Using natural language processing and machine learning to identify breast cancer local recurrence.

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Accurate identification of local breast cancer recurrences is critical for clinical research and patient care.
• Manual chart review is time-consuming and prone to errors.
• Automated methods can significantly improve efficiency and accuracy.

Purpose of the Study:

• To develop and validate a novel machine learning model for automated detection of local breast cancer recurrences.
• To leverage natural language processing (NLP) techniques for extracting relevant clinical information from electronic health records (EHRs).
• To reduce the burden of manual chart review in identifying cancer recurrences.

Main Methods:

• A concept-based filter and prediction model were designed using EHR data.
• A positive concept set was created by manually reviewing and extracting recurrence indicators from progress notes.
• MetaMap was used to process text, and a support vector machine (SVM) was trained with extracted concepts and pathology report counts.

Main Results:

• The developed model achieved a high Area Under the Curve (AUC) of 0.93 in cross-validation and 0.87 in held-out testing.
• The model outperformed baseline classifiers using full MetaMap concepts, filtered MetaMap concepts, or bag-of-words approaches.
• The concept-based filtering significantly improved the model's predictive performance.

Conclusions:

• The proposed automated model offers an efficient alternative to labor-intensive chart reviews for identifying breast cancer local recurrences.
• The model demonstrates potential for replication at other institutions with minor adaptations to the concept set and a suitable training dataset.
• This approach facilitates faster and more accurate clinical research and patient management.