Collider bias undermines our understanding of COVID-19 disease risk and severity

Abstract

Observational data on COVID-19 including hypothesised risk factors for infection and progression are accruing rapidly, often from non-random sampling such as hospital admissions, targeted testing or voluntary participation. Here, we highlight the challenge of interpreting observational evidence from such samples of the population, which may be affected by collider bias. We illustrate these issues using data from the UK Biobank in which individuals tested for COVID-19 are highly selected for a wide range of genetic, behavioural, cardiovascular, demographic, and anthropometric traits. We discuss the sampling mechanisms that leave aetiological studies of COVID-19 infection and progression particularly susceptible to collider bias. We also describe several tools and strategies that could help mitigate the effects of collider bias in extant studies of COVID-19 and make available a web app for performing sensitivity analyses. While bias due to non-random sampling should be explored in existing studies, the optimal way to mitigate the problem is to use appropriate sampling strategies at the study design stage.

Utilization of COVID-19 Treatments and Clinical Outcomes among Patients with Cancer: A COVID-19 and Cancer Consortium (CCC19) Cohort Study

“…Confounding by severity is a concern in this population, as patients with increased baseline severity were more likely to be treated with one or more therapies. Collider bias and channeling associated with treatment may also affect assessment of the associations (35,36). Although adjustments and varying methodologic techniques were applied, residual confounding may affect the results, and causality cannot be established.…”

Patterns of COVID-19 testing and mortality by race and ethnicity among United States veterans: A nationwide cohort study

“…Strengths of this study included that it was based on well-annotated electronic health record data from a team with decades of experience using VA data, enabling a rapid and reliable analysis of COVID-19 outcomes by race and ethnicity. This analysis utilized patients' records from an entire healthcare system, which made it less prone to collider bias (i.e., non-random selection of individuals into a study) than other COVID-19 studies limited to individuals testing positive or admitted to hospital [21]. Unlike other nationwide healthcare systems, linkage to COVID-19 testing data or outcomes was not required as the integrated nature of VA healthcare provided at over 1,200 sites allows all information to be stored in its Corporate Data Warehouse.…”

Ethnic differences in COVID-19 infection, hospitalisation, and mortality: an OpenSAFELY analysis of 17 million adults in England

“…Due to the selection of people who have the opportunity to be tested, it is currently impossible to disentangle whether observed differences in outcomes are due to ethnicity or due to having received a test (collider bias). 28…”