Surveillance-based indicators are a better predictor of subnational measles vaccination coverage than administrative coverage data.

Accurate and timely subnational vaccination coverage measures allow us to better understand disease dynamics, and plan targeted immunization campaigns.

Existing subnational measures of coverage are infrequent (e.g., the DHS survey) or inaccurate (e.g., administrative data provided to the WHO by member states). Furthermore, they are poorly correlated with each other.

However, coverage is also theoretically linked to characteristics of suspected measles cases, such as mean age of suspected cases, proportion of suspected cases without recent or active measles infection (i.e., IgM negative), and proportion of suspected cases vaccinated.

Therefore, we investigated our hypothesis: that mean age, proportion testing negative for IgM antibodies, and proportion reporting prior vaccination against measles would be positively correlated with and predict vaccine coverage at the subnational level.

We conducted the analysis on a first-level administrative subdivision (ADM1) and year level for 19 countries in the WHO African Region. The “gold standard” outcome variable, vaccine coverage, was derived from the Demographic and Health Surveys (DHS). The predictor variables (mean age, proportion IgM negative, proportion previously vaccinated) were derived from aggregated disease surveillance data routinely collected by member states and submitted to the WHO. The comparison measure, administrative vaccine coverage, was derived from separate WHO data. We included countries with 2 or more DHS iterations between 2007 and 2020.

Using the GLMMTMB package in R, we built beta regression models with random intercepts for each country on the first DHS iteration. We then predicted, for the subsequent iterations, vaccine coverage, and compared it to the DHS coverage.

Using information about historical country-level vaccination coverage and case characteristics yields future subnational-level predictions that are better correlated with coverage than administrative data or case characteristics alone (Pearson’s r = 0.75), p ≤ 0.001).

Correlation between administrative coverage data and observed coverage is poor (Pearson’s r = 0.13, p > 0.1).

Predictions between case characteristics and coverage outperform administrative data and improve over time.

High regional heterogeneity in vaccination coverage drives large outbreaks, even in the presence of high overall coverage.

Existing measures of coverage are inaccurate or infrequent.

A new source of information—routinely collected characteristics of suspected measles cases—can be used alone or in unison with existing data sources to estimate timely, accurate measures of coverage.

Presented at EEID2022 at Emory University in Atlanta, Georgia, US, on June, 2022.