County-level spatiotemporal patterns, forecasting, and non-fatal burden of human brucellosis in Hulunbuir, China, 2016–2025

Abstract

Background Human brucellosis remains an important zoonotic public health problem in northern China, especially in pastoral and agro-pastoral areas. County-level evidence is needed to distinguish large absolute service burden from high population-adjusted risk and to support seasonal preparedness. Methods We conducted a retrospective surveillance-based county-level analysis integrating descriptive epidemiology, population-adjusted spatial analysis, Bayesian smoothing, short-term forecasting, and approximate non-fatal burden estimation. County-level population denominators were available for 2016–2024. Because 2025 records had a high deletion rate and lacked reliable county-level denominator linkage, analyses involving 2025 were treated as provisional reported-case analyses. Results After excluding 2,615 deleted records, 12,220 reported cases were included; the denominator-linked 2016–2024 panel contained 10,824 cases and 21,892,827 person-years. Annual crude incidence peaked in 2021 at 68.96 per 100,000 population. The largest absolute burdens occurred in Zhalantun City, Arun Banner, and Morin Dawa Daur Autonomous Banner, whereas the highest population-adjusted incidence rates were observed in New Barag Right Banner, New Barag Left Banner, and Chen Barag Banner. Monte Carlo scan-statistic testing supported the leading population-adjusted excess windows in Arun Banner in 2019, Zhalantun City in 2021, and New Barag Right Banner in 2022 (all Monte Carlo p < 0.001). BYM smoothing produced consistent high-relative-risk areas, and posterior medians, credible intervals, and exceedance probabilities are reported in the Supplementary materials. In rolling-origin validation, SARIMA achieved the lowest average monthly MAE and MAPE; LightGBM produced a marginally lower mean annual total error across all four folds, but this advantage was not retained after excluding the provisional 2025 fold. Forecasting was interpreted as short-term preparedness support, not as evidence of a single superior model. Approximate non-fatal burden estimates varied across disability-weight and duration assumptions; the upper-bound proxy scenario produced approximately 1,530 DALYs for 2016–2025 and approximately 1,350 DALYs when provisional 2025 records were excluded. Conclusion Human brucellosis in Hulunbuir showed persistent seasonality and marked county-level heterogeneity. Formal population-adjusted interpretation was strongest for 2016–2024, while 2025 results should be read as provisional reported-case evidence.