Leveraging Machine Learning and Robotic Process Automation to Identify and Convert Unstructured Colonoscopy Results Into Actionable Data: Proof-of-Concept Study | JoVE Visualize

Area of Science:

Health Informatics
Artificial Intelligence in Healthcare
Clinical Workflow Optimization

Background:

Rising patient volumes and quality initiatives necessitate efficient colorectal cancer (CRC) screening interval documentation.
Current methods for extracting CRC screening follow-up dates from external colonoscopy reports are often manual and prone to errors.
A need exists for automated solutions to ensure accurate and timely patient follow-up for CRC screening.

Purpose of the Study:

To develop and implement an integrated, end-to-end workflow solution using machine learning (ML) and robotic process automation (RPA).
To automate the extraction and updating of colorectal cancer (CRC) screening follow-up dates from unstructured data in electronic health records (EHR).
To improve the accuracy and efficiency of documenting CRC screening intervals from inbound colonoscopy reports.

Main Methods:

Developed a proof-of-concept process involving six stages: gap identification, objective definition, technology selection, process architecture creation, validation, and health system-wide implementation.
Trained an ML model on colonoscopy reports within the Epic Systems EHR platform.
Integrated ML and RPA to create an automated workflow for extracting and updating CRC screening recall dates from external colonoscopy reports.

Main Results:

The automated process achieved 80.7% accuracy in identifying valid follow-up dates from 690 colonoscopy reports.
A false negative identification rate of 32.9% was observed for follow-up dates during process validation.
Post-implementation, the system processed 16,563 external colonoscopy reports, with 35.3% identified for RPA processing.

Conclusions:

Automated workflows leveraging ML and RPA are feasible for extracting and updating CRC screening follow-up dates, improving recall accuracy and reducing documentation burden.
This approach addresses EHR documentation deficiencies and data integration challenges for quality measures.
The solution offers practical methods to overcome healthcare interoperability issues and leverage unstructured data.