Epidemiological trends and comparative forecasting models of human brucellosis in inner mongolia autonomous region, mainland China, 2004-2024.

INTRODUCTION

Brucellosis remains a severe zoonotic threat in the Inner Mongolia Autonomous Region of China.

METHODOLOGY: This study integrates a comprehensive epidemiological trend analysis with a novel methodological comparison of forecasting techniques to inform control strategies.

RESULTS

Using reported human brucellosis surveillance data from Inner Mongolia for 2004-2024, joinpoint regression analysis revealed a persistently increasing yet fluctuating long-term trend (AAPC = 5.13%, P < 0.001), characterized by significant epidemic surges in 2004-2010 (APC = 22.43%, P < 0.001) and 2016-2021 (APC = 29.83%, P < 0.001), interrupted by a decline phase in 2010-2016 (APC = -17.17%, P < 0.001) and 2021-2024 (APC = -12.15). The disease demonstrated strong seasonality with June-August peaks, and predominance among farmers and herdsmen aged 30-60 years. Building on this epidemiological foundation, we rigorously compared the predictive performance of the standard Seasonal Autoregressive Integrated Moving Average (SARIMA) model against its Bootstrap-enhanced version for 24-month-ahead forecasting (2023-2024 validation). This finding offers a novel perspective on enhancing the predictive performance of brucellosis models. While the Bootstrap approach achieved superior point forecast accuracy by reducing Mean Absolute Error by 39.95% and Median Absolute Percentage Error by 33.55% compared to SARIMA, it produced severely overconfident prediction intervals, with only 33.33% empirical coverage compared to SARIMA's 91.67%. This study validates the SARIMA model as a robust baseline for brucellosis forecasting and introduces a Bootstrap ensemble method as a powerful tool for significantly enhancing point prediction accuracy.

CONCLUSION

The findings provide novel epidemiological insights that offer a scientific basis for disease control measures and decision-making. Future work should aim to develop hybrid models that bridge this gap, and delivering high accuracy in both point and interval forecasts.