risk, indicating that individuals are initiating antidepressant medication around the highest-risk period for a seizure visit. Careful interpretation of these results is essential, as these findings could reflect different processes. First, our findings could reflect a protective effect of antidepressants on seizures given that visits for seizures lower following antidepressant initiation, consistent with many who have argued that serotonergic antidepressants in particular should make seizures less likely.^{6, 20–23} Given that our data reflect health care records, our results could also reflect increased contact with the medical care system when patients have seizures, then also leading to screening and treatment for depression, especially since depressive symptoms appear increase with seizure risk..^1–5 This interpretation of our findings could make sense of previous mixed findings in the literature, particularly among other prior observational work..

Although our results may be seen as indicating that serotonergic antidepressants may be safe for the clinical care of depression in individuals with seizures , it important to remember that our data are observational, and we cannot be sure that our findings are a true protective effect. It is also possible that there is still a negative effect of antidepressants on seizures. Because the highest risk period is just prior to antidepressant initiation, and depression symptoms increase along with seizures,⁵ reliance on the contrast between the months leading up to initiation with higher seizure risk and subsequent post-initiation months with lower seizure visit risk may be biased to some degree due to regression to the mean (e.g., increased contact with the health care system leading to better screening for mood and anxiety symptoms).³⁷ A negative impact of antidepressants on seizures, combined with factors such as regression to the mean, would explain why individuals’ risk for seizure visits largely decreased after initiation of antidepressants but did not return to baseline following initiation (observed in both our relative risk and absolute risk estimates).

Comparing our findings to the only other within-individual study of the association between antidepressant use and seizures further highlights the importance of considering the fluctuation of seizures risk and depressive events over time when trying to understand if antidepressant initiation and usecauses seizures. Wu, Liu, Tsai, Liu ⁷ did not explore possible fluctuations in risk around treatment initiation, assuming that all treatment time is equivalent (as is all untreated time). They used the period of 91–120 days prior to a seizure event as their referent,⁷ finding higher risk of seizures related to antidepressant use. Importantly, if we had used a similar referent in our relative risk models, we would have observed similar findings. However, this snap shot in time would not have accurately captured the pattern and variation in seizure risk across our entire study, and we would not have captured data showing the highest risk period being just before medication initiation. This highlights the importance of documenting variation in seizures over time and careful consideration of these fluctuations around selection into treatment. It is possible that past research has observed such discrepant findings in part because most research has either not carefully considered referent selection or examined the time-varying risk of seizures.

This study is subject to several limitations. First, we could not sufficiently control for time-varying confounding by indication (nor the only other within-individual study)⁷ . We adjusted for stable confounding by design in our within-individual analyses, which would include important medical conditions that often cause seizure disorders (e.g., TBI, stroke, CNS infection, brain tumors). We also adjusted for concurrent psychiatric diagnoses (including depression, anxiety disorders, and substance use disorders) and psychotropic and ASM use. However, confounding is likely to still impact our findings, particularly given we were not able to capture time-varying events (e.g., depressive symptoms, substance use) that did not lead to diagnosis. Second, because SSRIs and SNRIs comprised the largest percentage of initiated medications, we were unable to study the individual associations with other subclasses of medications that have been implicated in increasing seizure risk, namely tricyclic antidepressants. For the same reason, we were unable to investigate the associations with individual drugs of concern (i.e., bupropion, clomipramine, amoxapine, and maprotiline).^7,11,38 Third, we cannot be sure that filled prescriptions were taken as recommended, meaning our results are analogous to intention-to-treat analyses. This also means that our findings do not rule out the possibility that seizure occurrence may be related to antidepressant non-adherence. Fourth, it is possible that our findings will not generalize to other contexts outside of Sweden, as prescribing patterns (and thus selection into medication initiation) may be different. It will be important for future research to examine if associations are robust across contexts. Fifth, we did not have comprehensive access to all seizure diagnoses that were made prior to an epilepsy diagnosis, and we also only captured seizure events in which individuals sought care. Further, it is possible that non-epileptic seizures may have been captured as seizures in our data. As such, we may only be capturing the most severe events and/or visits in which providers deemed epilepsy codes to be relevant to the hospitalization or visit. Sixth, due to the limited range of years of coverage, we were not able to use a longer wash-out period. Seventh, although it is standard to exclude older adults from study because increased seizure incidence at this age may be for different reasons than seizures in younger populations,^9,10 additional research is needed in adults above the age of 65.

The present study examined the extent to which seizure visits are related to antidepressant initiation in the general population and among individuals with a previous seizure history by combining the use of 1) a large, population-based cohort, 2) estimation of seizure visit risk across time, and 3) within-individual comparisons that adjusted for stable confounding by design combined with statistical adjustment for time-varying factors. Although it is important to highlight the need for additional work to address challenges around determining causality, our findings demonstrate that individuals initiate antidepressants around the time of highest risk for a hospitalization or an unplanned outpatient visit for a seizure. Risk was lower among individuals with a history of seizures, which may indicate lower risk among these patients or a lack of treatment seeking if seizures worsen following antidepressant initiation. Our study highlights the need for research to address possible bias in estimates related to the fluctuation of seizure risk and depressive symptoms over time. To our knowledge the current study is the first to attempt to reconcile discrepancies in the literature by considering dynamic influences using a within-individual design.

HIGHLIGHTS.

• Previous research examining risk for seizures related to antidepressant initiation has not considered the interdependent time-varying relationship between depression and seizures.

• Using a within-individual design, we examined risk for seizure-related hospitalizations and unplanned outpatient visits related to antidepressant initiation in a large national sample, with consideration of the fluctuation in seizure visit occurrence in months before and after antidepressant intitiation.

• Although we observed an association between antidepressant iniaition and risk for seizure visits, the highest risk period for seizure visits was actually the months leading up to antidepressant initiation, after which risk lowers; these results are consistent with no increase in seizures associated with the initiation of antidepressants and possibly even be a decrease in seizures, though findings may also reflect regression to the mean.

• Risk for seizure visits was lowest among individuals with a history of prior seizures, which may be reassuring for the clinical care of these patients or suggest lack of treatment-seeking following seizures.

• This study highlights the need to consider seizure risk across time; the failure to account for these dynamics may help account for discrepant findings in previous studies.