Automated Identification and Extraction of Exercise Treadmill Test Results

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Exercise treadmill tests (ETTs) are crucial for diagnosing acute coronary syndrome but manual review of results is time-consuming for research.
• Automated interpretation of ETTs is needed to overcome limitations in large-scale clinical studies and cost-effectiveness analyses.
• Developing and validating an NLP algorithm for ETT interpretation addresses the need for efficient data extraction from electronic health records.

Purpose of the Study:

• To develop and validate an automated natural language processing (NLP) algorithm for interpreting exercise treadmill test (ETT) results.
• To assess the clinical utility of the NLP-derived ETT categories by examining associations with 30-day patient outcomes.
• To demonstrate the feasibility of using NLP for large-scale outcome studies involving noninvasive cardiac tests.

Main Methods:

• Retrospective analysis of 5214 adult emergency department encounters with ETTs within 30 days.
• Validation of an NLP algorithm by comparing its categorization (normal, ischemic, nondiagnostic, equivocal) against double-blind physician reviews.
• Statistical analysis of 30-day outcomes (death or acute myocardial infarction) across NLP-categorized ETT results.

Main Results:

• The NLP algorithm demonstrated high accuracy in categorizing ETT results, with 96.4% sensitivity and 94.8% specificity for normal versus other categories.
• Significant variation in 30-day death or acute myocardial infarction rates was observed across ETT categories: normal (0.08%), ischemic (1.9%), nondiagnostic (0.77%), and equivocal (0.58%).
• The algorithm achieved good discrimination for predicting outcomes (C-statistic, 0.81).

Conclusions:

• Natural language processing (NLP) provides an accurate and efficient method for interpreting noninvasive cardiac tests, facilitating large-scale outcome research.
• Most patients undergoing ETTs in the emergency department exhibit normal results and low risk.
• Abnormal, nondiagnostic, or equivocal ETT results are associated with slightly elevated risks, warranting further investigation and clinical attention.