Machine learning-based prediction of fracture risk in patients with type 2 diabetes in Zhengzhou area
Objective To screen for fracture risk factors and apply various machine learning algorithms to the prediction of fracture risk in patients with type 2 diabetes,with the aim of achieving early detection and intervention for fracture risk in pa-tients with type 2 diabetes.Methods Using FRAX and fracture risk factors related to diabetes,risk prediction was carried out on 795 patients with type 2 diabetes complicated by osteoporosis from January 2018 to December 2022 included in the Stan-dardised-Metabolic Disease Management Center of People's Hospital of Zhengzhou,and different machine learning models were constructed to obtain different receiver operating characteristic(ROC)curve.Results Using the random forest algorithm to rank the importance of feature variables,the ranking of feature importance was obtained.All three machine earning models had certain effects on fracture risk prediction,among which the random forest and extreme gradient boosting(XGBoost)model performed the best,achieving accuracy of 0.94 and 0.93,respectively,and precision of 0.97 and 0.95,respectively,followed by the back propagation(BP)neural network,achieving accuracy and precision of 0.89 and 0.91,respectively.Conclusion Ranking the importance of feature variables reveal that the application of glucocorticoids,smoking and alcohol consumption,and family history are in the forefront,and the level of blood sugar control highlight its importance,with fasting blood sugar and glycated hemoglobin contributing significantly to fracture risk.The predictive effects of the three models all have high predictive value.
Machine learningType 2 diabetesFracture riskZhengzhou area
NETL
NSTL
万方数据
