Concomitant Use of Selective Serotonin Reuptake Inhibitors With Oral Anticoagulants and Risk of Major Bleeding

Key Points

Question  Is there an association between concomitant use of selective serotonin reuptake inhibitors (SSRIs) and oral anticoagulants (OACs) and the risk of major bleeding among patients with atrial fibrillation compared with OAC use alone?

Findings  In this nested case-control study comprising 42 190 cases with major bleeding matched to 1 156 641 controls, concomitant SSRI and OAC use was associated with a 33% increased risk of major bleeding compared with OAC use alone; this risk was highest in the first few months of concomitant use and was substantially lower after 6 months.

Meaning  This study suggests that concomitant use of SSRIs and OACs may be a risk factor for bleeding and should be closely monitored, particularly within the initial months of treatment.

Importance  Selective serotonin reuptake inhibitors (SSRIs) are commonly prescribed antidepressants associated with a small increased risk of major bleeding. However, the risk of bleeding associated with the concomitant use of SSRIs and oral anticoagulants (OACs) has not been well characterized.

Objectives  To assess whether concomitant use of SSRIs with OACs is associated with an increased risk of major bleeding compared with OAC use alone, describe how the risk varies with duration of use, and identify key clinical characteristics modifying this risk.

Design, Setting, and Participants  A population-based, nested case-control study was conducted among patients with atrial fibrillation initiating OACs between January 2, 1998, and March 29, 2021. Patients were from approximately 2000 general practices in the UK contributing to the Clinical Practice Research Datalink. With the use of risk-set sampling, for each case of major bleeding during follow-up, up to 30 controls were selected from risk sets defined by the case and matched on age, sex, cohort entry date, and follow-up duration.

Exposures  Concomitant use of SSRIs and OACs (direct OACs and vitamin K antagonists [VKAs]) compared with OAC use alone.

Main Outcomes and Measures  The main outcome was incidence rate ratios (IRRs) of hospitalization for bleeding or death due to bleeding.

Results  There were 42 190 patients with major bleeding (mean [SD] age, 74.2 [9.3] years; 59.8% men) matched to 1 156 641 controls (mean [SD] age, 74.2 [9.3] years; 59.8% men). Concomitant use of SSRIs and OACs was associated with an increased risk of major bleeding compared with OACs alone (IRR, 1.33; 95% CI, 1.24-1.42). The risk peaked during the initial months of treatment (first 30 days of use: IRR, 1.74; 95% CI, 1.37-2.22) and persisted for up to 6 months. The risk did not vary with age, sex, history of bleeding, chronic kidney disease, and potency of SSRIs. An association was present both with concomitant use of SSRIs and direct OACs compared with direct OAC use alone (IRR, 1.25; 95% CI, 1.12-1.40) and concomitant use of SSRIs and VKAs compared with VKA use alone (IRR, 1.36; 95% CI, 1.25-1.47).

Conclusions and Relevance  This study suggests that among patients with atrial fibrillation, concomitant use of SSRIs and OACs was associated with an increased risk of major bleeding compared with OAC use alone, requiring close monitoring and management of risk factors for bleeding, particularly in the first few months of use.

Antidepressant medications are among the most frequently prescribed class of drugs worldwide, with up to 19% of individuals aged 60 years or older in the US reporting use of an antidepressant over the past 30 days.¹ Selective serotonin reuptake inhibitors (SSRIs) are the most widely used antidepressant medications and are often recommended over other classes of antidepressants for the treatment of major depressive disorder due to their comparable efficacy and favorable safety profile.^2,3 However, SSRIs have been shown to increase the risk of major bleeding,^4-14 possibly owing to their inhibition of platelet activation during hemostasis.² Although the absolute risk remains low for most individuals who use SSRIs,^11,12,15 coprescription with drugs such as oral anticoagulants (OACs) may be consequential. Concomitant use of SSRIs and OACs is common given the prevalence of mental health disorders.¹⁶

Some observational studies have assessed the association between concomitant use of SSRIs and OACs and the risk of major bleeding. However, some had notable limitations, including exposure misclassification,¹⁷ possible informative censoring,^18,19 residual confounding,^19-21 and limited statistical power.^20,22-24 Gaps in evidence that may inform the coprescription of SSRIs and OACs include whether the risk varies with demographic or clinical characteristics or between direct OACs (DOACs) and vitamin K antagonists (VKAs). In addition, data on the risk of specific types of major bleeding are limited.²⁵

To address these knowledge gaps, we conducted a population-based, nested case-control study to assess whether the concomitant use of SSRIs and OACs was associated with the risk of major bleeding compared with OAC use alone among patients with atrial fibrillation (AF). We also assessed whether the risk varied by duration of use, relevant demographic and other risk factors, potency of SSRIs, and OAC type.

In this population-based, nested case-control study, we used the UK Clinical Practice Research Datalink (CPRD GOLD and Aurum databases), a large primary care database of electronic medical records that contains demographic and lifestyle information, medical diagnoses, prescriptions, and referrals for more than 60 million patients from more than 2000 general practices.^26,27 These data are representative of the UK population in terms of age, sex, and race and ethnicity.^26,27 Drug prescriptions issued by the general practitioner are automatically recorded at the time of prescription.^26,27 Quality control audits of the CPRD are regularly conducted to ensure the accuracy and completeness of data.^26,27 The CPRD was linked with the Hospital Episodes Statistics repository, which contains details of inpatient and day case admissions,²⁸ and the Office for National Statistics database, which contains electronic death certificates.²⁹ The study protocol was approved by the CPRD Research Data Governance (No. 22_001906) and the Research Ethics Board of the Jewish General Hospital in Montreal, Canada, which also waived the need for patient informed consent as the data were deidentified. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.³⁰

We conducted a population-based study with a nested case-control approach to analysis because of its computational efficiency compared with a full-cohort analysis given the time-varying nature of both medications of interest, the size of the cohort, and the long duration of follow-up.³¹ We first identified all patients aged 18 years or older with an incident diagnosis of AF between January 2, 1998, and March 29, 2021, and at least 1 year of registration with the practice before AF diagnosis. From this base cohort, we selected those with a prescription for an OAC (apixaban, dabigatran, edoxaban, rivaroxaban, or warfarin) after AF diagnosis, with the date of first prescription defined as study cohort entry. We excluded patients who received OACs any time before cohort entry or SSRIs 6 months prior to cohort entry. We also excluded patients with hyperthyroidism in the year prior to cohort entry because AF in association with hyperthyroidism rarely requires long-term oral anticoagulation. Patients meeting these criteria were followed up until a first major bleeding event, death, end of registration with the practice, or end of the study period (March 29, 2021), whichever occurred first.

We identified cases as patients with a first recorded diagnosis of major bleeding during follow-up, defined as hospitalization with a primary diagnosis of major bleeding or death with bleeding as the primary cause, using relevant International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes (eTable 1 in Supplement 1); elective hospitalizations were not considered. ICD-10 codes for bleeding have shown good positive predictive values between 81% and 95%.^32-34 The index date was the date of hospital admission. For each case, we randomly selected up to 30 controls among the cohort members from the risk sets defined by the case. Each risk set, at each case’s index date, included all individuals who did not experience major bleeding and thus were still at risk up to that point in follow-up time, matched on age, sex, calendar date of cohort entry (±6 months), and duration of follow-up. Thus, as per the risk-set sampling approach, cases were eligible for selection as controls prior to becoming a case, and patients may have been selected as controls for multiple cases.^35,36 The index date for controls was the date resulting in the same duration of follow-up for cases and controls.

We identified prescriptions of SSRIs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, or sertraline) and OACs for all cases and their matched controls between cohort entry and the index date. Exposure was defined in 4 mutually exclusive categories: concomitant use of SSRIs and OACs, OAC use alone, nonuse, and other use. We considered patients as concomitant users of SSRIs and OACs if the duration of their last prescription for both medications covered or ended 30 days before the index date. Similarly, we considered patients as users of OACs alone if their last prescription for an OAC covered or ended 30 days before the index date, without a prescription for an SSRI in this period—this was the reference category. Users of SSRIs alone, non-SSRI antidepressants alone, or non-SSRI antidepressants concomitantly with SSRIs and/or OACs were classified separately (other use). Finally, nonusers were those not exposed to any medications of interest on or 30 days before the index date.

We adjusted all models for the following comorbidities based on substantive knowledge, measured at or earlier than 365 days (1 year) before the index date: smoking, alcohol abuse, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) (<25, 25-29.9, or ≥30.0), depression, hypertension, diabetes, stroke or transient ischemic attack, coronary artery disease, congestive heart failure, peripheral arterial disease, disorders of hemostasis, cancer (other than nonmelanoma skin cancer), liver disease, chronic kidney disease, chronic obstructive pulmonary disease, anemia, and venous thromboembolism. We also included history of bleeding at any time before cohort entry and the time between incident AF diagnosis and first OAC prescription. Diabetes and hypertension were defined using diagnostic codes or relevant medications. All models were also adjusted for use of the following drugs measured between 365 days (1 year) and 730 days (2 years) prior to the index date: angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, β-blockers, calcium channel blockers, thiazide diuretics, other diuretics, antiplatelets, lipid-lowering drugs (including statins), antipsychotics, non-SSRI antidepressants, nonsteroidal anti-inflammatory drugs, proton pump inhibitors, and H₂ receptor blockers. We considered the number of hospitalizations in the year before cohort entry as a surrogate marker for overall health. Finally, we adjusted for socioeconomic status using the Index of Multiple Deprivation, categorized in deciles.³⁷

We used conditional logistic regression to compute odds ratios of major bleeding associated with concomitant use of SSRIs and OACs compared with OAC use alone, adjusting for the covariates listed. In a nested case-control approach, odds ratios are unbiased estimators of incidence rate ratios (IRRs) with very limited loss in precision.³⁶ In secondary analyses, we assessed whether the risk of major bleeding varied according to age, sex, chronic kidney disease, history of bleeding, type of OAC (DOACs or VKAs), and potency of SSRIs (strong or moderate serotonin reuptake inhibitors based on the dissociation constant).³⁸ Next, among patients continuously exposed to OACs and concomitantly exposed to SSRIs and OACs at the index date, we investigated whether the risk of major bleeding varied with the duration of continuous concomitant use of SSRIs in 3 prespecified categories (≤30 days, 31-180 days, or >180 days) compared with OAC use alone. These categories were selected because SSRIs were reported to exert antiplatelet action as early as 2 to 3 weeks after initiation.^39,40 We defined continuous exposure to SSRIs and OACs separately, allowing a 30-day grace period between consecutive prescriptions where patients were still considered exposed. Patients were then considered concomitant users on any given day if exposed to both drugs on that day. In addition, we used a restricted cubic spline with 5 interior knots to produce a smooth curve of the IRR as a function of continuous duration of use. We also estimated the risk in specific anatomical locations, including gastrointestinal bleeding, intracranial hemorrhage, and other major bleeding. We assessed the risk of any bleeding associated with concomitant use of SSRIs and OACs. For this analysis, we repeated the selection of cases and controls already described, with cases defined using relevant diagnostic codes in primary electronic medical records. Finally, we assessed whether an interaction was present between SSRIs and OACs with respect to the risk of major bleeding on both the additive and multiplicative scales (eMethods 1 in Supplement 1). In other words, we assessed whether the joint association of the 2 exposures departed from the sum or product of their individual associations with the risk of bleeding, although an additive interaction has been described as most indicative of biological or mechanistic interaction.^41,42

We performed 4 sensitivity analyses to assess the robustness of the results. First, to explore potential exposure misclassification, we considered only prescriptions that covered the index date and, next, those that covered or ended within 15 days before the index date. Second, to account for the potential adjustment for covariates affected by exposure, all covariates were measured at or prior to cohort entry. Third, we implemented multiple imputation by chained equations for missing values of BMI and smoking, combining results from 5 imputed datasets.⁴³ Fourth, we repeated the analysis by type of OAC, excluding patients with a history of valvular surgery or rheumatic valvular disease before cohort entry because DOACs are not indicated for patients with valvular AF.^44-46

We conducted a supplementary time-conditional propensity score–matched analysis to further explore the potential for residual confounding.^47,48 In brief, among the base cohort of patients with incident AF initiating OACs, we matched each patient initiating SSRIs to a patient using OACs alone up to that point in time with the same age (±1 year), sex, calendar date of OAC initiation (±1 year), and time-conditional propensity score (eMethods 2 in Supplement 1). Finally, we conducted a post hoc analysis, repeating the primary analysis with additional adjustment for the following comedications reported to interact with OACs, measured between 1 and 2 years before the index date: clarithromycin, erythromycin, penicillin, azole antifungals, quinidine, amiodarone, dronedarone, propafenone, allopurinol, oral corticosteroids, tamoxifen, valproic acid, cyclosporin, tacrolimus, disulfiram, methylphenidate, and sulfamethoxazole. All analyses were performed with a 2-sided hypothesis test, and P < .05 was considered statistically significant, without adjustment for multiple comparisons, using SAS, version 9.4 (SAS Institute Inc).

After applying all eligibility criteria, the cohort included 331 305 patients (mean [SD] age, 73.7 [10.8] years; 57.1% men) with incident AF initiating OACs (eFigure in Supplement 1). During a mean (SD) follow-up of 4.6 (4.0) years, 42 391 patients were hospitalized with major bleeding, yielding an incidence rate of 27.9 per 1000 person-years (95% CI, 27.7-28.2 per 1000 person-years). Among those, 42 190 cases (mean [SD] age, 74.2 [9.3] years; 59.8% men) were matched to 1 156 641 controls (mean [SD] age, 74.2 [9.3] years; 59.8% men). As anticipated, risk factors for major bleeding were more prevalent among cases than controls (Table 1).

Concomitant use of SSRIs and OACs was associated with an increased risk of major bleeding compared with OAC use alone (IRR, 1.33; 95% CI, 1.24-1.42) (Table 2). The risk was the highest during the first 30 days of continuous use (IRR, 1.74; 95% CI, 1.37-2.22), and decreased thereafter (eTable 2 in Supplement 1). This trend was also observed when modeling the IRR flexibly as a spline of the duration of continuous use (Figure 1). The risk did not vary according to age, sex, history of major bleeding, chronic kidney disease (Figure 2; eTable 3 in Supplement 1), or potency of SSRIs (eTable 4 in Supplement 1). The risk of major bleeding was associated with concomitant use of SSRIs and DOACs compared with DOACs alone (IRR, 1.25; 95% CI, 1.12-1.40) and with concomitant use of SSRIs and VKAs compared with VKAs alone (IRR, 1.36; 95% CI, 1.25-1.47) (Table 3). With respect to types of major bleeding, the association was present for intracranial hemorrhage, gastrointestinal bleeding, and other major bleeding (Table 2). Last, concomitant use of SSRIs and OACs was also associated with the risk of any bleeding (IRR, 1.22; 95% CI, 1.16-1.28) compared with OAC use alone (eTable 5 in Supplement 1).

In the assessment of interaction, a small superadditive interaction may have been present, although the estimate was not statistically significant (relative excess risk due to interaction [RERI], 0.10; 95% CI, −0.07 to 0.27) (eTable 6 in Supplement 1). Based on the estimated RERI, the interaction may be associated with approximately 5% of all major bleeding. In addition, there was limited evidence of a multiplicative interaction. Results from sensitivity analyses were consistent with those of the primary analysis (eTables 7-10 in Supplement 1). Finally, the association remained in the time-conditional propensity score–matched analysis, although slightly attenuated (adjusted hazard ratio, 1.23; 95% CI, 1.08-1.40) (eTable 11 in Supplement 1), and was consistent in the post hoc analysis (eTable 12 in Supplement 1).

In this population-based, nested case-control study, the concomitant use of SSRIs and OACs was associated with a 33% increased risk of major bleeding. The association was the strongest for the first few months of concomitant use. The overall risk remained consistent regardless of age, sex, potency of SSRIs, history of major bleeding, or chronic kidney disease as well as type of OAC. Concomitant use was individually associated with gastrointestinal bleeding, intracranial hemorrhage, and other major bleeding. Interaction between SSRIs and OACs, if any, was limited.

In light of these findings, the risk of major bleeding may be a pertinent safety consideration for patients using SSRIs and OACs concomitantly. This finding has been echoed in the summary of product characteristics for different OACs, which describes SSRIs as interacting drugs given that they independently increase the risk of bleeding. Although clinical guidelines for the management of major depressive disorder have acknowledged the risk of bleeding associated with SSRIs, the potential for interaction with OACs was either not discussed or based on very limited evidence.^49,50 Likewise, guidelines from Canadian, US, and European cardiology associations for the management of AF suggest consideration of drug-drug interactions when prescribing OACs,^44-46 with nonsteroidal anti-inflammatory drugs being the only class of drugs cited.⁴⁵ Although the European Heart Rhythm Association lists SSRIs as drugs with pharmacodynamic interactions with DOACs, no evidence was cited.⁵¹

The risk of major bleeding associated with the concomitant use of SSRIs and OACs has been assessed in previous observational studies, although results were inconsistent.^{7,17-20,22-24,52} Limitations of previous studies included residual confounding,^19,20,22,24 varying exposure definitions, limited statistical power,^20,22-24 and the assessment of the concomitant use of SSRIs and OACs being a secondary objective.^7,18 In line with our results, a systematic review and meta-analysis of 8 observational studies suggested an increased risk of major bleeding associated with the concomitant use of SSRIs and OACs (hazard ratio, 1.35; 95% CI, 1.14-1.58) compared with OAC use alone.²⁵ However, several knowledge gaps were identified, including the risk of major bleeding in important patient subgroups. The present study confirmed that compared with OACs alone, concomitant use of SSRIs and OACs increased the risk of major bleeding among patients 60 years of age or older and among both sexes. One study previously assessed this association in these patient subgroups; however, statistical power was limited.²² Furthermore, we showed that the association remained similar irrespective of patients’ history of major bleeding or chronic kidney disease, both important factors in the HAS-BLED (Hypertension, Abnormal Renal/Liver Function, Stroke, Bleeding History or Predisposition, Labile International Normalized Ratio, Elderly [>65 Years], Drugs/Alcohol Concomitantly) score for major bleeding risk.⁵³ Our findings also suggested that the concomitant use of SSRIs with both DOACs and VKAs was associated with an increased risk of major bleeding, with a possible lower risk with DOACs, although 95% CIs overlapped. A cohort study of patients from the ROCKET-AF (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Embolism and Stroke Trial in Atrial Fibrillation) trial suggested that the concomitant use of SSRIs and warfarin may increase the risk of major bleeding compared with rivaroxaban; however, the results were limited by high uncertainty and potential for selection bias and confounding.²⁰ In another nested case-control study of nursing home residents, similar increases in risk were associated with the concomitant use of SSRIs and DOACs and of SSRIs and VKAs; however, statistical power was low.²³

The increased risk of major bleeding with the concomitant use of SSRIs and OACs may occur through multiple mechanisms of action. During hemostasis, serotonin is released by platelets to enhance platelet activation and aggregation and prime them to interact with coagulation factors.⁵⁴ Selective serotonin reuptake inhibitors block the serotonin reuptake transporter on platelet membranes and reduce serotonin content within platelets by up to 80% to 90%,^39,40 decreasing the potency of hemostasis over time. In addition, some SSRIs, such as fluoxetine and fluvoxamine, inhibit the 1A2 and 2C9 isozymes of the hepatic cytochrome P450 enzyme, which play a key role in the metabolism of warfarin.⁵⁵ Nonetheless, the interaction analysis suggests that the joint association of SSRIs and OACs is mainly owing to their individual risks of major bleeding; hence, any additional risk posed by pharmacokinetic interaction is likely minimal.

Although the increased risk of major bleeding does not suggest withholding treatment with either SSRIs or OACs, measures can be taken to mitigate this risk. Studies suggest that DOACs have lower potential for pharmacokinetic interactions with SSRIs than VKAs, and guidelines also recommend them over VKAs for the management of nonvalvular AF.^44-46,55,56 Taken together with the findings in this study, DOACs may also be preferred for patients concomitantly using SSRIs. On the other hand, the risk of major bleeding was similar between SSRIs with more potent inhibition and SSRIs with less potent serotonin inhibition; thus; changing the SSRI may not be associated with bleeding risk. Finally, coprescription of proton pump inhibitors has also been suggested to prevent gastrointestinal bleeding.^51,57 Overall, risk factors for bleeding should be monitored and managed to improve the safety of the concomitant use of SSRIs and OACs.⁵¹ Close monitoring is particularly essential within the first few months of concomitant use.

This study has notable strengths. First, the selection of a large study population from routine care settings enhanced generalizability and provided sufficient statistical power to generate precise estimates in primary and secondary analyses. Second, selection bias was unlikely because we analyzed a well-defined cohort and used a nested case-control approach. Third, the assessment of additive and multiplicative interactions provided evidence suggesting that any biological interaction between use of SSRIs and OACs and the risk of major bleeding may only be marginally synergistic.⁴²

This study also has some limitations. Residual confounding may affect the results given the observational nature of the study. The baseline risk for major bleeding may differ between patients concomitantly using SSRIs and those who were not. To mitigate potential bias, we adjusted for several potential confounders, including some lifestyle risk factors (such as BMI, smoking, and alcohol abuse). Furthermore, the results remained consistent in the time-conditional propensity score–matched analysis and in a post hoc analysis adjusted for additional comedications. Another consideration is that prescriptions recorded in the CPRD are those issued by general practitioners; hence, misclassification of exposure is possible if patients do not follow the treatment regimen. Prescriptions also do not include those issued by specialists, although AF as well as mild and moderate depression are managed mainly by general practitioners in the UK.^58,59 To explore the potential for misclassification, we varied the exposure assessment window in sensitivity analyses, which produced results consistent with the main results. Finally, outcome misclassification through inaccurate recording of bleeding in the Hospital Episodes Statistics repository may occur. In addition, the physician’s judgment may be influenced by knowledge of patient treatment. To mitigate bias, we considered only primary diagnoses and did not include elective hospitalizations.

In this large population-based, nested case-control study of patients with AF, the concomitant use of SSRIs and OACs was associated with an increased risk of major bleeding compared with OACs alone. To minimize the risk of bleeding, individual modifiable risk factors should be controlled, and patients should be closely monitored, particularly during the first few months of concomitant use.

Accepted for Publication: January 26, 2024.

Published: March 22, 2024. doi:10.1001/jamanetworkopen.2024.3208

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2024 Rahman AA et al. JAMA Network Open.

Corresponding Author: Christel Renoux, MD, PhD, Centre for Clinical Epidemiology, Lady Davis Institute for Medical Research, Jewish General Hospital, 3755 Côte Sainte Catherine H 416.1, Montreal, QC H3T 1E2, Canada (christel.renoux@mcgill.ca).

Author Contributions: Mr Rahman and Dr Renoux had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Rahman, Platt, Boivin, Rej, Renoux.

Acquisition, analysis, or interpretation of data: Rahman, Platt, Beradid, Rej, Renoux.

Drafting of the manuscript: Rahman, Boivin, Renoux.

Critical review of the manuscript for important intellectual content: Rahman, Platt, Beradid, Rej, Renoux.

Statistical analysis: Rahman, Beradid, Renoux.

Administrative, technical, or material support: Renoux.

Supervision: Platt, Boivin, Rej, Renoux.

Conflict of Interest Disclosures: Dr Platt reported receiving personal fees from Biogen, Boehringer Ingelheim, Merck, Nant Pharma, and Pfizer outside the submitted work. Dr Rej reported receiving a Clinician-Scientist Salary Award from Fonds de Recherche Quebec Sante; receiving grants from Mitacs; serving on a steering committee for AbbVie; and holding shares in Aifred outside the submitted work. No other disclosures were reported.

Funding/Support: Mr Rahman was supported by a Tomlinson Doctoral Fellowship from McGill University, a Canada Graduate Scholarship–Doctoral from the Canadian Institutes of Health Research, and a stipend from the Drug Safety and Effectiveness Cross-Training Program.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 2.

Additional Contributions: The authors would like to thank Sophie Dell’Aniello, MSc, Centre for Clinical Epidemiology, Lady Davis Institute for Medical Research, Jewish General Hospital, for providing technical support in the analysis of data. She was not compensated for her contribution.