Risk prediction models for oral frailty in older adults: a scoping review

Oral frailty reflects age-related decline in oral function and is associated with systemic functional decline and adverse outcomes in older adults. Accurate risk identification may support nursing assessment, risk stratification, and targeted intervention. However, the quality, validation status, and transportability of existing oral frailty prediction models remain unclear.

This scoping review aimed to systematically map existing oral frailty risk prediction models for older adults, with particular attention to model development methods, outcome definitions, retained predictors, validation strategies, predictive performance, risk of bias, and applicability to clinical or community settings.

PubMed, Web of Science Core Collection, Embase, CINAHL, ProQuest, the Cochrane Library, China National Knowledge Infrastructure, Wanfang, VIP, and SinoMed were searched from inception to 10 April 2026. Eligible studies developed, validated, or updated multivariable prediction models for oral frailty in older adults. Data extraction was guided by CHARMS, and risk of bias and applicability were assessed using PROBAST.

Seventeen studies were included. Most studies were cross-sectional and geographically concentrated in China, and oral frailty definitions and assessment approaches varied across studies. The reported proportion of oral frailty ranged from 25.50 to 92.50%. Across the included studies, logistic regression formed the basis of model development, while nomograms were generally used to display the final prediction tools. Frequently retained predictors included age, nutrition-related factors, swallowing difficulty or choking, frailty or physical frailty, and denture-related factors. Although 15 studies reported some form of internal validation, external validation was available in only two. The reported area under the receiver operating characteristic curve ranged from 0.725 to 0.985. Although most studies reported calibration assessment and several evaluated clinical utility, all studies were judged to have a high overall risk of bias.

Existing models showed acceptable to excellent apparent discrimination, but clinical use remains constrained by heterogeneous outcome definitions, a predominance of cross-sectional designs, high risk of bias, limited external validation, and uncertain generalizability. Future studies should adopt prospective multicenter designs, standardize outcome definitions and reporting, define prediction horizons, and perform external validation before routine implementation.

The systematic review has been registered on the Open Science Framework. The unique identifier is 10.17605/OSF.IO/K2GM4 (https://doi.org/10.17605/OSF.IO/K2GM4).