Exploring regression dilution bias using repeat measurements of 2858 variables in ≤49 000 UK Biobank participants

Charlotte E Rutter; Louise A C Millard; Maria Carolina Borges; Deborah A Lawlor

doi:10.1093/ije/dyad082

Exploring regression dilution bias using repeat measurements of 2858 variables in ≤49 000 UK Biobank participants

Int J Epidemiol. 2023 Oct 5;52(5):1545-1556. doi: 10.1093/ije/dyad082.

Authors

Charlotte E Rutter^{1

2}, Louise A C Millard^{1

3}, Maria Carolina Borges^{1

3}, Deborah A Lawlor^{1

3

4}

Affiliations

¹ MRC Integrative Epidemiology Unit, University of Bristol, Bristol, UK.
² Department of Medical Statistics, London School of Hygiene and Tropical Medicine (LSHTM), London, UK.
³ Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, UK.
⁴ Bristol NIHR Biomedical Research Centre, University of Bristol, Bristol, UK.

Abstract

Background: Measurement error in exposures and confounders can bias exposure-outcome associations but is rarely considered. We aimed to assess random measurement error of all continuous variables in UK Biobank and explore approaches to mitigate its impact on exposure-outcome associations.

Methods: Random measurement error was assessed using intraclass correlation coefficients (ICCs) for all continuous variables with repeat measures. Regression calibration was used to correct for random error in exposures and confounders, using the associations of red blood cell distribution width (RDW), C-reactive protein (CRP) and 25-hydroxyvitamin D [25(OH)D] with mortality as illustrative examples.

Results: The 2858 continuous variables with repeat measures varied in sample size from 109 to 49 121. They fell into three groups: (i) baseline visit [529 variables; median (interquartile range) ICC = 0.64 (0.57, 0.83)]; (ii) online diet by 24-h recall [22 variables; 0.35 (0.30, 0.40)] and (iii) imaging measures [2307 variables; 0.85 (0.73, 0.94)]. Highest ICCs were for anthropometric and medical history measures, and lowest for dietary and heart magnetic resonance imaging.The ICCs (95% confidence interval) for RDW, CRP and 25(OH)D were 0.52 (0.51, 0.53), 0.29 (0.27, 0.30) and 0.55 (0.54, 0.56), respectively. Higher RDW and levels of CRP were associated with higher risk of all-cause mortality, and higher concentration of 25(OH)D with lower risk. After correction for random measurement error in the main exposure, the associations all strengthened. Confounder correction did not influence estimates.

Conclusions: Random measurement error varies widely and is often non-negligible. For UK Biobank we provide relevant statistics and adaptable code to help other researchers explore and correct for this.

Keywords: Regression dilution bias; UK Biobank; attenuation by errors; concordance; random measurement error; random variability.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Bias
Biological Specimen Banks*
Diet*
Humans
United Kingdom / epidemiology

Abstract

Publication types

MeSH terms

Grants and funding