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

Area of Science:

• Epidemiology
• Statistical Genetics
• Computational Biology

Background:

• Traditional epidemiological studies often focus on candidate risk factors, potentially introducing bias and overlooking novel associations.
• Genome-wide association studies (GWAS) excel at identifying genetic risk factors, but nongenetic risk factor discovery using big data lags.
• Modern biobanks offer vast potential risk factor data, necessitating advanced analytical methods to avoid bias and control for multiple testing.

Purpose of the Study:

• To implement a Doublethink-based exposome-wide association study (EWAS) to identify direct nongenetic risk factors for COVID-19 hospitalization.
• To leverage the UK Biobank dataset comprising 201,917 participants and 1,912 potential risk factors.
• To simultaneously control Bayesian False Discovery Rate (FDR) and frequentist Familywise Error Rate (FWER) using a novel hypothesis testing approach.

Main Methods:

• Utilized a Doublethink model-averaged hypothesis testing approach, incorporating Markov Chain Monte Carlo (MCMC) for analysis.
• Conducted an exposome-wide association study on 1,912 variables within the UK Biobank cohort.
• Focused on COVID-19 hospitalization data from the 2020 outbreak.

Main Results:

• Identified nine individual and seven groups of variables as exposome-wide significant for COVID-19 hospitalization.
• Found significant direct effects for factors including aging, dementia, and prior infection, alongside common factors like age, sex, and obesity.
• Observed that effects of hypertension, depression, and diabetes appeared mediated through general comorbidity, while cardiovascular disease did not show significant direct effects.

Conclusions:

• The Doublethink approach provides a powerful, agnostic method for identifying direct risk factors in large biobanks, controlling both Bayesian FDR and frequentist FWER.
• Overlooked factors like aging and dementia emerged as significant direct risk factors for COVID-19 hospitalization.
• This joint Bayesian-frequentist hypothesis testing framework offers flexible post hoc analysis and highlights the benefits of an unbiased, exposome-wide discovery strategy.