The Quest for Timely Insights into COVID-19 Should not Come at the Cost of Scientific Rigor

Abstract

To the Editor: The OpenSAFELY study of “factors associated with COVID-19-related mortality” illustrates the potential of linked administrative data to provide timely insights into COVID-19.1 The study confirms that the risk of death from COVID-19 is higher in older people, Black and minority ethnic people, people living in poorer areas, and people with pre-existing conditions. Beyond this, however, the study reveals relatively little. It does not say whether these characteristics are due to greater disease susceptibility or severity, nor why people with these characteristics experienced higher risks. Understanding why different characteristics lead to different risks of a death or disease are causal questions that requires causal methods.2 Williamson et al. instead present a mutually “adjusted model” that is neither suitable for causal interpretation nor was sufficiently validated for use as a prognostic instrument.3 These limitations were raised by several (#EpiTwitter) peers in response to their preprint, including a public letter from Westreich et al.4 In the final paper, Williamson et al. responded by cautioning against “interpreting… (the) estimates as causal effects,”1 but neglected to clarify what, if anything, the reported estimates actually estimate or how the results should instead be interpreted. They also drew clear inferences about the contribution of pre-existing conditions and deprivation to the higher risk of death in Black and minority ethnic people. What we will call Schrodinger’s inferences like this—where the authors caution against causal interpretations while themselves offering causal interpretations—should be avoided by using clear language and appropriate methods,5 else confusion is inevitable.6 The quest for timely insights into COVID-19 should not come at the cost of scientific rigor. Many of the issues with Williamson et al. could have been avoided with more extensive and transparent engagement with feedback on their preprint. Journal editors and reviewers should consider ways to incorporate such feedback into the formal review process. Peter W.G. TennantLeeds Institute for Data AnalyticsFaculty of Medicine and HealthUniversity of Leeds, LeedsAlan Turing InstituteLondon, United Kingdom[email protected] Eleanor J. MurrayDepartment of EpidemiologyBoston University School of Public HealthBoston, MA

Topics & Concepts