Abstract
OBJECTIVE: To develop and validate a machine learning-based nomogram for predicting heart failure (HF) in elderly patients with relapsed/refractory multiple myeloma (RRMM) receiving carfilzomib-based therapy, facilitating early identification and individualized clinical management. METHODS: This retrospective study analyzed clinical data from 192 elderly RRMM patients treated with carfilzomib-based therapy at Shaanxi Provincial Cancer Hospital (from January 1, 2023, to December 31, 2024). Machine learning algorithms, including the Least Absolute Shrinkage and Selection Operator (LASSO) regression, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), were used for variable selection. Robust predictors identified through cross-model consistency evaluation and bootstrap resampling were incorporated into a nomogram. Model performance was assessed using concordance index (C-index), calibration curves, and decision curve analysis (DCA). RESULTS: HF occurred in 25.5% (49/192) of patients. Machine learning models consistently identified coronary artery disease (CAD), hypertension, renal insufficiency, and albumin (Alb) levels as significant HF risk factors. The nomogram showed good predictive performance (C-index: 0.780, 95% CI: 0.704-0.841), internal calibration (Hosmer-Lemeshow χ² = 1.334, P = 0.970), and external validation (Hosmer-Lemeshow χ² = 1.054, P = 0.788). DCA confirmed clinical utility across a wide range of threshold probabilities (1% to 83%), with a peak net benefit of 0.248. CONCLUSION: This study provides a practical nomogram for cardiovascular risk assessment in elderly RRMM patients receiving carfilzomib-based therapy, which may assist clinicians in early risk stratification and support tailored monitoring and management throughout treatment.