Selective Serotonin Reuptake Inhibitor Use and Risk of Major Bleeding during Treatment with Vitamin K Antagonists: Results of A Cohort Study

Background  Selective serotonin reuptake inhibitors (SSRIs) may increase the risk of major bleeding by decreasing platelet function or decreasing vitamin K antagonist (VKA) metabolism via cytochrome P450 (CYP) inhibition. Aims  To determine whether SSRIs are associated with major bleeding during VKA treatment and investigate the possible mechanisms. Methods  In this cohort study, information on SSRI use and bleeding complications was obtained from patient records of VKA initiators between 2006 and 2018 from two anticoagulation clinics. Conditional logistic regression and time-dependent Cox regression were used to estimate the effect of SSRIs on a high international normalized ratio (INR ≥ 5) within 2 months after SSRI initiation and on major bleeding during the entire period of SSRI use, respectively. SSRI use was stratified for (non-)CYP2C9 inhibitors. Results  A total of 58,918 patients were included, of whom 1,504 were SSRI users. SSRI initiation versus nonuse was associated with a 2.41-fold (95% confidence interval [CI]: 2.01–2.89) increased risk for a high INR, which was 3.14-fold (95% CI: 1.33–7.43) among CYP2C9-inhibiting SSRI users. The adjusted hazard ratio of major bleeding was 1.22 (95% CI: 0.99–1.50) in all SSRI users and 1.31 (95% CI: 0.62–2.72) in CYP2C9-inhibiting SSRI users compared with nonusers. Conclusion  SSRI use is associated with an increased risk of high INR and might be associated with major bleeding. The risk of a high INR was slightly more elevated for CYP2C9-inhibiting SSRI users, suggesting there might be a pharmacokinetic interaction (by CYP2C9 inhibition) next to a pharmacodynamic effect of SSRIs on platelet activation.


Introduction
Vitamin K antagonists (VKAs) are used to treat and prevent thrombosis. In the Netherlands, patients using VKAs are monitored by anticoagulation clinics, which regularly measure the international normalized ratio (INR). Despite this intensive monitoring system, 1 to 3% of the VKA users suffer from major bleeding each year. 1,2 In addition to the effect of VKAs, co-medication such as selective serotonin reuptake inhibitors (SSRIs) may be associated with an increased risk of bleeding. [3][4][5][6][7] There are three possible causal mechanisms by which SSRIs could increase the risk of bleeding described in previous literature. First, SSRIs inhibit serotonin reuptake in platelets which might result in a decreased platelet function leading to an increased risk of bleeding. 8 Second, SSRIs might increase gastric acid secretion, which might increase the risk of upper gastrointestinal bleeding by increasing the risk of peptic ulcer disease. 9 Third, the SSRIs fluvoxamine and fluoxetine have a pharmacokinetic interaction with VKAs through inhibition of cytochrome P450 2C9 (CYP2C9), which is the most important enzyme for the metabolism of VKAs. 10,11 This inhibition could lead to a high INR and consequently an increased bleeding risk. Previous studies showed that use of fluvoxamine (a CYP2C9 inhibitor) is associated with an increased risk of a high INR compared with nonusers of SSRIs, 12 but have not studied major bleeding as an outcome. Other studies reported relative risk estimates of major bleeding between 1.0 and 1.7, but were underpowered as shown by large confidence intervals (CIs) around these point estimates and correction for potential confounders was not optimal. 4,6,7 Furthermore, none of these studies included phenprocoumon users, while phenprocoumon is metabolized by both CYP2C9 and CYP3A4, whereas acenocoumarol and warfarin are predominantly metabolized by CYP2C9. These differences in metabolism may make phenprocoumon less prone to the pharmacokinetic interaction with SSRIs through CYP2C9 inhibition as compared with acenocoumarol and warfarin. 13 Because of these limitations, there is still uncertainty regarding the increased risk of major bleeding due to SSRI use in patients treated with a VKA and possible mechanisms of these bleeds have not been studied extensively. To offer patients the most optimal treatment for both depression and thrombosis, these mechanisms may provide clues to what antidepressants are preferred in patients using VKAs and which type of VKA might be preferred in patients using SSRI. We therefore determined whether SSRIs are associated with major bleeding during VKA treatment and investigated possible mechanisms that might cause major bleeding. To establish this aim, we performed a cohort study, in which both the INR and major bleeding were studied as outcomes.

Study Population and Data Collection
In this cohort study, all patients aged 18 years or older who started using a VKA between January 1, 2006 and January 1, 2018 at the Anticoagulation Clinic Leiden or the Thrombosis Service of Star-shl, Rotterdam, were included. They were followed from the start of VKA use until they stopped using the VKA, moved to a city not covered by the anticoagulation clinics, died, or reached the end of the study period (January 2018).
Patient characteristics, exposures, and outcomes were derived from the electronic patient records of both anticoagulation clinics. 14 There, the INR of every patient is measured every 1 to 6 weeks; the interval between INR measurements is based on the stability of the INR and the previous INR measurement. At every monitoring visit, nurses take a standardized short questionnaire including questions on changes of co-medication and the occurrence of bleeds. Bleeding events are reported to the anticoagulation clinic by physicians and re-enlisting forms to the anticoagulation clinic after a hospital discharge. 15 Records of the cohort include age, sex, indication for VKA therapy, type of VKA (acenocoumarol or phenprocoumon), INR target range, INR measurements, the occurrence of bleedings, VKA dosage, and concomitant medication use (such as SSRIs). 16 Institutional review board approval was obtained through Medical Ethical Committee of the Leiden University Medical Centre; participant consent was waived because the analysis used preexisting, coded data.

Exposure and Outcomes
The starting date and end date of SSRI use were reported by the patient or the drug-dispensing pharmacy. Patients using SSRI at baseline were classified as prevalent users; they were considered exposed to SSRI from baseline. Patients who initiated a SSRI during the study period were classified as incident users and considered exposed from the start date of the SSRI. The exposure period ended 30 days after the date SSRI use was discontinued during follow-up or when the end of the study period was reached. We added these extra 30 days to the exposure period to account for the elimination of the SSRI. SSRI use was stratified by CYP2C9-inhibiting SSRIs.
The three studied outcomes were high INR, difference in VKA dosage, and major bleeding, which are described below. A high INR was defined as an INR above 4 or above 5. Both cutoffs were considered since an INR above 4 is associated with a slight increased risk of major bleeding and an INR above 5 is associated with an apparent increased risk of major bleeding. 17 For the analysis with high INR as an outcome, we matched every VKA user who initiated treatment with SSRIs with up to five non-SSRI users on age, sex, type of VKA, duration of VKA treatment, VKA indication, and use of proton-pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs (NSAIDs) and antiplatelet drugs. We matched on the duration of VKA use, since INR values outside the therapeutic range are more common in the first period after the initiation of VKA. When a SSRI was initiated within 30 days of VKA initiation, the matching nonusers used VKA for a maximum of 15 days longer. A maximum difference of 30 days was used, when the SSRI was initiated between day 30 and day 360 after VKA initiation. When a SSRI was initiated after day 360 of VKA use, a difference in the duration of VKA use with the matching nonusers of maximum 180 days was accepted. The presence of a high INR was considered during a period of 2 months after SSRI initiation for the SSRI initiators or match day for the matched non-SSRI users.
To verify the results of the analysis with high INR as an outcome, we included VKA dosage as an outcome. If a pharmacokinetic interaction occurs between VKA and SSRI, we expected that a high INR would occur, after which the dosage of VKAs would decrease. Therefore, for SSRI users the difference in VKA dosage before the start of the SSRI and 2 months after the start of the SSRI was also calculated. Dosage changes were considered in tablets (1 tablet acenocoumarol equals 1 mg and 1 tablet phenprocoumon 3 mg) and in percentages with the dosage before SSRI initiation as the reference category. Both measures were considered since relative dosage changes (percentages) are more comparable when the amounts of tablets differ at baseline, but dosage change in tablets might be more intuitive for VKA prescribers.
Major bleeding was defined as fatal bleeding, bleeding in a critical area or organ, or bleeding resulting in hospitalization or blood transfusion and classified according to the location of bleeding (i.e., intracranial bleeding, gastrointestinal bleeding, ocular bleeding, cutaneous bleeding, joint and muscular bleeding, urogenital bleeding, respiratory tract bleeding, traumatic bleeding, and other types of major bleeding). We stratified the analysis for site of bleeding to explore whether there was a marked increase in the risk of gastrointestinal bleeding which might point to the mechanism of increased gastric acid secretion.

Statistical Analyses
For the analyses with major bleeding as an outcome, we used a Cox proportional hazards model with time-dependent covariates to estimate crude and adjusted hazard ratios (HRs) and 95% CIs. We adjusted for sex and the time-varying confounders age, indication of VKA treatment, INR target range, PPIs, NSAIDs, and antiplatelet drugs in the analysis. For this analysis, patients were followed until the first occurrence of a major bleeding and censored at the moment they stopped using the VKA, moved to a city not covered by the anticoagulation clinics, died, or reached the end of the study period (January 2018).
To study the association of SSRI use with a high INR, we estimated odds ratios with 95% CIs by means of conditional logistic regression. Furthermore, the association of SSRI use with VKA dosage was studied by performing a paired t-test.
In all analyses, SSRI users were stratified by CYP2C9inhibiting SSRIs to be able to study a potential pharmacokinetic interaction. In addition, the results were stratified by type of VKA to explore a potential difference between phenprocoumon and acenocoumarol. To account for confounding by indication (i.e., the indication depression leads to the outcome of a high INR or major bleeding), we performed a sensitivity analysis in tricyclic antidepressant (TCA) users as a negative control. 18 Since mirtazapine and nortriptyline are commonly used antidepressants and have, just like maprotiline and doxepin, the lowest affinity for the serotonin transporter, these four TCAs were chosen. 19 Patients using both SSRI and TCA at the same time were classified as SSRI users. A second sensitivity analysis was conducted for the analysis on major bleeding to rule out prevalent user bias, 16 in which we stratified the analysis based on incident and prevalent SSRI users.
All analyses were performed with R, version 4.1.0, a language and environment for statistical computing and graphics.

Results
A total of 58,918 patients started VKA therapy within the study period. Of these, 57,019 were nonusers, 1,504 were SSRI users, and 395 were TCA users at baseline. The number of treatment periods was higher than the number of patients, since some patients were treated with a VKA multiple times and discontinued treatment in between. This resulted in 61,245 treatment periods for baseline nonusers, 1,557 treatment periods for baseline SSRI users, and 404 treatment periods for baseline TCA users. Of all patients at baseline, 52% (n ¼ 32,898) were male, while in the groups of SSRI users and TCA users, 31% (n ¼ 488) and 39% (n ¼ 159) patients were male, respectively. At baseline, the mean age was 69 years (standard deviation: 15). The most common indications for use of a VKA were atrial fibrillation (56%) and venous thromboembolism (27%), and a low INR target range was used in most treatment periods (91%). Acenocoumarol was used by 97% (n ¼ 41,394) of the patients at the anticoagulation clinic in Rotterdam as compared with 15% (n ¼ 3,016) of the patients at the anticoagulation clinic in Leiden. Concerning co-medication, users of a SSRI or a TCA used a NSAID twice as often (6 and 5%, respectively) compared with nonusers (3%). The prevalence of PPI use was higher among SSRI (44%) and TCA users (38%) compared with nonusers (15%) (►Table 1). A total of 1,182 patients initiated SSRI and 456 patients initiated TCA during the study period. At the time of initiation of SSRI and TCA for the initiators, the mean age at initiation was 76 and 79 years, respectively (►Supplementary Table S1 [online only] for all initiators, ►Supplementary Table S2 for initiators included in the major bleeding analysis [online only]).

Discussion
This study shows that the initiation of a SSRI during VKA treatment is associated with an increased risk of a high INR during the first 2 months after initiation. For all SSRI initiators, the risk was twofold increased, and when restricting to CYP2C9-inhibiting SSRIs, the initiation was associated with a threefold increased risk. Initiation of non-CYP2C9inhibiting SSRIs or TCAs was associated with a twofold increased risk of a high INR. Furthermore, we observed the risk of major bleeding for concomitant users of VKAs and all SSRIs is possibly slightly increased. When stratified for CYP2C9-inhibiting SSRIs and the non-CYP2C9-inhibiting SSRIs, similar results were observed for both groups. TCA users did not show increased risk estimates of major bleeding.

SSRI Use and a High INR and VKA Dosage
We found an increased risk of a high INR and a corresponding decrease in VKA dosage in all SSRI initiators compared with nonusers, which was in line with a previous study. 12 Beforehand, we expected this only for the CYP2C9-inhibiting SSRIs, due to the pharmacokinetic interaction. 10,11 When restricting to CYP2C9-inhibiting SSRIs, a higher risk of an increased INR and a stronger dosage decrease were observed compared with non-CYP2C9-inhibiting SSRIs, which might indicate that CYP interaction decreases the metabolism of VKAs, resulting in a higher INR (i.e., the pharmacokinetic mechanism), but a chance finding cannot be excluded. For TCA initiators the risk of a high INR was increased, but the mean dosage difference was around zero. These results seem contradictory, however, VKA dosage can be influenced by other factors. Since the risk of a high INR was also increased for TCA initiators (our negative control), the increased risk is possibly partly explained by confounding due to the indication depression. Depressed patients might take less care of themselves and eat less, including green vegetables containing vitamin K, which may result in increased occurrence of high INR. This hypothesis is supported by multiple studies that showed that patients suffering from a mental illness had a lower time in INR target range than healthy VKA users. [20][21][22]

SSRI Use and Major Bleeding
We observed a possible association between SSRI use and major bleeding in patients using VKA, with a small increased risk of major bleeding for SSRI users. To interpret this result, several aspects must be considered. Although a chance finding cannot be excluded entirely since the lower limit of the 95% CI for major bleeding in SSRI users is just below 1.0, the observed risk estimate for major bleeding is confirmed by the results of previous research showing an increased risk for major bleeding in SSRI, 4,6 making a chance finding less likely. Furthermore, as in all observational studies, residual confounding should be considered. However, our sensitivity analysis showed that TCA use was not associated with an increased risk of major bleeding compared with nonusers, which corresponds with results from previous research 5 and suggests absence of confounding by indication (i.e., depression) for the outcome of major bleeding. Therefore, it is possible that SSRI use is associated with an increased risk of major bleeding, which supports the hypothesis of platelet inhibition.
Although the risk estimate of major bleeding for patients using CYP2C9-inhibiting SSRIs was slightly more pronounced compared with non-CYP2C9-inhibiting SSRI users, the number of events in the patients using CYP2C9-inhibiting SSRIs is too low to draw formal conclusions regarding CYP interaction and the association with major bleeding.
After stratification for type of VKA, no association was found between SSRI use and major bleeding in phenprocoumon users, while for acenocoumarol users we observed the risk of major bleeding among the SSRI users was possibly elevated. This difference could be due to the long half-life of phenprocoumon, ensuring more stable INRs 23,24 and potentially fewer major bleedings. Furthermore, the inhibition of CYP2C9 is expected to influence the metabolism of phenprocoumon less than acenocoumarol, because the enzyme CYP3A4 is also involved in the metabolism of phenprocoumon. 1,13,25,26 This may explain why no difference in risk estimates was observed between CYP2C9-inhibiting SSRIs and the non-CYP2C9-inhibiting SSRIs in phenprocoumon users and may support that the pharmacokinetic interaction may explain the increased risk for major bleeding among SSRI users. However, the number of events in the analysis stratified for type of VKA is too low to draw formal conclusions. The analysis stratified for site of bleeding showed that the risk of gastrointestinal bleeding for SSRI users might possibly be increased compared with nonusers, which is in line with previous research. 27,28 In case SSRIs increase gastric secretion, we would have expected that the risk of gastrointestinal bleeding would be increased more, compared with other locations. However, the risks for gastrointestinal major bleeding and intracranial bleeding and cutaneous major bleeding were similar. Therefore, our results did not indicate that major bleedings are caused by increased gastric acid secretion by SSRIs. A possible explanation may be the fact that SSRIs increase gastric acid secreting has only been studied in vitro and in rodent models, 29,30 so it remains unclear to what extend this mechanism is relevant for the risk of bleeding in patients using a SSRI. However, a limitation of our data was that details regarding the location (i.e., upper or lower) of gastrointestinal bleeding were not available. Therefore, we were not able to estimate the risk for upper gastrointestinal bleeding separately, and it is possible we missed a potential association.

Prevalent User Bias
Due to prevalent user bias, we expected beforehand that we would find higher relative risk estimates of major bleeding among incident users than prevalent users. Incident users are more representative for the population who start using a drug (here SSRIs). Prevalent users on the other hand are expected to be a selected group of patients who are relatively healthier, since they are physically able to continue using the SSRI and do not stop using the SSRI early, caused by for example side effects. In addition, they have a higher compliance to drugs and suffer less from side effects such as major bleedings. 16 Surprisingly, in prevalent SSRI users a slightly higher risk estimate of major bleeding was observed than in incident SSRI users. A possible explanation for this result could be that longer use of a SSRI may indicate that these patients have a more severe depression. Depression is associated with conditions such as coronary heart disease, diabetes, hypertension, and atrial fibrillation, 3 which are also risk factors for major bleeding. 2 As mentioned before, the shorter half-life of acenocoumarol compared with phenprocoumon could result in less stability in INR 23,24 and consequently an increased risk of major bleeding when starting a new drug, which is supported by our results. Another explanation is that the difference can be caused by small numbers and statistical variation.

Strengths and Limitations
More than 58,000 patients were included in our study, which is a strength as robust risk estimates were found compared with other studies. Furthermore, we have included TCA users as the negative control to detect potential confounding by indication (i.e., depression). Another strength is that we included phenprocoumon users in our study, which has a different pharmacokinetic profile from acenocoumarol, and gave us the opportunity to further study the mechanism by which SSRIs may cause major bleeding. Next to the limitations already mentioned above, our study has the following additional limitations. First, the use of co-medication, under which SSRIs and TCAs fall, is reported by the patients or drug-dispensing pharmacies in case of interacting co-medication (such as the CYP2C9inhibiting SSRIs). This could have resulted in underreporting of SSRI and TCA use and thereby misclassification of the exposure status, which would result in underestimation of the observed risk estimates. Second, the bleeding events were reported by patients or physicians, which might result in underreporting. However, we do not expect this to be dependent on the exposure status. Therefore, this might result in underestimation of the incidence rates of major bleeding, but we do not expect this to influence our relative risk estimates. Third, information was not available on the indication and dosages of SSRIs and TCAs and adjustment for the indication depression was not possible. Therefore, we included TCA as a negative control to detect confounding by depression. However, SSRIs are also used for the indication anxiety, and TCAs in a low dosage also against neuropathic pain. This may have led to suboptimal detection of confounding by indication. However, it is likely though that most included patients used a SSRI or a TCA against depression. Fourth, our cohort consists of patients who started using a VKA between 2006 and 2018. During this period direct oral anticoagulants (DOACs) were introduced and since 2016 the majority of patients who initiated anticoagulant therapy were prescribed a DOAC. 31 Since nowadays DOACs are the first-choice anticoagulant for most indications, characteristics of current VKA users might be different compared with our study population, including elderly patients who started VKA therapy before the DOAC era as well as patients with contraindication for DOACs such as patients with renal insufficiency. Since these characteristics are associated with major bleeding, the observed incidence of major bleeding might be an underestimation of the current risk of major bleeding. Last, despite the large total number of included patients, some subgroups, especially the CYP2C9-inhibiting SSRI users, had a small sample size, which limited the statistical power and resulted in wide confidence intervals.

Clinical Implications
To take our results into account in clinical practice, several aspects must be considered. The risk of a high INR for the non-CYP2C9-inhibiting SSRI initiators or users was lower than that for CYP2C9-inhibiting SSRIs compared with nonusers and the dosage decrease was weaker. Therefore, when a patient has to start a SSRI when already using a VKA, we would suggest starting a non-CYP2C9-inhibiting SSRI, based on our results. Since the indication depression might play a role, it could be useful to have more intensive monitoring during the first 2 months of SSRI and TCA use.
Next to VKA, DOACs are frequently used to treat and prevent thrombosis nowadays. This might raise the question of whether our results apply to DOAC users or whether DOACs and SSRI also interact. CYP metabolism has an important role in the elimination of apixaban and rivaroxaban, but occurs mainly by CYP3A4. 32 Since CYP2C9 does not play a major role in the metabolism of DOACs, we do not expect a relevant pharmacokinetic interaction on CYP level between SSRI and DOACs. However, in contrast to VKA, DOACs are substrates of the Pglycoprotein (P-gp) transporter, a transport protein on the cell membrane which is expressed in the gut, liver, and kidneys, that has a role in pharmacokinetics. Paroxetine, fluvoxamine, and sertraline might inhibit P-gp activity and thereby might cause a pharmacokinetic interaction with DOACs. 33 Next to this, the pharmacodynamic effect of SSRIs on platelets is expected to be present in DOAC users as well. Therefore, users of both DOACs and SSRIs might be at increased risk of major bleeding. Indeed, Zhang et al reported a 1.68-fold (95% CI: 1.10-2.59) increased risk of major bleeding in SSRI users as compared with nonusers in a cohort with incident DOAC users. 34 However, the analysis was not stratified for different types of SSRIs, therefore it is not known whether a certain type of SSRI should be preferred with regard to the risk of major bleeding. As far as we are aware, the risks of major bleeding for SSRI users in DOAC versus VKA users have not been studied.

Conclusion
In conclusion, SSRI initiation and in particular the initiation of CYP2C9-inhibiting SSRIs is associated with an increased risk of a high INR during VKA use, which could result in major bleeding. Based on our results, an association between SSRI use and increased risk of major bleeding during VKA use is highly possible. These results suggest that platelet inhibition and a pharmacokinetic interaction (by CYP2C9 inhibition) both may play a role in the occurrence of a high INR and major bleedings after SSRI initiation and during SSRI use.
What is known about this topic?
• Selective serotonin reuptake inhibitors (SSRIs) may increase the risk of major bleeding by decreasing platelet function or interaction with vitamin K antagonist (VKA) metabolism via cytochrome P450 (CYP).
What does this paper add?
• SSRI use was associated with an increased risk of a high international normalized ratio (INR) and might be associated with major bleeding. • The risk of a high INR was slightly more increased after initiation of a CYP2C9-inhibiting SSRI, supporting the hypothesis of CYP interaction. • We recommend using a non-CYP-inhibiting SSRI and a close INR monitoring.

Data Availability Statement
The deidentified data that support the findings of this study are available from the corresponding author upon reasonable request.

Funding/Acknowledgments
This study was performed by S.B. within the master's degree program in Pharmacy of Leiden University. J.L.I.B. is funded by a ZonMw grant (848017007). This study was not funded.

Conflicts of Interest
None declared.