Aspirin has potential benefits for primary prevention of cardiovascular outcomes in diabetes: updated literature-based and individual participant data meta-analyses of randomized controlled trials

We conducted this literature-based review and IPD using an updated protocol registered in the PROSPERO International prospective register of systematic reviews (CRD42019122326) and in accordance with guidelines of PRISMA (Additional file 1: Appendix S1) [14], PRISMA-IPD (Additional file 1: Appendix S2) [15], methods recommended by the IPD Meta-analysis Methods Group of the Cochrane Collaboration [16], and guidance of Riley and colleagues [17]. We searched MEDLINE, Embase, and The Cochrane Library for randomised controlled trials (RCTs) published from November, 2015 (date of the last search of our previous review [9]) to 29 January 2019. The computer-based searches combined free text and medical subject headings and combination of key words related to aspirin and diabetes, with no restrictions placed on language. The full details of the search strategy are provided in Additional file 1: Appendix S3. The titles and abstracts of all articles identified by the literature search were initially screened to assess their suitability for inclusion, after which we acquired potentially relevant articles for detailed full text evaluation. Two reviewers (SKK and SS) independently conducted full text evaluation using the inclusion criteria and any disagreements regarding eligibility of an article was discussed. We also searched reference lists of selected studies and relevant reviews for additional publications. No separate ethical approval was required for the conduct of this study, as any necessary ethical approval was obtained for each of the individual studies contributing data to the meta-analysis.

Intervention studies were eligible if they met the following inclusion criteria: (i) were randomised controlled, open or blinded trials; (ii) assessed the effects of aspirin therapy compared to a placebo or no treatment; (iii) enrolled adults with diabetes mellitus (either exclusively or as a subgroup) without previous history or clinical evidence of CVD; (iv) reported data on cardiovascular endpoints, all-cause mortality, or other adverse events such as bleeding, cancer etc.; and (v) and had a follow-up duration of at least a year. Studies that were non-randomised comparing aspirin with another antiplatelet agent, included people with known CVD, or were secondary publications of trials already included in the analysis were excluded from the review.

One author (SKK) initially conducted the data extraction using a standardized data collection form and a second author (SS) independently checked the extracted data with that in the original articles. Data were extracted on the following characteristics: study design; patient characteristics, intervention and comparison, and outcomes. The risk of bias assessment was conducted using the Cochrane Collaboration’s risk of bias tool [18].

The primary outcomes were MACE) [defined as composite of nonfatal myocardial infarction (MI), nonfatal stroke, and cardiovascular death] and all-cause mortality. Secondary outcomes included (i) other cardiovascular endpoints (nonfatal MI, coronary heart disease death, fatal and nonfatal stroke, revascularization, sudden coronary death, and transient ischaemic attack) and adverse events (any bleeding, gastro-intestinal bleeding, cancer, allergic reactions, and arrhythmias). Cardiovascular outcomes used in trials and their ascertainment are reported in Additional file 1: Appendix S4.

Investigators of eligible trials identified by the literature search strategy and well-known investigators in the field, were contacted by email, letter or phone, provided with a summary of the study protocol, and invited to join the collaboration if they had the relevant data available. Investigators expressing interest to collaborate in this effort were then provided with full details of the study protocol and a list of relevant study variables that could be used in the analyses. Data from each study were obtained using a standardised spreadsheet or in STATA format. The raw data were examined and inconsistencies or irregularities were clarified with the investigators. Individual level data collected was coded and entered into a single database.

For the literature-based meta-analysis, summary measures were presented as relative risks (RRs) with 95% confidence intervals (CIs). Following Cornfield’s rare disease assumption [19], reported hazard ratios (HRs) and odds ratios (ORs) were assumed to approximate the same measure of RRs. We used reported RRs or calculated study specific unadjusted RRs based on raw events. To minimize the effect of heterogeneity, random-effects models was used to pool RRs. Heterogeneity across studies was assessed using the Cochrane χ² statistic and the I² statistic [20]. Study level characteristics including geographical location, sex differences, allocation concealment, baseline CVD risk, dose of aspirin, compliance, duration of treatment, and number of outcomes were prespecified as characteristics for assessment of heterogeneity, which were evaluated using stratified analysis and random effects meta-regression. For comparisons involving 10 or more studies, publication bias was assessed by visually inspecting a funnel plot and applying Egger’s regression symmetry test [21].

In the IPD meta-analysis, since not all trials provided time to event data, effects of the intervention on outcomes were expressed as ORs using logistic regression analysis in the main analysis. Cox proportional hazards models were used for time to event data in supplementary analyses. We employed a one-step IPD meta-analyses in which IPD from all studies were modelled simultaneously with fixed effects, adjusting for age and trial arm to obtain the pooled intervention effect with a 95% CI. To contextualise our results, we also calculated the number needed to treat (NNT) using the formula: NNT = 1/absolute risk reduction (ARR). The ARR was derived by multiplying the control risk by the relative risk reduction. All statistical analyses were performed with Stata release 15 (StataCorp, College Station, Texas, USA).

Relative risks for cardiovascular outcomes and all-cause mortality events comparing aspirin therapy with placebo or no treatment in pooled analyses are reported in Fig. 2 and Appendices S6–S12. In 10 trials of MACE (34,058 participants and 3104 events), aspirin therapy significantly reduced the risk of MACE compared with placebo or no treatment 0.89 (95% CI 0.83 to 0.95). Using a fixed effects model, the pooled RR remained unchanged (Additional file 1: Appendix S6). There was no evidence of heterogeneity between the contributing studies (I²= 0%, 0 to 62%; p > 0.99). On exclusion of the ETDRS trial which involved a small proportion of patients with previous CVD, the pooled RR remained the same 0.89 (95% CI 0.82 to 0.96) (Additional file 1: Appendix S7).

Aspirin therapy was not associated with a significant decrease in risk of all-cause mortality (eight trials; 27,102 participants and 2776 events) 0.95 (95% CI 0.88 to 1.02) and there was no evidence of heterogeneity between contributing studies (I²= 0%, 0 to 68%; p = 0.88) (Fig. 2; Additional file 1: Appendix S8).

In seven trials of MI and five trials of CHD, aspirin therapy was not associated with a significant reduction in risk of any of these outcomes: 0.84 (95% CI 0.64 to 1.11)and 0.98 (95% CI 0.79 to 1.21) respectively (Fig. 2; Additional file 1: Appendices S9, S10). There was evidence of moderate heterogeneity (I²= 57%, 1 to 82%; p = 0.03) for the MI analysis and no evidence of heterogeneity (I²= 0%, 0 to 79%; p = 0.75) for the CHD analysis.

There was no significant reduction in the risk of stroke or CVD death when aspirin therapy was compared with placebo or no treatment: 0.88 (95% CI 0.72 to 1.08) and 0.92 (95% CI 0.78 to 1.08;) respectively (Fig. 2; Additional file 1: Appendices S11, S12). There was no significant evidence of heterogeneity between the contributing studies for stroke or CVD death: (I²= 10%, 0 to 68%; p = 0.35) and (I²= 23%, 0 to 67%; p = 0.26) respectively.

There was no significant difference in the risk of other cardiovascular outcomes such as nonfatal MI, CHD death, fatal stroke, nonfatal stroke, ischaemic stroke, haemorrhagic stroke, CVD, revascularization, angina pectoris, sudden coronary death, and TIA, when aspirin was compared with placebo or no treatment (Fig. 3; Additional file 1: Appendix S13).

There was no evidence of effect modification by any of the clinically relevant study-level characteristics explored for the outcomes of MACE, all-cause mortality, and CVD death (Additional file 1: Appendices S14–S18). For MI, the moderate heterogeneity between studies contributing to pooled analysis seemed to be partly explained by treatment duration (p value for meta-regression = 0.01). Compared to participants with treatment duration more than 5 years, participants with treatment duration of 5 years or less had a significantly reduced risk of MI with aspirin use 0.70 (95% CI 0.53 to 0.93) (Additional file 1: Appendix S16). There was also evidence of effect modification by aspirin dosage (p value for meta-regression = 0.02) and treatment duration (p value for meta-regression = 0.02) for the effect of aspirin use on stroke. The risk of stroke was significantly reduced for trials using aspirin dosage of 100 mg per day or less compared to more than 100 mg per day. Similarly, compared to participants with treatment duration of 5 years or less, participants with treatment duration of more than 5 years had a significantly reduced risk of stroke with aspirin (Additional file 1: Appendix S17).

Figure 4 presents pooled RRs of the effects of aspirin therapy compared with placebo or no treatment on adverse outcomes such as major bleeding, gastrointestinal bleeding, non-gastrointestinal bleeding, cancer, venous thromboembolism, gastrointestinal symptoms, arrhythmias, and allergy. No statistically significant difference in risk was found for any of these outcomes.

The ARR of MACEin people with diabetes associated with aspirin use was 1.06% (95% CI 0.48 to 1.63)over the average follow-up period of 5 years, which translates into a NNT of 95 (95% CI 61 to 208) to prevent one MACE. For MI in participants with a treatment duration of 5 years or less, the ARR and NNT were 2.26 (95% CI 0.53 to 3.55) and 44 (95% CI 28 to 189) respectively. For the risk of stroke with aspirin dosage of 100 mg per day or less, the ARR and NNT were 0.99 (95% CI 0.20 to 1.62) and 101 (95% CI 62 to 507) respectively. For the risk of stroke with treatment duration of more than 5 years, the ARR and NNT were 2.41 (95% CI 0.62 to 3.64) and 42 (95% CI 27 to 172) respectively.

A funnel plot for the comparison that involved 10 studies (MACE) was symmetrical (Additional file 1: Appendix S19) which was consistent with Egger’s regression symmetry test (p = 0.99). In addition, there was no definitive evidence of selective reporting when studies were grouped by size in meta-regression analysis (Additional file 1: Appendix S14).

Investigators of 10 trials were contacted to contribute data to the APPRAISE consortium. Of these 10, we obtained IPD data from three RCTs namely: Physicians’ Health Study (PHS); Primary Prevention Project (PPP); and the Women’s Health Study (WHS). Details of contributing trials are presented in Additional file 1: Appendix S20. The three RCTs comprised of 2306 participants of which 1188 were randomised to the aspirin group and 1118 to the placebo group (Table 2). Overall the mean (standard deviation, SD) age was 60.1 (8.7) years and 38.5% were male. The median (interquartile range) duration of follow-up was 9.5 (5.2–10.4) was years (based on data from two trials). The trial groups had similar cardiovascular risk profiles. In pooled analysis of the three trials, there was no significant evidence of an effect of aspirin on MACE, all-cause mortality, MI, stroke, or CVD death (Fig. 5). In pooled analysis of two trials (PHS and WHS) with time to event data, there was no significant difference in risk for any of the outcomes assessed (Additional file 1: Appendix S21). There was significant evidence of a differential effect of aspirin on the risk of MACE by smoking status ((p value for meta-regression = 0.01) (Fig. 6). Aspirin use reduced the risk of MACE in non-smokers 0.70 (95% CI 0.51 to 0.96) compared to those who smoked 1.77 (95% CI 0.91 to 3.45). The ARR of MACE in non-smokers associated with aspirin use was 3.04% (95% CI 0.41 to 4.97) over the average follow-up period of 10 years, which translates into a NNT of 33 (95% CI 20 to 246) to prevent one MACE.

Using a literature-based meta-analysis and pooled analysis of individual level data from the PHS, PPP, and WHS trials, we have evaluated the efficacy and safety of aspirin for the primary prevention of cardiovascular outcomes and all-cause mortality events among people with diabetes. Compared with no aspirin, aspirin use was associated with an 11% reduction in the risk of MACE (which was a combination of cardiovascular mortality, nonfatal MI, and nonfatal stroke outcomes). Aspirin use however no effect on other cardiovascular endpoints and all-cause mortality. Heterogeneity was generally low or absent between contributing studies for majority of outcomes, except for the MI comparison which was characterised by moderate heterogeneity. Except for MI and stroke, there were no differential effects of aspirin on other outcomes by the pre-specified clinically relevant characteristics including baseline cardiovascular risk. Aspirin use was associated with a 30% reduction in the risk of MI for a treatment duration of 5 years or less, but no benefit for a treatment duration of more than 5 years. In addition, aspirin use reduced the risk of stroke for a lower intervention dose (≤ 100 mg/day) and longer treatment duration (> 5 years). In the context of multiple statistical tests for interaction conducted, the results of the subgroup analyses should be interpreted with caution and may require replication in further studies. There were suggestions that aspirin use might be associated with adverse effects such as major bleeding, gastrointestinal symptoms, and arrhythmias, but the estimates were imprecise and not significant. The effects of aspirin on cancer and venous thromboembolism outcomes were also not significant. Finally, pooled analysis of individual level data from three trials showed no significant evidence of an effect of aspirin on any of the outcomes evaluated. However, aspirin use was associated with a 30% reduction in the risk of MACE in non-smokers; with a suggestion of an increased risk of MACE in smokers, though the risk estimate in this group was based on only 46 events and was imprecise and should be interpreted with caution. These findings may reflect evidence that cigarette smoking causes an increase in platelet aggregability [32, 33], hence aspirin may not effectively inhibit platelet activity in smokers. It has been reported that smokers may be about 12 times resistant to the effects of aspirin. An established body of literature suggests that the reduced antiplatelet effect of aspirin is associated with a fourfold increase in the risk of adverse cardiovascular events [34, 35].

The current study builds on previous meta-analyses including one of ours [4‐7, 9]. Since the publication of the ASCEND-A Study of Cardiovascular Events in Diabetes and ASPREE-Aspirin in Reducing Events in the Elderly trials in 2018, a number of reviews have recently been published. In one review which was based on the overall population and not specifically in people with diabetes [36], Zheng and Roddick in a subsidiary analysis reported data for participants with diabetes based on 10 studies. In their report, aspirin therapy was associated with a reduction in the primary composite cardiovascular outcome (MACE) based on pooled analysis of seven trials, but not with any of the secondary cardiovascular outcomes. In addition, they reported increases in major and gastrointestinal intestinal bleeding. In a research letter, Fortuni and colleagues included only five trials in their pooled analysis and reported similar findings to that of Zheng and Roddick [37]. Our current study which is based on pooled analysis of 12 trials is therefore an updated assessment of the topic and also includes a pooled analysis of IPD contributed by three trials. Consistent with the recent and other previously published reviews on the topic [4‐7, 9, 36], aspirin use reduced MACE with no differential effects of aspirin use on other cardiovascular endpoints as well as all-cause mortality. New findings by our current study include important differences by aspirin dosage and treatment duration for the effect of aspirin use on the risk of MI and stroke and no effects of aspirin use on major bleeding and other adverse effects. Based on IPD contributed by three trials, we had the opportunity to re-evaluate the efficacy of aspirin for the specified outcomes in clinically relevant subgroups with suggestions of an important difference by smoking status for the effect of aspirin on the risk of MACE.

Compared with aspirin use for the secondary prevention of vascular disease, the use of aspirin for primary prevention has been a widely debated and controversial topic. Two recently published landmark trials, ASCEND, and ASPREE, [10, 11] evaluated the efficacy and safety of aspirin for the primary prevention of cardiovascular outcomes in people with diabetes and have been included in this review. Whereas the absolute benefits of aspirin on vascular benefits were counterbalanced by the bleeding hazard in the ASCEND study, there was no differential effect of aspirin on the risk of CVD and bleeding in people with diabetes in the ASPREE trial. Though our results show aspirin may have a beneficial effect in the primary prevention of cardiovascular disease in people with diabetes, aspirin had no differential effect on bleeding risk. Given the imprecise estimates reported for GI and non-GI bleeding, these results may be due to inadequate power of these trials to detect these events. Indeed, a wealth of data from general and secondary prevention populations suggests that the main adverse effect associated with aspirin use is GI bleeding [38]. Both low aspirin dosage and long term therapy is associated with an absolute excess of GI bleeding complications in these population groups [39]. A higher risk of bleeding events has also been reported among the elderly and people at low cardiovascular risk [40]. Though the current data suggests otherwise, real-world data in general populations have demonstrated higher rates of bleeding in people with diabetes on aspirin therapy [41]. Taking the overall findings together, two questions still remain on the role of aspirin therapy in primary prevention of CVD in diabetes: (i) are the absolute vascular benefits of aspirin counterbalanced by the potential for bleeding and (ii) in what groups of the population do the benefits of aspirin outweigh its bleeding hazards. Based on a recent comprehensive narrative review by Lippi and colleagues, the authors suggest that the harms of aspirin in primary prevention of CVD may be larger than the benefits, especially in the elderly general population [42]. Our study findings suggest there may be important differences in the effect of aspirin by treatment dosage, treatment duration, as well as smoking status, but these results are based on limited data from subgroup analyses. Though our findings showed no suggestions of differences in the effect of aspirin by gender on all outcomes evaluated, there is evidence suggesting that the variation in the effect of aspirin therapy on cardiovascular outcomes could be explained by gender. In a meta-analysis of 23 trials which aimed to evaluate whether gender might play a role in explaining the large variation of aspirin efficacy across primary and secondary MI prevention trials, the authors demonstrated that gender accounted for a substantial proportion of the variability in the efficacy of aspirin in reducing MI rates across these trials [43]. Emerging evidence points to the fact that women have an increased risk of aspirin resistance compared to men, which makes aspirin less effective in women [44]. Apart from the condition diabetes which is associated with reduced rates of responsiveness to aspirin [45], other factors that could potentially reduce the antiplatelet effect of aspirin include older age, obesity, renal insufficiency, and platelet count. To answer all these pertinent questions may require an IPD meta-analysis based on all contributing trials. An updated meta-analysis by the Antithrombotic Trialists’ Collaboration could help address some of these questions. There have been contradictory guideline recommendations on the role of aspirin in primary CVD prevention; however, the ADA recommends the use of low-dose aspirin for the primary prevention of CVD in adults with type 1 and 2 diabetes who are at increased CVD risk, whereas not recommended for people at low CVD risk as the potential for bleeds likely offsets potential benefits in these people [13]. Though the current findings do not justify these recommendations, it appears the clinical decision to initiate low dose aspirin for primary prevention is a complex process and should be individualised and tailored to each patient’s baseline CVD and bleeding risk, as these tend to differ from one individual to another. Patients preferences should also be taken into account when making the decision. Leggio and colleagues in their review recommend that when a decision is taken to initiate primary CVD prevention, uncoated aspirin formulations with higher bioavailability should be prescribed at the lowest effective lower dose and concurrently used with proton-pump inhibitors for those at high risk of gastrointestinal bleeding [46]. In addition, concurrent use of nonsteroidal anti-inflammatory drugs should be avoided and those with rapid platelet turnover should be considered for twice-daily dosing regimen [46].

Several strengths of this review deserve consideration. Compared to previous reviews including the most recent one [36], it is the largest and most comprehensive review on the topic to date. In total we included 12 trials with additional data contributed by three trials that shared their individual level data, hence we had the ability to examine the efficacy and hazards of aspirin therapy for a wider range of cardiovascular end points as well as adverse events. We systematically explored for possible sources of heterogeneity and tested for evidence of effect modification using several prespecified study level characteristics. This study also had limitations, several of which were inherent to the meta-analysis. Definitions of some of the clinical outcomes were not consistent across all studies, which could potentially have led to biased estimates. The definition of MACE used in several of the eligible studies included total stroke, which did not differentiate between ischaemic and haemorrhagic stroke events. Hence, this could have resulted in an underestimation of the potential protective effects of aspirin. There appeared to be selective reporting by some studies, as data on some cardiovascular endpoints and adverse events were not reported by some of the included studies. For example, MACE and all-cause mortality outcomes were reported in 10 and 8 trials respectively. We were unable to perform detailed subgroup analyses by clinically relevant subgroups such as appropriate baseline CVD risk groups, appropriate treatment dosages, type of diabetes, and duration of diabetes because of the limited data and inconsistent way of reporting. For example, the dosages and timing of aspirin varied considerably across the studies. These findings should therefore be interpreted with caution given the limitations.