Background

Acetylsalicylic acid (commonly referred to as “aspirin”) is an antithrombotic agent that inhibits platelets by irreversibly acetylating the serine residue of cyclooxygenase-1 (COX-1) in platelets with subsequently reduced levels of prothrombotic thromboxane A2 (TxA2) [1,2,3]. In patients with known cardiovascular disease (CVD), the potential for aspirin to reduce further cardiovascular (CV) events significantly outweighs the risks of major bleeding and thus aspirin has since become a mainstay in secondary prevention of CVD [4,5,6,7,8]. However, in primary prevention, its role is still under debate [9]. This is due to an as yet unclear balance between the benefits and risks of aspirin treatment in patients without a diagnosed atherosclerotic disease.

Previously published meta-analyses have indicated that aspirin significantly reduced myocardial infarction (MI) and major adverse cardiovascular events (MACE) without an impact on stroke and CV- or all-cause death [10,11,12,13,14]. Furthermore, an increased risk of major bleeding events under aspirin strongly outweighed the benefits of aspirin treatment in primary prevention [10, 12,13,14]. As a result, the current guidelines on CVD prevention from the European Society of Cardiology (ESC) do not recommend antiplatelet therapy in patients free of overt CVD [8]. On the contrary, the recently published 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease states that aspirin might be considered in selected adults aged 40 to 70 who are at higher CV risk but at no increased bleeding risk [15]. The U.S. Preventive Services Task Force recommends initiation of aspirin treatment depending on age and 10-year CVD risk [16].

Recently, three major trials (ARRIVE, ASCEND, and ASPREE) evaluating the use of aspirin in primary prevention of CVD were published [17,18,19]. The ARRIVE trial enrolled patients with moderate to high cardiovascular risk, the ASCEND trial patients with diabetes mellitus (DM) only, and the ASPREE trial elderly patients. Only the ASCEND trial [18] showed a significant reduction in the rate of major adverse CV events, but the effect was, once again, accompanied by a significant increase in major bleeding. Using the three recently published trials, we aimed to perform a meta-analysis with a particular focus on subgroups in order to potentially characterize patient populations with a more favorable benefit-risk ratio.

Methods

Protocol and registration, data extraction, and quality assessment

Our review was registered with PROSPERO under the registration number CRD42019118474. Two reviewers applied the selection criteria (GG and JMSM) independently and in duplicate. This study was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, as described previously [20,21,22,23,24].

Data sources and searches

We searched PubMed and Web of Science using predefined search terms (primary prevention AND aspirin AND clinical trial OR meta-analysis) until November 2018. Six additional trials [25,26,27,28,29,30] that were included in a previous meta-analysis [5] were also identified and included in our analysis. The titles and abstracts of suspected relevant citations were screened for eligibility, and the full text was acquired for further evaluation if the citation was deemed pertinent. References of retrieved meta-analyses and reviews were also checked for additional trials.

Study selection and outcomes

Included studies had to be randomized controlled trials (RCT) and include at least 1000 patients. Studies had to be controlled (placebo or control group), but could be open-label or blinded. The target patient population comprised patients without any history of CVD. Patients with a low ankle-brachial index (ABI) who had no symptoms and no diagnosis of peripheral arterial disease were considered as a primary prevention cohort. Exclusion criteria were non-RCTs, duplicate reports, ongoing studies, and studies that included patients with history of CVD.

The primary efficacy outcome was all-cause mortality. Secondary efficacy outcomes included cardiovascular mortality, the composite of major adverse cardiovascular events (MACE), MI, and ischemic stroke (IS). MACE was defined as a composite of nonfatal stroke, nonfatal MI, and CV mortality. In order to accurately assess the rate of MACE, we performed two analyses, one comparing the calculated rate of MACE as per our definition and one comparing the rate of the study defined primary outcome as a part of a sensitivity analysis. Stroke was defined as “ischemic stroke” but not all included studies reported on the incidence of IS alone. If not sufficiently specified, the number of reported strokes was used. We also reported on the incidence of hemorrhagic stroke. Bearing in mind the uncertain effect of aspirin on cancer outcomes, cancer risk was prespecified as an exploratory outcome. For further analysis of data, we performed four subgroup analyses involving diabetes, sex, concomitant statin treatment, and smoking.

Major bleeding was the primary safety endpoint. Definition of major bleeding varied between studies. If not defined as “major bleeding,” we used the following definitions: “bleeding requiring transfusion,” “bleeding rendering patients intensive care dependent,” “bleeding causing death,” or “intracranial bleeding.” The extracranial major bleeding analysis comprised the total of all major bleedings and some GI bleeding events that were classified as relevant in respect to the analysis. Intracranial hemorrhages and GI bleedings were also assessed as single endpoints.

Data synthesis and analysis

Variables are reported as numbers or percentages as appropriate. Risk ratios (RR) were calculated from individual studies and pooled according to the inverse variance model with 95% confidence intervals (95% CI) and reported as relative risk reduction or increase respectively (RRR/RRI) within a mean time frame of 6.4 years (which is the mean follow-up period of included studies). The statistical inconsistency test (I2) was used to assess heterogeneity vs. homogeneity between studies. If the I2 value was low (I2 < 50%), a fixed-effect model was additionally calculated, as reported previously [20, 22, 24]. The following sensitivity analyses were performed: (i) comparison of the results of the fixed vs. random-effect model, (ii) the influence of each study was assessed by testing whether deleting each in turn would have significantly changed the pooled results of the meta-analysis, (iii) sensitivity analysis of the date of publication before and after 2010, (iv) sensitivity analysis assessing the length of follow-up (< 5 vs. > 5 years), and (v) and analysis focusing on the study defined primary outcome parameter.

Absolute risk reduction or increase (ARR, ARI) and number needed to treat or harm (NNT, NNH) were calculated per 1 year of treatment. This was performed as follows: event incidence rates were divided by their respective mean follow-up periods and multiplied by 100 to obtain the incidence rate per 100 patient years. Out of these, the ARR or ARI were calculated by subtraction, and subsequently, the NNT or NNH were calculated according to the following formula: NNT or NNH = 1/(ARR or ARI). Events prevented/caused per 10,000 patients per year were calculated by dividing 10,000 by the NNT or NNH. This transformation of data allows for a better understanding of risks for doctors and patients.

The mortality-adjusted net clinical benefit was calculated as follows: [IRischemi stroke_aspirin + w1IRmyocardial infarction_aspirin + w2IRhemorrhagic stroke_aspirin + w3IRmajor extracranial bleeding_aspirin] − [IRischemic stroke_control + w1IRmyocardial infarction_control + w2IRhemorrhagic stroke_control + w3IRmajor extracranial bleeding_control], where w1, w2, and w3 are the death-related weights associated with each type of event. Weights were calculated as the impact of each event on mortality, as derived from recent analyses [31, 32], and related to IS (weight = 1). Weights were thus w1 = 0.89 for MI, w2 = 3.23 for hemorrhagic stroke, and w3 = 0.63 for major extracranial bleeding [31, 32]. In the mortality-adjusted net clinical benefit analysis, a lower estimate indicates a greater benefit of aspirin.

A two-tailed p value of < 0.05 was considered significant. Review Manager (Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) was used for statistical calculations.

Results

Description of studies

Our search retrieved 608 references. Five hundred ninety items were excluded based on title and abstracts that were not RCTs, investigated aspirin in secondary prevention of CVD, or were identified as non-pertinent studies (Additional file 1: Figure S1). Additionally, retrieved reviews and meta-analyses were thoroughly examined to identify further trials. One study was excluded as it contained a significant number of patients with definite or suspected CVD [33]. Thirteen trials [17,18,19, 25,26,27,28,29,30, 34,35,36,37] were eligible for analysis and comprised a total of 164,225 patients, 82,900 allocated to aspirin and 81,325 allocated to the control group. One included study [36] was a 10-year follow-up of a previously published trial [38]. The mean age of patients included in our meta-analysis was 62 years. The mean follow-up period was 6.4 years (ranged from 3.6 to 10.3 years). Three trials exclusively included patients with known diabetes [18, 36, 37]. Three trials included men only [25, 28, 30], and one trial included women only [29]. The dosage of aspirin ranged from 75 to 500 mg once daily. Two trials evaluated the effect of aspirin (325 mg and 100 mg) given on alternate days [29, 30]. Only two studies reported the use of proton-pump inhibitors (PPIs) [18, 19]. Included studies are characterized in Tables 1 and 2.

Table 1 Characteristics of included studies
Full size table
Table 2 Overview of study demographics
Full size table

Efficacy and safety outcomes

All trials reported on all-cause mortality [17,18,19, 25,26,27,28,29,30, 34,35,36,37] and included 164,225 patients. The incidence of all-cause mortality was similar between the aspirin and control groups (4.52% vs. 4.54%, respectively; RR 0.98; 95% CI, 0.93–1.02; p = 0.26; I2 = 0%; Fig. 1, Table 3, Additional file 1: Figure S2A). Use of aspirin was not associated with a reduction in CV mortality (RR 0.99; 95% CI, 0.90–1.08; p = 0.75; I2 = 0%; Fig. 1, Table 3; Additional file 1: Figure S3A) compared with no aspirin. Aspirin was associated with a RRR of MACE by 9% (RR 0.91; 95% CI, 0.86–0.95; p < 0.0001; I2 = 0%; ARR 0.052%; NNT 1908; Fig. 1, Additional file 1: Figure S2B), of MI by 14% (RR 0.86; 95% CI, 0.77–0.95; p = 0.005; I2 = 50%; ARR 0.041%; NNT 2452; Fig. 1, Table 3, Additional file 1: Figure S3B), and of IS by 10% (RR 0.90; 95% CI, 0.82–0.99; p = 0.03; I2 = 17%; ARR 0.022%; NNT 4448; Fig. 1, Table 3, Additional file 1: Figure S3C) compared with no aspirin.

Fig. 1
figure 1

Risk ratios (RR) of the major outcomes

Full size image
Table 3 Risk estimates: absolute risk reduction (ARR) or increase (ARI) and number needed to treat (NNT) or to harm (NNH) for the primary and secondary endpoints over a treatment period of 1 year, which were statistically different between groups
Full size table

Twelve studies, including a total of 159,086 patients, reported on the rate of major bleeding complications [17,18,19, 25,26,27, 29, 30, 34,35,36,37]. Aspirin use was associated with a 46% RRI of major bleeding complications (RR 1.46; 95% CI, 1.30–1.64; p < 0.00001; I2 = 31%; ARI 0.077%; NNH 1295; Fig. 1, Table 3, Additional file 1: Figure S2C) compared with no aspirin. Extracranial major bleedings and GI bleedings were the major driver of the composite of bleeding events, with intracranial bleedings and hemorrhagic stroke having no statistical impact (Additional file 1: Figure S4). Aspirin did not decrease the cancer incidence (Additional file 1: Figure S5).

The net clinical benefit

All trials [17,18,19, 25,26,27,28,29,30, 34,35,36,37] provided data for the estimation of the adjusted net clinical benefit. Aspirin was not associated with a net clinical benefit after adjustment for event-associated mortality risk (mean 0.034%; 95% CI, − 0.184 to 0.252%; Fig. 2).

Fig. 2
figure 2

Analysis of the mortality-adjusted net clinical benefit

Full size image

The crude net clinical benefit comprising MACE and major bleeding events was calculated with data from 12 studies [17,18,19, 25,26,27, 29, 30, 34,35,36,37], showing that aspirin did not lead to a net clinical benefit (RR 1.01; 95% CI, 0.97–1.05; p = 0.54; I2 = 0%; Additional file 1: Figure S6).

Analysis of subgroups

Statin treatment

MACE:

Two trials, comprising a total of 34,594 patients, reported on the MACE risk in patients with and without statin treatment [18, 19]. Aspirin-treated patients who were also treated with statins had a 12% RRR of MACE when compared with control plus statin (RR 0.88; 95% CI, 0.80–0.96; p = 0.007; I2 = 0%; Fig. 3). In contrast, aspirin without statin co-treatment did not reduce MACE risk when compared with control without statin (RR 0.94; 95% CI, 0.83–1.08; p = 0.39; I2 = 25%; Fig. 3).

Fig. 3
figure 3

Subgroup analysis: Forest plot depicting the relative risk (RR) of MACE in patients with and without statin treatment

Full size image

Smoking

MACE:

Five trials, comprising a total of 88,539 patients, reported on incidence of MACE in smokers and non-smokers [17, 19, 29, 35, 36]. In non-smokers, aspirin use was associated with a 10% RRR of MACE (RR 0.90; 95% CI, 0.82–0.99; p = 0.04; I2 = 23%; Fig. 4) compared with no aspirin. In smokers, aspirin did not affect the risk of MACE (RR 1.11; 95% CI, 0.96–1.28; p = 0.16; I2 = 0%; Fig. 4).

Fig. 4
figure 4

Subgroup analysis: Forest plot depicting the relative risk (RR) of MACE in smokers and non-smokers

Full size image

Sex

MACE:

Nine trials, including 59,337 patients, reported the incidence of MACE in men [18, 19, 25, 28, 30, 34,35,36,37], seven trials in women (69,164 patients) [18, 19, 29, 34,35,36,37]. There was a sex interaction for aspirin effect: the direction of the effect of aspirin on MACE tended to be similar in men and women, but the effect size differed, and did not reach statistical significance in females. Aspirin in men was associated with a RRR of MACE of 11% (RR 0.89; 95% CI, 0.83–0.95; p = 0.0008; I2 = 12%; Fig. 5) compared with controls. In women, aspirin did not significantly reduce the risk of MACE (RR 0.95; 95% CI, 0.88–1.02; p = 0.16; I2 = 0%; Fig. 5) compared with controls.

Fig. 5
figure 5

Subgroup analysis: Forest plot depicting the relative risk (RR) of MACE in males and females

Full size image

Myocardial infarction:

Three trials, including 32,295 patients, reported on MI in men [25, 28, 30], only one in women (39,876 patients) [29]. In men and in women, aspirin did not significantly reduce the risk of MI (RR 0.76; 95% CI, 0.57–1.01; p = 0.06; I2 = 76%; RR 1.03; 95% CI, 0.84–1.25; p = 0.26; respectively). This is in contrast to the overall population and underlines that the sex sub-analysis for MI is underpowered.

Ischemic stroke:

Three trials, including 32,295 patients, reported on IS in men [25, 28, 30], but only one trial reported these data for women (39,876 patients) [29]. Aspirin did not reduce the RR of IS in men (RR 1.02; 95% CI, 0.72–1.44; p = 0.93; I2 = 55%). In women, however, aspirin reduced IS by 23% (RR 0.77; 95% CI, 0.63–0.94; p = 0.010) compared to control as reported in one study.

Diabetes

MACE:

Six studies, including 27,292 patients, reported on the rate of primary endpoint in patients with diabetes, showing a RRR of MACE by 9% (RR 0.91; 95% CI, 0.85–0.99; p = 0.02; I2 = 0%, Fig. 6), which is consistent with the analysis of the overall population. No data for a non-diabetic subgroup were available.

Fig. 6
figure 6

Subgroup analysis: Forest plot depicting the relative risk (RR) of MACE in patients with diabetes

Full size image

Myocardial infarction:

Five studies, comprising 20,865 patients, provided data for MI in diabetic patients, showing no difference between aspirin vs. control (RR 0.94; 95% CI, 0.72–1.23; p = 0.65; I2 = 53%) [18, 29, 30, 36, 37].

Ischemic stroke:

Four studies, including 20,332 patients, reported on IS in diabetic patients, showing a RRR of IS by aspirin of 24% (RR 0.76; 95% CI, 0.59–0.98; p = 0.03; I2 = 43%) [18, 29, 36, 37].

Sensitivity analyses:

Sensitivity analysis assessing the date of publication showed that the direction of the effect on MACE remained unchanged. However, the magnitude of the effect tended to be greater in studies published before 2010 compared to studies published after this date (RRR 11% vs. 7%, respectively).

Due to low heterogeneity (I2 = 0%), a fixed-effect model was calculated in addition to the random-effect model for each outcome (Additional file 1: Table S1), which confirmed the robustness of our findings.

By sequentially excluding one single study from the pooled analysis, the direction and the magnitude of the effect on MACE remained unchanged.

Sensitivity analysis assessing the length of follow-up/length of study drug use showed that the direction of effect on MACE remained unchanged. However, the magnitude of the effect tended to be greater in studies with a shorter-term use of aspirin (≤ 5 years, RRR 13%) vs. longer-term use (> 5 years, RRR 8%).

We additionally analyzed the primary endpoint of each study according to the study definition (which in some studies slightly differed from the MACE definition used in our meta-analysis). In the aspirin group, 4.3% of patients (3601/82,900) reached the primary endpoint compared to 4.7% in the control group (3827/81,325). Treatment with aspirin, therefore, significantly reduced the RR of the primary endpoint by 9% (RR 0.91; 95% CI, 0.87–0.95; p < 0.0001; I2 = 0%), confirming the result of the MACE analysis.

Discussion

Our meta-analysis in over 160,000 patients without a history of CVD showed that aspirin did not reduce all-cause or CV mortality but reduced the risk of MACE, MI, and IS at the cost of an increased risk of major bleeding events. Hence, aspirin treatment was associated with a lower NNH than the NNT for the safety and efficacy outcomes: major bleeding and MACE: 1295 vs. 1908 respectively. Most importantly, our meta-analysis shows that there is a treatment interaction in three subgroups: non-smokers, male sex, and treatment with statins.

Two recently published meta-analyses have provided information about the use of aspirin in primary prevention of CVD [39, 40]. Our meta-analysis confirms previous findings and provides additional value with four distinct subgroup analyses and a mortality-adjusted net clinical benefit analysis.

One of the most important findings of our study is the net clinical benefit of aspirin, adjusted for the risk of event-associated mortality, which aims to balance the preventive impact of aspirin on risk for ischemic events such as MI and IS, versus the impact of increased risk of bleeding. The outcome of intracranial hemorrhage is generally worse than the outcome of IS or MI, with the best outcome following a GI bleeding event. Based on previous estimates [31, 32], we weighted hemorrhagic stroke threefold worse than IS. Our weighted analysis provides quantitative assessments of the net clinical benefit of aspirin among primary CVD prevention patients and confirms the result of the crude net clinical benefit estimation. Although models adjusting for event-associated mortality are commonly used [31, 32], weighting one nonfatal event against another is very difficult, as the risks might differ between patients. Therefore, it is still unclear how to properly weight an ischemic event against a bleeding event. Some people with a high risk of having an ischemic event will prefer to take the risk of having a GI-bleed on aspirin, in order to reduce the risk of IS or MI. As there was no significant difference in mortality, intracerebral hemorrhage, or hemorrhagic stroke between aspirin and control, patient preferences should be considered.

Considering upper GI bleeding, which is the most common complication in patients under antiplatelet therapy [41,42,43], PPIs have been proven effective in the prevention of GI bleeding and are recommended in patients at increased risk for this bleeding [44]. On the other hand, long-term treatment with PPIs is associated with increased risk of community-acquired pneumonia (CAP) [45], bone fractures [45, 46], and enteric infections, mainly by Salmonella and Campylobacter spp. [45]. Furthermore, PPI-related hypomagnesemia is of clinical significance as it is a known cause of cardiac arrhythmias [45]. Thus, in consideration of the benefits and risks of the respective treatments, the question arises as to whether patients without bleeding risk should receive long-term treatment with PPIs concomitantly with aspirin for primary prevention.

A population of special interest is patients treated with statins. Interestingly, our subgroup analysis comprising 18,000 patients who were concomitantly treated with statins and aspirin showed a benefit in terms of MACE reduction, whereas those treated with aspirin without statins did not. Remarkably, patients treated with aspirin and statins showed the highest RRR of MACE of 12% compared to the overall population and patients with DM. A possible explanation for this interaction might be the consideration that those taking statins are at higher risk for CVD because of hyperlipidemia, and therefore might benefit more. Another possible elaboration might be a direct plaque-stabilizing effect of statins, which, in combination with platelet inhibition by aspirin, improves ischemic outcome. Notably, statins are associated with reduced platelet reactivity and improved response to aspirin [47,48,49,50,51,52]. However, it is unclear whether the improved response to aspirin under statin treatment is caused directly by statin-platelet interaction, indirectly via reduced levels of lipids [47,48,49, 51, 52], or by a combination of the two. Elevated cholesterol levels have been linked to decreased aspirin-induced platelet acetylation, explaining the indirect effect of statins on platelet inhibition [53]. Two mechanisms have been identified as being involved in the direct effect of statins on platelets [54]. Administration of atorvastatin resulted in the downregulation of phospholipase A2 (PLA2) (after 24 h) and NOX2 (after 2 h) leading to reduced levels of TxA2 and prothrombotic platelet isoprostanes respectively [55]. Based on these findings, early and late antiplatelet effects of statins have been hypothesized [54].

Interestingly, our subgroup analysis showed aspirin use in non-smokers to reduce the risk of MACE by 10%, whereas smokers did not benefit from aspirin treatment. This confirms the result of a previous meta-analysis by Seidu et al., who describe a 30% risk reduction with aspirin in non-smokers [56]. Smoking has been linked to an attenuated antiplatelet effect of aspirin in the past [57,58,59], and our meta-analysis suggests a possible translation of this phenomenon into clinical practice. In current smokers, a treatment switch from aspirin to the P2Y12 receptor inhibitor clopidogrel seems to be an interesting alternative. Smoking is a known inducer of cytochrome P450 (CYP) 1A2, an essential isoenzyme that converts clopidogrel into its active metabolite, and thus may facilitate an adequate platelet inhibition [60]. Studies have demonstrated fewer ischemic events in smokers following clopidogrel administration [61, 62]; however, in primary prevention of CVD, the overall role of clopidogrel has not yet been investigated.

It is crucial to note that our meta-analysis has shown sex differences in aspirin effects. Aspirin showed a reduction of MACE in men but not in women. In contrast, aspirin reduced the risk of stroke in women as shown in a single study, but not in men. Results from previous meta-analyses have also detected a more pronounced effect of aspirin for MACE or MI in men and for stroke in women [5, 63,64,65]. Although sex differences in aspirin effects are of interest, it is currently unclear how they can be used in clinical decision-making [8, 16, 66].

Another population of special interest is patients with DM. Diabetes increases the risk of CVD, and aspirin is therefore expected to have a greater preventive effect in these patients [67]. In our subgroup analysis comprising over 20,000 patients with diabetes mellitus, aspirin showed a significant 9% RRR in MACE, which confirms the estimate in the overall population. While older guidelines have deemed the use of aspirin reasonable in certain patient populations with diabetes [66], current 2019 guidelines from the ACC/AHA do not specifically comment on the use of aspirin in diabetic patients in primary prevention of CVD [15]. The newly published ESC guidelines on diabetes, pre-diabetes, and CVD have stated aspirin may be used in patients with DM at high/very high risk of CVD and in the absence of clear contraindications (class IIb) [68].

In the general population, the U.S. Preventive Services Task Force’s guideline recommends aspirin for patients based on age and prediction tools such as the 10-year cardiovascular disease calculator [16]. Importantly, these recommendations are given with a moderate evidence level (B and C). The 2019 ACC/AHA guidelines acknowledge the controversy of aspirin in primary prevention of CVD, but state that aspirin might be considered in selected adults aged 40 to 70 who are at higher CV risk but at no increased bleeding risk [15]. Additionally, two cost-utility analyses suggest a clear benefit of aspirin [69, 70]. However, the ESC guidelines on CVD prevention do not recommend the general use of aspirin for the primary prevention of cardiovascular disease [8].

Limitations

The main limitation is that some studies did not differentiate between ischemic and hemorrhagic stroke. In such cases, the total of “all strokes” was included. The primary endpoint and follow-up periods also differed between some studies; we have adjusted for this in the sensitivity analyses. Another limitation of this meta-analysis was the use of heterogeneous definitions of major bleeding. One study used the GUSTO bleeding classification [17]; most others used a prespecified composite of bleeding events such as GI bleeding and major extracranial bleeding and defined their severity by hospitalization, prolongation of hospitalization, surgery, transfusion requirement, or fatality. The severity and definition of GI bleeding events were often not further detailed.

Furthermore, some trials included in our meta-analysis [25, 27, 28, 30] were performed several decades ago. Since then, there may have been changes in medical standards, the prevalence of risk factors, and access to early diagnostic services.

Conclusions

The increased risk of major bleeding and lack of reduction of mortality might outweigh the benefits of aspirin in primary prevention of CVD in the overall population. Three patient subgroups: non-smokers, patients treated with statins, and males, had the greatest risk reduction of MACE.