Development and validation of a risk prediction model for osteoporosis in elderly patients with type 2 diabetes mellitus: a retrospective and multicenter study

Abstract

Abstract Background This study aimed to construct a risk prediction model to estimate the odds of osteoporosis (OP) in elderly patients with type 2 diabetes mellitus (T2DM) and evaluate its prediction efficiency. Methods This study included 21,070 elderly patients with T2DM who were hospitalized at six tertiary hospitals in Southwest China between 2012 and 2022. Univariate logistic regression analysis was used to screen for potential influencing factors of OP and least absolute shrinkage. Further, selection operator regression (LASSO) and multivariate logistic regression analyses were performed to select variables for developing a novel predictive model. The area under the receiver operating characteristic curve (AUROC), calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC) were used to evaluate the performance and clinical utility of the model. Results The incidence of OP in elderly patients with T2DM was 7.01% (1,476/21,070). Age, sex, hypertension, coronary heart disease, cerebral infarction, hyperlipidemia, and surgical history were the influencing factors. The seven-variable model displayed an AUROC of 0.713 (95% confidence interval [CI]:0.697–0.730) in the training set, 0.716 (95% CI: 0.691–0.740) in the internal validation set, and 0.694 (95% CI: 0.653–0.735) in the external validation set. The optimal decision probability cut-off value was 0.075. The calibration curve (bootstrap = 1,000) showed good calibration. In addition, the DCA and CIC demonstrated good clinical practicality. An operating interface on a webpage (https://juntaotan.shinyapps.io/osteoporosis/) was developed to provide convenient access for users. Conclusions This study constructed a highly accurate model to predict OP in elderly patients with T2DM. This model incorporates demographic characteristics and clinical risk factors and may be easily used to facilitate individualized prediction.