Collider Bias Undermines Our Understanding of COVID-19 Disease Risk and Severity

The document discussed how collider bias affected the understanding of COVID-19 disease risk and severity in observational studies. It highlighted that studies often used non-representative samples, such as hospital patients or volunteers, which could lead to distorted associations between variables due to collider bias. This bias occurred when variables influencing the likelihood of being sampled were also associated with the outcomes of interest. The analysis of UK Biobank data showed that participants tested for COVID-19 were selected based on various traits, complicating the interpretation of results. The document suggested that while existing studies should explore collider bias, the best way to address it was through appropriate sampling strategies during the study design phase.