Overlooked Shortcomings of Observational Studies of Interventions in Coronavirus Disease 2019: An Illustrated Review for the Clinician

Abstract

The rapid spread of severe acute respiratory syndrome coronavirus 2 infection across the globe triggered an unprecedented increase in research activities that resulted in an astronomical publication output of observational studies. However, most studies failed to apply fully the necessary methodological techniques that systematically deal with different biases and confounding, which not only limits their scientific merit but may result in harm through misleading information. In this article, we address a few important biases that can seriously threaten the validity of observational studies of coronavirus disease 2019 (COVID-19). We focus on treatment selection bias due to patients' preference on goals of care, medical futility and disability bias, survivor bias, competing risks, and the misuse of propensity score analysis. We attempt to raise awareness and to help readers assess shortcomings of observational studies of interventions in COVID-19.

Keywords: bias; confounding; observational studies.

Figure 1.

A, Illustration of survivor bias due to misclassification of preintervention immortal time (time between admission to intervention). B, Illustration of appropriate classification of preintervention immortal time: absence of survivor bias.

Figure 2.

Modified from Sapir-Pichhadze et al [23]. A, Cause-specific hazard ratio (HR) for each event: the HR for death (the event of interest) and the HR for discharge (the competing event). These can be obtained from separate Cox proportional hazards regression models. This approach provides an etiological exploration of risk factors and shows how risk factors are associated with each event; direct and indirect effects can be distinguished. B, The subdistribution hazard function estimates the hazard rate for death at time t based on the risk set that remains at time t after accounting for all previously occurring event types, which includes competing events (death and discharge). As a time-averaged risk comparison, subdistribution HRs extend overall risk ratios. Abbreviations: CSH, cause-specific hazard; SDH, subdistribution hazard.