Comparison of the Spatial Variations in Ethiopia of Malaria Forecasting Skill Using Various Models in the Open Source EPIDEMIA Forecasting System

Abstract:

Over the past ten years, the Epidemic Prognosis Incorporating Disease and Environmental Monitoring for Integrated Assessment (EPIDEMIA) project has developed and tested a malaria forecasting system that integrates public health surveillance with monitoring of environmental and climate conditions. The system has been implemented with near real-time epidemiological data to generate malaria early warning reports in the Amhara region of Ethiopia. One of our current objectives is to develop strategies with USAID/Ethiopia, Ethiopia’s National Malaria Control Program (NMCP), the Ethiopian Public Health Institute (EPHI), and other national and regional partners for scale up and application of these forecasts to other regions in Ethiopia.

The EPIDEMIA forecasting system is implemented in R, a free software environment for statistical computing, and includes the developed R package epidemiar that provides a generalized set of functions for disease forecasting. We also designed workflows and wrote customized code for malaria forecasting in Ethiopia, including a Google Earth Engine script to capture the necessary summaries of the environmental variables and formatting scripts to create distributable reports with maps and graphs of the results. We built model validation tools into the epidemiar R package for on-demand evaluation for one through n-week ahead predictions.

Using the model validation reports, we identified areas where the forecasting model works well at one through 12-week ahead predictions. Spatial heterogeneity in forecast skill can arise from a variety of factors, including localized changes in malaria patterns, environmental conditions, or external influences such as shifts in geographic boundaries. We found spatial variation between two different modelling approaches, one utilizing the anomalies of the environmental variables as predictors as opposed to the observed values, and compared the forecast skill results.