Swaziland Health Clinic, Dr Kate Battle

Catchment modelling for area-averaged malaria data

Geospatial Modelling For Malaria Risk Stratification And Intervention Targeting

Until recently, the limited availability and low quality of routine health care data on the incidence of febrile malaria in low resource settings has meant that the vast majority of risk mapping has relied upon direct observations of parasite prevalence to predict disease burden, particularly those from the large-scale national surveys such as Demographic Health and Malaria Indication Surveys.

However, with the progressive roll-out of new digital record keeping tools and increasing access to care, there is a new rationale for approaches to risk mapping based on routine case reports from across the hierarchy of health facilities (hospitals, GP clinics, rural outposts). These new data types offer the potential for monthly, or even weekly, updating of risk maps—unlike maps made from large-scale parasite surveys, which are rarely feasible at sub-annual cadences due to logistical and financial constraints—but come with an additional modelling challenge: accounting for the treatment seeking decisions of febrile individuals (or their guardians). The malaria cases presenting at any given facility will be drawn from the surrounding region, yet we will rarely know which village a given patient came from, which villages tend to seek care at the facility, nor even how many villages (or total potential patients) lie within its catchment.

For this reason we are developing a statistical framework for simultaneous estimation of malaria risk maps and health facility catchments based on a modified ‘gravity model’ in which a patient’s treatment seeking behaviour is shaped by their travel time distance to each nearby facility and the relative ‘attractiveness’ of those facilities.

Already we have deployed versions of this approach to assist intervention planning for malaria control in Botswana, Mozambique, and Haiti; applications which are guiding our ongoing technical and theoretical developments in this field. In future work we will extend this modelling framework to the case of easy-access group surveys sampled at schools and churches, and we are collaborating with researchers conducting surveys of treatment seeking behaviour in various target countries to gather data for testing and improving the accuracy of our estimates.