Variability and agreement of frailty measures and risk of falls, hospital admissions and mortality in TILDA.

Little is known about the within-person variability of different frailty instruments, their agreement over time, and whether use of repeat assessments could improve the strength of associations with adverse health outcomes. Repeat measurements recorded in 2010-2011 (Wave 1) and 2012 (Wave 2) from The Irish Longitudinal Study on Ageing (TILDA) were used to classify individuals with frailty using the frailty phenotype (FP) and frailty index (FI). Within-person variability and agreement of frailty classifications were assessed using ANOVA and kappa (K) statistics, respectively. Associations of each frailty measure (wave 1, wave 2, or mean of both waves) with risk of falls, hospitalisations and all-cause mortality were assessed using logistic regression. Among 7455 individuals (mean age 64.7 [SD 9.9] years), within-person SD was 0.664 units (95% CI 0.654-0.671) for FP and 2 health deficits (SD 0.050 [0.048-0.051]) for FI. Agreement of frailty was modest for both measures, but higher for FI (K 0.600 [0.584-0.615]) than FP (K 0.370 [0.348-0.401]). The odds ratios (ORs) for all-cause mortality were higher for frailty assessed using the mean of two versus single measurements for FI (ORs for mortality 3.5 [2.6-4.9] vs. 2.7 [1.9-3.4], respectively) and FP (ORs for mortality 6.9 [4.6-10.3] vs. 4.0 [2.8-5.635], respectively). Frailty scores based on single measurements had substantial within-person variability, but the agreement in classification of frailty was higher for FI than FP. Frailty assessed using the mean of two or more measurements recorded at separate visits was more strongly associated with adverse health outcomes than those recorded at a single visit.