Probabilistic case detection for disease surveillance using data in electronic medical records

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Epidemiology

Background:

• Accurate and timely detection of influenza and influenza-like illness (ILI) is crucial for public health management.
• Traditional diagnostic methods can be time-consuming and may not capture the full clinical picture.
• Emergency department (ED) data offers a rich source for real-time disease surveillance.

Purpose of the Study:

• To develop and evaluate a probabilistic case detection system (CDS) for influenza and ILI.
• To leverage Bayesian network modeling and natural language processing (NLP) for enhanced diagnostic accuracy.
• To assess the performance of the CDS using emergency department dictated notes and laboratory results.

Main Methods:

• Development of a Bayesian network model for medical diagnosis.
• Integration of natural language processing (NLP) to extract relevant information from dictated ED notes.
• Computation of posterior probabilities for influenza and ILI.
• Validation of diagnostic accuracy using the area under the Receiver Operating Characteristic (ROC) curve.

Main Results:

• The case detection system (CDS) demonstrated high diagnostic accuracy for influenza and ILI.
• The area under the ROC curve for diagnostic accuracy reached 0.97.
• The overall accuracy of the natural language processing (NLP) component within the CDS was 0.91.

Conclusions:

• The developed probabilistic case detection system (CDS) is highly accurate in identifying influenza and ILI.
• The combination of Bayesian networks and NLP provides a powerful tool for syndromic surveillance.
• This approach can significantly improve the efficiency and accuracy of influenza case detection in emergency departments.