Predictive Accuracy of Frailty Tools for Adverse Outcomes in a Cohort of Adults 80 Years and Older: A Decision Curve Analysis

Abstract Objectives To compare the predictive performance of 3 frailty identification tools for mortality, hospitalization, and functional decline in adults aged ≥80 years using risk reclassification statistics and decision curve analysis. Design Population-based, prospective cohort. Setting BELFRAIL study, Belgium. Participants 560 community-dwelling adults aged ≥80 years. Measurements Frailty by Cardiovascular Health Study (CHS) phenotype, Longitudinal Aging Study Amsterdam (LASA) markers, and Groeningen Frailty Indicator (GFI); mortality until 5.1 ± 0.25 years from baseline and hospitalization until 3.0 ± 0.25 years; and functional status assessed by activities of daily living at baseline and after 1.7 ± 0.21 years. Results Frailty prevalence was 7.3% by CHS phenotype, 21.6% by LASA markers, and 22% by GFI. Participants determined to be frail by each tool had a significantly higher risk for all-cause mortality and first hospitalization. For functional decline, only frail by GFI had a higher adjusted odds ratio. Harrell 's C-statistic for mortality and hospitalization and area under receiver operating characteristic curve for functional decline were similar for all tools and Conclusions and Implications In a cohort of adults aged ≥80 years, 3 frailty tools based on different conceptualizations and assessment sources had comparable but unsatisfactory discrimination for predicting mortality, hospitalization, and functional decline. All showed clinical utility for predicting these outcomes over relevant risk thresholds, but none was significantly superior. Future research on frailty tools should include a focus on the specific group of adults aged ≥80 years, and the predictive accuracy for adverse outcomes of different tools needs a comprehensive assessment that includes decision curve analysis.