Comparison of alternate scoring of variables on the performance of the frailty index

The frailty index (FI) is used to measure the health status of ageing individuals. An FI is constructed as the proportion of deficits present in an individual out of the total number of age-related health variables considered. The purpose of this study was to systematically assess whether dichotomizing deficits included in an FI affects the information value of the whole index.

Secondary analysis of three population-based longitudinal studies of community dwelling individuals: Nova Scotia Health Survey (NSHS, n = 3227 aged 18+), Survey of Health, Ageing and Retirement in Europe (SHARE, n = 37546 aged 50+), and Yale Precipitating Events Project (Yale-PEP, n = 754 aged 70+). For each dataset, we constructed two FIs from baseline data using the deficit accumulation approach. In each dataset, both FIs included the same variables (23 in NSHS, 70 in SHARE, 33 in Yale-PEP). One FI was constructed with only dichotomous values (marking presence or absence of a deficit); in the other FI, as many variables as possible were coded as ordinal (graded severity of a deficit). Participants in each study were followed for different durations (NSHS: 10 years, SHARE: 5 years, Yale PEP: 12 years).

Within each dataset, the difference in mean scores between the ordinal and dichotomous-only FIs ranged from 0 to 1.5 deficits. Their ability to predict mortality was identical; their absolute difference in area under the ROC curve ranged from 0.00 to 0.02, and their absolute difference between Cox Hazard Ratios ranged from 0.001 to 0.009.

Analyses from three diverse datasets suggest that variables included in an FI can be coded either as dichotomous or ordinal, with negligible impact on the performance of the index in predicting mortality.