Prediction of subnational-level vaccination coverage estimates using routine surveillance data and survey data

Background: Measles vaccination has significantly reduced the global burden of the disease, but disparities in vaccination coverage persist. Accurate and timely estimates of subnational vaccination coverage are crucial for identifying high-risk areas and guiding targeted interventions. However, existing methods face limitations related to accuracy, timeliness, and spatial resolution. We explored the use of routinely collected case-based surveillance data to predict measles vaccination coverage at the subnational level. Methods: The study used aggregated case data from 18 countries in the WHO African region, obtained from the WHO measles surveillance database. Three surveillance-based indicators were derived: mean age of suspected measles cases, proportion of vaccinated suspected cases, and proportion of IgM-negative suspected cases. These indicators were used to build a beta regression model with measles vaccination coverage from the Demographic and Health Surveys (DHS) as the gold standard. We compared out-of-sample predictions created using this model to withheld DHS estimates using Pearson's rho. Findings: We found that each of the three surveillance-based indicators were more strongly correlated with DHS-based survey coverage than administrative estimates. Out-of-sample predictions achieved high correlation with DHS-based coverage, with a rho of 0.74. Interpretation: The findings suggest that routinely collected measles surveillance data can effectively predict subnational measles vaccination coverage. The approach addresses limitations of existing methods by providing yearly estimates that are more accurate than administrative data and more readily available than surveys. This enables timely identification of low-coverage areas and facilitates targeted interventions.