Examining gender-related differential item functioning of the Veterans Rand 12-item Health Survey.

Abstract

Previous research suggests that gender differences in patient-reported outcome measures (PROMs) may reflect measurement bias rather than true differences in underlying health status. The aim of this study is to examine whether the Veterans Rand 12-item Health Survey (VR-12) allows for unbiased comparisons of physical and mental health scores across gender. The VR-12 is a generic PROM consisting of 12 items with 3-6 response options for the measurement of mental and physical health. Study data were from the 2015 Health Outcomes Survey pertaining to the Medicare beneficiaries. A total of 277,518 participants included 116,817 (42.1%) males and 160,701 (57.9%) females. Scale-level and item-level differential functioning methods were applied using multiple-group confirmatory factor analysis and ordinal logistic regression, respectively. The scale-level differential functioning showed support for strict invariance (RMSEA=0.045; CFI=0.995) across gender. Although we found statistically significant differential item functioning for several items, the magnitude was negligible (maximum ΔR 2=0.007). The VR-12 physical and mental health status scores are unbiased with respect to gender.