骨质疏松性椎体压缩性骨折的风险因素分析及预测模型构建OA

Objective To develop a risk prediction model for osteoporotic vertebral compression fracture(OVCF)using machine learning(ML)techniques and to interpret its decision-making mechanism,and thereby facilitating early risk identification of OVCF.Methods A total of 188 patients confirmed with OVCF via lumbar MRI at General Hospital of Northern Theater Command from January 2022 to December 2024 were enrolled as the case group,while 394 patients without fractures during the same period were included as the control group.Data were integrated using R language and randomly divided into training cohort(408 patients)and validation cohort(174 patients)at a ratio of 7∶3.The key predictive factors were selected to construct ten machine learning models.The optimal model was analyzed in combination with SHAP to visualize model predictions,and interpret its decision-making logic.Results Univariate and multivariate analysis identified that gender,history of trauma,age,history of alcohol consumption,and body mass index(BMI)were the predictive factors for OVCF.Ten machine learning models demonstrated that the Logistic regression model achieved the highest value of area under the receiver operating characteristic curve(AUC)(AUC=0.851,95%CI:0.790 to 0.911),with stable standardized net benefits across the risk threshold range and high consistency between the predicted probabilities and actual probabilities.The Logistic regression model was identified as the optimal model.Additionally,SHAP-based visualization model revealed that the history of trauma,gender,history of alcohol consumption,age,and BMI were the key factors.Conclusion Machine learning-based models can significantly improve the diagnostic efficacy of OVCF,and Logistic regression combined with a nomogram tool is conducive to clinical decision support.