Received: 26 May 2021 | Accepted: 6 October 2021 DOI: 10.1111/ene.15137 ORIGINAL ARTICLE Antidepressant use and stroke or mortality risk in the elderly Begüm Irmak Ön1 | Xavier Vidal2 | Ursula Berger1 | Mònica Sabaté2 | Elena Ballarín2 | Olga Maisterra3 | Antonio San-­Jose4 | Luisa Ibáñez2 Abstract 1 Institute for Medical Information Processing, Biometry and Epidemiology, LMU Munich, Munich, Germany Background and purpose: Current evidence on antidepressant-­related stroke or mortal- 2 Clinical Pharmacology Service, ity risk is inconsistent. Because the elderly have the highest exposure to antidepressants, Department of Pharmacology the aim was to quantify their association with stroke and mortality risks in this vulnerable Therapeutics and Toxicology, Vall d'Hebron University Hospital, Fundació population. Institut Català de Farmacologia, Methods: Persons over 65 years old and registered in the Information System for Autonomous University of Barcelona, Barcelona, Spain Research in Primary Care of Catalonia during 2010–­2015 comprised the study popu- 3 Vall d'Hebron University Hospital lation. Antidepressant exposure was categorized into current-­users, recent-­users, past-­ Neurology Service, Barcelona, Spain 4 users and antidepressant non-­users (controls). The effect of antidepressant exposure on Vall d'Hebron University Hospital Internal Medicine Service, Barcelona, Spain stroke or death, whichever came first, was analyzed by Cox regression adjusted for es- tablished risk factors. Correspondence Luisa Ibáñez, Clinical Pharmacology Results: Of the 1,068,117 participants included, 20% had antidepressant reimbursements Service, Department of Pharmacology during follow-­up, 17% had a stroke and 3% died. The risk of experiencing stroke or death Therapeutics and Toxicology, Vall d'Hebron University Hospital, Fundació was higher in antidepressant current-­users (hazard ratio [HR] 1.04; 95% confidence in- Institut Català de Farmacologia, terval [CI] 1.02–­1.06), recent-­users (HR 3.34; 95% CI 3.27–­3.41) and past-­users (HR 2.06; Autonomous University of Barcelona, Pg. Vall d'Hebron 119–­129, Barcelona 08035, 95% CI 2.02–­2.10) compared to antidepressant non-­users. Antidepressant current-­use Spain. was associated with increased stroke (HR 1.56; 95% CI 1.50–­1.61) but decreased mortal- Email: li@icf.uab.cat ity risk (HR 0.93; 95% CI 0.91–­0.94). During antidepressant recent-­use and past-­use, both Funding information stroke and mortality risks were significantly increased compared to no antidepressant This project received funding from the Instituto de Salud Carlos III (grant use. PI15/01429) and is co-­funded by the Conclusions: Antidepressant use may be associated with increased stroke risk in the European Regional Development Fund (FEDER). The funding sources did not elderly. When using antidepressants in this population, the potential risks should be interfere with the study conduct considered. KEYWORDS aged, antidepressive agents, mortality, pharmacoepidemiology, stroke I NTRO D U C TI O N prevalence of antidepressant use in all European databases analyzed in many pharmacoepidemiology studies [1]. Antidepressant use has been rising in developed countries [1]. In Although the elderly have the highest exposure to antidepres- Catalonia, antidepressant consumption increased by 22% between sants, limited data are available to justify their use in this vulnerable 2007 and 2011 [2]. The prevalence of antidepressant consumption population. They are usually excluded from randomized controlled in the Spanish elderly population significantly increased from 4% trials [4] and meta-­analyses show small effect sizes for antidepres- in 2003 to 8% in 2009 [3]. Age was positively correlated with the sants compared to placebo with no substantial efficacy difference This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2021 The Authors. European Journal of Neurology published by John Wiley & Sons Ltd on behalf of European Academy of Neurology. Eur J Neurol. 2022;29:469–477. wileyonlinelibrary.com/journal/ene | 469 470 | ÖN et al. between antidepressant classes [4–­6]. Considering their favorable 74% of the Catalan population, Spain [16]. The analysis dataset, tolerability, guidelines recommend selective serotonin reuptake in- retrieved in 2017, consisted of variables related to demographics, hibitors (SSRIs) as first-­line therapy for depression, reflecting results diagnoses (International Classification of Diseases [ICD] 10), smok- from randomized controlled trials in younger adults [6]. Age-­related ing and number of practitioner visits from primary care records on a polypharmacy, multimorbidity and physiological changes affecting daily basis. Monthly drug reimbursements (Anatomical Therapeutic drug metabolism and action in the elderly make such an extrapola- Chemical [ATC]) by the National Health System dispensed to outpa- tion problematic [4]. tients with prescriptions from CIH physicians were available without Stroke is the second cause of death worldwide [7]. Each 10-­year a specific date. The date of death from the Health Department was increase in age doubles the annual stroke risk [8]. SSRIs and other linked to the demographics table. The dataset covered the period antidepressants, which include but are not limited to serotonin–­ from 1 January 2009 to 31 December 2015. norepinephrine reuptake inhibitors (SNRIs), were associated with higher risk of stroke in a geriatric population, whereas tricyclic antidepressants (TCAs) were not [9]. Yet, another study had contradicting conclusions: Study population SSRIs, but not TCAs, were protective against stroke [10]. In terms of mortality, elderly men from a French cohort were more likely to die if Persons registered in SIDIAP between 1 January 2010 and 31 they were on antidepressant medication versus not on antidepressant December 2015 were eligible if they were born between 1900 and medication [11]. These results were replicated in Australia [12]. A data- 1950. Persons who exited the CIH healthcare system during the base study found similar results, without differentiating the sex effect study period were considered ineligible to prevent potential dupli- [9]. However, current antidepressant treatment was found to reduce cates due to reentry. mortality in a user-­only study analyzing all age groups [13]. The cohort was formed from eligible persons who entered the co- With increasing life expectancy and reduced fertility, the pro- hort in the initial month they met the inclusion criteria. Persons were portion of the population aged over 65 years is projected to increase included when they were over 65 years old. Participants were included from 19% in 2020 to 27% by 2050 and to 30% by 2100 in developed if they were registered in SIDIAP for at least 12 months before in- countries [14]. Therefore, elucidating the relationship between com- clusion, provided they had at least one primary care visit [17] and no mon exposures and serious health issues within this population is antidepressant reimbursement during this period. The cohort was fol- crucial. Our primary objective was to estimate the risk of stroke or lowed until death or end of the study period. mortality related to antidepressant use compared to no antidepres- sant use in the elderly. As subgroup analyses, this risk was assessed for separate antidepressant classes. The aim was also to separately Exposure quantify the risks for ischaemic stroke, hemorrhagic stroke, unspec- ified stroke, death and fatal stroke. The follow-­up time was divided into antidepressant treatment epi- sodes based on the reimbursement data. The current-­use episode was formed by joining consecutive months of antidepressant reim- M E TH O D S bursement including 1 month gaps (the grace period) with no reim- bursement. Switching to different antidepressant agents was also Study design allowed. When a current-­user stopped antidepressant treatment, the first 2 months following the grace period were categorized as An observational cohort study was conducted, including persons recent-­use. Past-­use was defined as the 9 months immediately fol- aged over 65 and registered in the Information System for Research lowing recent-­ use. Consecutive months without antidepressant in Primary Care (SIDIAP). Ethics Committee approval was received reimbursement at the current month and for at least 12 preced- from the University Institute for Primary Care Research Jordi Gol ing months were joined to create a no antidepressant use episode. Clinical Research on 12 May 2016. Informed consent from the Graphical presentation of the treatment episode construction is participants was not necessary because of the retrospective na- available online in Appendix S1. For subgroup analysis, the antide- ture of the anonymized data collection. This study is registered in pressants reimbursed during a use episode were categorized into the the European Union Post-­Authorization Studies Register (number following classes: TCAs, SSRIs, antidepressants other than TCAs or EUPAS19188) [15]. SSRIs, and multiple (drugs from multiple antidepressant classes dur- ing a single use episode). Data source Outcomes The Information System for Research in Primary Care database con- tains electronic medical records of persons receiving health services The outcome was detected from the fields of diagnoses and the from the Catalan Institute of Health (CIH), comprising approximately death registry in the SIDIAP dataset. Primary outcome was the ANTIDEPRESSANTS AND STROKE OR DEATH RISK | 471 composite of stroke and all-­cause mortality (as date of death) to ac- of stroke events that occurred during the initial month of current-­ count for competing risks [18]. For secondary analyses, stroke was use due to the level of information on dispensations being to the categorized as ischaemic (I63), hemorrhagic (I60–­I62) or unspeci- month rather than to the date; (3) if lacking data on inpatient med- fied (G46, I64). Stroke was considered fatal if death occurred within ications would explain the observed results; (4) if drugs other than 30 days. antidepressants showed a similar trend in terms of the results; (5) the effect of conceptualizing the outcome with a subdistributional hazards model (Fine and Gray) rather than a cause-­specific hazards Covariates model; (6) if combining SSRIs with SNRIs to form a single group tar- geting the serotonin reuptake mechanism and considering acetyl- Age, sex and urbanity (demographic variables) were assessed at co- salicylic acid as an independent comedication for adjustment rather hort entry. Smoking status was determined during the study period. than as part of other comedication groups would change the results; The number of primary care visits during the year before episode-­ and (7) if multiple imputation of missing urbanity level rather than start was used as a proxy for health-­seeking behavior. Comorbidities complete-­case analysis would change the results. Details of the sen- and comedications that are established or possible risk factors for sitivity analyses are available online in Appendix S1. stroke or antidepressant use were selected. Comorbidities were collected from primary care diagnoses records any time before an- tidepressant treatment episode-­start. Comedication reimbursements R E S U LT S were determined during 6 months before the episode-­start. All covar- iates are described in Table 1. Lists of all the variables and their defini- From 1,307,316 eligible persons, 1,069,121 persons were recruited tions based on ATC or ICD codes are available online in Appendix S1. the first month that they met the inclusion criteria (Figure 1). From these, 1004 (0.09%) were further excluded because of missing ur- banity level, leaving 1,068,117 (82% of 1,307,316) persons in the Statistical analysis analysis cohort. Of the final cohort, 212,621 persons (20%) had at least one Time to the composite outcome was analyzed by the time-­varying (ever-­users) and 855,496 (80%) had no antidepressant reimburse- Cox proportional hazards model. The composite outcome was cho- ment (never-­users) during follow-­up. Baseline characteristics are sen as the main outcome because both death and stroke are clinically presented in Table 1. At recruitment, ever-­users were older, predom- relevant. Exposure was the time-­dependent variable. No antidepres- inantly female, more likely to be non-­smokers, had more primary sant use episode was the reference category. The effect estimates care visits during the previous year, and were more predisposed to were adjusted for time-­constant recruitment month, demographic stroke risk factors, compared to never-­users. There were 4,303,527 variables and smoking. Further adjustment for comorbidities, comed- person-­ years in no antidepressant use, 264,476 person-­ years in ications and number of primary care visits was employed by allowing current-­use, 47,963 person-­years in recent-­use and 98,459 person-­ them to be updated whenever a person changed from one exposure years in past-­use episodes. episode to another. In subgroup analyses, use episodes of individual During follow-­up, 19% of the participants experienced the com- antidepressant classes were compared to the episodes of non-­use. posite outcome, 17% died, and 3% experienced a stroke, of which When the composite outcome was the dependent variable, a 3% were fatal. Of the stroke events, 8% were hemorrhagic, 17% person's observation was terminated if the person experienced were ischaemic, whereas the type was unspecified in 75%. Crude stroke or death. For the analysis of stroke, a person's observation incidence rates of death and stroke were significantly higher in use was censored on the date of death, a cause-­specific hazard model or episodes than in the no antidepressant use episode. Amongst par- at the end of the study. For the analysis of death, the observations ticipants on antidepressants, 60% were using SSRIs, 8% were using of persons who survived up to the end of the study were censored. TCAs, 25% were using an antidepressant other than SSRIs or TCAs, Following recruitment, persons with missing urbanity level were ex- and 7% were using antidepressants from more than one class in a cluded from the analysis because their proportion in the total cohort single episode. Crude outcome statistics and unadjusted effect esti- was negligible. mates are available in Appendix S1. An alpha level of 0.05 was used in the Wilcoxon rank-­sum test, In the fully adjusted analysis (Table 2), risk of composite outcome the chi-­squared test and the proportion test to compare baseline during antidepressant current-­use was significantly higher (HR 1.04; characteristics. Effect sizes were reported as crude incidence rates 95% CI 1.02–­1.06) than during no antidepressant use. The composite per 1000 person-­years and adjusted hazard ratios (HRs) with 95% outcome was even more likely during recent-­use (HR 3.34; 95% CI confidence intervals (CIs). Analyses were performed using SAS 9.4. 3.27–­3.41) and past-­use (HR 2.06; 95% CI 2.02–­2.10). Risk of death Various post hoc sensitivity analyses were performed to address was significantly lower during antidepressant current-­use (HR 0.93; and explore the effects of the data source and our assumptions on 95% CI 0.91–­0.94) than no antidepressant use. Stroke risk was higher the results. These were (1) updating covariates regularly instead of during antidepressant current-­use than during no antidepressant use only at episode changes; (2) the effect of potential misclassification (HR 1.56, 95% CI 1.50–­1.61) and gradually decreased in recent-­use 472 | ÖN et al. TA B L E 1 Baseline characteristics Characteristic Ever-­users Never-­users p value Number of people 212,621 855,496 Age (years), median (interquartile range) 74.8 (68.3–­8 0.8) 70.4 (65.0–­78.3) <0.0001a Sex Female 67% 50% <0.0001b Urbanity Rural 19% 20% <0.0001b Smoking Current smoker 15% 20% <0.0001b Ex-­smoker 10% 13% Non-­smoker 73% 63% Missing 3% 4% Primary care visits, median (interquartile 7 (4–­12) 5 (3–­9) <0.0001a range) Comorbidities <0.0001b No comorbiditiesc 8% 12% <0.0001d Atrial fibrillation 7% 6% <0.0001d Cardiovascular diseases 68% 62% <0.0001d Chronic kidney diseases 7% 6% <0.0001d Chronic lower respiratory diseases 9% 8% <0.0001d Coagulation/platelet defects 1% 1% 0.45d Dementia 3% 2% <0.0001d Diabetes 22% 21% <0.0001d Diseases of the digestive system 0.5% 0.6% <0.01d Lipidemia disorders 45% 42% <0.0001d Mental and behavioral disorders 43% 32% <0.0001d Neoplasms 14% 13% <0.0001d Nervous system diseases 5% 3% <0.0001d Neuropathic pain and fibromyalgia 3% 2% <0.0001d Pneumonia 3% 3% 0.69d Stroke 6% 4% <0.0001d Transient cerebral ischaemic attacks 2% 2% <0.0001d Comedications <0.0001b No comedicationse 11% 18% <0.0001d Analgesics 50% 36% <0.0001d Anti-­dementia drugs 3% 2% <0.0001d Anti-­inflammatory and antirheumatic 37% 28% <0.0001d drugs Antipsychotics 4% 3% <0.0001d Antithrombotic drugs 31% 27% <0.0001d Systemic corticosteroid use 5% 4% <0.0001d Drugs of cardiovascular system 72% 66% <0.0001d Hypoglycemics 18% 16% <0.0001d Hypnotics and sedatives 40% 20% <0.0001d a Wilcoxon rank-­sum test. b Chi-­squared test. c Patients with none of the selected comorbidities. d Single proportion test. e Patients with none of the selected comedications. ANTIDEPRESSANTS AND STROKE OR DEATH RISK | 473 1,307,316 eligible persons: current-­use, it became non-­significant during recent-­use and past-­ • born between 1900-1950, use episodes. Low numbers of fatal stroke events did not allow the • not transferred out of the SIDIAPa system comparison of antidepressant classes. Excluded n = 97,738 Persons who died before recruitment Sensitivity analyses Persons who were alive at the time of recruitment n = 1,209,578 The assumptions of and the results from the main analysis were Excluded n = 2,761 robust to updating covariates regularly instead of only at episode Persons in SIDIAP for less than one year changes (sensitivity analysis 1), considering SSRIs and SNRIs as Persons in SIDIAP for more than a single group and acetylsalicylic acid as a separate comedication one year n = 1,206,817 for adjustment (sensitivity analysis 6) and the multiple imputation Excluded n = 33,017 of missing urbanity level (sensitivity analysis 7). When the stroke Persons with no primary care outcome was conceptualized with a subdistributional hazards model visits during previous year (sensitivity analysis 5), which is the preferred method of taking into Persons with at least one primary care visit during previous year account competing risks when the aim is prediction but not causal n = 1,173,800 inference, the risk of stroke during current-­use of antidepressants Excluded n = 104,679 was slightly but non-­significantly increased and the risk of stroke Persons with antidepressant reimbursed during previous year during recent-­use and past-­use of antidepressants was decreased Persons recruited the first month compared to the main analysis. they met the inclusion criteria When all stroke events that occurred during the initial month n = 1,069,121 of antidepressant use were considered to occur before the dispen- Excluded from analysis n = 1,004 Persons with missing information sation rather than after (sensitivity analysis 2), the direction of the on urbanity effect of antidepressant current-­ use on the composite outcome Persons in the analysis cohort n= 1,068,117 changed direction and the risk became slightly decreased compared to no antidepressant use, but the current-­use of antidepressants F I G U R E 1 Flow-­chart with final reason for exclusion. aSIDIAP, still posed harm with regard to the stroke event. The pattern of Information System for Research in Primary Care decreased death during current-­use and increased death following discontinuation of antidepressants may be attributable to the limita- (HR 1.45; 95% CI 1.33–­1.58) and past-­use (HR 1.21; 95% CI 1.13–­ tion of our data source of dispensation to outpatients because the 1.30) but remained significant. odds of being hospitalized within current-­use was more likely than Compared to no antidepressant use, current-­and recent-­ use being hospitalized during other time periods (sensitivity analysis 3). were associated with increased risk of not only fatal stroke (HR Also, two drug classes tested other than antidepressants, nonste- 1.31; 95% CI 1.09–­1.59) but also all stroke subtypes (ischaemic, roidal anti-­inflammatory drugs and antipsychotics, showed a similar hemorrhagic and unspecified). The simultaneous occurrence of trend in terms of trend of death, in support of the above-­mentioned hemorrhagic and ischaemic stroke was not analyzed because of an hypothesis (sensitivity analysis 4). All statistics from the main unad- insufficient number of events (n = 16). justed and fully adjusted analyses, sensitivity analyses and analysis Analysis of the composite outcome risk stratified by antide- for effect modification by sex are available in Appendix S1. pressant class (Figure 2) revealed significantly decreased risk for current-­use of the TCAs, SSRIs and multiple antidepressants categories but significantly increased risk for the class of other DISCUSSION antidepressants(HR 1.21; 95% CI 1.18–­1.25) compared to no an- tidepressant use. In contrast, antidepressant recent-­use and past-­ The risk of stroke or death was significantly higher during antide- use increased this risk significantly. The highest risk of composite pressant use than with no antidepressant use. The use of antide- outcome was associated with the use of multiple antidepressants pressants other than TCAs or SSRIs increased the risk of stroke or (more than 1.5 times compared to all antidepressants during death. Also, previous use of antidepressants from multiple classes recent-­use and past-­use), whilst TCAs showed the lowest risk in all highly increased this risk. Amongst antidepressant classes, the categories of use episodes. hazard of death or stroke was the lowest with TCA use. Risk of A pattern similar to that of the composite outcome is observed antidepressant-­related stroke was highest whilst on treatment and in the analysis regarding the isolated outcome of death (Figure 3). gradually decreased but stayed significantly high up to 12 months Use of SSRIs or the class of other antidepressants significantly in- following discontinuation. Risk of antidepressant-­related death was creased stroke risk compared to no antidepressant use in all expo- significantly lower than no antidepressant use during treatment but sure windows (Figure 4). Although stroke risk was high during TCA significantly increased after withdrawal. 474 | ÖN et al. Composite 10,430 10,439 Current-use 859 672 118 28 69 n <0.0001 <0.0001 <0.0001 <0.0001 Recent-use p value 1.00 0.89 0.02 Note: Adjusted hazard ratios (HRs) with 95% confidence intervals (CIs), and number of outcome events (n) for episodes, including the reference category of no antidepressant use. Past-use 1.00 (0.83–­1.20) 1.03 (0.70–­1.50) 2.17 (2.13–­2.22) 1.32 (1.04–­1.69) 2.06 (2.02–­2.10) 1.21 (1.13–­1.30) 1.25 (1.15–­1.35) HR (95% CI) null 0 2 4 6 Past-­use Hazard Ratio Class All SSRI TCA Others Multiple F I G U R E 2 Risk of antidepressant use compared to no antidepressant use by antidepressant class—­composite outcome. Summary graphics of hazard ratios (with 95% confidence intervals) 9441 9165 400 546 84 62 26 of use episodes to no antidepressant use by antidepressant class. n SSRI, selective serotonin reuptake inhibitors; TCA, tricyclics; Others, other antidepressants; Multiple, more than one class of <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 antidepressant in a single use episode [Colour figure can be viewed p value 0.02 <0.01 at wileyonlinelibrary.com] Death 3.59 (3.52–­3.67) 1.39 (1.26–­1.54) 1.45 (1.33–­1.58) 1.63 (1.10–­2.43) 3.34 (3.27–­3.41) 1.36 (1.10–­1.69) 2.27 (1.76–­2.94) HR (95% CI) Recent-­use Current-use Recent-use 13,183 15,755 2868 3847 665 138 312 Past-use n TA B L E 2 Risk of antidepressant use compared to no antidepressant use null 0 2 4 6 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 p value <0.01 Hazard Ratio Class All SSRI TCA Others Multiple F I G U R E 3 Risk of antidepressant use compared to no 1.52 (1.46–­1.58) 1.56 (1.50–­1.61) 0.93 (0.91–­0.94) 1.04 (1.02–­1.06) 1.31 (1.09–­1.59) 1.73 (1.53–­1.96) 1.65 (1.51–­1.80) antidepressant use by antidepressant class—­outcome death. Current-­use HR (95% CI) Summary graphics of hazard ratios (with 95% confidence intervals) of use episodes to no antidepressant use by antidepressant class. SSRI, selective serotonin reuptake inhibitors; TCA, tricyclics; Others, other antidepressants; Multiple, more than one class of antidepressants in a single use episode [Colour figure can be viewed at wileyonlinelibrary.com] 168,922 146,954 22,337 No use 29,400 4879 2170 959 n In a database of outpatients being reimbursed, antidepressant use was associated with significantly lower mortality compared to no antidepressant use in individuals aged over 10 [13]. Prospective Composite outcome Hemorrhagic stroke Unspecified stroke Ischaemic stroke studies in elderly cohorts found no such decrease in risk [11,12]. In Stroke (total) our study, compared to those who were not using antidepressants, Fatal stroke 10% fewer deaths were observed amongst patients using antide- Death pressants. However, following treatment discontinuation, 3.6 times more patients died during the initial 2 months and 2.2 times more ANTIDEPRESSANTS AND STROKE OR DEATH RISK | 475 Stroke common. The literature on prophylactic antidepressant treatment in stroke survivors makes both options probable [24]. Current-use High population coverage of our real-­world data source and not restricting the indication during cohort selection make our results generalizable. Another strength of this study is its reliance on re- Recent-use imbursement data to determine the antidepressant exposure, con- sidered an accurate source in pharmacoepidemiology [25]. Although medications paid out-­of-­pocket are absent and secondary treatment Past-use adherence is unclear in reimbursement datasets, these limitations would lead to bias towards the null and not undermine our results. 0.50 0.75 1.00 1.25 1.50 1.75 Data from death registries are also considered valid. However, di- Hazard Ratio agnoses of stroke may be incompletely reflected in primary care Class All SSRI TCA Others Multiple records. Still, after adjusting for health-­seeking behavior, no differ- ential recording of diagnoses within antidepressant exposure and F I G U R E 4 Risk of antidepressant use compared to no non-­exposure periods was expected. antidepressant use by antidepressant class—­outcome stroke. From our analysis, it cannot be inferred by which mechanism Summary graphics of hazard ratios (with 95% confidence intervals) antidepressants affect the risk of stroke or mortality. It has been of use episodes to no antidepressant use by antidepressant class. suggested that increased fibrinolytic activity and weight gain may SSRI, selective serotonin reuptake inhibitors; TCA, tricyclics; mediate the effect of antidepressants on bleeding or thrombotic Others, other antidepressants; Multiple, more than one class of antidepressants in a single use episode [Colour figure can be events [26]. These are interesting hypotheses for future studies. viewed at wileyonlinelibrary.com] Antithrombotic drugs, like serotonergic antidepressants, are known to increase the risk of bleeding. Also, atrial fibrillation is known to patients died during the following period up to 12 months. Drug increase the risk of stroke. In our cohort, significantly more anti- reimbursement information was only available from outpatients, depressant users had atrial fibrillation and were on antithrombotic which might have caused misclassification of exposure. Persons drugs than antidepressant never-­users, although the significance with worsening health are likely to be hospitalized. This might have of these differences is due to the high number of participants. Our caused informative stopping of reimbursement and create an arti- analysis and reported effect estimates are adjusted for these and ficial risk increase afterwards. This unexpected finding might also other possibly confounding comorbidities and comedications. The be explained by the tendency of the elderly to be deprescribed the sensitivity analyses taking into account acetylsalicylic acid as a sepa- drugs for less critical indications like mental health problems when rate category of comedication and serotonergic antidepressants as a they are close to death [19]. Or, elderly individuals with health prob- single category did not substantially change the results. Yet, another lems requiring antidepressant treatment would be at increased risk potential limitation may be due to the lack of data on testosterone of death after discontinuing antidepressants. use, which may have acted as a confounding factor and contributed Based on a total of 10 observational studies, a meta-­analysis to stroke or death, especially in men [27]. concluded that, compared to no antidepressant treatment, SSRI use In conclusion, antidepressant use may be associated with in- significantly increased cerebrovascular disease risk, whilst TCAs creased risk of stroke or mortality in the elderly. These results did not [20]. This result was confirmed by a later study in an elderly should be treated with caution because residual confounding or Italian cohort with existing cardiovascular diseases [21]. In our study, misclassification cannot be ruled out. The highly increased mortality controlling for history of cardiovascular diseases, a higher stroke risk risk following the discontinuation of antidepressant treatment war- associated with the use of all antidepressant classes was found, ex- rants further analysis. Prior to initiating antidepressant therapy in cept for current simultaneous use of multiple antidepressant classes. the elderly, physicians should evaluate its possible benefits against Similarly, a case-­crossover study found increased stroke risk with the possible risks of stroke or death. all antidepressants, the most pronounced being SSRIs, followed by other antidepressants, and TCAs. Yet the class differences were not AC K N OW L E D G E M E N T S significant [22], similar to a study investigating different cardiovas- SIDIAP and the Department of Health provided the data. This article cular event risks between SSRI and non-­SSRI users [23]. Although has not been written in conjunction with the professionals of the da- our results indicate an increased risk of stroke with current antide- tabase. The results and conclusions are the sole responsibility of the pressant treatment, any conclusion should be treated with caution. authors of this study. Drs Pili Ferrer and Lina Camacho contributed One-­third of stroke events occurred during the initial month of an- to the study's conception. tidepressant reimbursement and may have been misclassified as an event because only the month of exposure, not the exact date, was C O N FL I C T O F I N T E R E S T available to us. Either antidepressant exposure has an immediate The authors have no conflicts of interest in connection with this effect on risk of stroke or post-­stroke antidepressant treatment is article. 476 | ÖN et al. AU T H O R C O N T R I B U T I O N S Psychogeriatrics. 2014;14(4):261-­268. https://doi.org/10.1111/ Irmak B. Ön: Conceptualization (equal); data curation (equal); psyg.12057 7. Feigin VL, Krishnamurthi RV. Global Burden of Stroke. In Grotta formal analysis (lead); methodology (equal); visualization (lead); JC, Albers GW, Broderick JP, Kasner S, Lo E, Sacco R, Wong L, eds. writing—­original draft (lead); writing—­review and editing (equal). Stroke, 6th edn. Elsevier; 2016:165-­206:chap 13. Xavier Vidal: Conceptualization (equal); formal analysis (support- 8. Seshadri S, Wolf PA. Modifiable Risk Factors and Determinants of ing); methodology (equal); supervision (equal); writing—­original Stroke. In Grotta JC, Albers GW, Broderick JP, Kasner S, Lo E, Sacco R, Wong L, eds. Stroke, 6th edn. Elsevier; 2016:217-­233:chap 15. draft (supporting); writing—­review and editing (supporting). 9. Coupland C, Dhiman P, Morriss R, Arthur A, Barton G, Hippisley-­ Ursula Berger: Conceptualization (equal); methodology (equal); Cox J. Antidepressant use and risk of adverse outcomes in older supervision (equal); writing—­original draft (supporting); writing—­ people: population based cohort study. BMJ. 2011;343:d4551. review and editing (supporting). Mònica Sabaté: Conceptualization https://doi.org/10.1136/bmj.d4551 10. Lee YC, Lin CH, Lin MS, Lin JW, Chang CH, Lai MS. Effects of selec- (equal); methodology (equal); supervision (equal); writing—­original tive serotonin reuptake inhibitors versus tricyclic antidepressants draft (supporting); writing—­review and editing (supporting). Elena on cerebrovascular events: a nationwide population-­based cohort Ballarín: Conceptualization (equal); methodology (equal); supervi- study. J Clin Psychopharmacol. 2013;33(6):782-­789. https://doi. sion (equal); writing—­original draft (supporting); writing—­review org/10.1097/JCP.0b013​e3182​9c970e 11. Ryan J, Carriere I, Ritchie K, et al. Late-­life depression and mor- and editing (supporting). Olga Maisterra: Conceptualization tality: influence of gender and antidepressant use. Br J Psychiatry. (equal); methodology (equal); supervision (equal); writing—­original 2008;192(1):12-­18. https://doi.org/10.1192/bjp.bp.107.039164 draft (supporting); writing—­review and editing (supporting). 12. Almeida OP, Alfonso H, Hankey GJ, Flicker L. Depression, antide- Antonio San-­Jose: Conceptualization (equal); methodology (equal); pressant use and mortality in later life: the Health in Men Study. PLoS ONE. 2010;5(6):e11266. https://doi.org/10.1371/journ​ supervision (equal); writing—­original draft (supporting); writing—­ al.pone.0011266 review and editing (supporting). Luisa Ibáñez: Conceptualization 13. Haukka J, Arffman M, Partonen T, et al. Antidepressant use and (equal); funding acquisition (lead); methodology (equal); supervi- mortality in Finland: a register-­linkage study from a nationwide sion (equal); writing—­original draft (supporting); writing—­review cohort. Eur J Clin Pharmacol. 2009;65(7):715-­720. https://doi. and editing (supporting). org/10.1007/s0022​8-­0 09-­0616-­9 14. World Population Prospects 2019. United Nations DoEaSA, Population Division; 2019:240–­ 241. Volume I: Comprehensive DATA AVA I L A B I L I T Y S TAT E M E N T Tables (ST/ESA/SER.A/426). www.unpop​ulati​on.org. Accessed The data that support the findings of this study belong to the Jordi January 03, 2021. Gol Institut and SIDIAP database. Restrictions apply to the availabil- 15. Ibañez L. Use of antidepressant drugs and its association with risk of stroke, frequency of hospitalization, and mortality in an elderly ity of these data, which were used under license for this study. population: a descriptive and analytic cohort study. The European Union electronic Register of Post-­Authorisation Studies (EU PAS ORCID Register). Updated June 1, 2017. http://www.encepp.eu/encep​p/ Begüm Irmak Ön https://orcid.org/0000-0002-9008-0141 viewR​esour​ce.htm?id=19189. Accessed July 1, 2019. 16. Database. The Information System for the Development of Primary Luisa Ibáñez https://orcid.org/0000-0002-1175-0574 Care Research (SIDIAP). http://www.sidiap.org/index.php/. Accessed May 30, 2018. REFERENCES 17. Hernan MA, Robins JM. Using big data to emulate a target trial when 1. Abbing-­ Karahagopian V, Huerta C, Souverein PC, et al. a randomized trial is not available. Am J Epidemiol. 2016;183(8):758-­ Antidepressant prescribing in five European countries: applica- 764. https://doi.org/10.1093/aje/kwv254 tion of common definitions to assess the prevalence, clinical ob- 18. Austin PC, Lee DS, Fine JP. Introduction to the analysis of survival servations, and methodological implications. Eur J Clin Pharmacol. data in the presence of competing risks. Circulation. 2016;133(6):601-­ 2014;70(7):849-­857. https://doi.org/10.1007/s0022​8-­014-­1676-­z 609. https://doi.org/10.1161/circu​latio​naha.115.017719 2. Gomez-­Lumbreras A, Ferrer P, Ballarín E, et al. Study of antidepres- 19. Glynn RJ, Knight EL, Levin R, Avorn J. Paradoxical relations of sant use in 5 European settings. Could economic, sociodemographic drug treatment with mortality in older persons. Epidemiology. and cultural determinants be related to their use? J Affect Disord. 2001;12(6):682-­689. https://doi.org/10.1097/00001​6 48-­20011​ 2019;249:278-­285. https://doi.org/10.1016/j.jad.2019.01.039 1000-­0 0017 3. Carrasco-­Garrido P, López de Andrés A, Hernández Barrera V, 20. Biffi A, Scotti L, Corrao G. Use of antidepressants and the risk of Jiménez-­Trujillo I, Jiménez-­García R. National trends (2003–­2009) cardiovascular and cerebrovascular disease: a meta-­ analysis of and factors related to psychotropic medication use in community-­ observational studies. Eur J Clin Pharmacol. 2017;73(4):487-­497. dwelling elderly population. Int Psychogeriatr. 2013;25(2):328-­338. https://doi.org/10.1007/s0022​8-­016-­2187-­x https://doi.org/10.1017/S1041​61021​200169X 21. Biffi A, Rea F, Scotti L, et al. Antidepressants and the risk of cardiovas- 4. Schwarz S, Frölich L. Depression. In: Wehling M, ed. Drug Therapy cular events in elderly affected by cardiovascular disease: a real-­life for the Elderly. Springer-­Verlag; 2013:197-­213. investigation from Italy. J Clin Psychopharmacol. 2020;40(2):112-­121. 5. Tham A, Jonsson U, Andersson G, Soderlund A, Allard P, Bertilsson https://doi.org/10.1097/jcp.00000​0 0000​0 01189 G. Efficacy and tolerability of antidepressants in people aged 65 22. Wu CS, Wang SC, Cheng YC, Gau SSF. Association of cerebrovas- years or older with major depressive disorder—­a systematic review cular events with antidepressant use: a case-­crossover study. Am and a meta-­ analysis. J Affect Disord. 2016;205:1-­12. https://doi. J Psychiatry. 2011;168(5):511-­521. https://doi.org/10.1176/appi. org/10.1016/j.jad.2016.06.013 ajp.2010.10071064 6. Alamo C, Lopez-­Munoz F, Garcia-­Garcia P, Garcia-­Ramos S. Risk–­ 23. Almuwaqqat Z, Jokhadar M, Norby FL, et al. Association of anti- benefit analysis of antidepressant drug treatment in the elderly. depressant medication type with the incidence of cardiovascular ANTIDEPRESSANTS AND STROKE OR DEATH RISK | 477 disease in the ARIC study. J Am Heart Assoc. 2019;8(11):e012503. endothelial cells. Biochem Cell Biol. 2007;85(2):246-­251. https://doi. https://doi.org/10.1161/jaha.119.012503 org/10.1139/o07-­011 24. Flaster M, Sharma A, Rao M. Poststroke depression: a review emphasizing the role of prophylactic treatment and synergy with treatment for motor recovery. Top Stroke Rehabil. 2013;20(2):139-­ S U P P O R T I N G I N FO R M AT I O N 150. https://doi.org/10.1310/tsr20​02-­139 Additional supporting information may be found in the online ver- 25. Schneeweiss S, Avorn J. A review of uses of health care utiliza- tion databases for epidemiologic research on therapeutics. J Clin sion of the article at the publisher’s website. Epidemiol. 2005;58(4):323-­337. https://doi.org/10.1016/j.jclin​ epi.2004.10.012 26. Hoirisch-­Clapauch S, Nardi AE. Antidepressants: bleeding or How to cite this article: Ön BI, Vidal X, Berger U, et al. thrombosis? Thromb Res. 2019;181(Suppl 1):S23-­s28. https://doi. Antidepressant use and stroke or mortality risk in the elderly. org/10.1016/s0049​-­3848(19)30362​-­7 27. Jin H, Lin J, Fu L, et al. Physiological testosterone stimulates tis- Eur J Neurol. 2022;29:469–­477. https://doi.org/10.1111/ sue plasminogen activator and tissue factor pathway inhibitor ene.15137 and inhibits plasminogen activator inhibitor type 1 release in