Abstract
INTRODUCTION: Risk assessment for high-risk populations is critical for preventing Type 2 Diabetes Mellitus (T2DM). Although China's public health services have continuously contributed to early grass-roots diagnosis of diabetes for years, universally applicable tools for identifying latent high-risk elderly populations urgently need to account for heterogeneity, robustness, and generalizability. Therefore, this study developed and validated the integrated Chinese Adapted Risk Evaluation for Diabetes Mellitus (iCARE-DM) model for elderly Chinese individuals. METHODS: The iCARE-DM model was developed based on pooled effect estimates from a meta-analysis of cohort studies that identified T2DM risk factors in East Asian populations and validated in three multicenter Chinese populations. Predictive performance was evaluated using area under the curve (AUC), sensitivity, specificity, accuracy, log-rank tests, and compared with the guideline-recommended model (i.e., New Chinese Diabetes Risk Score, NCDRS) as well as four machine learning (ML) models. RESULTS: The iCARE-DM model achieved AUC values of 0.741, 0.783, and 0.766, outperforming the NCDRS model by at least 12%. Although the best-performing ML model achieved AUC values comparable to the iCARE-DM model, its performance varied significantly across populations (with a range as high as 9%). Subgroup analyses of the iCARE-DM model confirmed consistent performance across age, gender and rural-urban groups. CONCLUSION: The iCARE-DM model demonstrated higher accuracy than the NCDRS model and exhibited superior robustness and generalizability compared to the ML models. The iCARE-DM model provides a robust, culturally adapted tool for T2DM risk assessment in elderly Chinese individuals.