Predictive, preventive, and personalized medicine in statin-treated patients with NSTE-ACS, HFpEF, and type 2 diabetes: a data-driven model for long-term residual risk assessment.

BACKGROUND: Despite significant advances in guideline-directed medical therapy (GDMT), statin-treated patients with non-ST-elevation acute coronary syndrome (NSTE-ACS), heart failure with preserved ejection fraction (HFpEF), and type 2 diabetes mellitus remain at substantial residual risk of major adverse cardiac and cerebrovascular events (MACCEs). Within the framework of predictive, preventive, and personalized medicine (3PM/PPPM), this study aimed to develop and externally validate a data-driven survival model for individualized long-term risk stratification to support the transition from reactive management to proactive and targeted prevention. METHOD: In this multicenter retrospective cohort study, 1,206 NSTE-ACS patients with HFpEF and type 2 diabetes who underwent successful percutaneous coronary intervention (PCI) were enrolled from six Chinese tertiary hospitals and geographically divided into a training cohort (n = 763) and an independent validation cohort (n = 443). Sixty clinical, laboratory, and echocardiographic variables were screened using wrapper-based feature selection across five machine learning algorithms with stratified 5-fold cross-validation. Thirty-one survival models spanning four methodological categories were developed and compared. The primary endpoint was MACCEs, defined as a composite of recurrent acute coronary syndrome, stroke, and hospitalization for heart failure. RESULTS: Seven consensus biomarkers were identified: age, albumin, creatinine, monocyte-to-lymphocyte ratio, non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (NHHR), N-terminal pro-B-type natriuretic peptide (NT-proBNP), and non-high-density lipoprotein cholesterol (non-HDL-C). Among 31 survival models, Xgboost.Cox demonstrated the best overall performance, achieving a mean Uno's C-index of 0.879 across cohorts (training, 0.947; validation, 0.810). Time-dependent SurvSHAP(t) (survival Shapley additive explanations) was employed to provide instance-specific interpretability. An interactive web-based calculator implementing the Xgboost.Cox model was developed to facilitate clinical translation and individualized risk assessment. CONCLUSION AND EXPERT RECOMMENDATIONS: This 3PM/PPPM-guided framework enables early identification of high-risk phenotypes among NSTE-ACS patients with HFpEF and type 2 diabetes before irreversible clinical deterioration occurs. We recommend: (1) implementing multidimensional baseline profiling including inflammatory, metabolic, and cardiac functional bio-marker panels for the statin-treated patients with NSTE-ACS to facilitate risk stratification at discharge; (2) integrating the web-based XGBoost-Cox calculator into clinical workflows to generate individualized risk estimates and support shared decision-making; and (3) tailoring follow-up intensity and preventive strategies to predicted risk, with closer monitoring and more intensive GDMT optimization considered for high-risk patients, while ensuring guideline-recommended therapy for all eligible individuals. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at https://doi.org/10.1007/s13167-026-00467-2.