Does concurrent prescription of selective serotonin reuptake inhibitors and non-steroidal anti-inflammatory drugs substantially increase the risk of upper gastrointestinal bleeding?
L. J. Tata, Division of Epidemiology & Public Health, University of Nottingham, Clinical Sciences Building, City Hospital, Hucknall Road, NG5 1PB, UK.
Background: A 15-fold increased risk of gastrointestinal bleeding has been reported with concurrent use of selective serotonin reuptake inhibitors and non-steroidal anti-inflammatory drugs. Recent guidance cautions against concurrent prescription, particularly in older people.
Aim: To quantify the risk of gastrointestinal bleeding associated with current exposure to non-steroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, and both drugs concurrently.
Methods: We conducted a case–control analysis of 11 261 cases with upper gastrointestinal bleeding and 53 156 controls matched by gender, age and general practice from computerized primary care data. We coupled this with self-controlled case series analysis.
Results: Both drugs were associated with a twofold increased risk of gastrointestinal bleeding (odds ratio =2.38, 95% confidence interval 2.08–2.72 for selective serotonin reuptake inhibitors and odds ratio = 2.15, 95% confidence interval 2.02–2.28 for non-steroidal anti-inflammatory drugs). This increased risk was marginally higher for concurrent prescription (odds ratio = 2.93, 95%confidence interval 2.25–3.82). The self-controlled analysis showed a greater incidence rate ratio for gastrointestinal bleeding with non-steroidal anti-inflammatory drugs (2.71, 95% confidence interval 2.51–2.91) and lower incidence rate ratio with selective serotonin reuptake inhibitors (1.71, 95% confidence interval 1.48–1.98). The incidence rate ratio when both drugs were combined was 3.25, 95% confidence interval 1.95–5.42. Estimates were similar after restricting to people over 80 years of age. Increased risk of gastrointestinal bleeding was not specifically related to class of non-steroidal anti-inflammatory drugs and was similar when we looked at tricyclic anti-depressants.
Conclusions: Our study suggests that the risk of gastrointestinal bleeding is not substantially increased when non-steroidal anti-inflammatory drugs and selective serotonin reuptake inhibitors are prescribed together, compared with their use alone.
Depression in older people is an important public health problem affecting at least 7% of people over the age of 75 years.1 Anti-depressant drugs provide the most effective treatment for depression, but there is concern that selective serotonin reuptake inhibitors (SSRIs) may interact unfavourably with non-steroidal anti-inflammatory drugs (NSAIDs), both of which are commonly prescribed in this age group, leading to an excess risk of gastrointestinal bleeding.2
The evidence base supporting the cautionary advice in a recent Drugs and Therapeutics Bulletin is limited. The largest population based case–control study from general practices in the UK found a 15-fold increase in risk of gastrointestinal bleeding,3 although there were only 9 controls and 16 cases with current exposure to both drugs. A further limitation of this study was the failure to control fully for comorbidity. A more recent cohort study from Denmark found a 12-fold increase in risk of hospital admission for gastrointestinal bleeding, based on 17 events, in concurrent users of SSRIs and non-aspirin NSAIDs, yet indication bias, where these drugs are channelled towards individuals perceived to be at higher risk of complication, could provide an alternative explanation for this finding.4 Furthermore, both of these studies used selected populations in which people with common risk factors for gastrointestinal bleeding were excluded. The results therefore are not widely generalizable and so guidelines based on their findings may be over cautious.
To precisely quantify the risk of gastrointestinal bleeding with concurrent prescription of specific anti-depressants and NSAIDs, we carried out a large, population based, case-control study using The Health Improvement Network (THIN) database. To minimize potential confounding by comorbidity or indication bias unaccounted for in the classical approach we additionally carried out a self-controlled case series analysis.
We used THIN, a database of computerized medical records from general practices across England and Wales.5 Data for THIN are collected using Vision software and the Value Added Medical Products system, which are also used in the General Practice Research Database.6, 7
Study population, exposures and confounders
We identified individuals with a first diagnosis of upper gastrointestinal bleed between 1 January 1990 and 1 November 2003 from 176 general practices. All cases were 18 years of age or older at the time of the gastrointestinal bleed and had to have at least 6 months of general practice data prior to the event, to ensure adequate time was available for possible exposure to our drugs of interest. Our definition of upper gastrointestinal bleeding included Read diagnostic codes for duodenal bleeding, haematemesis and melaena, but excluded oesophageal and jejunal bleeding codes. We defined the date of the gastrointestinal bleeding event as the index date and, for each case, we randomly selected up to six controls matched by gender, general practice and age at the index date.
We categorized our anti-depressant drug exposures as SSRIs or tricyclic anti-depressants (TCA), and our NSAID drug exposures as non-selective NSAIDs, cyclo-oxygenase-2 selective inhibitors (COX-2 inhibitor), or aspirin. We extracted data on smoking status, body mass index (kg/m2) and comorbidity preceding the index date based on the Charlson Index,8 a validated composite score based on the type and number of diagnosed morbidities prior to the bleed event (higher score = more severe and/or increasing number of illnesses). This index is a weighted score which gives comorbidites associated with a greater risk of death a higher individual value. It has been shown in many studies to measure comorbidity with reasonable accuracy.
Using conditional logistic regression we estimated odds of first gastrointestinal bleed associated with current exposure (at least one prescription within the period 30 days before the index date) to anti-depressants and NSAIDS or to both. In multivariable analyses, we explored the effect of comorbidity and assessed the possibility of confounding by smoking status and body mass index, by adding these variables to the model and retaining only those covariates that affected the odds ratio (OR) by 10% or more. We fitted multiplicative interaction terms where appropriate. Some individuals in our study population had an indication of a prior gastrointestinal bleed (retrospective data entry of medical events occurring before their general practice registration date), or upper gastrointestinal ulcer with no bleeding complication, including perforation, erosion, or obstruction (retrospective data entry as well as events during their general practice record). To check the robustness of our findings we re-analysed final models excluding these individuals.
For our self-controlled case series analyses we compared the relative incidence of gastrointestinal bleeding in predefined high risk periods to other control periods, within-person. The main advantage of this method is the elimination of inter-individual confounding, such as differences in comorbidity.9 Using age-adjusted conditional Poisson regression we estimated the risk of gastrointestinal bleeding by comparing incidence during periods of exposure to anti-depressants, NSAIDs and both drugs concurrently, compared with incidence during periods when people were not exposed to either drug. To account for possible contraindication of anti-depressant or NSAID prescriptions shortly following a gastrointestinal bleed, i.e. that a change in underlying disease might affect attitudes to prescribing these drugs, or prescribing certain types of drugs rather than other types, we excluded the 30 days prior to the first prescription of each drug exposure episode from the baseline risk period. All analyses were carried out in Stata SE8 (StataCorp. 2003. Stata Statistical Software: Release 8.0; Stata Corporation, College Station, TX, USA).
We identified 11 261 people with a first gastrointestinal bleed and 53 156 matched controls who contributed a median of 5 years (inter-quartile range 2–9) of data prior to the case's gastrointestinal bleed. More cases and controls were male (54%) and the mean age at the gastrointestinal bleed was 61 years (standard deviation 21 years) (Table 1). Cases were more likely to be current or ex-smokers compared with controls (χ2 test for trend P < 0.001 across categories of smoking status) and were more likely to be underweight [OR 1.28 95% confidence interval (CI) 1.08–1.51] or obese (OR 1.30 95% CI 1.20–1.40). Cases were considerably more likely to have a CharIson Index of 2 or greater.
Table 1. Characteristics of cases and controls
|Male||6109 (54.2)||28 901 (54.4)||–|
|Age at first Gl bleed*|
| <40.0||2112 (18.8)||11 516 (21.7)||–|
| 40.0–59.9||4059 (36.0)||17 745 (33.4)||–|
| 60.0–79.9||2268 (20.1)||11 696 (22.0)||–|
| ≥80||2822 (25.1)||12 199 (22.9)||–|
| Non-smoker||3888 (34.5)||19 340 (36.4)||Reference|
| Ex-smoker||578 (5.1)||2526 (4.8)||1.15 (1.04–1.28)|
| Current smoker||3868 (34.3)||15 498 (29.2)||1.40 (1.33–1.48)|
| Missing||2927 (26.0)||15 792 (29.7)||0.86 (0.81–0.91)|
|Body mass index (kg/m2)|
| Underweight (<1 8. 5)||186 (1.7)||728 (1.4)||1.28 (1.08–1.51)|
| Normal (18.5–24.9)||2954 (26.2)||14 861 (28.0)||Reference|
| Overweight (25.0–29.9)||2496 (22.2)||11 496 (21.6)||1.06 (1.00–1.13)|
| Obese (>30)||1118 (9.9)||4311 (8.1)||1.30 (1.20–1.40)|
| Missing||4507 (40.0)||21 760 (40.9)||0.98 (0.93–1.04)|
| 0||5384 (47.8)||36 748 (69.1)||Reference|
| 1||2981 (26.5)||9897 (18.6)||2.21 (2.10–2.33)|
| ≥2||2896 (25.7)||6511 (12.2)||3.51 (3.30–3.73)|
Table 2 shows the prevalence of being prescribed a drug of interest in the 30 days prior to the case's gastrointestinal bleed. NSAIDs were most commonly prescribed, with 19% of cases and 10% of controls having current prescriptions compared with anti-depressant use of 7% and 4% respectively. For NSAID exposure and gastrointestinal bleeding, the effects were similar for non-selective drugs (OR 2.71 95% CI 2.50–2.94) and COX-2 inhibitors (OR 2.67 95% CI 2.12–3.36), but lower for aspirin (OR 1.54 95% CI 1.42–1.68). Current exposure to individual anti-depressant drugs was associated with an increased risk of gastrointestinal bleeding with a slightly higher risk during current SSRI use (OR 2.38 95% CI 2.08–2.72) compared with TCAs (OR 1.87 95% CI 1.67–2.09). When we assessed the confounding effect of smoking habit and body mass index, there was no material change in the ORs, indicating that they did not contribute to the analyses, so the estimates presented are not adjusted for these factors. After taking comorbidity into account we observed that the risk of gastrointestinal bleeding with current drug exposures decreased considerably with increasing Charlson Index, reaching null for most drugs in individuals with an index of 2 or greater (χ2 test for trend P < 0.001 for all drugs). For example, the ORs for gastrointestinal bleeding following exposure to antidepressants were 2.48 (95% CI 2.13–2.89) for individuals with a Charlson Index of 0, 1.88 (95% CI 1.44–2.46) for individuals with a Charlson Index of 1 and 1.22 (95% CI 0.95–1.56) for individuals with a Charlson Index of 2 or greater. This pattern of decreasing relative risk with increasing Charlson Index was similar in people exposed to NSAIDs. To investigate whether this gradient of decreasing relative risk was because of differential prescribing of acid suppressive drugs in people with a higher Charlson Index, we extracted all prescriptions for proton pump inhibitors and H2 antagonists. We found that these patterns were not explained by increased prescribing of gastro-protective medication in those people with higher Charlson Index. As these findings represent an interaction, we have not presented results adjusted for this factor.
Table 2. Association between anti-depressant and NSAID exposures and first Gl bleed
|Any anti-depressant||819 (7.3)||1855 (3.5)||2.09 (1.91–2.28)|
|Any SSRI||362 (3.2)||715 (1.3)||2.38 (2.08–2.72)|
|Any TCA||474 (4.2)||1165 (2.2)||1.87 (1.67–2.09)|
|Any NSAID||2109 (18.7)||5147 (9.7)||2.15 (2.02–2.28)|
|Any non-selective NSAID||1058 (9.4)||1943 (3.7)||2.71 (2.50–2.94)|
|Any COX-2 inhibitor||125 (1.1)||222 (0.4)||2.67 (2.12–3.36)|
|Aspirin||991 (8.8)||2951 (5.6)||1.54 (1.42–1.68)|
Table 3 shows that the OR for gastrointestinal bleeding was slightly higher for concurrent prescription of an anti-depressant and a NSAID (OR 2.83 95% CI 2.39–3.34) compared with being prescribed only one of these drug types (OR 2.23 95% CI 2.01–2.47 for anti-depressants and OR 2.19 95% CI 2.05–2.33 for NSAIDs). The ORs were not greater in people over 80 years of age. There was little variation when we explored the risks associated with various combinations of SSRIs/TCAs and non-selective NSAIDS/COX-2 Inhibitors. The estimates were similar when we repeated our case–control analyses excluding individuals with indications of previous ulcer or gastrointestinal bleeding. Concurrent prescription of gastro-protective drugs did not substantially change the risk estimates associated with anti-depressants or NSAIDs or the combined effect of these two drugs.
Table 3. Association between anti-depressant and NSAID interactions and first Gl bleed
|Anti-depressant only||581||1408||2.23 (2.01–2.47)||148||414||1.80 (1.47–2.21)|
|NSAID only||1871||4700||2.19 (2.05–2.33)||738||2030||1.93 (1.74–2.14)|
|Anti-depressant and NSAID||238||447||2.83 (2.39– 3.34)||93||191||2.37 (1.82– 3.10)|
|SSRI only||253||522||2.63 (2.24–3.07)||59||130||2.31 (1.66– 3.20)|
|SSRI and NSAID||92||168||2.93 (2.25–3.82)||41||78||2.64 (1.76–3.94)|
|TCA only||318||870||1.96 (1.72–2.25)||87||279||1.57 (1.22–2.02)|
|TCA and NSAID||139||270||2.70 (2.18– 3.35)||51||109||2.23 (1.57– 3.17)|
|Non-selective NSAID only||801||1551||2.84 (2.59– 3.11)||251||414||2.96 (2.50– 3.51)|
|Anti-depressant and non-selective NSAID||101||140||3.80 (2.91–4.96)||25||36||3.26 (1.90– 5.60)|
|COX-2 inhibitor only||69||140||2.97 (2.20–4.00)||27||49||3.19 (1.95– 5.23)|
|Anti-depressant and COX-2 inhibitor||18||28||3.43 (1.86–6.34)||6||6||5.14 (1.58– 16.75)|
In our self-controlled analysis, we identified 8130 cases with at least one exposure to an anti-depressant or an NSAID. The incidence rate ratios (IRR) for gastrointestinal bleeding occurring during drug exposure were 1.71 (95% CI 1.48–1.98) for SSRIs, 1.23 (95% CI 1.06–1.42) for TCAs, and 2.71 (95% CI 2.51–2.91) for NSAIDs. With concurrent prescription, the IRRs were 3.25 (95% CI 1.95–5.42) for SSRIs combined with NSAIDs and 2.88 (95% CI 1.78–4.66) for TCAs combined with NSAIDs. When we restricted the analysis to the 2028 individuals who were 80 years of age or older at the case's gastrointestinal bleed, the IRRs were 1.85 (95% CI 1.47–2.34) for anti-depressants, 3.03 (95% CI 2.65–3.46) for NSAIDs and 4.35 (95% CI 2.08–9.09) for concurrent prescription.
In contrast to previous studies, we have found only a threefold increase in risk of gastrointestinal bleeding associated with concurrent prescription of anti-depressants and NSAIDs. The risk with both drugs was only marginally higher than the risk associated with separate exposure to anti-depressants or NSAIDs in our case–control analyses. Our self-controlled case series analysis showed a similar increase in risk of gastrointestinal bleeding with concurrent prescription, however the excess risk with anti-depressant prescription was smaller and not specific to SSRIs, suggesting that this association might not be causal. Our findings are important as they are widely generalizable and suggest that current advice to avoid combining these drugs so as to avoid an excess risk of gastrointestinal bleeding is over cautious.
Strengths and limitations
Our study is large and has the statistical power to address the specific hypothesis of an interaction between two commonly prescribed drugs. Unlike previous studies we did not exclude individuals on the basis of prior comorbidity but investigated the role of comorbidity in our analysis by using a validated weighted comorbidity score and by conducting a separate self-controlled case-series analysis. We found that the increased risk of gastrointestinal bleeding associated with both drug exposures of interest were slightly smaller for individuals in the highest category of comorbidity. Although this may appear counter intuitive these individuals represent a select group already at high risk of gastrointestinal bleeding, for whom there is greater consideration of possible drug contraindication by the physician, likely relating to possible gastrointestinal bleeding. Our finding of an increased risk of gastrointestinal bleeding with concurrent use of NSAIDs and anti-depressants in people with no comorbidity is consistent with other studies of these drugs, however, we have estimated a lower increased risk. Should however residual confounding by comorbidity remain in our case–control analysis, then our use of the self controlled case series analysis will have eliminated this problem. A limitation of this approach, however, is that, where disease is present, any change in disease severity may affect attitudes to prescribing of drugs considered to increase the risk of bleeding. Our use of a 30-day period of exclusion following the index gastrointestinal bleed to some extent avoids this problem. Nonetheless our estimation of the gastrointestinal bleeding risks associated with NSAIDs are similar to numerous other studies and are replicated in both our case-control and case series analyses.10 The increased risks associated with anti-depressants in our study are less consistent between our two methods of analysis. One explanation of this inconsistency is that as the case series analysis minimizes confounding by severity of the underlying disease or other unmeasured risk factors such as alcohol intake, residual confounding has inflated our case–control results. People receiving anti-depressant treatment may be at greater risk of gastrointestinal bleeding for reasons relating to the underlying severity of depression or related health behaviours. We believe that the estimates obtained from our case series analysis, with respect to the risk of gastrointestinal bleeding associated with anti-depressant prescriptions, are a closer representation of the true magnitude of risk and suggest that the observed relationship may not be causal. It seems likely that the increased risk of gastrointestinal bleeding in those people prescribed both anti-depressants and NSAIDs concurrently is therefore mostly due to the NSAID use.
In a population based case–control study from general practices in the UK, current exposure to SSRIs was the same as in our population (3% in cases and 1% in controls), and the results were a similar risk of upper gastrointestinal bleeding (OR 3.0, 95% CI 2.1–4.4) associated with SSRI use within 30 days prior to the bleed event. Our results were also similar to those found with their non-SSRIs, which included many of the drugs we classified as TCAs (4 and 2%, OR 1.4, 95% CI 1.1–1.9).3 In marked contrast to our study they found a 15-fold increase in risk of gastrointestinal bleeding with concurrent exposure to NSAIDs and anti-depressants, although they had considerably fewer people exposed to both drugs. The differences between the Danish cohort study, where they found a 12-fold increase in risk of hospital admission for gastrointestinal bleeding associated with use of both SSRIs and non-aspirin NSAIDs together, and our results could also be due to chance, although indication bias, or channelling, where newer drugs that are perceived to be safer are given to frailer individuals, might also explain their findings.4 By using the case series analysis we have minimized this problem in our study. Both of these previous studies used selected populations in which people with common risk factors for gastrointestinal bleeding were excluded. The results, in contrast to our own, are therefore not widely generalizable.
Our study shows that concurrent prescription of NSAIDs and either SSRI or TCAs does not confer a substantially increased risk of gastrointestinal bleeding compared with each drug alone. This finding is of importance as depression in the elderly is a current and common problem for which anti-depressants provide effective treatment. Our study suggests that there is no pressing need to avoid concurrent prescription of anti-depressants and NSAIDs in older people.
This study was completed as part of employment paid by the University of Nottingham. No external funding was required for this study.
Competing interest statement
LJ Tata, PJ Fortun, RB Hubbard, L Smeeth, CJP Smith, HJ Whitaker, TR Card, J West have no competing interests to declare.
CJ Hawkey has previously received research funding and/or honoraria from AstraZeneca, GlaxoSmithKline, Merck, Nitromed, Novartis, Pfizer, Takeda, Serono International SA, Wyeth.
CP Farrington has declared that in 1992–1993 he acted as expert witness in litigation over the safety of MMR vaccines, and received payment indirectly from the defendants (GlaxoSmithkline, Merck and Co. and Aventis Pasteur MSD). Also, his Department (Statistics, The Open University) is in receipt of an EPSRC CASE studentship partly funded by GlaxoSmithKline.