Baseline Event Rates and Model Validation

Baseline rates of CVD events are generated by the combination of population characteristics at model initiation, the model's estimation of the natural progression of CVD risk factors as individuals age, and the model's risk equations for disease. Appendix A contains additional tables and figures. Appendix A Table 1 presents prevalence rates of myocardial infarction and ischemic stroke generated by the model for a birth cohort starting at age 40 and compares these values to corresponding rates observed in NHANES^29-33 as a benchmark for the external validity of the ModelHealth: CVD natural history engine. Baseline rates of major GI bleeding in the non-elevated risk population (e.g., excluding persons with prior bleeding history or other contraindications) were estimated using data from a large Italian population-based cohort study,³⁴ with adjustments made for the U.S. age and sex distribution (Table 1). GI bleed case-fatality rates, based on patients without complicating comorbidities, were derived from a 74-hospital prospective study in the United Kingdom.³⁵ Baseline CRC incidence rates used in the model reflect contemporary use of screening technologies, such as colonoscopy, which can prevent CRC by the identification and removal of precursor adenomatous polyps or adenoma.

Table 1

Key Model Parameters and Sensitivity Values.

Integration of Systematic Review Results Into the Model

Findings from the three coordinated systematic evidence reviews on aspirin conducted on behalf of the USPSTF were integral to the parameter assumptions and model design in this study.^13-15 These reviews incorporated the latest evidence on aspirin's potential benefits and harms in the primary prevention of CVD, CRC, and all cancers combined. The reviews found evidence that daily aspirin use reduces the risk of non-fatal myocardial infarction, non-fatal stroke, and, after 10 years of use, CRC incidence and mortality. Aspirin also was found to increase the risk of fatal and non-fatal hemorrhagic stroke and major GI bleeding. The best balance of cardiovascular benefits to harms was reflected in daily aspirin doses of 100mg or less. Benefits with respect to CRC incidence were not strongly correlated with dose. There was not clear or compelling evidence that aspirin changes the relative risk of CVD death or fatal GI bleeds. Nor was there evidence that aspirin effects are differential by age or, in contrast to prior findings from the prior USPSTF review,^9,36 by sex. Evidence review findings also were used to inform baseline levels of GI bleeding risk and the selection of the American College of Cardiology and American Heart Association (ACC/AHA) risk calculator to determine baseline CVD risk in the model.³⁷ The systematic reviews did not assess the evidence for aspirin use in secondary prevention of CVD.

Aspirin Benefits and Harms

All aspirin effects were modeled as relative risk modifications to the annual probability of an event. Model parameters for primary prevention are summarized in Table 1. CVD and bleeding relative risks were derived from seven low-dose primary prevention trials, defined as 100mg of aspirin per day or less, identified by the systematic evidence review.^13,38-44 The effect of aspirin on the relative risk of developing colorectal cancer was estimated from three randomized clinical trials identified by the systematic evidence review,^14,45,46 but is restricted to a benefit observed after 10 years of continuous use. All non-CRC benefits and harms are assumed to take effect immediately, and all relative risks are assumed to return to 1.00 with discontinuation of aspirin. Indirect effects of aspirin on CVD incidence and mortality may arise when the prevention or occurrence of an initial event alters the disease progression probabilities for subsequent events, as determined by the Framingham-derived risk equations internal to the model (Appendix B Table 3). Effects of aspirin after experiencing a non-fatal CVD event are derived from secondary prevention trials (Appendix B Table 6).

Quality-of-Life Weights

Health utilities for the major outcomes affected by aspirin use were estimated using literature sources^47-53 and are summarized in Table 2. Living without a CVD condition or CRC was given a health utility of 0.872. All other health utility weights were applied multiplicatively to that baseline. Disutilities from myocardial infarction and GI bleeding events were applied only during the year an event occurs. In the base-case analysis, no disutility was applied to taking aspirin daily, but two alternative scenarios with aspirin disutilities included were considered in sensitivity analysis. Quality-of-life reductions for congestive heart failure were included because, as a major sequela to myocardial infarction, incidence may be indirectly affected by aspirin use in the model.

Table 2

Health Utility Weights.

Patient Population

The key questions were assessed independently for men and women across four 10-year age bands (40-49, 50-59, 60-69, and 70-79 years old) and across baseline 10-year CVD risk bands ranging from 1-20%. Baseline 10-year CVD risk was rounded to the nearest integer and estimated using the ACC/AHA risk calculator for the first hard atherosclerotic cardiovascular disease (ASCVD) event (non-fatal MI, non-fatal stroke, or coronary death).³⁷ The calculation of CVD risk at baseline is independent from the event rates predicted by the model and mirrors CVD risk identification as it may be practiced in clinical settings. For each age, sex, and baseline CVD risk band, simulated persons were randomly oversampled from population characteristics representative of the U.S. population. For men aged 60-79 and women aged 70-79, low 10-year risk bands that are rarely or never observed were excluded. To define the representative U.S. population, initial demographic characteristics—including age, sex, and race/ethnicity—were drawn from the United States Census.⁵⁴ Initial CVD risk factors, including BMI, SBP, LDL, HDL, and diabetes status, were derived from the combined 2001-2010 NHANES surveys.^29-33 Initial smoking status is derived from the 2007 National Health Interview Survey²³ and calibrated to estimates by the Congressional Budget Office.²⁶ All persons for the decision analysis were assumed to be free of CVD and CRC at baseline. Table 3 illustrates how the ACC/AHA risk bands are distributed across the U.S. population.

Table 3

Estimated Distribution of ACC/AHA 10-Year ASCVD Risk by Sex and Age.

Base-Case Analysis

All analyses compared outcomes of a simulated population routinely using aspirin for the primary prevention of CVD (i.e., prior to any major events) to the same population, all else held equal, not using aspirin for primary prevention. For secondary prevention (e.g., after a major CVD event), aspirin was initiated at contemporary rates of adherence (Appendix B Table 11) in both simulation arms. To align with common clinical practice, aspirin use was discontinued permanently in both arms after any major GI bleeding or hemorrhagic stroke event. Life years and quality-adjusted life years (QALYs) were the primary outcomes of interest, but all modeled benefit and harm events also were measured. Decision analysis criteria were limited to an assessment of positive or negative net balance of life years, QALYs, and event counts. Model simulations were independently conducted with a sample population of 100,000 persons for each age, sex, and baseline CVD risk group.

Uncertainty and Sensitivity Analysis

Two sources of uncertainty were considered in this study: stochastic heterogeneity resulting from the variability in outcomes experienced by a randomly selected sample population and parameter uncertainty resulting from the imprecision of model parameter estimates.⁵⁵ Confidence intervals reflecting stochastic heterogeneity were estimated by bootstrap resampling the simulated population for each stratified outcome 100,000 times with replacement.

Deterministic (one-way) sensitivity analyses of key parameters were conducted by replicating simulations with all other parameters, probabilities, and population characteristics held equal. Monte Carlo methods were used to perform probabilistic sensitivity analyses, with parameter values approximated using a triangle distribution. Table 1 presents the parameter value ranges used in the deterministic and probabilistic sensitivity analyses. Uncertainty in aspirin's effect to reduce CVD mortality risk was included among the sensitivity analysis parameters because a small but not statistically significant effect was observed in the systematic review. Parameter values for the relative risk of CVD mortality and hemorrhagic stroke were capped at 1.00 to maintain consistency in the directionality of aspirin benefits and harms.