Abstract
AIM
To investigate whether selective serotonin re-uptake inhibitor (SSRI) use is associated with an increased risk of haemorrhagic stroke in a cohort of antidepressant users.
METHODS
We conducted a case control study, nested within a cohort of antidepressant users in the United Kingdom General Practice Research Database. A cohort of 365 195 patients prescribed either an SSRI or tricyclic antidepressant between 1992 and 2006 was identified. Three hundred and fifty-seven cases of haemorrhagic stroke were observed and 1631 control patients without haemorrhagic stroke were selected.
RESULTS
The primary analysis showed no evidence of an association between current SSRI or TCA use and haemorrhagic stroke. Current use of an SSRI compared with no use at the time of haemorrhagic stroke was associated with an adjusted odds ratio of 1.11 (95% confidence interval (CI) 0.82, 1.50). For current tricyclic use the equivalent odds ratio was 0.73 (0.52, 1.02). There was no evidence that prior cerebrovascular events modified the effect of either SSRIs or TCAs.
CONCLUSIONS
We found no evidence that SSRIs are associated with an increased risk of haemorrhagic stroke, regardless of prior history of cerebrovascular events.
Keywords: antidepressive agents, haemorrhagic stroke, pharmacoepidemiology, serotonin
WHAT THIS STUDY ADDS
We found no association between SSRI use and haemorrhagic stroke.
The large number of patients involved in the study allowed us to rule out any substantial effect.
The results were similar in people with and without a previous history of cerebrovascular events.
WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT
Selective serotonin re-uptake inhibitors (SSRI) are associated with an increased risk of bleeding disorders at a number of sites.
It is currently unclear whether they increase the risk of haemorrhagic stroke, with conflicting results reported.
Introduction
Selective serotonin re-uptake inhibitors (SSRI) are associated with an increased risk of bleeding, most notably upper gastrointestinal bleeding [1–4]. Whether SSRIs are associated with haemorrhagic stroke is unclear. Results from the Women's Health Initiative Study suggest a doubling in the risk of haemorrhagic stroke amongst post-menopausal SSRI recipients [5]. Other studies found no association, although low power and the inability to adjust for potential confounders meant a clinically relevant increased risk remained possible [6–8]. The UK General Practice Research Database (GPRD) provides an ideal setting to conduct a large observational study to assess a possible association.
Methods
The general practice research database
The GPRD contains information from over 10 million patients registered at ∼500 UK general practices [9, 10]. Continuous patient records include details of all consultations, diagnoses, prescribed medicines and basic demographics data. Patients are representative of the England and Wales population in terms of age, sex and geographical distribution [11]. Data quality is subject to rigorous checks and audits and >500 peer reviewed studies have been published using GPRD data. A previous GPRD study of stroke estimated the validity of recorded stroke diagnoses to be ∼90% [12]. GPRD data are provided entirely anonymized. Ethical approval was obtained from the London School of Hygiene and Tropical Medicine Ethics Committee.
Study design
This was a matched case control study nested in a cohort of GPRD-registered patients who had been prescribed an antidepressant at some time. Inclusion criteria for the cohort were registered in the GPRD between 1 January 1998 and 31 December 2006, two or more prescriptions for either an SSRI/related antidepressant or tricyclic/related antidepressant (TCA), registered at least 12 months before the first SSRI or TCA prescription and no previous haemorrhagic stroke. Patients receiving both an SSRI and TCA, those receiving an anticoagulant at any point and patients with evidence of trauma within 30 days before haemorrhagic stroke were excluded. For each patient the start date was the earliest recorded SSRI or TCA prescription and the end date was the earliest of leaving the practice, haemorrhagic stroke or 31 December 2006.
Cases and controls
Cases had a diagnosis of haemorrhagic stroke between 1 January 1998 and 31 December 2006. Up to five controls (people with no diagnosis of haemorrhagic stroke at any time) were selected per case matched on age (within 2 years), sex, practice, and had to be currently registered on the date of diagnosis of their matched case (the index date).
Cases and controls were excluded if they had a record of ischaemic stroke, unspecified stroke, TIA or other cerebrovascular event within 30 days before the index date. Patients with a history of these events earlier than 30 days before the index were retained, allowing the investigation of prior cerebrovascular events as a possible effect modifier. SSRI and TCA exposure was classified as ‘current’ if the prescription date was within 30 days before the index date. ‘Recent’ use was defined between 31 and 60 days before the index date. Patients without prescriptions between 1 and 60 days before the index date were considered unexposed.
Conditional logistic regression was performed, with adjustment for smoking, alcohol consumption, body mass index (BMI), prior history of transient ischaemic attack (TIA) or other stroke, hypertension, diabetes, NSAID use, aspirin use, clopidogrel or dipyridamole use, year of first prescription (SSRI or TCA) and total observation time.
Results
A cohort of 396 195 patients was identified, 124 373 received two or more TCA prescriptions and 271 822 received two or more SSRI prescriptions. Within the cohort, 365 cases of haemorrhagic stroke were identified. For eight cases no controls meeting the matching criteria could be identified leaving 357 cases for whom 1631 controls were included.
Table 1 summarizes the background characteristics of the participants. The median age was 75 years for cases and 73 years for controls and 65% of cases were women compared with 66% of controls. Differences between cases and controls included a higher prevalence of hypertension amongst cases (53% vs. 41%) and a higher proportion of cases with a history of TIA or ischaemic/unspecified stroke (28% vs. 9%). A higher proportion of cases had received aspirin (51% vs. 36%).
Table 1.
Baseline characteristics of haemorrhagic stroke case control study patients
| Cases n = 357 (%) | Controls n = 1631 (%) | |
|---|---|---|
| Age at index date (years) | ||
| ≤30 | 2 (0.6) | 6 (0.4) |
| >30–40 | 10 (2.8) | 49 (3.0) |
| >40–50 | 26 (7.3) | 129 (7.9) |
| >50–60 | 40 (11.2) | 197 (12.1) |
| >60–70 | 72 (20.2) | 344 (21.1) |
| >70–80 | 95 (26.6) | 463 (28.4) |
| >80 | 112 (31.4) | 443 (27.2) |
| Sex | ||
| Female | 232 (65.0) | 1,082 (66.3) |
| Male | 125 (35.0) | 549 (33.7) |
| Observation time in GPRD (quintiles) | ||
| 1 (<3.4 years) | 70 (19.6) | 307 (18.8) |
| 2 (3.4–6.3 years) | 82 (23.0) | 371 (22.8) |
| 3 (6.3–9.9 years) | 70 (19.6) | 326 (20.0) |
| 4 (9.9–13.2 years) | 68 (19.1) | 332 (20.4) |
| 5 (>13.2 years) | 67 (18.8) | 295 (18.1) |
| Smoking status | ||
| Non-smoker | 147 (41.2) | 712 (43.7) |
| Current smoker | 104 (29.1) | 376 (23.1) |
| Ex-smoker | 82 (23.0) | 491 (30.1) |
| Unknown status | 24 (6.7) | 52 (3.2) |
| Alcohol consumption | ||
| Non-drinker | 67 (18.8) | 280 (17.2) |
| Ex-drinker | 6 (1.7) | 10 (0.6) |
| Current drinker (amount unknown) | 6 (1.7) | 33 (2.0) |
| <2 units/day | 182 (51.0) | 972 (59.6) |
| 3–6 units/day | 25 (7.0) | 124 (7.6) |
| >6 units/day | 11 (3.1) | 20 (1.2) |
| Unknown status | 60 (16.8) | 192 (11.8) |
| BMI (kg m−2) | ||
| <20 | 27 (7.6) | 100 (6.1) |
| 20–25 | 106 (29.7) | 520 (31.9) |
| >25 | 143 (40.1) | 756 (46.4) |
| Unknown | 81 (22.7) | 255 (15.6) |
| Medical history | ||
| TIA/other stroke | 101 (28.3) | 147 (9.0) |
| Hypertension | 188 (52.7) | 673 (41.3) |
| Diabetes | 54 (15.1) | 209 (12.8) |
| Medication history | ||
| NSAIDs | 267 (74.8) | 1,166 (71.5) |
| Aspirin | 182 (51.0) | 592 (36.3) |
| Clopidogrel or dipyridamole | 44 (4.3) | 34 (3.6) |
One thousand and thirty-four patients received SSRIs before the index with fluoxetine the most frequently prescribed (22%; Table 2). TCAs were prescribed to 954 patients, most frequently amitriptyline (45%).
Table 2.
Frequency of antidepressant use; no patients used both SSRIs and TCAs, all patients are counted only once
| Antidepressant type | n (%) |
|---|---|
| SSRI and related | |
| Fluoxetine | 229 (22.2) |
| Citalopram | 202 (19.5) |
| Paroxetine | 159 (15.4) |
| Sertraline | 77 (7.5) |
| Venlafaxine | 23 (2.2) |
| Mirtazapine | 15 (1.5) |
| Escitalopram | 13 (1.3) |
| Fluvoxamine | 3 (0.3) |
| Duloxetine | 2 (0.2) |
| Mixed SSRI usage | 311 (30.1) |
| Total | 1034 |
| TCA and related | |
| Amitritptyline | 429 (45.0) |
| Dosulepin | 187 (19.6) |
| Lofepramine | 32 (3.4) |
| Trazodone | 23 (2.4) |
| Imipramine | 16 (1.7) |
| Clomipramine | 14 (1.5) |
| Trimipramine | 11 (1.2) |
| Doxepin | 7 (0.7) |
| Nortriptyline | 4 (0.4) |
| Fluphenazine | 4 (0.4) |
| Mianserin | 3 (0.3) |
| Mixed TCA usage | 223 (23.4) |
| Total | 954 |
One hundred and two cases (29%) were current SSRI users at the index date, compared with 386 controls (24%); crude odds ratio (OR) 1.25 (95% CI 0.96, 1.63) (Table 3). After adjustment for smoking, alcohol, observation time in the database, prior cerebrovascular events, BMI, hypertension, diabetes, non-steroidal anti-inflammatory drugs (NSAID) and aspirin use, clopidogrel or dipyridamole use and calendar year of first SSRI or TCA prescription the OR was 1.11 (95% CI 0.82, 1.50). When current or recent SSRI exposure was considered, the crude OR was 1.29 (95% CI 1.00, 1.68), with an adjusted OR of 1.15 (95% CI 0.86, 1.54). We found no evidence of effect modification by prior cerebrovascular events (P = 0.95, results not shown).
Table 3.
Association between SSRI exposure and haemorrhagic stroke
| Antidepressant exposure | Cases (%) | Controls (%) | Crude OR | Adjusted OR* |
|---|---|---|---|---|
| No SSRI current use | 255 (71) | 1245 (76) | – | – |
| SSRI current use | 102 (29) | 386 (24) | 1.25 (0.96–1.63) | 1.11 (0.82–1.50) |
| No TCA current use | 290 (81) | 1252 (77) | – | – |
| TCA current use | 67 (19) | 379 (23) | 0.77 (0.57–1.04) | 0.73 (0.52–1.02) |
| No SSRI current/recent use | 245 (69) | 1215 (74) | – | – |
| SSRI current/recent use | 112 (31) | 416 (25) | 1.29 (1.00–1.68) | 1.15 (0.85–1.54) |
| No TCA current/recent use | 283 (79) | 1222 (75) | – | – |
| TCA current/recent use | 74 (21) | 409 (25) | 0.77 (0.58–1.04) | 0.72 (0.52–1.01) |
| No SSRI ever use | 164 (46) | 795 (48) | – | – |
| SSRI ever use | 193 (54) | 836 (51) | 1.10 (0.86–1.41) | 1.03 (0.77–1.37) |
| No TCA ever use | 198 (55) | 836 (51) | – | – |
| TCA ever use | 159 (45) | 795 (49) | 0.85 (0.66–1.09) | 0.91 (0.68–1.21) |
Adjusted for smoking, alcohol, BMI, prior history of TIA or other stroke, hypertension, diabetes, NSAID use, aspirin use, clopidogrel or dipyridamole use, year of first prescription (SSRI or TCA), observation time.
Sixty-seven (19%) cases were current TCA users at the index date, compared with 379 (23%) controls, with a crude OR of 0.77 (95% CI 0.57, 1.04). The adjusted OR was 0.73 (95% CI 0.52, 1.10). For current or recent TCA exposure, the crude OR was 0.77 (95% CI 0.58, 1.04) and the adjusted OR was 0.72 (95% CI 0.52, 1.01). There was no evidence of effect modification by prior cerebrovascular events (P = 0.99; results not shown).
Discussion
This study found no evidence of increased risk of haemorrhagic stroke associated with current or recent use of either SSRIs or TCAs.
We nested this case control study within a cohort of patients prescribed antidepressants at some stage to limit the effects of confounding by indication, as the underlying indications for antidepressant treatment could themselves be associated with haemorrhagic stroke. However, non-depression SSRI and TCA indications differ. For example, SSRIs are used for obsessive compulsive disorder and TCAs are used for neuropathic pain. The results of a sensitivity analysis restricted to patients with a record of depression or anxiety also found no association with either drug class (data not shown).
Exposure was determined using prescription data, with no guarantee that the drug was obtained from the pharmacy or taken. This could lead to exposure misclassification, likely to be non-differential by case or control status and leading to a dilution of any effect. We reduced this possibility by excluding patients with only a single TCA or SSRI prescription, assuming repeat prescriptions are likely to indicate at least some use of the drug. In addition, we classified SSRI-related antidepressants such as venlafaxine and mirtazapine together with traditional SSRIs; if any effect is limited to traditional SSRIs, this would again lead to misclassification of exposure and a possible dilution of any effect. However, less than 5% of SSRI exposure was accounted for by the SSRI-related antidepressants and so bias was likely to be minimized.
It is possible that some haemorrhagic strokes were recorded as non-specific stroke and therefore such strokes could have affected the control group. However, any such misclassification is likely to be non-differential with respect to antidepressant exposure and could reduce any measure of association. Since haemorrhagic stroke is relatively rare, the proportion of controls affected is likely to be small, limiting the impact of any misclassification.
Although the association between SSRIs and haemorrhagic stroke has been investigated previously we believe this study adds further to evidence already generated. A previous study conducted in the GPRD[6] included only 65 cases, and was unable to exclude the possibility of a large increase in risk (OR 0.8, 95% CI 0.3, 2.3). We were able to include 357 cases and obtain much greater precision. GPRD data quality, particularly for lifestyle factors such as smoking, has also improved since the introduction of the Quality and Outcomes Framework in 2004. This meant we were better able to control for key confounding factors compared with previous studies [6, 7].
In contrast to the results seen here, the Women's Health Initiative Study reported a hazard ratio of 2.12 (95% CI 1.10–4.07) for haemorrhagic stroke, comparing prior SSRI users with non-users in post-menopausal women [5]. However the study only determined antidepressant exposure at a single time point. It is unknown whether any association was present with current SSRI use.
Of note, the estimates of association for SSRIs and TCAs are in opposite directions, although the confidence intervals for these estimates overlapped considerably and did not allow conclusions to be drawn about any difference between the two. A direct comparison between SSRIs and TCAs was not an a priori objective in this study, but could be a useful starting point for other investigators using a different dataset to explore this question further.
The study has limitations in terms of generalizability. We excluded patients with recent trauma and those with prior anticoagulant treatment and so these results may not apply to these patient groups.
The results of the present study largely agree with a number of studies done in different datasets and at different points in time. To date, no clear evidence of a substantial association between SSRIs and haemorrhagic stroke has been detected.
Acknowledgments
Ian Douglas is funded by a Medical Research Council Methodology Fellowship and Liam Smeeth is funded by a Wellcome Trust Senior Clinical Research Fellowship.
We are grateful to Lesley Wise from the UK Medicines and Healthcare products Regulatory Agency (MHRA) who contributed to the study design and provided valuable comments on the draft manuscript. We also thank the MHRA for providing the data from the General Practice Research Database for this study.
Competing Interest
Ian Douglas consults for and holds stock in GlaxoSmithKline. Liam Smeeth has no conflicts of interest. David Irvine works for the Medicines and Healthcare products Regulatory Agency (MHRA); the views expressed in this article are personal and are the author's own and do not reflect the official policy or position of the MHRA.
REFERENCES
- 1.Dalton SO, Sørensen HT, Johansen C. SSRIs and upper gastrointestinal bleeding: what is known and how should it influence prescribing? CNS Drugs. 2006;20:143–51. doi: 10.2165/00023210-200620020-00005. [DOI] [PubMed] [Google Scholar]
- 2.Layton D, Clark DW, Pearce GL, Shakir SA. Is there an association between selective serotonin reuptake inhibitors and risk of abnormal bleeding? Results from a cohort study based on prescription event monitoring in England. Eur J Clin Pharmacol. 2001;57:167–76. doi: 10.1007/s002280100263. [DOI] [PubMed] [Google Scholar]
- 3.Tata LJ, Fortun PJ, Hubbard RB, Smeeth L, Hawkey CJ, Smith CJ, Whitaker HJ, Farrington CP, Card TR, West J. Does concurrent prescription of selective serotonin reuptake inhibitors and non-steroidal anti-inflammatory drugs substantially increase the risk of upper gastrointestinal bleeding? Aliment Pharmacol Ther. 2005;22:175–81. doi: 10.1111/j.1365-2036.2005.02543.x. [DOI] [PubMed] [Google Scholar]
- 4.Meijer WE, Heerdink ER, Nolen WA, Herings RM, Leufkens HG, Egberts AC. Association of risk of abnormal bleeding with degree of serotonin reuptake inhibition by antidepressants. Arch Intern Med. 2004;164:2367–70. doi: 10.1001/archinte.164.21.2367. [DOI] [PubMed] [Google Scholar]
- 5.Smoller JW, Allison M, Cochrane BB, Curd JD, Perlis RH, Robinson JG, Rosal MC, Wenger NK, Wassertheil-Smoller S. Antidepressant use and risk of incident cardiovascular morbidity and mortality among postmenopausal women in the Women's Health Initiative Study. Arch Intern Med. 2009;169:2128–39. doi: 10.1001/archinternmed.2009.436. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.de Abajo FJ, Jick H, Derby L, Jick S, Schmitz S. Haemorrhagic stroke and use of selective serotonin reuptake inhibitors. Br J Clin Pharmacol. 2000;50:43–7. doi: 10.1046/j.1365-2125.2000.00216.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Bak S, Tsiropoulos I, Kjaersgaard JO, Andersen M, Mellerup E, Hallas J, García Rodríguez LA, Christensen K, Gaist D. Selective serotonin reuptake inhibitors and the risk of stroke: a population-based case-control study. Stroke. 2002;33:1465–73. doi: 10.1161/01.str.0000018589.56991.ba. [DOI] [PubMed] [Google Scholar]
- 8.Kharofa J, Sekar P, Haverbusch M, Moomaw C, Flaherty M, Kissela B, Broderick J, Woo D. Selective serotonin reuptake inhibitors and risk of haemorrhagic stroke. Stroke. 2007;38:3049–51. doi: 10.1161/STROKEAHA.107.491472. [DOI] [PubMed] [Google Scholar]
- 9.Lawson DH, Sherman V, Hollowell J. The general practice research database. Q J Med. 1998;91:445–52. doi: 10.1093/qjmed/91.6.445. [DOI] [PubMed] [Google Scholar]
- 10.The GPRD Practice & Patient Populations – July 2010 Release. Available at http://www.gprd.com/_docs/GPRD%20Practice%20Patient%20Populations%20Jul2010.pdf (last accessed 29 September 2010)
- 11.Office for National Statistics. Key Health Statistics from General Practice 1998. London: Office for National Statistics; 2000. p. 149. Series MB6 No. 2. [Google Scholar]
- 12.Andersohn F, Schade R, Suissa S, Garbe E. Cyclooxygenase-2 selective nonsteroidal anti-inflammatory drugs and the risk of ischemic stroke: a nested case-control study. Stroke. 2006;37:1725–30. doi: 10.1161/01.STR.0000226642.55207.94. [DOI] [PubMed] [Google Scholar]