Spatial co-clustering of facility-recorded malaria and sickle cell disease admissions across health facility catchments in Uganda: a Bayesian shared-component disease mapping analysis

Malaria and sickle cell disease (SCD) remain major causes of morbidity and mortality in Uganda, yet their joint spatial organization across routine service geographies has rarely been examined. We assessed whether facility-recorded malaria and SCD admissions shared residual spatial structure across modeled health-facility catchments. We conducted a national ecological study using higher-level health-facility catchments as spatial units. Monthly facility-level inpatient malaria and SCD admissions reported in Uganda’s DHIS2 from January 2020 to December 2024 were aggregated over 5 years to stabilize sparse catchment-level counts. Catchments were delineated using a two-step floating catchment area approach based on nearest higher-level facility assignment and a 60-minute travel-time threshold. We summarized crude cumulative admission rates per 10,000 modeled catchment population. We then fitted Bayesian shared-component disease-mapping models, including baseline, extended, and covariate-adjusted specifications. Expected counts were derived from total catchment population as a pragmatic standardization baseline. Mean crude malaria admission burden was 293.6 per 10,000 catchment population, mean crude SCD admission burden was 27.9 per 10,000, and mean descriptive composite admission burden was 321.5 per 10,000. The extended model fit better than the baseline model (DIC 3326.33 vs. 3400.12; WAIC 4964.56 vs. 5908.24). In the extended model, the shared spatial precision was low (posterior mean 0.010, 95% CrI 0.007 to 0.012), consistent with substantial shared residual spatial heterogeneity across the two recorded admission outcomes, whereas disease-specific spatial precision parameters were very large and highly uncertain, suggesting limited stable evidence for additional disease-specific spatial structure. Positive shared spatial effects were concentrated most consistently across health-facility catchments in northern and eastern Uganda. The SCD copy-loading parameter was close to zero (mean − 0.084, 95% CrI − 0.232 to 0.066). In covariate-adjusted analyses, model fit improved further (the lowest DIC in the fully adjusted model, 3023.63) with the same residual shared signal persisting. Facility-recorded malaria and SCD admissions exhibited residual spatial co-patterning across some service geographies, where integrated approaches to surveillance and service strengthening may be evaluated, while recognizing that the shared component may also reflect common referral, diagnostic, and reporting processes.