Optimizing A syndromic surveillance text classifier for influenza-like illness: Does document source matter?

Area of Science:

• Public Health Informatics
• Computational Epidemiology
• Natural Language Processing in Healthcare

Background:

• Syndromic surveillance systems often use electronic free-text data from chief complaints and triage notes.
• Limited research has explored using the full text of electronic health records for biosurveillance compared to specific document types.

Purpose of the Study:

• To evaluate a text classifier's performance in detecting influenza-like illness (ILI) across different electronic document sources.
• To compare the effectiveness of commonly used biosurveillance documents against routine visit notes and full electronic note corpora.

Main Methods:

• Developed and evaluated an automated text classifier for ILI detection.
• Compared classifier performance using data from chief complaints, emergency department notes, nurse triage notes, routine visit notes, and a comprehensive electronic note corpus.

Main Results:

• Commonly available surveillance source documents demonstrated acceptable statistical performance for automated ILI surveillance.
• The performance of these standard sources was comparable to using routine visit notes and a full electronic note corpus.

Conclusions:

• Standard electronic health record documents are sufficient for effective automated ILI surveillance.
• These findings inform decisions on selecting appropriate electronic data sources for biosurveillance systems.