Global Health
Evaluating Waning Vaccine Effectiveness: Methods Applied to Rotavirus Vaccine in Low-Resource Settings Using Observational Data Adiba Hassan* Adiba Hassan Hassan Hassan Hassan Centers for Disease Control and Prevention
Background: Accurately assessing waning vaccine effectiveness (VE), or declining VE post-vaccination, is crucial for public health policy and implementation. However, statistical approaches have relied on simulated or trial data with limited application to real-world data. We applied two survival-based methods to detect waning VE using observational rotavirus data and compared findings with traditional approaches.
Methods: We pooled data from 22 low- and middle-income countries with test-negative case-control VE studies including 13,863 children aged <2 years hospitalized for acute diarrhea with confirmed rotavirus vaccination status. Case-patients had rotavirus-positive stool tests and control-patients were rotavirus-negative. VE was assessed for a period of 18-months following vaccination by first fitting a logistic model (1-odds ratio) at 3-month time intervals since vaccination and a Cox model for a fixed VE estimate (1-hazard ratio). We then varied the hazard ratio over time using: 1) Scaled Schoenfeld residuals averaged over time categories, and 2) Time-vaccine interaction term, then plotted using a LOESS curve. All models were adjusted for country and stratified by age group at hospitalization (<1 vs. 1 to <2 years) to account for confounding and effect modification, respectively. Logistic models were additionally adjusted for continuous age at hospitalization.
Results: Waning was identified only in age-stratified logistic model among children aged 1 to <2 years, while the Cox model didn’t differ in VE by age groups. Conversely, Schoenfeld residuals and time-vaccine interaction methods identified comparable waning VE among both age groups. VE in children <1 year peaked at 59% at 6 months of age and waned to 21% by 11 months; in children 1 to <2 years, VE waned by >40 percentage points.
Conclusions: Both time-varying survival frameworks detected waning VE compared with traditional approaches. To improve VE accuracy using observational data, investigators should incorporate time-varying confounders into these frameworks to further differentiate effects of waning from such biases.
