Preserving the discreteness of deficits during coding leads to a lower frailty index in individuals living in long-term care

Abstract

This study assesses two coding approaches on the frailty index (FI).

Two FI were calculated using 43 variables from 29,758 older adults (84.6 ± 8 years old; 64 % female) in long-term care. Scores were coded as 0, 0.5, or 1 regardless of the number of levels (grouped), or preserved (e.g., a 4 level variable was coded as 0, 0.33, 0.67, or 1; discrete). Grouped and discrete FI were compared with each ordinal variable removed but all other ordinal variables included. This was repeated until 28 unique (14 grouped, 14 discrete) FI had been constructed each with one ordinal variable removed per FI. FI was correlated to age and mortality separated by sex.

The median grouped (0.302 (0.221–0.372)) was higher relative to the discrete (0.237 (0.170–0.307)) FI. The discrete (r = 0.91, r = 0.87) and grouped (r = 0.93, r = 0.87) FI showed similar relationships to age and mortality. Removal of any ordinal variable reduced grouped FI by 0.004 or 0.016, whereas removal led to both increases (range: 0.003–0.001) and reductions (range: 0.002–0.008) for discrete FI.

A grouped approach inflates FI. A discrete approach provides a more accurate measure of frailty.