Personalized Prediction of Cardiovascular Benefits and Bleeding Harms From Aspirin for Primary Prevention

Abstract

Aspirin reduces the risk for cardiovascular disease (CVD) but also increases the risk for bleeding (1, 2). For persons who have already had a cardiovascular event, the benefits of aspirin generally outweigh its harms, but the balance of benefits and risks is unclear in primary prevention (3). It was hoped that the results of 3 major trials published last year would determine whether aspirin has a role in the primary prevention of CVD among persons at intermediate risk for CVD, but these studies recruited participants with a lower CVD risk than expected (46). An updated meta-analysis (7) of aspirin for the primary prevention of CVD, which incorporated the findings from these 3 trials, confirmed that aspirin reduces the relative risk for CVD and increases the relative risk for bleeding. However, the key questionwhether there are persons without established CVD in whom the absolute benefits of aspirin outweigh its absolute harmsremains unanswered. Cardiovascular disease risk assessment using prognostic CVD risk models is now an accepted strategy for estimating the absolute CVD-related benefits of primary preventive interventions internationally (8, 9). Several prognostic CVD risk models are available (8, 1013) to estimate the absolute benefits of aspirin for an individual, and a prognostic bleeding risk model (14) was published recently that enables an individualized estimate of the absolute harms of aspirin. The aim of this study was to investigate, using an individualized assessment of the absolute cardiovascular benefits of aspirin and its bleeding harms, whether there are persons without established CVD for whom the absolute cardiovascular benefits of aspirin are likely to outweigh its absolute bleeding harms. Our hypothesis was that there are some persons without CVD for whom aspirin is likely to result in net benefit. Methods Design We conducted an individualized benefitharm analysis in which we estimated the net effect of aspirin for each participant in a previously described open cohort study (14, 15). The study was approved by the Northern Y Health and Disability Ethics Committee in 2003 (AKY/03/12/314), with annual approval by the National Multi Region Ethics Committee since 2007 (MEC07/19/EXP). Study Population Participants were automatically recruited into the cohort when their CVD risk was assessed with PREDICT (Enigma Solutions), a Web-based decision support program integrated with electronic primary care practice management systems in New Zealand (13, 16). For the purpose of this study, only data at study entry were used and cohort entry was restricted to the most recent 5-year period (1 January 2012 to 31 December 2016). When PREDICT is used to assess CVD risk, an electronic CVD risk profile is stored both in the practice management system and anonymously in a central database. With the permission of health providers, this central database profile was linked to an encrypted National Health Index number, which was used to link CVD risk profiles anonymously to national and regional databases. National databases were used to obtain or confirm data on demographic characteristics, deaths, publicly funded hospitalizations, cancer, and subsidized pharmaceutical dispensing. Laboratory test results were obtained from a regional laboratory repository. All participants were included unless they met any of the following exclusion criteria at the time of risk assessment: age less than 30 years or greater than 79 years; history of CVD, congestive heart failure, atrial fibrillation, chronic kidney disease (estimated glomerular filtration rate <30 mL/min/1.73 m2 on 2 occasions 90 days apart), diabetes (with overt nephropathy or other renal disease), or intracerebral bleeding; or treatment with an antithrombotic agent (aspirin, antiplatelet drug, or anticoagulant) in the preceding 6 months. Approximately 80% of persons eligible for CVD risk assessment (according to New Zealand guidelines) in practices using the PREDICT program have had their CVD risk assessed with this software (15). The PREDICT program is used by 35% to 40% of New Zealand primary care practices, and these practices serve approximately 35% of the New Zealand population (15). Further information regarding the setting, study entry, data sources, data linkage, and participant eligibility and exclusion criteria is available in earlier publications (14, 15). Absolute CVD and Bleeding Risk Absolute CVD and bleeding risk were estimated for each participant at study entry by using sex-specific risk scores derived from Cox proportional hazards models for time to first CVD event and bleeding event, respectively. These models were recently developed from and validated in PREDICT cohort data (13, 14). A CVD event was defined as an acute hospitalization or death from ischemic heart disease, ischemic or hemorrhagic cerebrovascular event, peripheral vascular disease, or congestive heart failure (13). The proportion of each type of first CVD event in the CVD risk models was as follows: myocardial infarction, 34%; unstable angina, 15%; other coronary heart disease, 5%; ischemic stroke, 15%; hemorrhagic stroke, 4%; transient ischemic attack, 7%; peripheral vascular disease, 6%; congestive heart failure, 12%; and other ischemic CVD-related deaths, 2% (13). Overall, 10% of the CVD events were fatal (13). A major bleed was defined as a hospitalization (principal diagnosis only, unless a transfusion of whole blood or packed cells also occurred during the admission) or death from a nontraumatic, nonprocedural gastrointestinal, intracranial (including hemorrhagic stroke), or other (respiratory, ocular, joint, pericardial, peritoneal) bleed (14, 15). The proportion of each type of first major bleed in the bleeding risk models was as follows: gastrointestinal, 69%; intracranial, 16%; and other, 15%. Overall, 7% of the major bleeds were fatal (14). Proportional Effect of Aspirin on CVD and Bleeding Risk The proportional (relative) effect of aspirin on CVD and bleeding risk was obtained from an updated systematic review and meta-analysis of 13 randomized controlled trials that included 164225 participants and compared aspirin (daily dose, 50 to 500 mg) with an inactive control among adults without established CVD (7). The meta-analysis found that over a median of 5.0 years (interquartile range, 4.7 to 6.7 years), aspirin was associated with an 11% proportional reduction in fatal CVD or nonfatal myocardial infarction or nonfatal stroke (hazard ratio [HR], 0.89 [95% CI, 0.84 to 0.95]); a 43% proportional increase in major bleeding, as defined by each contributing study (HR, 1.43 [CI, 1.30 to 1.56]); and no difference in all-cause mortality (HR, 0.94 [CI, 0.88 to 1.01]) compared with the inactive control. Heterogeneity of treatment effect among studies was low for each of these outcomes. The Antithrombotic Trialists Collaboration also found that the effect of aspirin on CVD events and major noncerebral bleeding was consistent across several subgroups in its 2009 meta-analysis of individual-participant data (1). Personalized Prediction of the Effect of Aspirin on CVD and Bleeding Risk Benefits and harms from aspirin were estimated as the differences between the pretreatment risk (absolute risk) for each event type (CVD or major bleed, separately) and the posttreatment risk (absolute riskthe proportional effect of aspirin) for each participant (17). Individual benefits and harms of aspirin were expressed as the predicted difference in the number of events among 1000 persons receiving aspirin for 5 years, as illustrated in Table 1. The main calculation for this study used the point estimate of the HR for the effect of aspirin on CVD and major bleeding. Table 1 illustrates the effect of using the upper (worst-case scenario) and lower (best-case scenario) limits of the 95% CI for this HR. The net effect of aspirin was calculated by subtracting the predicted reduction in CVD events from the predicted increase in major bleeds among 1000 persons receiving aspirin for 5 years; that is, a negative value represented net benefit (more CVD events avoided than major bleeds caused), and a positive value represented net harm (more major bleeds caused than CVD events avoided). The assumption in this calculation was that 1 CVD event was equivalent to 1 major bleed. A sensitivity analysis was conducted assuming that 1 CVD event was equivalent to 2 major bleeds. Participants were classified into the following mutually exclusive subgroups depending on the net effect of aspirin for 5 years among 1000 persons: net benefit (net effect, less than 1), equipoise (net effect, 1 to 1), and net harm (net effect, greater than 1). In some analyses, the net benefit subgroup was further divided into some net benefit (net effect, less than 1 and greater than 5) and substantial net benefit (net effect, 5 or less), and the net harm subgroup was divided into some net harm (net effect, greater than 1 and less than 5) and substantial net harm (net effect, 5 or greater). Table 1. Clinical Example* and Interpretation of Personalized BenefitHarm Calculation Statistical Analysis Women and men were described in total and by net effect subgroup according to a range of risk factors for CVD or major bleeding. Continuous variables were summarized as means with SDs and medians with interquartile ranges, and categorical data were summarized as frequencies and percentages. Data were compared among net effect subgroups by using 1-way analysis of variance (for continuous data) or the Pearson 2 test (for categorical data), with the significance level for a difference between subgroups set at P < 0.050 (2-sided). The distribution of the net effect of aspirin was illustrated in histograms according to net effect subgroups. The net effect of aspirin was plotted for a random selection of participants in scatter plots of pretreatment 5-year CVD risk by pretreatment major bleeding risk. Both plot types present net effect on the basis of