Development and validation of a machine learning-based risk prediction model for cancer-related fatigue in ovarian cancer patients.

Background: Cancer-related fatigue (CRF) substantially compromises quality of life in ovarian cancer, yet reliable early detection tools remain inadequate. This study sought to develop a machine learning-based predictive model for CRF risk. Methods: We consecutively recruited 407 ovarian cancer patients from three tertiary hospitals in Lanzhou, China (October 2024-August 2025). Data were randomly partitioned into training (70%) and testing (30%) sets. Least Absolute Shrinkage and Selection Operator (LASSO) regression was applied for feature selection. Seven machine learning algorithms were developed, with the optimal model selected through comparative evaluation and subjected to SHAP interpretability analysis. Results: CRF prevalence was 39.6%. The Support Vector Machine (SVM) demonstrated superior overall predictive performance: AUC of 0.884, accuracy of 0.829, sensitivity of 0.816, specificity of 0.838, and F1 score of 0.792; good calibration (Brier score = 0.132); and decision curve analysis showed the highest net benefit across a wide range of threshold probabilities (0.05-0.85), indicating strong clinical utility. SHAP analysis identified serum calcium level, anxiety-depression status, red blood cell count, education level, cancer stage, medical payment method, and marital status as top predictive features. Conclusions: The SVM model exhibits robust predictive efficacy and good clinical utility, serving as a valuable tool for CRF risk stratification in ovarian cancer care. Early identification of high-risk patients enables targeted interventions to improve outcomes.