Research on Tuberculosis Incidence Prediction Based on the ST-GNN Model

Abstract

Tuberculosis (TB) Is a Chronic Infectious Disease Caused by Mycobacterium Tuberculosis. Its Epidemic Process Is Influenced by Long-Term Socioeconomic Differences, Population Mobility, and Metabolic and Behavioral Risk Factors, Exhibiting Significant Spatiotemporal Heterogeneity in My Country. To More Accurately Characterize the Spatiotemporal Evolution of TB Incidence and Improve Predictive Performance, This Paper Constructs a Spatio-Temporal Graph Neural Network (ST-GNN) Prediction Model Based on Monitoring Data from 31 Provincial-Level Administrative Regions in My Country. This Model Integrates Spatial Correlation, Temporal Dynamics, and Attention Mechanisms, and Incorporates a Comparative Risk Assessment (CRA) Framework to Conduct Multi-Dimensional Risk Attribution Analysis. This Study Treats Provincial-Level Administrative Regions as Graph Nodes, Constructs a Spatial Association Network by Integrating Geographical Adjacency Relationships and Population Flow Information, Uses Graph Convolutional Networks (GCNs) to Learn Inter-Provincial Spatial Dependence Features, Employs Temporal Convolutional Networks (TCNs) to Characterize the Long-Term Dynamic Changes in Tuberculosis Incidence, and Introduces a Spatiotemporal Attention Mechanism to Enhance the Model's Ability to Identify Key Regions and Key Time Periods. Simultaneously, Bayesian Optimization Is Used to Automatically Adjust the Model's Hyperparameters to Improve Its Stability and Generalization Performance. Experimental Results Show That, Compared with Traditional Time Series Models and Various Deep Learning Models, the Constructed ST-GNN Model Exhibits Higher Prediction Accuracy and More Robust Spatiotemporal Feature Capture Capabilities Across Multiple Evaluation Metrics, and Can More Effectively Identify Regional Clustering Trends and Abnormal Fluctuations. Based on the Predictive Analysis, This Paper Uses the CRA Framework Proposed by the Global Burden of Disease Study to Conduct Attribution Analysis on Modifiable Risk Factors Such as High Body Mass Index (BMI), Smoking, and Insufficient Physical Activity. The Results Show That the Attribution Contributions of Different Risk Factors to Tuberculosis Incidence Vary Significantly Across Regions, with Some Behavioral and Metabolic-Related Factors Having a More Prominent Impact in High-Burden Areas. The Relevant Research Results Provide Quantitative Evidence for the Precise Prevention and Control of Tuberculosis, the Formulation of Regionally Differentiated Intervention Strategies, and the Rational Allocation of Public Health Resources.