The first epidemiological study.
The hypothesis that the use of SSRIs would increase the risk of bleeding received the first epidemiological support in 1999, when our group published a nested case-control study reporting a three times increased risk of upper gastrointestinal bleeding (de Abajo et al. 1999). For this study we used data from the UK-based General Practice Research Database (GPRD), a widely-known data source for pharmacoepidemiological research (García-Rodríguez & Pérez-Guthann 1998). Among the 1,651 cases confirmed as upper gastrointestinal bleeding, 52 (3.1%) were exposed to SSRIs during the 30-day time window before the index date (the date of the upper gastrointestinal bleeding diagnosis), while among the sample of 10,000 controls (a density-based random sample of the source population) we found 95 patients (1.0%) exposed to these drugs. This yielded an odds ratio of 3.0 (95% confidence interval (CI), 2.1–4.4), after adjustment for sex, age, past history of upper gastrointestinal disorder (including dyspepsia, peptic ulcer, and peptic ulcer complications), smoking, and use of NSAIDs, low-dose aspirin, anticoagulants and steroids (table 3). The exposure to drugs, including antidepressants, was assessed according to the computerised records of prescriptions which is considered complete at the GPRD. Additional adjustment for use of antiulcer drugs (antacids, omeprazole and H2 blockers) hardly modified the results (OR=2.9; 2.0–4.2). In the study we also gathered information from 248 cases presenting with ulcer perforation. Interestingly, no association was found with SSRIs (OR=1.3; 0.4–3.7).
Table 3. Risk of upper gastrointestinal bleeding associated with the use of antidepressants (from de Abajo et al. 1999).
| ||Cases (N=1,651)||Controls (N=10,000)||Adjusted RR* (95% CI)|
| SSRI#||52||95||3.0 (2.1–4.4)|
| NSRI§||74||241||1.4 (1.1–1.9)|
| Others$||4||25||0.8 (0.2–2.4)|
| Multiple||3||14||1.0 (0.3–3.7)|
| SSRI#||27||140||1.2 (0.8–1.9)|
| NSRI§||158||688||1.2 (1.0–1.5)|
| Others$||6||37||1.0 (0.4–2.6)|
All the individual SSRIs with proper information showed an increased risk ranging from 2.1 (0.9–5.1) of clomipramine to 4.3 (2.2–8.3) of paroxetine. We also included trazodone in this group, a drug with a rather selective but weak inhibitory action on the serotonin reuptake (table 1). Surprisingly, trazodone presented the highest increased risk (OR=8.6; 2.1–35.1). However, as trazodone only contributed with four cases, its exclusion from the analysis did not materially change the estimates for the SSRIs as a group (OR=2.9; 2.0–4.2). The main action of trazodone may be to block the 5-HT2 receptors, which, interestingly, are supposed to mediate the platelet aggregation effect of serotonin (Skop & Brown 1996).
A less increased risk was also found for some of the antidepressants considered also to have an inhibitory action, though not selective, on serotonin-reuptake mechanism: amitriptyline (OR=1.9; 1.2–2.9), imipramine (OR=2.1; 0.7–6.2) and lofepramine (OR=1.8; 0.8–4.0), while no association was found (OR=0.8; 0.2–2.4) with the group of antidepressants known to have no action on serotonin reuptake mechanism (nortryptiline, trimipramine, amoxapine, maprotiline, and mianserine). This observation was rather consistent with the mechanistic hypothesis linking the risk of bleeding with a depletion of serotonin from platelets. However, a discordant result was obtained: dothiepin, the tricyclic antidepressant most widely used in the UK, showed no increased risk (OR=1.0, 0.6–1.6). We advanced the explanation that in some experimental models dothiepin has shown a considerable selectivity on norepinephrine reuptake (Ishikawa et al. 1986). However, as far as we know, no in vivo studies have been published on the effect of dothiepin on platelet serotonin content at therapeutic doses.
The 52 cases considered as current users of SSRIs arose from a cohort of 69,593 persons who received a total of 435,021 prescriptions for any of these drugs, a total exposure of 35,731 person-years (assuming that one prescription lasts on average 30 days). The crude incidence rate of upper gastrointestinal bleeding for this population was estimated to be 1.2 per 10,000 prescriptions (1.5 per 1,000 person-years), two thirds of which (0.8 per 10,000 prescriptions or 1 per 1,000 person-years) would be attributable to SSRIs.
We also assessed the possible effect of dose and duration, but apparently these factors had no bearing on the risk of upper gastrointestinal bleeding (table 4). The majority of patients were prescribed medium or low doses, especially in the non-selective serotonin reuptake group where only 12% of current-user controls used equivalent amitriptyline doses over 75 mg. The duration of treatment lasts less than three months in over 40% of current-user controls. As shown in pharmacological studies, 20 mg fluoxetine or equivalents are enough in most patients to reduce serotonin content in more than 80% patients (Narayan et al. 1998). Therefore, although the effect must be dose-dependent, it is possible that the effect of most SSRIs is at therapeutic doses already near to maximum leaving little room to observe a dose-effect in an epidemiological study. The same can be said from the duration of treatment, bearing in mind that the effect is already maximum after 4–6 weeks.
Table 4. Effect of dose and duration of use among current single users of antidepressants as compared to non use.
| ||Cases (N=1,651)||Controls (N=10,000)||Adjusted RR* (95% CI)|
| Low/medium+||41||72||3.0 (2.0–4.6)|
| High||11||23||3.2 (1.5–6.8)|
| Low/medium$||67||211 ||1.5 (1.1–2.0)|
| High|| 7||30||1.0 (0.4–2.4)|
|Duration of treatment |
| Less than 91 days||18||40||2.7 (1.5–4.9)|
| 91 days or longer||34||55||3.3 (2.1–5.2)|
| Less than 91 days||27||100 ||1.2 (0.7–1.9)|
| 91 days or longer||47||142 ||1.6 (1.1–2.3)|
Perhaps the most striking result of our study was that the concurrent use of SSRIs with NSAIDs greatly increased the risk of upper gastrointestinal bleeding (OR=15.6; 6.6–36.6) (table 5), which shows a strong interaction (e.g. the relative risk of the combination is greater than the sum of the independent relative risks). The concomitant use of NSAIDs with non-selective serotonin-reuptake inhibitors gave rise to an OR of 4.6 as compared to non-use (95%CI, 2.8–7.9), which represents additivity of the independent effects. Likewise, a smaller interaction was observed between SSRIs and aspirin, OR=7.2 (95%CI, 3.1–17.1).
Table 5. Interaction between current use of SSRIs and current use of NSAIDs as compared to non-use of either drug (from de Abajo et al. 1999).
| ||Cases (N=1,651)||Controls (N=10,000)||Adjusted RR* (95% CI)|
| NSAIDs only|| 295||652||3.7 (3.2–4.4)|
| SSRIs only|| 38|| 93||2.6 (1.7–3.8)|
| NSAIDs and SSRIs|| 16|| 9||15.6 (6.6–36.6)|
Further epidemiological evidence on upper gastrointestinal bleeding.
Up to July 2005, we have obtained information from 5 other published epidemiological studies which provide data concerning the association between the use of SSRIs and upper gastrointestinal bleeding (table 6). All of them showed an increased risk, although in the one performed by Dunn et al. (2000), reported as a Letter to the Editor, the result did not reach statistical significance.
Table 6. Epidemiological studies which tested the hypothesis of an association between SSRIs use and an increase of upper gastrointestinal bleeding.
|Study||Design||Population||Exposed to SSRIs||Exposed cases||RR (IC95%) versus non-use||RR (IC95%) versus other antidepressants|
|de Abajo et al. (1999)||Nested case-control||General|| 69,593|| 52||3.0 (2.1–4.4)|| 2.2 (1.4–3.5) |
|Dunn et al. (2000)||Cohort (PEM)||General||237,609§||103φ||–||1.2 (0.9–1.7)|
|Van Walraven et al. (2001)||Cohort||Elderly (>65 y)||101,397||404||–|| 1.1 (1.0–1.2)*|
|Dalton et al. (2003)||Cohort||General|| 17,320|| 55||3.6 (2.7–4.7)||2.0 (1.0–4.6)|
|Tata et al. (2005)||Nested case-control||General||–||362||2.4 (2.1–2.7)||–|
|Wessinger et al. (2005)||Case-control||Hospital||–|| 71||1.5 (1.0–2.1)||–|
The study by Van Walraven et al. (2001) was a large cohort study performed in 383,911 elderly patients in Ontario, Canada exposed to antidepressants. They classified the antidepressants in 3 groups according to the dissociation constant (Kd) for the serotonin transporter provided by Tatsumi et al. (1997) (table 1). The only difference from our classification was that fluvoxamine appeared in the intermediate-affinity group and trazodone in the low-affinity group. They built a statistical model in which the natural log of bleeding rates was a linear function of serotonin reuptake inhibition grade (three groups). After adjusting for age, they found a significantly increased risk of 10.7% (95%CI: 4.6–17.2) per one unit of serotonin reuptake inhibition grade. They also found an effect modification by age, the octogenarians presenting the highest risk: 14.7 cases per 1.000 person-years in the high-affinity group versus 10.6 cases per 1,000 person-years in the low-affinity group. This rate difference corresponded to a number needed to treat in one year for having one extra case harmed (NNH) of 244. In a Cox-regression model, adjusting for sex, age, calendar year, diabetes, antecedents of upper gastrointestinal bleeding, and the use of NSAIDs, aspirin, corticoids, anticoagulants and antiulcer drugs, the hazard ratio of upper gastrointestinal bleeding associated with an increase of one unit in the affinity grade was 1.10 (95% CI: 1.02–1.19). This may be translated into a hazard ratio of 1.22 (1.13–1.32) for high affinity as compared to low-affinity. The interaction with NSAIDs was not studied.
Dalton et al. (2003), performed a cohort study linking the prescription and hospital discharge records from the North Jutland County, Denmark. A total of 26,005 patients using an antidepressant over the study period were included (giving a total follow-up of 28,751 for current use and 29,823 for former use). They compared the standardized rates of upper gastrointestinal bleeding obtained per each of the different subcohorts (SSRIs, Non-SSRIs, and others) with the ones from the population not receiving any antidepressant drugs nor any of the drugs thought to cause upper gastrointestinal bleeding. They found a rate ratio (standardized by sex, 5-year age and calendar year) of 3.6 (95%CI: 2.7–4.7) for current users of SSRIs as compared to non-users, with a rate difference of 3.1 per 1,000 person-years (a NNH of 323 per year). Current users of SSRIs and NSAIDs showed an increased risk of 12.2 (95%CI: 7.1–19.5) (excess risk due to interaction=4.9) with a rate difference of 16.3 per 1,000 person-yeras (NNH of 61 per year), giving strong support to the interaction we first described. The increased risk associated with low-dose aspirin plus SSRIs was reported as 5.2 (95%CI: 3.2–8.0). All of the individual agents showed an increased risk, but the results reached statistical significance only for citolopram (relative risk (RR)=4.1; 2.8–5.7), fluoxetine (RR=3.9; 1.2–9.0) and clomipramine (RR=2.3; 1.0–4.3). An increased risk was also found with amitriptiline (RR=2.5; 1.4–4.1).
Tata et al. (2005) performed a nested case-control study using data from a new database called THIN from the UK, similar in a sense to the GPRD database. In the period of 1990–2003 they found 11,261 people with a first diagnostic code of upper gastrointestinal bleeding and randomly selected 53,156 controls from the source population matched by gender, general practice and age with the cases. Unfortunately, it appears that they did not perform an individual case review nor a clinical validation against the paper-based clinical records, which is a standard procedure in pharmacoepidemiology. This may have led to a gross positive misclassification of cases, considering as valid cases patients who present a suggestive episode of bleeding for the general practitioner but is not confirmed later as a true episode (a selection of only those cases referred to a consultant or a hospital may have helped reduce such a potential misclassification). They found a relative risk of upper gastrointestinal bleeding associated with the use of any SSRI of 2.4 (95% CI: 2.1–2.7). The use of any tricyclic antidepressant was also found to be significantly associated with an increased risk (RR=1.9; 1.7–2.1). But in opposition to the data from de Abajo et al. (1999) and Dalton et al. (2003), they were not able to detect any relevant interaction between SSRIs and NSAIDs (the RR of the combination=2.9; 2.3–3.8). Surprisingly, the relative risk associated with NSAIDs alone (RR=2.2, 2.1–2.3) was much lower than the one reported by many other publications (RR around 4) (Hernández-Díaz & García-Rodríguez 2000), perhaps due to the misclassification mentioned before. No data on individual SSRIs were provided.
While preparing the present MiniReview, the study by Wessinger et al. (2005) has only been reported in abstract form. It refers to the preliminary results of a traditional hospital-based case-control study, including 417 cases with “acute gastrointestinal bleeding” and 500 controls. The use of SSRIs was present in 17% of the cases and 12% of the controls, giving an OR of 1.5 (95%CI: 1.02–2.14), very similar to the one found for NSAIDs/COX2 agents (OR=1.4; 95%CI: 0.99–2.06).
Epidemiological studies concerning other bleeding outcomes.
Five epidemiological studies have been published aiming at assessing the risk of bleeding outcome other than upper gastrointestinal bleeding, among the users of SSRIs and other antidepressants (table 7). Three of them, focusing on abnormal bleeding and perioperative blood transfusion, suggested an increased risk, while two others, both in intracranial haemorrhage, did not.
Table 7. Epidemiological studies performed to assess the association between the use of SSRIs and bleeding disorders other than upper gastrointestinal bleeding.
|Layton et al. (2001)||Cohort (PEM)||Abnormal bleeding|| 50,150||486||SSRI versus non-psychiatric drugs: 1.38 (0.82–2.34)|
|Movig et al. (2003)||Case-control||Perioperative blood transfusion||–|| 59||SSRIs versus non-use: 3.7 (1.4–10.2)|
|Meijer et al. (2004)||Nested case-control||Abnormal bleeding|| 64,647||196||High versus low affinity: 2.6 (1.4–4.8) High versus intermediate: 1.9 (1.1–3.5)|
|de Abajo et al. (2000)||Nested case-control||Intracranial bleeding||247,234|| 65||SSRIs versus non-use: 0.8 (0.3–2.3)|
|Bak et al. (2002)||Nested case-control||Intracranial bleeding||–||659||SSRIs versus non-use: 1.0 (0.6–1.6)|
Of special interest is the nested case-control study carried out by Meijer et al. (2004), as they managed to show a moderate linear relationship (R2=0.5085) between the affinity of the antidepressant on the serotonin reuptake transporter and the risk of hospitalization for bleeding, reinforcing the mechanistic hypothesis of an impaired platelet aggregation due to a blockade of serotonin reuptake. The odds ratio of being hospitalised for abnormal bleeding was estimated to be 2.6 (95%CI: 1.4–4.8) among the users of high-affinity serotonin antidepressants as compared to users of low-affinity drugs. Users of intermediate-affinity drugs also showed an increased risk (OR=1.9; 95%CI: 1.1–3.5).
Apparently, the lack of association between SSRI use and intracranial haemorrhage observed in two independent studies is not consistent with the hypothesis of an increased bleeding. However, the pathogenic mechanism leading to intracerebral bleeding may well be different from the one leading to gastrointestinal bleeding, and further studies are certainly warranted.