Abstract

Background. As the world population continues to age, there is an increasing need to identify elderly patients at risk of developing frailty to provide early, effective interventions. During the past few years, several risk prediction models have been developed to assess the risk of developing frailty in older adults. However, to our knowledge, no comprehensive scoping review has been conducted to assess the quality and performance of these models. Therefore, in this scoping review, we aimed to comprehensively compare the methods used to construct, validate, and assess the performance of geriatric frailty risk prediction models.

Methodology. A systematic literature search of Chinese and English databases was performed to identify all relevant studies evaluating predictive frailty risk models. The study characteristics, modeling technique, model predictors, testing methods, and performance were extracted from the research articles. This study was conducted according to the methodological framework proposed by Arksey and O'Malley and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) specifications. This scoping review was registered on the Open Science Framework website.

Results. A total of 17 studies involving 18 predictive models of frailty risk in older adults were included. All studies were published between 2018 to 2024. Most of the studies (n= 14) were conducted in China and within a community research setting (n= 13). The majority of the studies involved multicenter studies (n=15). Thirteen out of the 18 models were developed using regression techniques, while the remaining 5 employed machine learning algorithms. Age was used as a predictor in 11 out of the 18 models. In terms of model performance, 15 out of the 18 models reported measures of discrimination (AUC: 0.7-0.99), and 8 reported calibrations. For model evaluation, 13 models underwent internal validation, while only 5 were externally validated.

Conclusions. While the proposed models had a high AUC and some predictive power, the number of predictors used to develop the model was limited, and most studies lacked external validation. The higher quality predictive models should be developed by following the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis + Artificial Intelligence (TRIPOD+AI) recommendations.