Rotasiil vaccine effectiveness against rotavirus-associated hospitalizations in the Democratic Republic of the Congo: comparison of multivariate logistic regression and inverse probability treatment weighting methods in a test-negative design

Background: Multinational rotavirus vaccines (Rotarix and RotaTeq) are 43 %–65 % effective in preventing hospitalizations due to rotavirus disease in African countries. The Rotasiil vaccine (Serum Institute of India) offers a potentially cost-effective option but lacks vaccine effectiveness (VE) estimates in African settings. We estimated Rotasiil VE against rotavirus hospitalizations in the Democratic Republic of Congo (DRC) using unconditional multivariate logistic regression and inverse probability of treatment weighting (IPTW) to adjust confounding. Methods: We examined sentinel surveillance data from two hospitals in Kinshasa (January 1, 2020 to December 31, 2023) for children aged ≤59 months who sought care for acute diarrhea. We used a test-negative case-control design and calculated VE as [1 – adjusted odds ratio (OR) × 100] using unconditional multivariate logistic regression. We compared two approaches to adjust the OR for confounding. First, the OR was adjusted by including the covariates in the model. Second, the OR was adjusted using stabilized IPTW. Results: In total, 310 children aged 6 to 30 months were included in the full rotavirus vaccine dose analysis, of which 118 (38 %) were rotavirus positive. The median age was 10 months (range: 8–14), and the proportion of fully vaccinated children was 87 %. The VE measured between fully vaccinated versus unvaccinated children was 60 % (95 % CI: 20–80) by multivariate logistic regression and 79 % (95 % CI: 71–85) by IPTW. Conclusion: Rotasiil vaccination under conditions of routine use effectively reduces rotavirus-associated diarrhea hospitalizations among children, supporting the ongoing use of rotavirus vaccine. The IPTW VE estimate was more precise than multivariate logistic regression but the confidence intervals overlapped, suggesting that the resulting VE estimates are not statistically different and the methods are complementary.