How group structure impacts the numbers at risk for coronary artery disease: polygenic risk scores and non-genetic risk factors in the UK Biobank cohort.

The UK Biobank is a large cohort study that recruited over 500,000 British participants aged 40-69 in 2006-2010 at 22 assessment centres from across the UK. Self-reported health outcomes and hospital admission data are two types of records that include participants' disease status. Coronary artery disease (CAD) is the most common cause of death in the UK Biobank cohort. After distinguishing between prevalence and incidence CAD events for all UK Biobank participants, we identified geographical variations in age-standardised rates of CAD between assessment centres. Significant distributional differences were found between the pooled cohort equation scores of UK Biobank participants from England and Scotland using the Mann-Whitney test. Polygenic risk scores of UK Biobank participants from England and Scotland and from different assessment centres differed significantly using permutation tests. Our aim was to discriminate between assessment centres with different disease rates by collecting data on disease-related risk factors. However, relying solely on individual-level predictions and averaging them to obtain group-level predictions proved ineffective, particularly due to the presence of correlated covariates resulting from participation bias. By using the Mundlak model, which estimates a random effects regression by including the group means of the independent variables in the model, we effectively addressed these issues. In addition, we designed a simulation experiment to demonstrate the functionality of the Mundlak model. Our findings have applications in public health funding and strategy, as our approach can be used to predict case rates in the future, as both population structure and lifestyle changes are uncertain.