The incremental design of a machine learning framework for medical records processing.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Clinical Workflow Optimization

Background:

• Fax processing in outpatient clinics is time-consuming and inefficient.
• Automating clinical administrative tasks is crucial for improving healthcare delivery.
• Existing solutions often lack the adaptability required for diverse clinical settings.

Purpose of the Study:

• To develop and evaluate coordn8, a web-based application for streamlining fax processing in outpatient clinics.
• To implement a human-in-the-loop machine learning framework for automated fax analysis.
• To assess the impact of coordn8 on processing time and machine learning inference accuracy.

Main Methods:

• Deployment of coordn8 in 11 outpatient clinics.
• Time savings analysis using user observation and fax processing event logs.
• Statistical evaluation of machine learning model generalizability and performance.
• Time series analysis to monitor and mitigate model drift as new clinics onboarded.

Main Results:

• Mean reduction of 147.5 seconds in individual fax processing time.
• High accuracy in machine learning tasks: 81.6% for document classification, 83.7% for patient identification, 98.4% for spam classification, and 81.0% precision for duplicate detection.
• Document classification accuracy improved by 10.2% after retraining.

Conclusions:

• coordn8 significantly decreases fax processing time and provides accurate machine learning inferences.
• The human-in-the-loop framework effectively collects high-quality data for model training.
• Model retraining successfully mitigated performance declines associated with expanding to new clinics.
• The developed framework serves as a template for health systems implementing similar AI technologies.