Evaluation of the Risk Prediction Model for Frailty in Diabetic Patients: A Systematic Review and Meta-Analysis

Frailty is highly prevalent among patients with diabetes and is associated with an increased risk of disability, hospitalization, and mortality. Although several frailty prediction models have been developed for this population, their methodological quality, predictive performance, and clinical applicability remain unclear. This systematic review and meta-analysis was therefore conducted to comprehensively evaluate existing prediction models for frailty in patients with diabetes.

PubMed, Web of Science, Embase, Cochrane Library, China National Knowledge Infrastructure (CNKI), Wanfang Data, China Biology Medicine Database (CBM), and China Science and Technology Journal Database (VIP) were searched for eligible prediction model studies from database inception to April 28, 2026. Two reviewers independently screened studies, extracted data, and assessed methodological quality using the CHARMS checklist and the PROBAST tool. Random-effects or fixed-effect models were applied according to heterogeneity. Meta-analyses of pooled model discrimination (area under the curve [AUC]) and common predictors were performed using RevMan 5.4 and MedCalc 23.6.1 software.

Of the 3492 identified articles, 19 studies were included, comprising a total of 36 prediction models. Sample sizes ranged from 152 to 1436 participants, and the AUC values varied from 0.703 to 0.975. The random forest model demonstrated the highest discriminative performance (AUC = 0.975). Frequently identified predictors included age, depression, activities of daily living (ADL), nutritional status, duration of diabetes, physical activity, polypharmacy, glycated hemoglobin (HbA1c), cognitive function, and marital status. All studies were judged to have a high risk of bias due to insufficient reporting of participants, predictors, outcomes, and analytical methods, although their overall applicability was considered high.

Existing frailty prediction models for patients with diabetes demonstrated good overall predictive performance and potential clinical utility. Nevertheless, substantial methodological limitations and a high risk of bias were identified across all included studies. Future model development should emphasize methodological rigor, external validation, and transparent reporting to improve reliability and facilitate clinical implementation.