Building a Low-Cost Model for Predicting Infectious Disease Burden: Dengue in Brazil

Abstract

Disease surveillance is an essential function of public health ministries, but the process is expensive and slow. Here, I consider the potential of a new low-cost early warning method that policy makers could use to immediately estimate the incidence of Dengue in Brazil. Using statewide data on the frequency of “Dengue” in Google search trends, lagged epidemiological data, climate data, and state and time fixed effects, I predict values for Dengue for each month in each state. Comparing predicted values to observed Dengue incidence, this model explained 72.2% of the variation in observed Dengue but tended to underestimate Dengue incidence. Accuracy could be greatly improved using more localized data with smaller time scales. This technique uses free and instantly available data, is quite accurate, and can be modified for other diseases and geographic contexts. Google trends data, when combined with publicly available metrics, provides a viable and productive alternative or complement to current epidemiological surveillance infrastructure and could expedite public health response at little to no cost to national health ministries.