Fig. 1. The effect of collider bias on COVID-19 epidemiology studies.

Collider bias occurs when both the risk factor or exposure of interest and the factors on the pathway to the outcome of interest influence the mechanisms behind selection into a study sample population. This can result in biased associations between the exposure and outcome. a | Collider bias can occur in studies of the association between kidney transplantation (the risk factor, red box) and death (the outcome) in people hospitalized with COVID-19 (the sample population). Hospitalization is related to unmeasured COVID-19 severity (blue circle). By restricting the sample population to those who are hospitalized (grey box), collider bias may alter associations between kidney transplantation and death that cannot be generalized to the wider population (dotted lines), because the indications for hospitalization may differ between transplant recipients and other patient groups. Similar problems arise when investigating associations in populations admitted to intensive care. b | Collider bias can also occur when investigations of long-term reductions in estimated glomerular filtration rate (eGFR) following SARS-CoV-2 infection are restricted to those with available eGFR measurements and SARS-CoV-2 test results (grey box). In such instances, infection is only partially observed as a consequence of limited access to testing (in most cases early in the pandemic, based on disease severity). Serum creatinine testing is also more likely in those who are at risk of declining kidney function (for example, people with diabetes or cardiovascular disease, or those on certain drugs), or in those at risk of, or suspected to have acute kidney injury (AKI). Collider bias can induce and/or alter associations between the variables (indicated by dotted lines). c | Autopsy studies in patients who have died with COVID-19 are also at risk of collider bias. As only people who died after developing COVID-19 are selected (grey box), and because pre-existing chronic kidney disease (CKD) is a risk factor for severe COVID-19 and death more generally, collider bias can alter associations between COVID-19 and histological features of CKD at autopsy (dotted lines).