Validating and Leveraging Non-SARS-CoV-2 Respiratory Infection as a Negative Control Outcome in a Phase 3 COVID-19 Vaccine Trial with Extended Observational Follow-Up

Negative control outcomes (NCOs) are useful tools for hidden bias detection, but empirical evidence validating NCOs for COVID-19 is lacking. To address this gap, we examined the blinded phase of the COVE study of the two-dose mRNA-1273 COVID-19 vaccine, which used symptom-triggered multiplex PCR testing to identify infections with SARS-CoV-2 and 20 other respiratory pathogens. We confirmed that acute respiratory infection (ARI) caused by non-SARS-CoV-2 pathogens was unaffected by vaccination yet positively associated with COVID-19 risk, highlighting the utility of non-SARS-CoV-2 ARI as a valid negative control. Subsequently, we leveraged non-SARS-CoV-2 ARI to detect bias in hazard-based, time-varying vaccine efficacy (VE) estimates in the blinded phase and an observational, open-label extension of the trial which evaluated a single mRNA-1273 booster. While the blinded phase analysis yielded minimal evidence of depletion of susceptibles bias, the booster phase analysis suggested substantial downward bias in time-varying VE estimates which may be due to some combination of latent confounding or selection bias. Our study suggests that ARIs caused by irrelevant respiratory pathogens are valuable bias detection tools in analysis of COVID-19 vaccines disposed to confounding and selection effects.

Our paper was accepted in American Journal of Epidemiology (to appear soon).