Operationalization of Frailty Using Eight Commonly Used Scales and Comparison of Their Ability to Predict All‐Cause Mortality

TLDR: To operationalize frailty using eight scales and to compare their content validity, feasibility, prevalence estimates of frailty, and ability to predict all‐cause mortality, it is shown that the former are more reliable than the latter.

Abstract: Objectives: To operationalize frailty using eight scales and to compare their content validity, feasibility, prevalence estimates of frailty, and ability to predict all-cause mortality. Design: Secondary analysis of the Survey of Health, Ageing and Retirement in Europe (SHARE). Setting: Eleven European countries. Participants: Individuals aged 50 to 104 (mean age 65.3 ± 10.5, 54.8% female, N = 27,527). Measurements: Frailty was operationalized using SHARE data based on the Groningen Frailty Indicator, the Tilburg Frailty Indicator, a 70-item Frailty Index (FI), a 44-item FI based on a Comprehensive Geriatric Assessment (FI-CGA), the Clinical Frailty Scale, frailty phenotype (weighted and unweighted versions), the Edmonton Frail Scale, and the FRAIL scale. Results: All scales had fewer than 6% of cases with at least one missing item, except the SHARE-frailty phenotype (11.1%) and the SHARE-Tilburg (12.2%). In the SHARE-Groningen, SHARE-Tilburg, SHARE-frailty phenotype, and SHARE-FRAIL scales, death rates were 3 to 5 times as high in excluded cases as in included ones. Frailty prevalence estimates ranged from 6% (SHARE-FRAIL) to 44% (SHARE-Groningen). All scales categorized 2.4% of participants as frail. Of unweighted scales, the SHARE-FI and SHARE-Edmonton scales most accurately predicted mortality at 2 (SHARE-FI area under the receiver operating characteristic curve (AUC) = 0.77, 95% confidence interval (CI) = 0.75�0.79); SHARE-Edmonton AUC = 0.76, 95% CI = 0.74�0.79) and 5 (both AUC = 0.75, 95% CI = 0.74�0.77) years. The continuous score of the weighted SHARE-frailty phenotype (AUC = 0.77, 95% CI = 0.75�0.78) predicted 5-year mortality better than the unweighted SHARE-frailty phenotype (AUC = 0.70, 95% CI = 0.68�0.71), but the categorical score of the weighted SHARE-frailty phenotype did not (AUC = 0.70, 95% CI = 0.68�0.72). Conclusion: Substantive differences exist between scales in their content validity, feasibility, and ability to predict all-cause mortality. These frailty scales capture related but distinct groups. Weighting items in frailty scales can improve their predictive ability, but the trade-off between specificity, predictive power, and generalizability requires additional evaluation.