Emerging infectious disease surveillance using a hierarchical diagnosis model and the Knox algorithm

Area of Science:

• Public Health
• Infectious Disease Epidemiology
• Computational Medicine

Background:

• Emerging infectious diseases pose significant global health and economic threats.
• Early detection and monitoring are essential for effective public health interventions.
• Existing surveillance systems may lack the speed and accuracy required for novel pathogens.

Purpose of the Study:

• To develop and evaluate a novel computational model for the early detection of emerging infectious diseases.
• To assess the model's performance in distinguishing infectious from non-infectious diseases and identifying known versus unknown infectious agents.
• To determine the model's suitability for real-time clinical record analysis and outbreak monitoring.

Main Methods:

• A two-layer Emerging Infectious Disease Detection Model (EIDDM) was proposed, utilizing TextCNN-Attention for binary classification (infectious vs. non-infectious) and LightGBM with a one-vs-rest strategy for multi-classification (known infectious diseases).
• The model was trained and validated on a large dataset of 37,422 infectious disease and 56,133 non-infectious disease records from five Beijing medical institutions.
• Performance was evaluated using accuracy, sensitivity, specificity, positive predictive value, and negative predictive value, with comparisons against XGBoost and Random Forest models. Spatiotemporal analysis using the Knox method was also performed.

Main Results:

• The first-layer TextCNN-Attention model achieved high specificity (97.57%) for non-infectious diseases and a positive predictive value of 95.07% for infectious diseases.
• The second-layer LightGBM model demonstrated an average prediction accuracy of 90.44% for emerging infectious diseases.
• The EIDDM exhibited an overall average accuracy of 86.11% and a rapid response time of approximately 27 ms for single online reasoning, alongside observed spatiotemporal clustering of infectious diseases (P < 0.05).

Conclusions:

• The EIDDM is a fast, accurate, and effective tool for the early detection and monitoring of emerging infectious diseases in real-world hospital settings.
• The model's high predictive performance minimizes misdiagnosis risks and supports timely public health responses.
• Spatiotemporal analysis revealed clustering patterns, highlighting the utility of integrated epidemiological and computational approaches for disease surveillance.