Construction of a novel online calculator for prediction of osteoporosis risk in Chinese type 2 diabetes patients

Background

Type 2 diabetes (T2D) has been established as an independent risk factor for osteoporosis, often resulting in a poor prognosis. Thus, it is crucial for clinicians to diagnose osteoporosis in diabetic patients. This study aimed to develop a prediction model for osteoporosis in people with T2D from China.

Methods

A clinical analysis was retrospectively carried out using our hospital database for patients with definite T2D diagnosed between January 1, 2012, and December 31, 2020. All patients were randomly divided into a training set (70 %) and a test set (30 %). Then, univariate and multivariate logistic regression analyses were used to screen independent risk factors for osteoporosis. Machine learning(ML) models were developed to predict osteoporosis risk using different methods such as logistic regression (LR), naive Bayes (NB), neural network (NNET), support vector machine (SVM), gradient boosting machine (GBM), and k-nearest neighbor (KNN). In addition, Shapley additivity explanations (SHAP) were employed to determine the significance of selected features and interpret the ML models.

Results

A total of 2029 patients were enrolled in the study, of which 457 suffered from osteoporosis. Based on the analysis, five characteristic variables were selected to construct the predictive model for osteoporosis in diabetics, comprising gender, age, BMI, heart rate, and alkaline phosphatase. The GBM model revealed an AUC of 0.79 in the test set and 0.89 in the external validation set. Furthermore, the calibration curves, decision curve analysis, and precision-recall curves highlighted the satisfactory clinical applicability of the GBM model. According to this model, an online calculator was built for clinicians to diagnose diabetes-related osteoporosis patients.

Conclusion

Age, sex, BMI, heart rate, and ALP are significantly associated with osteoporosis in people with T2D. The screening model provides an accurate, user-friendly, and low-cost tool for the early diagnosis of osteoporosis in people with T2D from China.