Exploring predictive frameworks for malaria in Burundi

Abstract

Malaria is a major public health concern in Burundi. The infection has been increasing in the last decade despite efforts to increase access to health services, and the deployment of several intervention programs. The use of different data sources can help to build predictive models of malaria cases in different sub-populations. We built predictive frameworks using generalized linear model (GLM), and artificial neural network to predict malaria cases in four sub-populations (pregnant women and children under 5 years, pregnant women, children between 0 and 11 months, children between 12 and 59 months), and the overall general population. The results showed that almost half malaria infections are observed in pregnant women and children under 5 years, but children between 12 and 59 months carry the highest burden. Neural network model performed better in predicting total cases compared to GLM. But the latter provided information on the e ect of predictors, which is an important source of information to mainstream target interventions. Early prediction of cases can provide timely information needed to be proactive for intervention strategies, and it can help to mitigate the epidemics and reduce its impact on populations and the economy.