Innovative Self-controlled Risk Interval Analysis of Effectiveness of COVID-19 mRNA Booster Vaccination in the Veterans Health Administration, Omicron Predominant Period

Background: Self-controlled risk interval (SCRI) analyses may be advantageous for vaccine effectiveness evaluations as they implicitly control for all time invariant confounders. However, time-varying changes in viral circulation may introduce bias.

Objectives: To evaluate short-term relative vaccine effectiveness (rVE) of COVID-19 mRNA monovalent vaccine booster compared with prior primary series vaccination using an SCRI design accounting for time-varying viral circulation (primary analysis); and test-negative design (TND) case-control study (secondary comparator analysis).

Methods: We used Veterans Health Administration data for all analyses. For the SCRI, we included individuals who received a monovalent booster between December 15, 2021-July 26, 2022 (overlap of Omicron dominant periods and availability of monovalent booster), had no vaccination 60 days prior, and no COVID-19 diagnosis nor SARS-CoV-2 positivity in 90 days prior. We identified laboratory confirmed SARS-CoV-2 cases: (a) on days 3-7 post-booster (control period), a time before immunity presumably has been induced, and (b) on days 17-21, 24-28, and 31-35 post-vaccination (booster period), a time when immunity has presumably been induced. Using data from non-boosted individuals, we estimated the daily rate of infection for each of 80 geographically distinct areas in the country to account for changes in viral circulation. We used logistic regression to calculate relative risk (RR) (95% confidence interval [CI]) of disease in the booster vs control period, including an offset term to account for the longer booster than control period and for differences in viral circulation during the average booster vs control periods. For the TND analysis, we identified individuals tested for SARS-CoV-2 infection during the same study period among those with either 2-dose primary series only or recent booster (17-35 days before test), applied similar criteria to what we used in the SCRI study, and matched with up to 4 controls per case within 2 weeks of test date and geographic area. We conducted conditional logistic regression with confounder adjustment.

Results: For the SCRI analysis, we identified 1,430 and 1,258 SARS-CoV-2 cases in the booster and control intervals, respectively, yielding a rVE of 47% (95%CI 43%, 51%). For the TND analysis, we identified 33,414 cases and 133,332 controls, estimating a rVE of 59% (95%CI 57%, 61%).

Conclusions: The study showed significant rVE in both the SCRI and TND analyses, with a slightly higher VE estimate for the TND. Future work will explore assumptions in the SCRI analysis, including the relationship between positivity date and infection date through medical record review, and potential residual confounding in TND.