Using Natural Language Processing to Extract Abnormal Results From Cancer Screening Reports.

Area of Science:

• Medical Informatics
• Computational Linguistics
• Oncology

Background:

• Timely follow-up of abnormal cancer screening results (mammography, Pap smears) is often delayed.
• Abnormal findings in free-text electronic medical records (EMRs) can be missed, leading to patient loss to follow-up.
• Natural Language Processing (NLP) offers a potential solution for identifying critical results within EMRs.

Purpose of the Study:

• To evaluate the performance of NLP software in extracting abnormal results from free-text mammography and Pap smear reports.
• To assess the accuracy, precision, and recall of NLP compared to manual physician review.

Main Methods:

• A physician manually reviewed 421 mammography and 500 Pap smear reports.
• NLP software was used to extract results from the same reports.
• Performance metrics (precision, recall, accuracy) were calculated by comparing NLP extraction against the physician's assessment.

Main Results:

• For mammography reports, NLP achieved 98% precision, 100% recall, and 98% accuracy.
• For Pap smear reports, NLP demonstrated 100% precision, 100% recall, and 100% accuracy.
• NLP models accurately identified abnormal results in both mammography and Pap smear reports.

Conclusions:

• Developed NLP models accurately extract abnormal findings from mammography and Pap smear reports.
• Future implementation involves using NLP for real-time alerts to clinicians.
• This technology aims to ensure timely follow-up for patients with abnormal cancer screening results.