Low-dose aspirin is associated with an increased risk of haemorrhage; but not in people with diabetes who have a high baseline rate of bleeding
- Correspondence to: Sonia Butalia
Division of Endocrinology and Metabolism, University of Calgary, 1820 Richmond Road SW, Richmond Road Diagnostic and Treatment Center, Calgary, Alberta T2T 5C7, Canada;
Commentary on: De Berardis G, Lucisano G, D'Ettorre A, et al. Association of aspirin use with major bleeding in patients with and without diabetes. JAMA 2012;307:2286–94.
Implications for practice and research
Low-dose aspirin (≤300 mg/day) increases the risk of major bleeding.
Individuals with diabetes have a higher risk of bleeding independent of aspirin exposure.
Future work should include comparative effectiveness studies to help inform treatment strategies based on individual risks, benefits, patient preferences and values.
Aspirin has a proven role in the secondary prevention of cardiovascular disease.1 Further interest has emerged on extending its use for primary prevention. However, the desired benefits of aspirin need to be carefully balanced against the potential harms.1 Although the cardioprotective benefit of aspirin has been well defined, the general risk of haemorrhage remains uncertain as existing data are mostly limited to highly selected populations from trials.1 Addressing this evidence gap, De Berardis and colleagues assessed the risk of major bleeding in people with and without diabetes who were prescribed low-dose aspirin for routine care.
An Italian population-based cohort study was conducted of adults using national administrative data linking hospital discharge records with prescription databases. New aspirin users (patients receiving a new prescription for aspirin ≤300 mg/day) and non-users were followed from the index date to the first hospitalisation for gastrointestinal or cerebral haemorrhage, death or the end of the study, for a follow-up period of up to 6 years. The primary outcome was hospitalisation for major gastrointestinal bleeding or cerebral haemorrhage. Propensity score matching was performed to account for potential confounding by indication. Poisson regression was then used to estimate adjusted incidence rate ratios (IRRs) for the outcome comparing exposed and non-exposed individuals.
A cohort of 372 850 individuals was assembled with a median follow-up of 5.7 years. Aspirin use was associated with an increased risk of major bleeding (IRR 1.55, 95% CI 1.48 to 1.63). Aspirin exposure was associated with increased risk of major bleeding in males, older age, hypertension, use of other antiplatelet or anticoagulants and history of previous cardiovascular events, but not among patients with diabetes. Individuals with diabetes, even in the absence of aspirin, were at an increased risk of major bleeding (IRR 1.36, 95% CI 1.28 to 1.44), and this relationship was independent of aspirin exposure.
In this population-based cohort study, the authors provide an estimate of bleeding risk associated with routine aspirin use. Previous safety data were chiefly derived from randomised controlled trials (RCTs) which may underestimate harm because of shorter follow-up times and stringent eligibility criteria that tend to select healthier patients.1 ,2 In comparison with a previous meta-analysis of RCT data, this study demonstrated a fivefold higher incidence of hospitalisations for major bleeding among both those receiving and not receiving aspirin.1
There are, however, limitations to this study. First, the authors could not account for adherence or over-the-counter use of aspirin and non-steroidal anti-inflammatory drugs. Second, there was potential for misclassification bias (ie, diabetes was defined using prescription data), measurement bias (outcome ascertainment was based on billing codes) and selection bias (284 463 patients exposed to aspirin but not continuing treatment were labelled ‘former users’ and excluded).3 ‘Former users’, in reality, contribute person-time to both the exposed and non-exposed groups. By restricting the cohort in this way, all relative risk estimates will be biased.3 Accordingly, a more ideal approach would have been to include ‘former users’ by using an inverse probability weighted marginal structural model or a Cox model to incorporate aspirin use as a time-varying exposure.4 ,5 Finally, although propensity matching was used there was still potential for residual confounding.6
The results of this study are important as it provides large population-based risks for haemorrhage associated with aspirin and further demonstrates that an increased risk of bleeding among individuals with diabetes irrespective of aspirin exposure. It goes beyond previous data by providing estimates of harm in ‘real-world’ settings.2 The results of this study prompt future comparative effectiveness research to help inform recommended strategies for individuals at differing risks, but receiving aspirin for primary or secondary prevention.