Big Data/Machine Learning/AI
BMI polygenic risk score as a tool for quantifying the magnitude of bias in the association between cross-sectional measures of BMI and osteoarthritis in the UK Biobank Patrick Carry* Patrick Carry Jyothi Lokanadham Mike Zuscik Cheryl Ackert-Bicknell
Osteoarthritis (OA) is associated with long-term disability and reduced quality of life. Due to lack of preventative interventions, there is a critical need to identify modifiable risk factors. Obesity has been hypothesized as a risk factor but, associations between BMI and OA are often measured in cross-sectional studies that are prone to bias due to residual confounding and reverse causation. Genetic measures of BMI may provide an unbiased estimate of the association between BMI and OA. The purpose of this study was to quantify differences in the strength of association between genetic measures of BMI and OA vs cross-sectional measures of BMI and OA.
We queried the UK Biobank (UKB) to identify individuals with health outcome and genetic data. Cross-sectional BMI was based on BMI at enrollment. Genetic BMI was based on a BMI polygenic risk score (PRS) from prior meta-analyses. Multivariable logistic regression models, adjusted for genetic batch, ancestry, age, prior joint injury, sex, and community economic deprivation (Townsend index) were used to test the association between BMI and OA. We also included interaction terms between BMI and covariates. We compared odds ratios (ORs) representing OA and cross-sectional BMI (biased) vs OA and PRS BMI (unbiased). All ORs represent odds of OA per 1 standard deviation increase, facilitating direct comparisons between the BMI measures.
There was a significant association between the BMI PRS and knee OA (OR: 1.22, 95% CI: 1.20-1.24) and hip OA (OR: 1.13, 95% CI: 1.11-1.32). Cross-sectional BMI ORs were higher than PRS BMI for knee OA (OR: 1.81, 95% CI: 1.78-1.84, +48%) and hip OA (OR: 1.27, 95% CI: 1.25-1.30, +13%). In the subgroup analysis, ORs for cross-sectional BMI were consistently higher for knee OA (range: +29% to +52%) and hip OA (range: +6% to +22%, Figure 1).
There was a strong interrelationship between BMI, affected joint, and demographics. Cross-sectional BMI consistently overestimated the association between BMI and OA. Bias, as represented by percent difference in genetic vs cross-sectional BMI, was highest in relation to knee OA, older age, male sex, and higher affluence. Genetic based measures of exposures are a useful tool for quantifying the magnitude of bias in the strength of the association between environmental risk factors and disease in cross-sectional studies.
