A Novel Approach Considering Spatial Heterogeneity in Tuberculosis Prediction of China

Background Present research has seldom comprehensively considered the impact factors’ spatial heterogeneities of tuberculosis(TB) in constructing artificial intelligence(AI) model. Objective This study aimed to construct optimal simulation and prediction models for TB in different provinces of mainland China, based on a novel modeling framework for integrating the spatial heterogeneities of both impact factors and artificial intelligence algorithms. Main results will hold significant value for achieving the global target of ending TB under Sustainable Development Goal 3.3. Methods Monthly TB incidence rate, six meteorological factors and three air quality factors are obtained from 31 provinces during 2004–2020. The spatiotemporal variations of TB and the corresponding impact factors were analysed for linear trends. Spearman's rank correlation analysis was used to quantify the relationships between TB and the impact factors. Significantly correlated factors were selected to build a model library that incorporates the spatial heterogeneity of predictors and six AI models. The Distance Between Indices of Simulation and Observation (DISO) was used to evaluate the comprehensive performance of models, and obtain the optimal one. Results TB incidence exhibited a significant declining trend during 2004–2020 (2.42 per 100,000, p < 0.05). The monthly incidence was significantly positively correlated with PM _2.5 (3-month lag, r = 0.72), PM ₁₀ (3-month lag, r = 0.73), and pressure(PRE, 3-month lag, r = 0.66)(p < 0.05). It was significantly negatively correlated with O ₃ (3-month lag, r=-0.49), temperature (TEM, 3-month lag, r=-0.46), relative humidity (RHU, 3-month lag, r=-0.39), precipitation (PRE, 3-month lag, r=-0.46), and sunshine duration (SSD, 3-month lag, r=-0.30) (p < 0.05). Among the AI models, the Autoregressive Integrated Moving Average with Exogenous variables(ARIMAX) demonstrated the best performance in Inner Mongolia (DISO = 0.20), while Random Forest(RF) was most prominent in Qinghai Province (DISO = 0.24). Conclusion In different provinces, ARIMAX and RF demonstrated superior predictive performance, which provide a valuable scientific foundation for prevention and control.