AI-Enabled Diagnostic Prediction within Electronic Health Records to Enhance Biosurveillance and Early Outbreak Detection

TL;DR

This paper introduces a machine learning method that improves early detection of infectious disease outbreaks by analyzing electronic health records.

Contribution

The novel contribution is integrating ML-based diagnostic predictions with traditional surveillance to enhance biosurveillance and outbreak detection.

Findings

33.3% of outbreaks were detected earlier with lead times of 1 to 24 days.

The system detected an average of 1.33 false positive outbreaks annually.

Combining ML predictions with traditional data improved biosurveillance effectiveness.

Abstract

Detecting infectious disease outbreaks promptly is crucial for effective public health responses, minimizing transmission, and enabling critical interventions. This study introduces a method that integrates machine learning (ML)-based diagnostic predictions with traditional epidemiological surveillance to enhance biosurveillance systems. Using 4.5 million patient records from 2010 to 2022, ML models were trained to predict, within 24-hour intervals, the likelihood of patients being diagnosed with infectious or unspecified gastrointestinal, respiratory, or neurological diseases. High-confidence predictions were combined with final diagnoses and analyzed using spatiotemporal outbreak detection techniques. Among diseases with five or more outbreaks between 2014 and 2022, 33.3% (41 of 123 outbreaks) were detected earlier, with lead times ranging from 1 to 24 days and an average of 1.33 false positive outbreaks detected annually. This approach demonstrates the potential of integrating ML with conventional methods for faster outbreak detection, provided adequate disease-specific training data is available.

Figures