Construction and verification of risk prediction model of osteoporotic fractures in patients with osteoporosis in China

Objective: To explore the influencing factors of osteoporotic fractures (OPF) in patients with osteoporosis, construct a prediction model, and verify the model internally and externally, so as to provide reference for early screening and intervention of OPF in patients with osteoporosis. Methods: Osteoporosis patients in the First Affiliated Hospital of Soochow University were selected, and the medical records of patients were consulted through the Hospital Information System (HIS) and the data management platform of osteoporosis patients, so as to screen patients who met the criteria for admission and discharge and collect data. SPSS 26.0 software was used for single factor analysis to screen statistically significant variables (p < 0.05). The influencing factors of OPF were determined by multivariate analysis, and a binary Logistic regression model was established according to the results of multivariate analysis. Hosmer-Lemeshow (H-L) goodness of fit and receiver operating characteristic curve (ROC) were used to test the model's efficiency, and Stata 16.0 software was used to verify the Bootstrap model, draw the model calibration curve, clinical applicability curve and nomogram. Results: In this study, the data of modeling set and verification set were 1,435 and 580, respectively. There were 493 (34.4%) cases with OPF and 942 (65.6%) cases without OPF in the modeling set. There were 204 (35.2%) cases with OPF and 376 (64.8%) cases without OPF. The variables with statistically significant differences in univariate analysis are Age, BMI, History of falls, Usage of glucocorticoid, ALP, Serum Calcium, BMD of lumbar, BMD of feminist neck, T value of feminist neck, BMD of total hip and T value of total hip. The area under ROC curve of the risk prediction model constructed this time is 0.817 [95%CI (0.794 similar to 0.839)], which shows that the model has a good discrimination in predicting the occurrence of OPF. The optimal threshold of the model is 0.373, the specificity is 0.741, the sensitivity is 0.746, and the AUC values of the modeling set and the verification set are 0.8165 and 0.8646, respectively. The results of Hosmer and Lemeshow test are modeling set: (chi(2) = 6.551, p = 0.586); validation set: [(chi(2) = 8.075, p = 0.426)]. The calibration curve of the model shows that the reference line of the fitted curve and the calibration curve is highly coincident, and the model has a good calibration degree for predicting the occurrence of fractures. The net benefit value of the risk model of osteoporosis patients complicated with OPF is high, which shows that the model is effective. Conclusion: In this study, a OPF risk prediction model is established and its prediction efficiency is verified, which can help identify the high fracture risk subgroup of osteoporosis patients in order to choose stronger intervention measures and management.