Subnational burden of childhood diarrhoeal disease associated with non-optimal climatic exposures in Ghana: a spatiotemporal Bayesian modelling study

Background Climate change is a major threat to global public health, with disproportionate impacts on children under five in low- and middle-income countries. Diarrhoeal disease, a leading cause of under-five mortality, is sensitive to temperature and rainfall through their associations with pathogen survival, water availability, and sanitation. Subnational estimates of the climate-attributable burden remain limited in official health statistics, constraining adaptation planning. This study applies the Standards for Official Statistics on Climate–Health Interactions (SOSCHI) framework, a scalable approach for generating subnational climate–health burden indicators from routine surveillance data. Methods Monthly diarrhoea cases for children under five from 261 Ghanaian districts (2013–2024) were analysed using a spatiotemporal Bayesian hierarchical model with embedded distributed lag non-linear models, fitted by integrated nested Laplace approximation (INLA), with a negative binomial likelihood for markedly overdispersed counts. ERA5 reanalysis provided monthly mean temperature and cumulative rainfall, with relative humidity and surface runoff as confounders, over two-month (temperature) and three-month (rainfall) lags. The model reproduced observed counts well ( r = 0.915) and outperformed a linear-exposure specification by 410 WAIC units. Attributable numbers (AN), fractions (AF), and rates (AR) were estimated for non-optimal exposures, against a reference selected by a pre-specified rule that resolved to the 95th percentile. Results Diarrhoeal risk peaked at a mean temperature of 26.2 °C (peak RR 1.079, 95% credible interval [CrI] 1.031–1.130) and at cumulative rainfall of 60 mm (1.082, 1.035–1.132), declining at higher values of both. An estimated 916, 176 cases (95% CrI 432, 418–1, 392, 238) were attributable to non-optimal mean temperature (AF 5.5%, 2.6–8.3; AR 184.8 per 100, 000, 87.2–280.8) and 927, 352 (531, 314–1, 323, 541) to non-optimal cumulative rainfall (AF 5.5%, 3.2–7.9; AR 187.0, 107.1–266.9); jointly, 1, 667, 998 cases (1, 015, 181–2, 308, 052) were attributable to the two combined. Temperature-associated risk was concentrated in the far north and rainfall-associated risk in the wetter south. Seasonal burden peaked between April and July. Conclusions SOSCHI provides a standardised, scalable framework for estimating subnational climate–health burden from routine surveillance data, here yielding district-level estimates for Ghana over 12 years. The non-linear, spatially heterogeneous associations reported provide a basis for prioritising adaptation planning and for embedding climate–health indicators into official health statistics. Clinical trial number Not applicable.