Study population and setting
This case-control study of vaccine effectiveness (VE) against COVID-19 hospitalization included 3,689 adults aged 18 and over (median age 58 years, 52% male) admitted to 21 hospitals in the Influenza and Other Viruses in the Acutely Ill Network across 18 US states from March 11 to August 15, 2021. The authors estimated VE for the Pfizer-BioNTech, Moderna, and Janssen vaccines; they compared VE across vaccines and estimated VE separately for each vaccine by time since full vaccination (14-120 days after vaccination vs. >120 days after vaccination). This was a test-negative study design in which cases (n=1,682) were admitted to hospital with a COVID-19-like illness and had a positive SARS-CoV-2 test result (PCR or antigen), and controls (n=2,007) were admitted to hospital with a negative SARS-CoV-2 test result via PCR. Patients’ vaccination status was assessed via interview and verified with external sources (e.g., medical records or vaccination registries). Patients were considered fully vaccinated if they had received the final dose at least 14 days before symptom onset. Partially vaccinated patients and those who received doses of two different vaccines were excluded. VE was estimated with multivariable logistic regression, adjusting for admission date, geographic region, age, sex, race, and ethnicity. Additionally, the authors compared serum antibody levels (anti-spike IgG and anti-receptor binding domain IgG) with the Wilcoxon rank sum test in 100 healthy individuals without prior SARS-CoV-2 infection who had been fully vaccinated with one of the three vaccines 2-6 weeks before measurement.
Summary of Main Findings
Among all VE study participants, 64% were unvaccinated, 20% were fully vaccinated with the Pfizer-BioNTech vaccine, 13% were fully vaccinated with the Moderna vaccine, and 3% were fully vaccinated with the Janssen vaccine. Estimated VE against COVID-19 hospitalization during the full study period was highest (p=0.01) for the Moderna vaccine (93%, 95% CI: 91% to 95%), followed by the Pfizer-BioNTech vaccine (88%, 85% to 91%) and the Janssen vaccine (71%, 56% to 81%). Comparing the period from 14-120 days after full vaccination to more than 120 days after vaccination, estimated VE for Moderna remained similar at 93% (90% to 95%) and 92% (87% to 96%), respectively. However, estimated VE for Pfizer declined from 91% (88% to 93%) to 77% (67% to 84%). Too few individuals had been fully vaccinated with Janssen for longer than 120 days to permit a similar analysis. Anti-RBD IgG and anti-spike IgG antibody levels were highest among Moderna vaccine recipients (median 4,333 and 3,236 BAU/mL respectively), followed by Pfizer vaccine recipients (median 3,217 and 2,983 BAU/mL) and Janssen vaccine recipients (median 57 and 59 BAU/mL); the difference between Moderna and Pfizer-BioNTech was not statistically significant (p=0.22).
The study employed a test-negative design, which is a standard means of addressing possible bias in observational studies of vaccine effectiveness.
No variant-specific analysis was conducted, and the time since full vaccination may have been confounded with the relative prevalence of the Delta variant. Additionally, no changes over time in antibody levels were assessed. It is therefore not possible to infer the relative contributions of waning immunity and any differences by variant in the changes in VE over time. Additionally, time since vaccination was categorized as a binary variable, which may have caused residual confounding and prevented a more detailed examination of changes in VE. Controls included patients both with and without COVID-19-like illness, and the authors did not present detailed information on characteristics of cases vs. controls. While a test-negative design improves the likelihood of drawing cases and controls from the same source population (a critical prerequisite for validity in case-control studies), it does not guarantee this. Bias may have affected results if vaccination status influenced the probability of being hospitalized for a non-SARS-CoV-2 infection (e.g., if vaccination made it more likely for individuals to have close contacts with others). Similarly, if vaccination status was correlated with seeking care conditional on COVID-19 (e.g., if vaccinated people were less inclined to seek care upon becoming sick), results could be biased. VE was only assessed for adults and for non-immunocompromised patients. Finally, as with any observational study, unmeasured confounding (e.g., by occupation) is possible.
This study provides useful evidence on changes over time in vaccine effectiveness against COVID-19 hospitalization.