Machine learning-based high-benefit approach versus traditional high-risk approach in statin therapy: the Shizuoka Kokuho database study.
Journal:
Scientific reports
Published Date:
Jul 29, 2025
Abstract
Statins are widely prescribed for the primary prevention of cardiovascular diseases, yet individual responses vary, necessitating personalized treatment strategies. Conventional approaches prioritize treating high-risk patients, but advancements in machine learning now enable the estimation of conditional average treatment effects (CATE), offering opportunities to enhance treatment effectiveness. This study utilized the Shizuoka Kokuho Database to investigate heterogeneity in statin treatment effects. A 1:1 propensity score-matched cohort design was employed to evaluate the effect of statins in preventing a composite outcome, cardiovasuclar and cerebrovascular events and all-cause mortality. CATE was estimated using the causal forest model, an advanced ensemble machine learning technique. The effectiveness of a novel high-benefit treatment approach was compared with the traditional high-risk strategy. The propensity score-matched cohort included 8,792 individuals (mean age 67.4 years, 68.6% women). The causal forest model identified substantial heterogeneity in treatment effects. The high-benefit approach achieved a number needed to treat (NNT) of 15.1 (95% confidence interval [CI]: 9.4-23.4), significantly outperforming the high-risk approach (NNT: 29.5, 95% CI: 17.2-235.3). These findings demonstrate that leveraging machine learning to estimate CATE can enhance statin therapy by personalizing treatment, minimizing unnecessary medication, and improving population health outcomes.
