Spatial Regression Analysis of eBL Risk.

We used our spatial estimates of age-specific lifetime P. falciparum infections to model the rate of eBL diagnosis by age and sex, in a given country and district, during a given year, conditional on no prior eBL diagnosis. Because eBL is rare, and the incidence rate is highly variable across locations, a negative binomial regression was used (Eq. 2, SI Appendix, Fig. S1, see p. 37) to account for possible overdispersion. A random intercept was included for each district because there are repeated measurements of each spatial unit over multiple years. Fixed effects were also included for the relationship between age, sex, country, and calendar year. To ensure that our cumulative-exposure metric was preferable to a cross-sectional measurement of malaria exposure, we compared the performance of this model to those using cross-sectional annual exposures by age and location. Results from this analysis are described in the Appendix. All models were fitted using the “brms” package for Bayesian hierarchical regression modeling for R 4.0.0, using default priors for both regression and variance parameters (57, 58). Code for the analysis can be found at https://github.com/broenk/eBL. Malariometric data are available through the MAP, and data on eBL cases can be obtained on request from SMM.