Risk prediction models for frailty in older adults: A systematic review and critical appraisal

Abstract

Background

Frailty can lead to increased adverse health outcomes in older adults. Risk prediction models for frailty have benefits in guiding the prevention. Studies have increasingly focused on the development of risk prediction models for frailty in older adults. The quality and clinical applicability of these models remain unknown.

Objectives

To systematically review and critically appraise the current risk prediction models for frailty in older adults.

Methods

PubMed, Embase, CINAHL, and Cochrane Library were searched from inception to June, 2024 to identify published studies focusing on developing or validating risk prediction models for frailty in older adults. Data extraction was independently conducted by two reviewers based on the checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies. The quality of included models was assessed using the Prediction Model Risk of Bias Assessment Tool.

Results

Of 5421 retrieved studies, 19 studies with 22 risk prediction models for frailty were included. The included models focused on community-dwelling and hospitalized older adults. Logistic regression and machine learning methods were employed to develop risk prediction models. The frequently used predictors were age (77.3 %), cognitive function (31.8 %), self-rated health (27.3 %), sex (22.7 %), activities of daily living (22.7 %), and depression (22.7 %). Internal and external validation were conducted in 17 (77.3 %) and four (18.2 %) models, respectively. Twenty-one (95.5 %) models evaluated model discrimination, with the AUC or c-index ranging from 0.707 to 0.920 in the internal validation and from 0.612 to 0.889 in the external validation. Fifteen (68.2 %) models assessed model calibration using the calibration curve, Hosmer-Lemeshow test, and Brier score and showed good calibration. All risk prediction models had high risk of bias primarily due to problems in the analysis domain and nine (40.9 %) models had high concern regarding applicability.

Conclusions

Current risk prediction models for frailty in older adults demonstrated poor validation and evaluation. Future research should focus on improving current models to aid their implementation and developing and validating new models with rigorous methodology.

PROSPERO registration

CRD42024582883.