Monitoring free-text data using medical language processing

Area of Science:

• Medical Informatics
• Radiology
• Natural Language Processing

Background:

• Effective patient follow-up is critical for managing serious conditions like cancer.
• Manual review of radiology reports is time-consuming and prone to errors.
• Automated systems can enhance the efficiency and accuracy of medical data analysis.

Purpose of the Study:

• To describe and evaluate RadTRAC, a software system for automated monitoring of free-text radiology reports.
• To identify chest X-ray (CXR) reports indicating new or expanding neoplasms for patient follow-up.
• To assess the performance of RadTRAC compared to expert radiologists and existing clinical records.

Main Methods:

• Developed RadTRAC using medical language processing and statistical rules.
• Processed 470 free-text chest X-ray (CXR) reports.
• Compared RadTRAC's classification against expert review and a clinical logbook (gold standard).
• Reviewed patient charts and subsequent reports for cases flagged by RadTRAC or the logbook.

Main Results:

• RadTRAC achieved 90% sensitivity and 82% specificity against the logbook.
• Performance was comparable to expert radiologists (92% sensitivity, 90% specificity).
• Identified six new tumors/metastases missed by the logbook.
• Highlighted six cases with suspicious findings lacking adequate follow-up.

Conclusions:

• RadTRAC demonstrates high accuracy in identifying critical findings in radiology reports.
• An automated monitoring system based on RadTRAC technology could significantly improve patient follow-up.
• This technology has the potential to enhance patient care by ensuring timely interventions for new or progressing neoplasms.