Machine learning-enabled multitrust audit of stroke comorbidities using natural language processing.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Clinical Data Curation

Background:

• Electronic health records are increasingly common in healthcare systems.
• Machine learning offers advanced methods for data curation in audits and research.
• Evaluating new tools against traditional methods is crucial for improving data quality.

Purpose of the Study:

• To compare the consistency of traditional clinical curation methods with a machine learning tool, MedCAT.
• To assess MedCAT's effectiveness in extracting stroke comorbidities from the Sentinel Stroke National Audit Programme (SSNAP) data.
• To determine if machine learning can enhance the accuracy and efficiency of clinical data extraction.

Main Methods:

• Utilized 2327 stroke admission episodes from three NHS hospitals (Jan 2019 - Apr 2020).
• Compared MedCAT against traditional SSNAP curation methods using a manually reviewed subsample of 200 episodes.
• Reported performance metrics including sensitivity, specificity, precision, negative predictive value, and F1 scores.

Main Results:

• Traditional methods showed good reporting for atrial fibrillation, hypertension, and diabetes, but poor reporting for congestive cardiac failure.
• MedCAT demonstrated superior performance across all four assessed stroke comorbidities.
• MedCAT's improved performance was primarily due to higher sensitivity and F1 scores.

Conclusions:

• Machine learning-enabled data collection can significantly support clinical initiatives.
• MedCAT has the potential to enhance the quality and speed of data extraction from clinical repositories.
• The scalability of machine learning tools can revolutionize audit and research methodologies in healthcare.