Machine learning and SHAP method for fracture risk prediction in multiple myeloma patients
10.3969/j.issn.1673-9701.2025.24.001
- VernacularTitle:机器学习模型结合SHAP法预测多发性骨髓瘤患者骨折风险的研究
- Author:
Luxing WANG
1
;
Hui JIANG
Author Information
1. 浙江中医药大学护理学院,浙江 杭州 310053
- Publication Type:Journal Article
- Keywords:
Multiple myeloma;
Fracture risk;
Machine learning;
SHAP
- From:
China Modern Doctor
2025;63(24):1-5
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop and assess a machine learning model using the Shapley additive explanations(SHAP)method to predict fracture risk in multiple myeloma(MM)patients.Methods A retrospective study analyzed 181 MM patients in Zhejiang University Medical School Affiliated First Hospital from June 2021 to June 2024.Data included patient information,lab tests,medical history,and disease staging.Univariate analysis and recursive feature elimination(RFE)were employed for the purpose of variable selection.Predictive models were developed utilizing extreme gradient boosting(XGBoost),random forest(RF),light gradient boosting machine(LightGBM),and Logistic regression(LR).The performance of these models was evaluated through 5-fold cross-validation,and SHAP values were utilized to assess variable contributions in the optimal model.Results A total of 181 MM patients were included,with 50 in fracture group and 131 in non fracture group.RFE identified five key variables,notably including ferritin and B-type natriuretic peptide.The area under receiver operating characteristic curve values for the XGBoost,RF,LightGBM,and LR models were 0.861,0.846,0.755,and 0.780,respectively,with XGBoost demonstrating superior performance.SHAP analysis revealed that B-type natriuretic peptide was the most influential variable in the XGBoost model.Conclusion The XGBoost model demonstrates efficacy in predicting fracture risk among MM patients,with SHAP values enhancing its interpretability.
