Prognostic Risk Assessment of HFpEF: A Study on Feature Selection and Prediction Model Optimization Based on Machine Learning

Background Heart failure with preserved ejection fraction (HFpEF) is common but its prognostic factors remain poorly understood. Identifying key determinants of poor outcomes and the role of treatment adherence is critical for improving patient management. Methods and Results This study followed 461 HFpEF patients for one year. LASSO regression was used for feature selection, and predictive models were developed using decision trees, random forests, XGBoost, and a stacking ensemble approach. Structural equation modeling (SEM) was used to assess the impact of treatment adherence on prognosis. Results Among the participants, 38.2% (n = 176) experienced adverse outcomes, including death and major adverse cardiovascular events (MACE). Multivariate analysis identified anemia, atrial fibrillation, and elevated blood urea nitrogen as major risk factors. Higher BMI, plasma albumin, and good treatment adherence were found to be protective factors. The XGBoost model demonstrated the highest performance (AUC = 0.861), and the stacked ensemble model further improved accuracy (AUC = 0.864). SEM showed that treatment adherence significantly influenced prognosis, with inflammatory status, metabolic disorders, and BMI acting as mediating factors. Conclusions Treatment adherence plays a critical role in the prognosis of HFpEF. The stacked ensemble model enhances predictive accuracy, highlighting the importance of adherence interventions and metabolic management to improve long-term outcomes in HFpEF patients.