Identifying direct risk factors in UK Biobank via simultaneous Bayesian-frequentist model-averaged hypothesis testing using Doublethink

Big data approaches to discovering nongenetic risk factors have lagged behind genome-wide association studies that routinely uncover novel genetic risk factors for diverse diseases. Instead, epidemiology typically focuses on candidate risk factors. Since modern biobanks contain thousands of potential risk factors, candidate approaches may introduce bias, inadequately control for multiple testing, and overlook important signals. Doublethink, a model-averaged hypothesis testing approach, offers a solution that simultaneously controls the Bayesian false discovery rate (FDR) and frequentist familywise error rate (FWER) while accounting for uncertainty in variable selection. Here, we investigate direct risk factors for COVID-19 hospitalization from among 1,912 variables in 201,917 UK Biobank participants by implementing a Doublethink-based exposome-wide association study using Markov Chain Monte Carlo. Focusing on the 2020 outbreak, we find nine individual variables and seven groups of variables exposome-wide significant at 9% FDR and 0.05% FWER. We identify significant direct effects among relatively overlooked risk factors including aging, dementia, and prior infection, which we evaluate in relation to studies of other populations. We detect significant direct effects among some commonly reported risk factors like age, sex, and obesity, but not others like cardiovascular disease. The effects of hypertension, depression, and diabetes appeared to be mediated via general comorbidity. Doublethink produces interchangeable posterior odds and P -values for individual variables and arbitrary groups, facilitating flexible and powerful post hoc hypothesis testing. We discuss the potential for impact and limitations of joint Bayesian-frequentist hypothesis testing, including the benefits of an agnostic exposome-wide approach to discovery.