Methods/Statistics
Analyzing Irregularly Age-Binned Data Using Catalytic Models of Seroprevalence Robert Dettmann* Robert Dettmann Dettmann Dettmann Dettmann Dettmann Brown University School of Public Health
Seroprevalence data are often reported in aggregated age bins that differ across studies, limiting comparability and complicating pooled analyses. This heterogeneity often forces meta-analyses to collapse age categories, reducing resolution, or exclude informative studies with incompatible binning. Accurate age-specific seroprevalence estimates are critical for characterizing infection risk, identifying windows of susceptibility, and informing vaccination policy. We present a flexible framework for analyzing heterogeneously binned seroprevalence data
We develop a likelihood-based modeling approach in which expected seroprevalence within each age bin is obtained by integrating a parametric age-specific seroprevalence function, enabling binomial likelihoods to be specified directly from aggregated counts. Study-specific likelihoods are combined into a joint likelihood, allowing estimation of pooled and study-level parameters without harmonizing age bins or excluding studies. Applying this framework to varicella seroprevalence data from 18 countries (N = 35,898 participants in 16 studies), we compare competing models that vary assumptions about age specific hazards of infection (HOI). Models incorporating maternal antibody protection improved fits to the data. Under constant HOI assumptions, we estimated the pooled HOI across all studies to be 0.215, (95% CI: 0.210-0.220) per year. For the piecewise model, we estimated HOI of 0.247 (CI: 0.241-0.253) for ages <9.5y and 0.09 (CI: 0.06-0.13) for ages ≥9.5y, with optimal change-point at 9.5y (CI: 8.7-10.5). Likelihood-based comparisons favored the piecewise HOI model (AIC: 2860.4) over the constant HOI model (AIC: 3016.4).
This framework enables synthesis of seroprevalence data across studies with inconsistent age binning and supports formal comparison of competing HOI models. The approach generalizes to other pathogens and provides a strategy for meta-analytic inference using heterogeneously aggregated serological data.
