Modifiers of COVID-19 vaccine efficacy: Results from four COVID-19 prevention network efficacy trials

Abstract

Questions remain regarding the effect of baseline host and exposure factors on vaccine efficacy (VE) across pathogens and vaccine platforms. We report placebo-controlled data from four Phase 3 COVID-19 trials during the early period of the pandemic. This was a cross-protocol analysis of four randomized, placebo-controlled efficacy trials (Moderna/mRNA1273, AstraZeneca/AZD1222, Janssen/Ad26.COV2.S, and Novavax/NVX-CoV2373) using a harmonized design. Trials were conducted in the United States and international sites in adults ≥ 18 years of age. VE was assessed for symptomatic and severe COVID-19. We analyzed 114,480 participants from both placebo and vaccine arms, enrolled July 2020 to February 2021, with follow up through July 2021. VE against symptomatic COVID-19 showed little heterogeneity across baseline socio-demographic, clinical or exposure characteristics, in either univariate or multivariate analysis, regardless of vaccine platform. Similarly, VE against severe COVID-19 in the single trial (Janssen) with sufficient endpoints for analysis showed little evidence of heterogeneity. COVID-19 VE is not influenced by baseline host or exposure characteristics across efficacy trials of different vaccine platforms and countries when well matched to circulating virus strains. This supports use of these vaccines, regardless of platform type, as effective tools in the near term for reducing symptomatic and severe COVID-19, particularly for older individuals and those with common co-morbidities during major variant shifts. Clinical trial registration numbers: NCT04470427, NCT04516746, NCT04505722, and NCT04611802.

Keywords: Comorbidity; Effect modifier; Environmental exposure; Epidemiologic; Occupational exposure; SARS-CoV-2; Vaccine efficacy.

Fig. 1

Estimates of vaccine efficacy against symptomatic Covid-19 within subgroups defined by categorical baseline covariates, with corresponding 95 % confidence intervals. In the AstraZeneca an Novavax studies, the number of events among Black/African-American participants was too low to estimate VE.

Fig. 2

Estimated vaccine efficacy by age and body mass index (BMI) across the four trials, with corresponding 95 % confidence intervals. Estimates and intervals are derived according to a working model that enforced that the relative risk of a Covid-19 endpoint on vaccine versus on placebo must be log-linear.

Fig. 3

Estimated tertiles of vaccine efficacy (VE) against Covid-19 endpoint as defined by all baseline covariates, with corresponding 95 % confidence intervals (CIs). For each vaccine, vaccine efficacy was first predicted using all baseline covariates, and then participants were broken into three subgroups as defined by this prediction: the 33·33 % with lowest predicted VE (Lowest VE), the 33·33 % with the next lowest predicted VE (Middle VE), and the 33·33 % with the highest predicted VE (Highest VE). Vaccine efficacy was then estimated in these subgroups using a cross-validation method. If needed, estimates were projected to satisfy the population-level constraint that efficacy should be nondecreasing when moving from the first to the third tertile of participants.