Automated interpretation of stress echocardiography reports using natural language processing

Area of Science:

• Cardiology
• Medical Informatics
• Natural Language Processing

Background:

• Stress echocardiography (SE) reports are often unstructured free text, making data extraction for research and clinical review challenging.
• Manual review of SE reports is time-consuming and labor-intensive, limiting the scalability of data reuse.
• Developing automated methods is crucial for efficient analysis of large SE datasets.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for automated abstraction of SE reports.
• To classify overall SE results into predefined categories: normal, non-diagnostic, infarction, and ischemia.
• To assess the accuracy and efficiency of the NLP algorithm in a large clinical cohort.

Main Methods:

• An NLP algorithm was developed to process SE reports from adult patients undergoing SE within 30 days of an emergency department visit for suspected acute coronary syndrome.
• Criterion validity was assessed by comparing NLP-abstracted results with double-blinded cardiologist reviews of 140 randomly selected reports.
• Construct validity was evaluated using abstracted SE data and other clinical variables across the entire cohort.

Main Results:

• The NLP algorithm successfully abstracted 6346 SE reports, demonstrating high agreement with cardiologist interpretations (Kappa=0.83, ICC=0.89).
• The algorithm achieved high performance metrics: 98.6% specificity and NPV, 95.7% sensitivity and PPV, and F-score for overall SE results, with near-perfect scores for ischemia findings.
• Patients with ischemia findings had the highest 30-day risk of acute myocardial infarction or death (5.0%), highlighting the clinical significance of NLP-identified results.

Conclusions:

• Natural language processing provides an accurate and efficient solution for abstracting unstructured SE reports.
• Automated SE report abstraction using NLP facilitates large-scale research, public health initiatives, and quality improvement in cardiovascular care.
• The study highlights the potential of NLP to unlock valuable clinical information embedded in free-text medical records.