Spatiotemporal Modelling of Bacterial Meningitis Outbreaks using Climatic Drivers in Africa

The African meningitis belt, spanning 26 countries, experiences regular seasonal outbreaks of bacterial meningitis. Outbreaks predominantly occur during the October-April dry season, characterised by low humidity, high temperatures, and increased dust levels driven by the movement of the Harmattan winds. Whilst many studies have examined the relationship between bacterial meningitis outbreaks and climate across Africa, this has rarely been examined on a continental level. Additionally, previous continental analyses have been unable to account for spatiotemporal changes in bacterial meningitis epidemiology. We developed a district-level Bayesian spatiotemporal model to estimate monthly meningitis risk across Africa using climatic variables.

For district-months between 2003-2022 we generated a Boolean variable to identify districts affected by meningitis outbreaks, defined as an incidence of ≥10 suspected cases per 100,000 population per week. Covariates included rainfall, humidity, windspeed and direction, MenAfriVac campaign status, population density, aerosol optical depth (AOD), temperature and land cover type. We used a Bayesian Integrated Nested Laplace Approximation (INLA) model to account for spatial and temporal heterogeneity. Covariate inclusion was decided using the Watanabe Akaike Information Criterion (WAIC), the Deviance Information Criterion (DIC), as well as considering variable epidemiological relevance. Our final model included linear effects for humidity, AOD, zonal wind strength and direction, an interaction term between zonal wind direction and AOD, and a non-linear temperature effect.

Zonal wind strength had a statistically significant positive relationship with outbreaks whilst humidity had a negative association. The interaction term between zonal wind direction and AOD had a highly positive association with outbreak occurrence. When zonal wind moved westwards, higher AOD levels increased the likelihood of outbreak occurrence. Furthermore, whilst the log odds of meningitis outbreak occurrence remained low at cooler temperatures, the effect peaked between 37–38°C before beginning to plateau. The non-linear association between meningitis incidence and temperature is suggestive of an optimal temperature for bacterial transmission. Additionally, the significance of the AOD and wind direction interaction highlights the role of the Harmattan winds in exacerbating meningitis risk, rather than outbreaks solely being driven by higher dust levels. Despite these variables all having a well-established relationship with meningitis outbreaks, our study shows that such climatic associations are still present at the continental level, when accounting for spatiotemporal effects