Finding the Balance Between Benefits and Harms When Using Statins for Primary Prevention of Cardiovascular Disease

Most guidelines recommend statins for primary prevention of cardiovascular disease (CVD) if 10-year risk exceeds 7.5% to 10%, often in addition to such other criteria as high cholesterol level or presence of at least 1 specific risk factor (15). However, use of statins for primary prevention of CVD is controversial and varies greatly among countries because of several factors, such as uncertainty about the applicability of the results of randomized controlled trials (RCTs) on primary prevention to real-world populations, the definition of eligible persons likely to benefit, and potential differential reporting of harms (69). Nevertheless, guidelines must make recommendations and deal with the uncertainties of the existing evidence base. Guidelines should consider multiple factors that influence the balance of benefits and harms of statins, including preventive effects and harms, baseline risks for benefit and harm outcomes, and outcome preferences of persons who may benefit from statins for primary prevention of CVD (10, 11). However, none of the current guidelines used a systematic assessment of the benefitharm balance of statins that considered these factors (12). In addition, whether the currently recommended risk thresholds of 7.5% and 10% are justified is unclear. Therefore, this study aimed to assess the balance of benefits and harms of statins for primary prevention of CVD and determine age- and sex-specific 10-year risk thresholds at which the net benefits of statins outweigh the net harms. Because not all statins have the same benefit and harm profiles (1315), we analyzed 4 commonly used statins separately. Methods Target Population and Setting We performed a quantitative benefitharm balance modeling study on use of statins for primary prevention of CVD for persons in the general population aged 40 to 75 years with no history of CVD events. We excluded persons older than 75 years because of scarce data in this age group. Our study evaluated the balance of benefits and harms and accounted for baseline risks for the benefit and harm outcomes, the magnitude of the increase or decrease in risk due to statins, the relative importance of the outcomes, and a specific time horizon. Statins and Benefit and Harm Outcomes This analysis focused on low- or moderate-dose statins, which are frequently prescribed for primary prevention of CVD (3), and excluded high-dose statins. We performed the benefitharm analysis for 4 statins (atorvastatin, simvastatin, pravastatin, and rosuvastatin) for which there were sufficient data from RCTs about their effects. We selected clinically relevant benefit and harm outcomes from systematic reviews (14, 16) and quantified their relevance in a preference-eliciting survey reported elsewhere (17). Benefit outcomes (those favoring statin use) were fatal and nonfatal CVD events. Harm outcomes (adverse effects of statins) were myopathy, hepatic and renal dysfunction, cataracts, hemorrhagic stroke, type 2 diabetes, any cancer, nausea or headache, and treatment discontinuation due to adverse effects. We considered non-CVD mortality as a competing risk because we found insufficient evidence that statins reduce deaths due to causes other than CVD. Although it would be possible to use specific CVD events in the benefitharm balance modeling, we considered a composite outcome of CVD events as the end point because most clinical guidelines and risk scores refer to such a composite outcome. This study did not consider costs associated with statins. Data Sources We systematically selected evidence on the following input parameters needed for the quantitative benefitharm balance modeling (18). Preventive or Adverse Effects of Statins We based estimates of statin effects on a network meta-analysis of CVD events and harm outcomes in a primary prevention population that we performed previously for the purpose of the current study (Figures 1 to 3 of the Supplement). We obtained information on preventive effects on cataracts and hemorrhagic stroke from 2 RCTs (HOPE-3 [Heart Outcomes Prevention Evaluation-3] [19] and MEGA [Management of Elevated Cholesterol in the Primary Prevention Group of Adult Japanese] [20], respectively) because there were not enough trials for a meta-analysis. Data on harm outcomes from trials might be incomplete because of short follow-up or limited external validity or may be affected by differential reporting bias (8, 21). Observational studies may be more suitable for providing real-world estimates of harms, but they are often inherently biased (8, 21). We thus considered combined statin effect estimates for the harm outcomes from large observational studies and RCTs using a Bayesian inverse varianceweighted averaging method (Appendix Table 1 and Table 1 of the Supplement) (2225). The preventive effects from observational studies contributed less to the pooled effect due to higher variances. We also performed a sensitivity analysis using estimates of statin effects on harms from RCTs only. Supplement. Supplementary Information Appendix Table 1. Effect Estimates of Statins on Benefit and Harm Outcomes Considered From Randomized Controlled Trials and Observational Studies* Baseline Outcome Risks in the Population Population-based baseline risks are favored over event rates in control groups of RCTs because they better reflect the real-world population (18). We extracted age- and sex-specific baseline rates of type 2 diabetes, any cancer, hemorrhagic stroke, and non-CVD mortality from Global Burden of Disease estimates for Switzerland (26) and rates of myopathy, renal and hepatic dysfunction, and cataracts from other observational data (age- or country-specific rates were not available for these) (23) (Table 2 of the Supplement). We chose Switzerland for some of these age- and sex-specific risks because the weights of the outcomes were elicited there and because outcome risks are moderate there, except for competing risk for death, which is low compared with other countries. However, we also performed the analysis for the United States and the United Kingdom (see Sensitivity Analyses). Because baseline risks for nausea or headache and treatment discontinuation were not available, we used rates from control groups of RCTs. Outcome Preferences We considered population average preference weights from a preference-eliciting study designed to inform the current study. The bestworst scaling survey elicited preferences for benefit and harm outcomes of statins from a primary prevention population in Switzerland and Ethiopia (17). We considered preference weights reported as the surface under the cumulative ranking curve (SUCRA) (analyzed using a network meta-analytic approach) for the base-case analysis (27) and additional preference scales, including bestworst score and log-odds, for sensitivity analyses. The SUCRA indicates the probability of an outcome that patients would rather avoid or a weight showing the relative importance of an outcome to patients (larger values correspond to more important outcomes) (Appendix Table 2). Appendix Table 2. Preference Weights for Benefit and Harm Outcomes Considered for the BenefitHarm Balance Modeling* Time Horizon Although data from RCTs are available for follow-up less than 5 years, we extended the time horizon to 10 years with the assumption of smaller CVD risk accumulation and similar effects of statins over the period in low- to moderate-risk persons. Subgroups We modeled the balance of benefits and harms and determined risk thresholds for 350 subgroups according to age and sex across 10-year CVD risks ranging from 1% to the value above which the benefits consistently outweighed the harms. We did not need to predict 10-year CVD risks because we repeated the analysis for each percentage point increase up to 25% (all statins) or 40% (specific statins). Although not relevant to this study, the use of well-calibrated risk scores is important for clinical practice to determine a person's risk and whether this risk justifies use of statins. Statistical Analysis BenefitHarm Balance Index We used a benefitharm balance model developed by Gail and colleagues at the National Cancer Institute (NCI) (28, 29) and extended it to fit our research question of determining risk thresholds. A detailed description of our model is provided in the Appendix. In brief, we estimated the expected age- and sex-specific number of benefit and harm outcome events per 1000 persons not using statins over 10 years by using an exponential model that assumed constant risk rates over the time horizon and accounted for competing risk for non-CVD death. We then calculated the corresponding number of expected events for each outcome among statin recipients by using the same model but with consideration of estimates of statin effects. The differences in expected events with and without statin use provided attributable absolute events for each benefit and harm outcome. We then weighted the differences by their respective preference weights and summed them to yield a single benefitharm index. The resulting index could have any negative (harms outweighed benefits), zero (harms equaled benefits), or positive (benefits outweighed harms) value on an arbitrary scale. To account for statistical uncertainty of the input parameters, we repeated the analysis 100000 times for each subgroup with resampling of the parameters independently from normal distributions defined by their mean estimates (log risk ratio, SUCRA, and baseline risks) and their SEs. Determining 10-Year Risk Thresholds for Net Benefits We calculated the probability of net benefits as the proportion of repetitions for which the benefitharm index was positive (benefits outweighed harms). This probability could have any value between 0% and 100%. We defined statins as having net benefits if the probability of the index exceeding zero was at least 60% and net harms if the probability was less than 40%. Thus, probabilities of at