Improving national level spatial mapping of malaria through alternative spatial and spatio-temporal models

Area of Science:

• Epidemiology
• Geospatial statistics
• Computational modeling

Background:

• Spatial prediction of malaria prevalence is crucial for public health interventions.
• Stochastic partial differential equations (SPDE) are commonly used for Gaussian process approximation in spatial modeling.
• Faster alternatives like generalized additive models (GAM) and gradient boosted trees (GBM) warrant investigation for malaria mapping.

Purpose of the Study:

• To compare the predictive skill of SPDE, GAM, and GBM for country-level malaria prevalence mapping.
• To evaluate the impact of incorporating past data and using spatio-temporal models on malaria prediction accuracy.
• To identify the strengths and weaknesses of different modeling approaches in various settings.

Main Methods:

• Comparative analysis of predictive performance for SPDE, GAM, and GBM.
• Assessment of spatio-temporal modeling incorporating historical malaria prevalence data.
• Evaluation of model accuracy across different countries and geographical contexts.

Main Results:

• Model performance varied by country and setting; SPDE and GAM generally showed strong predictive skills.
• Incorporating past data enhanced predictive accuracy for GAM and GBM, but not for SPDE.
• SPDE exhibited weaknesses in spatio-temporal settings, while GAM struggled at country borders.

Conclusions:

• SPDE and GAM are robust methods for malaria prevalence mapping, with varying performance depending on the context.
• The utility of incorporating historical data differs across modeling approaches.
• Spatial/spatio-temporal SPDE models should be considered alongside alternatives like GAM for comprehensive malaria prediction analysis.