Misclassification of yellow fever vaccination status revealed through hierarchical Bayesian modeling.

Abstract

Vaccination coverage estimates are crucial inputs to decisions about investments in vaccination, yet they can be prone to inaccuracies. At the individual level, inaccuracies can be described in terms of the sensitivity and specificity of vaccination status. We estimated these quantities using a hierarchical Bayesian analysis of data from a test-negative study design with reported yellow fever vaccination status as the exposure. Our analysis accounted for the possibility of misclassification of both the exposure and the test at the country level. Across all countries, our median estimates of the sensitivity and specificity of vaccination status were 0.69 (95% credible interval [CrI]: 0.21-0.98) and 0.70 (95% CrI: 0.21-0.98), respectively. Median estimates at the country level ranged from 0.06 (95% CrI: 0.04-0.09) to 0.96 (95% CrI: 0.94-0.98) for sensitivity, and from 0.15 (95% CrI: 0.09-0.23) to 0.98 (95% CrI: 0.90-1.00) for specificity. This suggests that there is substantial misclassification of yellow fever vaccination status in general and extensive variation in misclassification across countries. Taking into account misclassification in vaccination status, we made adjustments to reported vaccination coverage and showed that reported coverage may be significantly underestimated in 10 out of 20 countries and significantly overestimated in 5 out of 20.