Abstract
PURPOSE: Early clinical recognition of postmenopausal osteoporosis (PMOP) can be challenging. With the advancement of machine learning, several prediction models for PMOP have been developed. This study assessed their performance by carrying out a systematic review and meta-analysis. METHODS: The PubMed, Embase, Cochrane Library, Web of Science, China National Knowledge Infrastructure (CNKI), WanFang database and China Science and Technology Journal Database (VIP) were systematically searched. Studies with a sample size of at least 100 and involving postmenopausal women were included Included models were descriptively summarized, and meta-analyses were conducted to derive discrimination estimates. Homogeneous results from different studies were pooled using MedCalc software. RESULTS: Out of 37,115 identified studies, 21 were included. Most of the models were developed using data from cross-sectional studies and the sample size of included models ranged from 103 to 12,175, totaling 45,383 participants with 16,008 positive events. Several models contained some similar predictors, including age, prior fractures, and body mass index (BMI). We also conducted a meta-analysis that included 22 models with reported AUC and its 95% confidence interval (95% CI), which demonstrated that the prediction models have good discriminative performance. The most frequently observed predictive variables include age, weight, body mass index (BMI), menopause status, height, fracture history, lower limb cramps, fatigue, waist circumference，diabetes mellitus，hyperlipidemia and glucocorticoids. CONCLUSIONS: We found that the PMOP prediction models demonstrated promising performance. However, this review also highlights several potential limitations of current approaches, including a high risk of bias and limited external validation. Future research should aim to refine these models using larger and more diverse populations, as well as by incorporating additional risk factors to improve their clinical applicability.