Predicting poor response to anti-osteoporosis therapy: a machine learning model integrating clinical and novel
Yannan Bi1, Maolin Zhang1, Weiqiong Zhang1
1Department of Orthopedics, The Fourth Affiliated Hospital, Guangzhou Medical University, Guangzhou, Guangdong, China.
Objective:
This study was conducted to develop and validate a prediction model integrating clinical characteristics and novel biomarkers. The goal was to identify patients at high risk for a poor response to standard anti-osteoporosis therapy prior to treatment initiation, thereby supporting personalized therapeutic decision-making.
Methods:
A retrospective analysis was performed on 543 patients with primary osteoporosis admitted between January 2021 and December 2024. All patients received 12 months of standard treatment. Participants were randomly allocated to a training set (n = 380) and a validation set (n = 163) in a 7:3 ratio. In the training set, univariate analysis, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and multivariate logistic regression were used to determine independent predictors. Three machine learning models-Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)-were then constructed. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) values were used to interpret the optimal model.
Results:
Baseline characteristics were comparable between the training and validation sets (P > 0.05). Eight independent predictors of poor treatment response were identified: comorbid diabetes, history of fragility fracture, glucocorticoid use for ≥ 6 months, femoral neck bone mineral density T-score, and serum levels of osteocalcin, procollagen type I N-terminal propeptide, β-CrossLaps of type I collagen (β-CTX), and 25-Hydroxyvitamin D. Among the models, the RF algorithm demonstrated superior performance, with an AUC of 0.856 (95% CI: 0.808-0.905) in the training set and 0.825 (95% CI: 0.718-0.933) in the validation set. The model was well-calibrated, and DCA indicated a high net benefit. SHAP analysis confirmed serum β-CTX as the most significant predictive variable.
Conclusion:
A predictive model integrating multi-dimensional factors was successfully developed and validated for assessing osteoporosis treatment efficacy. The RF-based model exhibited robust predictive performance and clinical utility. It shows potential for pre-therapeutic identification of high-risk patients, facilitating precision management in osteoporosis.