Aspirin for the Primary Prevention of Cardiovascular Events: A systematic review and meta-analysis comparing patients with and without diabetes Open Access

The report of this protocol-driven systematic review adheres to the Quality of Reporting of Meta-analyses (QUOROM) standards for reporting systematic reviews of randomized clinical trials (11).

Eligible studies were RCTs that enrolled patients with diabetes without a prior history of myocardial infarction or stroke and assessed the efficacy of aspirin at any dosage.

An expert reference librarian designed and conducted the electronic search strategy. Our systematic search included MEDLINE, EMBASE, Cochrane Library, Web of Science, and Scopus, since their inceptions until November 2008. We used database-specific controlled language and terms that describe the key concepts: aspirin, diabetes, cardiovascular events, prevention, and randomized trials. We also reviewed the reference sections of identified reviews, published guidelines, and published manuscripts known to the authors. Study inclusion was not limited by publication status or language.

Pairs of reviewers, working independently and in duplicate, identified potentially eligible records for full-text retrieval from the titles and abstracts. Records in which the reviewers disagreed were also retrieved in full text. The reviewers assessed the retrieved full-text articles for eligibility. An independent reviewer resolved disagreements.

The outcomes of interest were ischemic stroke, myocardial infarction, and all-cause mortality.

Two reviewers extracted the following data from each eligible article: description of study participants, study characteristics, and outcome data.

Independent reviewers working in duplicate determined the extent to which trials reported concealment of allocation; blinding of patients, providers, and outcome assessors; and the extent of loss to follow-up.

We contacted the corresponding and/or first author of each eligible article identified in the screening process by e-mail to confirm data extraction and quality assessment and to request missing information.

We used the κ statistic to quantify chance-adjusted agreement between reviewers for article selection. Random-effects meta-analyses estimated the pooled treatment effects (relative risks [RRs] and their 95% CIs) of aspirin in preventing cardiovascular events and all-cause death in patients with diabetes and without diabetes. The I² statistic estimated the percentage of inconsistency across studies that was due to heterogeneity rather than chance or random error (12).

Given that subjects included in the Antiphospholipid Antibody–Acetylsalicylic Acid (APLASA) study were receiving aspirin for antiphospholipid antibody syndrome and had a short duration of follow-up and that a small number of subjects in ETDRS had suffered a prior stroke or myocardial infarction, we conducted sensitivity analysis to determine whether the exclusion of these studies affected the conclusions of this review.

We implemented three methods to determine whether a difference exists in aspirin effect in patients with and without diabetes. First, we estimated the ratio of RRs and its 95% CI using the method of Altman and Bland (13). This estimate compares the pooled RRs across subgroups of patients with diabetes and without across trials. Pooled estimates that exclude a ratio of 1 provide evidence of an aspirin-diabetes interaction. Second, we estimated the ratio of RRs comparing the aspirin effect in patients with and without diabetes within trials and then pooled these ratios. Again, pooled estimates that exclude a ratio of RRs of 1 indicate a statistically significant difference in the treatment effect of aspirin in patients with and without diabetes. We used Comprehensive Meta-Analysis version 2 software (Biostat, Englewood, NJ) and StatsDirect 2.5.4 (StatsDirect, Cheshire, U.K.) to conduct these analyses. Finally, we conducted a Bayesian random-effects logistic regression with aspirin use and diabetes status as the main effect factors, and an interaction term of these two was also applied to the included trials. A 95% credible interval of the regression coefficient of the interaction term that excludes 0 indicates that the effect of aspirin is different in patients with and without diabetes. This analysis was carried out using WinBUGS 1.4 (Medical Research Council and Imperial College of Science, Technology and Medicine, U.K.) with noninformative prior distributions and three parallel Markov Chain Monte Carlo chains with overdispersed initials. The Brooks-Gelman-Rubin (14) convergence diagnostic was applied to examine the convergence of the Markov Chain Monte Carlo samplers. Statistical inferences were then drawn based on the resulting posterior distributions of parameters of interest. For RCTs with zero events in one study arm, we used a continuity correction factor of 0.5 and conducted sensitivity analysis using the treatment arm method to test whether the choice of statistical methods affected study conclusions (15).

Figure 1 describes the process of study selection that yielded eight eligible RCTs for meta-analysis. The study by Zabel-Langhennig et al. (16) did not report any events of interest and therefore was not included in the main analysis. Table 1 describes the included studies showing important differences in the studies. Authors from 50% of eligible RCTs responded to our data requests (9,17,–19).

Table 2 shows the methodological quality of the included RCTs. The κ statistic for quality assessment identified through the screening process was 0.70 (range 0.64–0.87). Overall, trials had high methodological quality, with more than half of them concealing allocation and blinding patients and caregivers, most reporting minimal loss to follow-up, and only one was stopped earlier than planned because of finding significant benefit during an interim analysis.

Figure 2 A–C demonstrates the pooled RRs and 95% CI for the effect of aspirin versus control on death, myocardial infarction, and ischemic stroke, respectively, stratified by diabetes status. The estimates of RR and 95% CI among all patients regardless of diabetes status for the outcomes of mortality, myocardial infarction, and ischemic stroke were 0.93 (95% CI 0.85–1.03; I² 0%), 0.79 (0.66–0.95; I² 63%), and 0.73 (0.43–1.22; I² 39%), respectively. Estimates among patients with diabetes were 0.97 (0.87–1.08; I² 0%), 0.86 (0.67–1.11; I² 53%), and 0.62 (0.31–1.24; I² 67%), respectively. The corresponding estimates among patients without diabetes were 0.87 (0.75–1.02; I² 0%), 0.72 (0.55–0.95; I² 70%), and 0.89 (0.41–1.94; I² 0%), respectively.

Table 3 describes the results of between-study subgroup analysis, within-study subgroup analyses, and Bayesian random-effects regression analyses. All three methods yielded nonsignificant ratios of RR of aspirin in patients with and without diabetes across all outcomes examined. The choice of continuity correction method for RCTs with zero events in one arm did not substantially affect inferences.

Our conclusions were not significantly altered by the exclusion of the APLASA (P_interaction 0.28, 0.36, and 0.63 for mortality, myocardial infarction, and ischemic stroke, respectively) or ETDRS trials (P_interaction 0.18, 0.62, and 0.48 for mortality, myocardial infarction, and ischemic stroke, respectively).

Consistent with prior studies (7,9,10), we found no significant benefit of aspirin therapy for the primary prevention of cardiovascular events among patients with diabetes. Even for the outcome of myocardial infarction, in which the overall risk reduction was statistically significant, our estimated risk reduction among patients with diabetes remained nonsignificant. However, our estimates were imprecise and unable to exclude a treatment benefit of 13, 33, and 69% for the outcomes of mortality, myocardial infarction, and ischemic stroke, respectively. Using a between-studies approach, a within-studies approach, and a Bayesian analysis, we found evidence that the effect of aspirin among patients with diabetes was not significantly different from patients without diabetes. Thus, the data we present here provide imprecise evidence of benefit from using aspirin in patients with diabetes without a history of cardiovascular events; this treatment effect does not differ significantly from that observed in patients without diabetes.

The main limitation of our study relates to our use of published aggregate data and lack of statistical power. The possibility of publication bias cannot be adequately assessed and corrected given the small number of RCTs (20). Our comprehensive literature search minimized the risk of publication and reporting biases, although our author response rate was relatively low and resulted in less complete data. It is possible that other aspirin RCTs have examined diabetes subgroups but have not reported these results because of lack of a significant aspirin-diabetes interaction. If this were to be the case, their inclusion would likely strengthen our conclusions. The small number of included trials and the small number of events within these trials reduced the statistical power and therefore the precision of the estimates in our main analysis and in our tests for interaction between aspirin effect and diabetes status. Our analyses using multiple statistical methods to examine the diabetes-aspirin interaction and their consistent results strengthen our conclusions. While the use of composite end points would have increased our statistical power, we chose to pool individual end points because the composite end points in the included trials represent outcomes with large gradients in importance to patients and in potential magnitude of the effect of treatment across component end points (21). We did not examine the potential harm from aspirin therapy such as gastrointestinal bleeding, since these events in the small population of patients with diabetes included in RCTs are usually rare, leading to imprecise estimates. Hence, evidence about harm is better obtained from larger studies that include patients without diabetes.

Our inferences should be considered in the context of proposed criteria to evaluate the validity of a subgroup effect (22). Good evidence of a valid subgroup effect results from testing few a priori hypotheses based on strong biological rationale. The extent to which this is true for the diabetes subgroup analyses reported in the literature is unclear to us. Our analyses, however, do not support other criteria for a valid subgroup effect. We found that differences in aspirin treatment effect between patients with and without diabetes, both between and within RCTs, were inconsistent and of small magnitude. Thus, the current best evidence suggests that the efficacy of aspirin for the primary prevention of cardiovascular events in patients with diabetes is similar to that in patients without diabetes.

The decision to use aspirin for primary prevention in individual patients with diabetes remains a complex issue. Ultimately, the decision to use aspirin for an individual patient should consider the advantages and disadvantages of this preventive treatment given the patient's context and preferences. We suggest first estimating the 10-year risk of coronary heart disease (23,24). Multiplying this estimate by 0.88 (12% relative risk reduction with the use of aspirin based on data from the Antithrombotic Trialists' Collaborative [7]) would provide the revised 10-year risk estimate for a patient using aspirin. This should be presented alongside the risk of major gastrointestinal and extracranial bleeds with and without aspirin (7,25). A patient decision aid could present these data to help patients and clinicians consider the relative merits and downsides of aspirin use, as was done recently for use of statin therapy in patients with diabetes (26).

While there are insufficient data among patients with diabetes to conclusively show a benefit of aspirin therapy for the primary prevention of cardiovascular events, our data suggest, but do not confirm, that the relative benefit of aspirin is similar in patients with and without diabetes. Additional evidence from RCTs and individual-patient-data meta-analyses may help to further clarify this issue.

A.D.C. was supported by the Mayo Clinic Clinician–Investigator Training Program. M.M.F.-B. received grant support from the Instituto de Salud Carlos III, Ministerio de Sanidad y Consumo (BA08/90035), Government of Spain.

No potential conflicts of interest relevant to this article were reported.

We wish to sincerely thank all of the study authors who promptly and graciously responded to our requests for information.