Integrating vulnerability and hazard in malaria risk mapping: the elimination context of Senegal

Significant efforts over the past decades have successfully reduced the global burden of malaria. However, progress has stalled since 2015. In low-transmission settings, the traditional distribution of malaria along vector suitability gradients is shifting to a new profile, with the emergence of hotspots where the disease persists. To support elimination in this context, it is essential that malaria risk maps consider not only environmental and climatic factors, but also societal vulnerabilities, in order to identify remaining hotspots and ensure that no contributing factors are overlooked. In this paper, we present an integrated approach to malaria risk mapping based on the decomposition of malaria risk into two components: 'hazard', which refers to the potential presence of infected vectors (e.g. influenced by rainfall and temperature), and 'vulnerability', which is the predisposition of the population to the burden of malaria (e.g. related to health care access and housing conditions). We focus on Senegal, which has a heterogeneous malaria epidemiological profile, ranging from high transmission in the south-east to very low transmission in the north, and which aims to eliminate malaria by 2030.

We combined data from several sources: the 2017 Demographic and Health Survey (DHS) (national coverage) and the 2020-21 Malaria Indicator Survey (MIS) (south-east regions), as well as remotely sensed, high-resolution covariate data. Using Bayesian geostatistical models, we predicted the prevalence of malaria in children under five years of age with a spatial resolution of 1 km.

Including vulnerability factors alongside hazard factors in the 2017 DHS data model improved the accuracy of predictive maps, achieving a median predictive R² of 0.64. Furthermore, models including only vulnerability factors outperformed those including only hazard factors. However, the models trained on the 2020-21 MIS data performed poorly, achieving a median R² of 0.13 at best for the model based on hazard factors, likely due to data collection during the dry season.

These findings highlight the importance of integrating both vulnerability and hazard factors into predictive maps. Future work could validate this approach further using routine malaria data from health management information systems, such as DHIS2.