Integrating Environmental Monitoring and Disease Surveillance to Forecast Malaria Epidemics

Advance information about the timing and locations of malaria epidemics can facilitate the targeting of resources for prevention and emergency response. Early detection methods can detect incipient outbreaks by identifying deviations from expected seasonal patterns, whereas early warning approaches typically forecast future malaria risk based on lagged responses to meteorological factors. A critical limiting factor for implementing either of these approaches is the need for timely and consistent acquisition, processing and analysis of both environmental and epidemiological data. To address this need, we have developed EPIDEMIA – an integrated system for surveillance and forecasting of malaria epidemics. The EPIDEMIA system includes a public health interface for uploading and querying weekly surveillance reports as well as algorithms for validating incoming data, updating the epidemiological surveillance database, and harmonizing epidemiological data with environmental data from multiple sources. We used EPIDEMIA to evaluate the effectiveness of several malaria forecasting approaches using five years of weekly malaria case surveillance collected across 47 districts in the Amhara region of Ethiopia. We explored the consequences of using (1) different geographic groupings of districts for model calibration, (2) different sources of environmental data, including remote sensing data from earth observing satellites and gridded meteorological fields from the FEWS NET Land Data Assimilation System, (3) separate model for different parasite species (Plasmodium falciparum versus Plasmodium vivax and (4) different functional forms of the underlying statistical models. Districts where malaria incidence was declining rapidly because of malaria interventions were generally decoupled from climatic fluctuations, but several other groups of districts were sensitive to climatic variability. Neither remote sensing data nor gridded meteorological data was consistently superior, and temperature measurements from both data sources were associated with malaria outbreaks. Plasmodium falciparum malaria was more strongly associated with climatic variability than Plasmodium vivax. Models were improved by allowing environmental parameters to vary with season of year, reflecting different environmental triggers at the beginning versus end of the rainy season. Overall, the results demonstrate that it is possible to increase foreknowledge of malaria outbreaks using models based on routinely-collected malaria surveillance and environmental monitoring datasets. However, it is critical to develop a geographic stratification of the predictions based on physical and social environments, consider the responses of different malaria parasites, and account for the fact that different environmental triggers may be important in different seasons of the year.