Identification and risk prediction models for social frailty in older adults: a systematic review

Abstract Background Social frailty is closely associated with adverse health outcomes in older adults. Risk prediction models may facilitate its early identification; however, the methodological quality and clinical applicability of existing models remain unclear. This systematic review aimed to evaluate the methodological quality, predictive performance, and clinical applicability of identification and risk prediction models for social frailty in older adults. Methods Ten Chinese- and English-language databases, including the Cochrane Library, PubMed, Embase, CINAHL, Web of Science, Scopus, CNKI, Wanfang, CQVIP, and CBM, were searched from January 2017, when social frailty emerged as a distinct concept, to 30 June 2026. Eligible studies developed or validated identification or risk prediction models for social frailty in adults aged 60 years or older. Two reviewers independently screened studies and extracted data, with disagreements adjudicated by a third reviewer. Methodological quality was assessed with PROBAST, and the findings were synthesised narratively. Results Nine studies reporting 18 models and 4830 participants were included. Eight studies (88.9%) were cross-sectional and yielded identification models; only one was a prospective cohort study. Models were built with logistic regression or machine learning algorithms and included 5–14 final predictors, most frequently depression, educational level, age, family function, and social support. Internally validated AUCs were 0.771–0.993 for the identification models and 0.825 for the prognostic model; these were not pooled because their interpretations differ. Eight studies had insufficient outcome events (events per variable < 20); only two performed external validation, and only one reported the calibration intercept and slope. All nine studies were rated at high risk of bias, concentrated in the analysis domain; applicability concerns were low throughout. Conclusions The overall methodological quality of identification and risk prediction models for social frailty in older adults is suboptimal, with bias concentrated in the analysis domain. Future studies should adopt prospective cohort designs, adhere strictly to the TRIPOD reporting guideline, and strengthen external validation and clinical translation so that these models can move from research into practice. Registration PROSPERO registration number: CRD420261348102.