Abstract / Summary
Abstract Objective: This study aims to build a social frailty risk prediction model for cancer patients undergoing chemotherapy based on machine learning. Methods: From June 2025 to June 2026, 253 cancer patients who received their first chemotherapy at a certain tertiary grade A hospital in Shaanxi Province were selected as the research subjects. The relevant clinical data of the patients were collected, and the patients were randomly divided into a training set (178 cases) and a validation set (75 cases) in a ratio of 7:3. The outcome variable was whether social frailty occurred after the end of the chemotherapy cycle. Five machine learning algorithms, including logistic regression, random forest, XGBoost, decision tree, and naive Bayes, were used to build the prediction model. The performance of the model was evaluated using indicators such as the area under the receiver operating characteristic curve (AUC), Brier score, etc., and the optimal model was selected. The SHAP algorithm was used to conduct interpretive analysis of the optimal model. Results: Among the 253 cancer patients who received their first chemotherapy, 83 cases (32.81%) experienced social frailty after the end of the chemotherapy cycle.Among the five models, XGBoost exhibited high AUC (0.889), accuracy (0.867), sensitivity (0.708), and specificity (0.941), with the lowest Brier score (0.110). The SHAP values of XGBoost were calculated, and it was found that PHQ-9, PSSS, and Sleep disorder were the top 3 key characteristic variables for predicting social frailty in cancer patients undergoing chemotherapy. Conclusion: The social frailty risk prediction model for cancer patients undergoing chemotherapy based on XGBoost has good predictive ability and can provide a reference for optimizing the management of social frailty in cancer patients undergoing chemotherapy.