Challenges in interpreting Mendelian randomization studies with a disease as the exposure: Using COVID-19 liability studies as an exemplar.

Mendelian randomization (MR) studies using diseases as exposures are increasingly prevalent although any observed associations do not necessarily imply effect of diseases. To illustrate this challenge, we conducted a systematic review of MR studies focusing on COVID-19 consequence. We hypothesized if outcome genome-wide association studies (GWAS) were conducted before COVID-19 pandemic in late 2019, any observed associations in these studies were unlikely to be driven by COVID-19. We systematically searched PubMed, EMBASE, and MEDLINE for all MR studies published between 1 January 2019 and 20 May 2023. Inclusion criteria included MR studies which used COVID-19 as the exposure and designed to assess COVID-19's impact on health outcomes. We extracted relevant information, such as result interpretation and relevance assumption assessment. This review was registered at PROSPERO (CRD42023421079). Amongst 57 included studies, 45 studies used outcome GWAS published prior to 2019 whilst the remaining studies likely used outcome GWAS containing data collected before 2019. Relevance assumption was assessed mainly by p values. A total of 35 studies showed an association of COVID-19 liability with health outcomes. Regardless of the results, 45 studies attributed these as evidence (or lack of evidence) of COVID-19 consequence. In MR studies using disease liability as exposure, relevance assumption should consider the prevalence of the disease in the outcome GWAS in the context of 2 sample Mendelian randomization study rather than p values/F-statistic alone. Even when these are verified, these studies likely suffered from pleiotropy, making corresponding interpretation as effect of disease challenging.