Measles Burden Estimation Using Local Gaussian Process Classifiers

Abstract

AbstractMeasles is a highly contagious viral disease and remains a severe public health problem. Monitoring measles cases provides a powerful tool to identify outbreaks and epidemics. However, burden estimates of measles are challenging to obtain because of heterogeneous surveillance systems and a lack of resources for rapid laboratory tests means that many cases are reported based on symptoms alone, which has low specificity. We consider diagnostically confirmed measles case data in Ethiopia between 2009 and 2017 and propose a local Gaussian process binary classifier with spatial dependence to provide case predictions for untested individuals based on age, vaccination status, and location. By applying our modeling framework to untested suspected reported cases, we provide more accurate burden estimates at the district level. We validate our methods through simulation studies. We also find that our approach, which provides burden estimates at the district level, highlights temporal variation in the specificity of symptom-based diagnosis.