Aspirin for Primary Prevention of Cardiovascular Events: Meta-Analysis of Randomized Controlled Trials and Subgroup Analysis by Sex and Diabetes Status

Objective To evaluate the benefits and harms of aspirin for the primary prevention of CVD and determine whether the effects vary by sex and diabetes status. Methods We searched Medline, Embase, and Cochrane databases for randomized controlled trials comparing the effects of aspirin with placebo or control in people with no pre-existing CVD. Two investigators independently extracted data and assessed the study quality. Analyses were performed using Stata version 12. Results Fourteen trials (107,686 participants) were eligible. Aspirin was associated with reductions in major cardiovascular events (risk ratio, 0.90; 95% confidence interval, 0.85–0.95), myocardial infarction (0.86; 0.75–0.93), ischemic stroke (0.86; 0.75–0.98) and all-cause mortality (0.94; 0.89–0.99). There were also increases in hemorrhagic stroke (1.34; 1.01–1.79) and major bleeding (1.55; 1.35–1.78) with aspirin. The number needed to treat to prevent 1 major cardiovascular event over a mean follow-up of 6.8 years was 284. By comparison, the numbers needed to harm to cause 1 major bleeding is 299. In subgroup analyses, pooled results demonstrated a reduction in myocardial infarction among men (0.71; 0.59–0.85) and ischemic stroke among women (0.77; 0.63–0.93). Aspirin use was associated with a reduction (0.65; 0.51–0.82) in myocardial infarction among diabetic men. In meta-regression analyses, the results suggested that aspirin therapy might be associated with a decrease in stroke among diabetic women and a decrease in MI among diabetic men and risk reductions achieved with low doses (75 mg/day) were as large as those obtained with higher doses (650 mg/day). Conclusions The use of low-dose aspirin was beneficial for primary prevention of CVD and the decision regarding an aspirin regimen should be made on an individual patient basis. The effects of aspirin therapy varied by sex and diabetes status. A clear benefit of aspirin in the primary prevention of CVD in people with diabetes needs more trials.


Introduction
The burden of cardiovascular disease (CVD) is substantial. The most recent (2013) statistics on heart disease and stroke from the American Heart Association (AHA) estimate that the annual direct and indirect cost of CVD and stroke in the United States alone are $523 billion [1]. From 2000 to 2010, the total number of inpatient cardiovascular operations and procedures increased 28%, from 5,939,000 to 7,588,000. By 2030, 40.8% of the US population is projected to have some form of CVD, and the annual cost will increase to $1.13 trillion [1]. These strong upward trends underline the importance of primary prevention for those who are already at high risk of CVD.
The use of low-dose aspirin for primary prevention of CVD is recommended by many key guidelines [2,3]. However, a recent published study and a review [4,5] stated that the benefit of aspirin for the primary prevention of cardiovascular events was relatively small for individuals regardless of diabetic status and could easily be offset by the risk of hemorrhage. These studies challenge current recommendations, which are based on outcomes from several meta-analyses [6][7][8][9][10], prompting re-evaluation of the efficacy of aspirin. An important sex-specific meta-analysis showed that the effects of aspirin varied by sex [6]. However, it was conducted in 2006 and included only six primary prevention trials. In addition, the results were not confirmed in the Antithrombotic Trialists' Collaboration meta-analysis and a recent publication which did not find significant sex different in treatment effect [7,11]. Several guidelines [12,13] recommend aspirin for the primary prevention of cardiovascular events in patients with diabetes at risk of CVD, but others [2,14] do not. This conflict reflects the lack of definitive evidence. Existing recommendations are primarily based on extrapolations from indirect evidence, given the absence of statistically significant results in published meta-analyses in diabetics [15][16][17][18][19][20][21][22][23].
Therefore, we performed a new meta-analysis to re-assess the effects of aspirin for primary prevention of CVD and to investigate whether the effects vary by sex and diabetes status. Compared to the previous sex-specific meta-analyses, we enrolled almost twice that of previously published data. Given the limited power to detect interactions, even in a meta-analysis combining the results from several studies [24], we used multiple statistical methods to examine the diabetes-aspirin interaction and sex-aspirin interaction and their consistent results strengthen our conclusions.

Methods
For this meta-analysis, we used methods and definitions from previous meta-analyses [6] and performed our meta-analysis in line with approach recommended by the PRISMA statement [25]. Full study protocol is provided as Text S1.

Data Sources and Searches
Randomized controlled trials (RCTs) comparing the effect of aspirin with placebo or control in people without pre-existing CVD on outcomes of interest were eligible for inclusion. We identified trials by searching Medline, Embase, and Central (the Cochrane Central Register of Controlled Trials) from inception to December 2012. Reference lists from previously published relevant systematic reviews were also screened for additional studies [6,8,15]. The search strategies are as follows: First we searched terms ''aspirin*'' [MeSH] and term ''primary prevention''. Then the Boolean term ''AND'' was used to combine these two terms. Highly sensitive filters were used to limit results to randomized controlled trials and human studies. We searched only studies published in English. A similar search strategy was used for Embase and Central.

Study Selection
Two authors independently reviewed search results by title and abstract, then full text to identify eligible trials. Selection criteria included: (1) Prospective, randomized, controlled, open, or blinded trials. (2) Participants without clinical CVD (e.g., established or symptomatic) were randomly assigned to aspirin (any dose) versus placebo or control group for the primary prevention of CVD. (3) Trials carried out on a background of anticoagulation were eligible. (4) Follow-up had to exceed 90 days, because such short follow-up would not permit detection of cardiovascular outcomes related to aspirin treatment for primary prevention.

Outcomes
The outcomes of interest for both aspirin and control groups included major cardiovascular events (MCE, defined as death from cardiovascular causes, nonfatal myocardial infarction, and nonfatal stroke); myocardial infarction (MI, fatal and nonfatal); stroke (fatal and nonfatal; ischemic and hemorrhagic); ischemic stroke; cardiovascular mortality; total mortality (death from any cause); hemorrhagic stroke and major bleeding. Definitions for major bleeding varied across studies. However, participant-level data was unavailable to allow reclassification according to standard criteria [26][27][28]. Among all bleeding events, the gastrointestinal hemorrhage is one of the most common and serious complications of long-term aspirin use.

Data Extraction and Quality Assessment
Two investigators independently extracted data and evaluated the methodological quality using criteria previously published [29]. An arbitrator settled discrepancies by discussion in accordance with our selection criteria. We collected some basic information on the studies and outcomes of interest listed above. Data were collected from the original articles, previously published meta-analyses, and through contact with study authors.

Data Synthesis and Analysis
Analyses were performed using Stata version 12 (Stata Corp). Heterogeneity was assessed by Cochran's Q-test and the I 2 statistic [30]. A P value less than 0.10 indicated significant heterogeneity. For the I 2 metric, we defined low, moderate, and high I 2 values as 25%, 50%, and 75%, respectively [31]. We estimated the results with pooled relative risks (RR) and 95% confidence intervals (CI) using a Mantel-Haenszel fixed-effect model when the heterogeneity was negligible or moderate and a DerSimonian and Laird random-effects model when heterogeneity was significant [32]. All analyses were based on the intension-to-treat principle. A 2-tailed P-value 0.05 was considered statistically significant.
To explore potential sources of heterogeneity, we collected sufficient information to conduct particular subgroup analyses to determine the sex-aspirin interaction and diabetes-aspirin interaction. Because there is limited power to detect interactions, even in a meta-analysis combining the results from several studies, and it is not sufficient to conclude that the relative risks from the subgroups significantly different from each other when the two estimates and P values seem very different [24]. Thus, we implemented two methods to determine whether a difference exists in subgroup analysis. First, we estimated the pooled ratio of RRs comparing the aspirin effect in patients with and without diabetes and in patients with different genders across trials. Second, we used the method of Altman and Bland to compare the pooled RR and its 95% CI across subgroups [24]. In addition, we calculated numbers needed to treat (NNT) and numbers needed to harm (NNH) to examine the risk vs. benefit of aspirin therapy for some endpoints [33]. Values of NNT and NNH provided herein represent the number of persons that need to be treated with aspirin for 6.8 years (the overall mean follow-up time in our study) to avert or incur, respectively, 1 event.
We also performed meta-regression analyses to appraise the impact of gender and the daily dose of aspirin on outcomes [34]. Publication bias was assessed by the funnel plot and the Begg's and Egger's tests. We performed a sensitivity analysis to examine the robustness of the results, systematically removing one study from the analysis and recalculating the results.

Risk of Bias in Individual Trials
The risk of bias in trials is presented in Table S2. Randomization was stated in all studies, but the allocation concealment was adequately described in only eight studies and unclear in the remainder. Two trials were open-labeled [41,47], and placebo tables were not used in the control group in one trial [35]. Outcome assessment was not blinded in one trial [35] and unclear in two [42,45]. The description of incomplete outcome data was not adequate in two trials [36,43]. Three studies had a vitamin component [41,43,44], one had a beta carotene component [36], one had a anti-oxidant component [46], and one had a warfarin component [39].

Subgroup Analysis
The effects of aspirin by gender. Details of the included studies in the subgroup analyses by sex appear in Table S3.
When we used the method of Altman and Bland to compare the pooled RR and its 95% CI of MI (P = 0.02) and stroke (P = 0.01), the results also provide strong support for gender difference in the reduction of MI and stroke.
The effects of aspirin by diabetes status. Details of the included studies in the subgroup analyses by diabetes status appear in Table S4-A and Table S4    Among diabetic patients, we also conducted stratified analysis by sex. Pooled results demonstrated a 35% reduction in MI among men (RR, 0.65; 95% CI, 0.51 to 0.82; P,0.01; Q = 3.21, P = 0.20; I 2 = 37.6%), but the results were not significant in women (RR, 0.90; 95% CI, 0.71 to 1.14; P = 0.37; Q = 4.01, P = 0.14; I 2 = 50.1%). When we used the method of Altman and Bland to compare the pooled RR and its 95% CI (P = 0.06), the results demonstrated that difference in the reduction of MI was not significant. The data was insufficient to estimate other endpoints in diabetic patients.

Meta-regression
First we performed meta-regression in all populations to appraise the impact of the percentage of males on the incidence of endpoints. There was a statistically significant relationship between percentage of males and the effect of aspirin on stroke (P = 0.04) ( Figure 3A), which supported the conclusions from subgroup analysis. With respect to the negative results among patients with diabetes, we then performed meta-regression in diabetic patients. The result demonstrated a significant association between percentage of males and the effects of aspirin on MI (P = 0.03) ( Figure 3B) and stroke (P = 0.02) ( Figure 3C), suggesting that aspirin therapy might be associated with a decrease in stroke among diabetic women and a decrease in MI among diabetic men.
We also performed meta-regression to test for a linear relation of the positive effects (reduction in MCE, MI, ischemic stroke) or complications (major bleeding and hemorrhagic stroke) with daily dose of aspirin. The results suggested that risk reductions achieved with low doses (75 mg/day) were as large as those obtained with higher doses (650 mg/day), and the risk for bleeding did not increase with dose ( Figure S1).

Sensitivity Analysis
The heterogeneity was significant for the outcome of MI (Q = 28.17, P = 0.01; I 2 = 53.9%), which may be explained by the fact that WHS enrolled only women (weight 13.61%) and the PHS enrolled only men (weight 16.86%). The RR was 0.84 and 0.91 when these two studies were respectively removed from the model ( Figure S2). After excluding these two studies from our analyses, heterogeneity between trials substantially decreased (921/23704 vs 956/22035; RR, 0.88; 95%CI, 0.81 to 0.96; P,0.01, Q = 11.14, P = 0.43; I 2 = 1.3%).

Publication Bias
We used a comprehensive literature search strategy to minimize the risk of publication bias. Results of Begg's and Egger's tests for asymmetry were not statistically significant ( Figure S3).

Discussion
In a comparison of aspirin with placebo or control for the prevention of CVD, we found significant benefits of 10, 14, and 14% risk reduction for the outcomes of MCEs, MI, and ischemic stroke respectively in the overall population. Meanwhile, there were also clear harms of 34% relative increase in hemorrhagic stroke and 55% relative increase in major bleeding events. Our subgroup and meta-regression analyses indicated that the effects of aspirin therapy varied by sex and diabetes status. Aspirin use was associated with a significant reduction in the risk of cardiovascular events in both sexes but different specific types of benefits: a reduction in MI among men and a reduction in ischemic stroke among women. Aspirin had no significant effect on CVD in the overall diabetic population, but was associated with a reduction in MI among men with diabetes.
Although the results indicate a significant increase in bleeding complications, it is not sensible to conclude that the benefit of aspirin is offset by the risk of bleeding. First, we should estimate not only the incidence of benefits and harms, but also take into account the consequences of both harms and benefits on quality of life [49]. Setting aside the potentially fatal MI or stroke, it is clear that a non-fatal stroke or MI is more likely to result in long-term disability than a non-fatal gastrointestinal or other extracranial bleed. Although serious intracranial and extracranial bleed may also cause serious results, our results suggest that the benefit of reducing risk of ischemic stroke outweigh the harm hemorrhagic stroke. In addition, there are several methods to mitigate these adverse effects, for example, clinicians can remind those patients who decide to begin or continue an aspirin regiment for primary prevention of CVD, of the early recognition of the signs and symptoms as well as the risk factors of gastrointestinal bleeding. These risk factors include age, gender, upper gastrointestinal tract pain, gastrointestinal ulcers, NSAID use, uncontrolled hypertension, concomitant use of anticoagulants, and family history of gastrointestinal ulcers and so on [50]. Second, some of the previous published trials were criticized that few subjects exceeded the threshold for aspirin prophylaxis recommended by the American Heart Association [51]. In addition, evidence shows that.60% of aspirin users were above 60 years of age, 4-6% had a recent history of peptic ulcers, and over 13% used other non-steroidal antiinflammatory drugs [52]. It is obvious that the gastrointestinal harms would outweigh the cardiovascular benefits in certain groups whose gastrointestinal risk is high but cardiovascular risk is low. Thus, some of the previous published trials may overestimate the harm effect of the aspirin.
Third, the Antithrombotic Trialists' Collaborative, an individual-level meta-analysis of RCTs, indicated that the absolute benefits of aspirin were on a small order of magnitude in primary prevention and the effects of aspirin do not significantly depend on smoking history, blood pressure, total cholesterol, body-mass index, history of diabetes, or predicted risk of coronary heart disease [7]. However, the small number and rare events in these particular subgroups are not sufficient for precise estimate, and thus this paper provides insufficient evidence to answer the question of which particular category of individuals derive the most benefit from aspirin therapy. More highly powered analyses for specific populations are expected based on two major ongoing trials: the Aspirin to Reduce Risk of Initial Vascular Events (ARRIVE) Study (http://www.arrive-study.com/EN/study.cfm) and Aspirin in Reducing Events in the Elderly (ASPREE) [53].
The results of our subgroup analysis consistent with prior studies indicate that there is no significant benefit of aspirin therapy among patients with diabetes, but this may be due to inadequate power because the point estimate was similar to that among nondiabetics but with a wider confidence interval. It is well established that diabetes mellitus is associated with an increased risk of CVD [54]. Among diabetes patients, the coagulation system is altered, because plasma levels of procoagulant factors are increased while fibrinolytic capacity is decreased [55].
The mechanisms of the antithrombotic effects of low-does aspirin involve two aspects: cyclooxygenase (COX)-dependent actions and COX-independent actions [56]. Low-dose aspirin is considered to induce a permanent inactivation of COX-1 which results in the inhibition of platelet aggregation [57,58]. In many people, generation of new platelets and recovery of COXdependent platelet aggregation can reverse to a certain degree this effect within 24 hours after administration of aspirin [59]. Thus successive and low-dose daily administration of aspirin is essential to maintain inactivation of platelet COX-1. However, patients with type 2 diabetes have been demonstrated to be characterized by a large inter-individual variability in the recovery of COX-1 activity and enhanced platelet turnover rate which represents an important determinant of the extent and duration of platelet inhibition on repeated dosing with low-dose aspirin [60][61][62][63]. Thus it is possible that the current use of a once-a-day and low-dose regimen may not be sufficient to induce clinical benefits among diabetic patients [64][65][66][67]. More studies are needed to demonstrate whether a higher frequency of aspirin administration and possibly a higher daily dosage can optimize treatment with aspirin in diabetic patients. In addition, considerable efforts are needed to illuminate the relation between decreased responsiveness to aspirin and the COX-independent antithrombotic effects [56].
Platelet dysfunction, increased platelet aggregation and aspirin insensitivity were more common in patients with type 2 diabetes compared to nondiabetic people [68] [62]. In addition, insulin resistance and hyperglycaemia are reported to contribute to these alterations [55]. Among our eligible 14 studies, 6 were published before 2000. The management and treatments of diabetes have been improved over the decades. Thus, the treatments of diabetes are much different between the old studies and recent studies. These differences of the treatments may have impact on the effect of aspirin in diabetes subgroup analysis.
Our meta-regression analyses indicate that aspirin therapy may have different effects between the sexes in diabetic patients. Although there is no evidence that the pharmacodynamics of platelet inhibition by aspirin is any different in women than in men, the overall risk of CVD for people with diabetes is reported to be increased two-to threefold in men, and three-to fivefold in women [23]. More highly powered subgroup analyses for specific populations are awaited based on two major ongoing trials: A Study of Cardiovascular Events in Diabetes (ASCEND, International Standard Randomised Controlled Trial Number ISRCTN60635500, http://www.ctsu.ox.ac.uk/ascend/) and the Aspirin and Simvastatin Combination for Cardiovascular Events Prevention Trial in Diabetes (ACCEPT-D, Current Controlled Trials ISRCTN48110081) [69], which enrolled more than 15,000 diabetic patients without prior cardiovascular events to assess the effect of aspirin in the prevention of cardiovascular events. These trials may provide sufficient data to identify patients who derive the most benefit from aspirin therapy.
In sex subgroup analysis, our results are consistent with the previous sex-specific meta-analysis [6], but our findings conflicted with a recent publication which did not find any significant sex different in treatment effect [7,11]. There are several potential explanations for the different effects between sexes.
First, the different epidemiologic characteristics of cardiovascular disease between men and women may contribute to the different benefits. After age 40 years, men have a 49% lifetime risk for a coronary heart disease event, while women have a 32% risk. Men have a higher risk for MI, while the lifetime risk for ischemic stroke is greater in women than men from age 55 to 75 (17-18% in women and 13-14% in men) [2], and the risk of gastrointestinal bleeding is approximately twice as high in men than women [50].
Second, although some evidence indicates that there is no difference in pharmacodynamics of platelet inhibition by aspirin between the sexes and the 'aspirin resistance' may not exist [70][71][72], there is still insufficient evidence to support these conclusions. In fact, few randomized trials have measured 'aspirin resistance' directly, whether gender plays an important part in 'aspirin resistance' remains a question for future research [73][74][75][76].
The range of the dosage varies from 75 to 650 mg/day in eligible trials. Our analysis suggests that risk reductions achieved with low doses (75 mg/day) were similar to those obtained with higher doses (650 mg/day). In fact, it is reported that the successive daily administration of 30 mg of aspirin is sufficient to result in virtually inactivation of COX-1 [77]. However, there is no trial using this dosage. It is reported that apart from the inhibition of platelet aggregation, the impairment of cytoprotection in the gastrointestinal mucosa which is clearly dose-dependent also increases risk of upper gastrointestinal bleeding associated with aspirin therapy [72]. Thus, it is very likely that lower dosage of aspirin would decrease the bleeding complications. There is a need for additional placebo-controlled trials to demonstrate whether lower dosage of aspirin should be recommended for primary prevention of CVD in the future. However, because of the varied definitions of major bleeding among the studies, the result of our meta-regression does not indicate a clear does-effect relation which conflicts with previous published studies [78,79].
We acknowledge several limitations of our studies. First, we observed moderate heterogeneity among trials for some outcomes of interest. However, we have no access to patient-level data and our author response rate was relatively low, which may have led to limited statistical power. Second, the data were insufficient to report separate outcomes for type 1 and type 2 diabetes and different sexes in diabetic patients. Finally, the data on bleeding in our analyses were not sufficient to estimate whether changes in the dose of aspirin might reduce the risk of hemorrhage and whether further attempts at dosage reduction may compromise therapeutic efficacy.
In conclusion, our results demonstrate a significant net benefit to risk of aspirin for the primary prevention of CVD, and the decision regarding an aspirin regimen should be made on an individual patient basis, after careful evaluation of the trade-off between benefits and harms by the physician and patient. The effects of aspirin therapy vary by sex. Additional evidence is necessary before we make specific recommendations for aspirin use according to diabetes status.   Text S1 Study protocol.

(DOC)
File S1 Forest plot of effect of aspirin. Figure S4 in File S1. Figure S4-A Effect of aspirin on the prevention of MCEs. Figure  S4-B MI. Figure S4-C Stroke. Figure S4-D Ischemic stroke. Figure  S4-E All-cause mortality. Figure S4-F Cardiovascular mortality. Figure S4-G Hemorrhagic stroke. Figure S4