Calibration of the Active Australia questionnaire and application to a logistic regression model

Abstract

Objectives: To estimate the extent of measurement error in the Active Australia questionnaire, and to examine the impact of measurement error on the association of moderate-vigorous physical activity (MVPA) with obesity.

Design: Accelerometer Validation Study, cross-sectional; data from the third wave of a prospective cohort (Australian Diabetes, Obesity, and Lifestyle (AusDiab) Study)).

Methods: Self-reported physical activity data were obtained from 4005 participants of the third wave of the AusDiab study via the Active Australia questionnaire. Accelerometer-derived physical activity data were obtained from a subsample of 670 participants. Validity coefficients and attenuation factors were estimated from a measurement error model. A regression calibration method was applied to a logistic regression model examining the association between self-reported MVPA and obesity to adjust observed odds ratios (OR) for measurement error.

Results: The validity coefficient was 0.35 (0.28, 0.43) and the attenuation factor was 0.16 (0.13, 0.20) in models adjusted for age and sex. The uncorrected OR for obesity for 210min/week of MVPA (50th percentile) relative to 80min/week (25th percentile) was 0.87 (0.85, 0.90). The attenuation factor was used to adjust this OR for measurement error, giving a corrected OR of 0.43 (0.32, 0.55).

Conclusions: Substantial measurement error (relative to accelerometry) was evident in the Active Australia questionnaire, leading to attenuation of the association of MVPA with obesity. A regression-calibration method can be used to adjust risk estimates for associations between self-reported MVPA and health-related outcomes for measurement error specific to self-report. These corrected risk estimates reflect associations that would be expected if MVPA were measured by accelerometry.

Keywords: Accelerometry; Active Australia questionnaire; Measurement error; Regression calibration.