Potential “Healthy Vaccinee Bias” in a Study of BNT162b2 Vaccine against Covid-19

Source: Jorunal of New England Medicine

Add higher baseline mortality for the placebo group to the list of biases in observational studies

Using observational methods, Arbel et al. (Dec. 23, 2021, issue)1 calculated an adjusted 90% lower mortality due to Covid-19 among participants who received a first BNT162b2 vaccine (Pfizer–BioNTech) booster than among those who did not receive a booster. They found 65 Covid-19–associated deaths (reported as 0.16 per 100,000 persons per day) among participants in the booster group and 137 (reported as 2.98 per 100,000 persons per day) among those in the nonbooster group — a 94.6% difference. In a subsequent letter (March 10, 2022, issue),2 Arbel et al. reported 441 deaths not related to Covid-19 in the booster group and 963 deaths not related to Covid-19 in the nonbooster group.

We did not have access to the data and could not account for the timing of the receipt of boosters or adjust for the covariates included in the analyses. However, using the person-days of exposure included in the 2021 article by Arbel et al. and the deaths not related to Covid-19 reported in the subsequent letter, we estimated the mortality not related to Covid-19, according to vaccination status, with the following formula: the ratios of total deaths not related to Covid-19 to Covid-19–related deaths, according to vaccination group, multiplied by mortality due to Covid-19, according to vaccination group, which accounts for person-days of exposure. The mortality not related to Covid-19 was calculated as (441/65)×0.16=1.09 per 100,000 persons per day in the booster group as compared with (963/137)×2.98=20.95 per 100,000 persons per day in the nonbooster group. This corresponds to a 94.8% lower mortality not related to Covid-19 among participants in the booster group and indicates a markedly lower incidence of adverse health outcomes in the booster group.

Underlying health plays a substantial role in Covid-19–related mortality. The unadjusted differences in mortality related to Covid-19 and mortality not related to Covid-19, according to vaccination status, were essentially the same in the 2021 study by Arbel and colleagues. These findings arouse strong concern regarding unadjusted confounding. The adjusted 90% lower mortality due to Covid-19 reported among the participants who received a booster cannot, with certainty, be attributed to boosting. “Healthy vaccinee bias” in this population may have also led to overestimates of vaccine effectiveness in similar studies from Clalit Health Services. Inclusion of mortality not related to Covid-19 in all observational Covid-19 vaccine studies would provide important context.