A Semiautomated Chart Review for Assessing the Development of Radiation Pneumonitis Using Natural Language...

Area of Science:

• Medical Informatics
• Computational Linguistics
• Oncology

Background:

• Manual chart review is time-consuming for health research.
• Identifying patient cohorts and outcomes requires significant researcher effort.
• Automating parts of this process can enhance research efficiency.

Purpose of the Study:

• To evaluate a Python-based text-extraction program for identifying radiation pneumonitis (RP).
• To assess the feasibility and accuracy of assisted chart review for RP detection.
• To compare automated extraction against manual review for clinical research.

Main Methods:

• Retrospective manual chart review for lung cancer patients receiving radiotherapy.
• Utilized Natural Language Toolkit (NLTK) and regular expressions in Python for text extraction.
• Compared program output against a gold standard manual review of 50 patients and 1413 documents.

Main Results:

• The program identified 92% of patients with grade ≥1 RP (sensitivity 0.92).
• It correctly identified 100% of patients with clinically significant grade ≥2 RP (sensitivity 1.0).
• Automated review avoided unnecessary manual review for 22% of patients.

Conclusions:

• The text-extraction program effectively assists in identifying patients with RP post-radiotherapy.
• This automated approach streamlines chart review, saving time compared to manual methods.
• The program shows potential to improve the efficiency of future clinical research.