Efficient identification of nationally mandated reportable cancer cases using natural language processing and machine...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Cancer Registries

Background:

• Accurate identification of reportable cancer cases is crucial for cancer surveillance and research.
• Manual review of clinical notes by cancer registrars is time-consuming and prone to errors.
• Automated systems are needed to improve the efficiency and accuracy of cancer case detection.

Purpose of the Study:

• To develop and evaluate the Cancer Registry Control Panel (CRCP) for automated detection of reportable cancer cases.
• To assist cancer registrars in efficiently and accurately identifying cancer cases from clinical documentation.

Main Methods:

• Developed the Cancer Registry Control Panel (CRCP) using the Unstructured Information Management Architecture - Asynchronous Scaleout (UIMA-AS).
• Integrated UIMA MetaMap and rule-based annotators to identify cancer concepts and filter non-reportable cases.
• Employed supervised machine learning, combining pathology reports and diagnosis codes to identify candidate patients.
• Utilized natural language processing (NLP) to highlight cancer mentions in clinical notes for registrar validation.

Main Results:

• CRCP achieved an accuracy of 0.872, with a precision of 0.843 and recall of 0.848.
• The system increased throughput by 22.6% compared to manual review, with higher precision and recall.
• Recall could be increased to 0.939 by incorporating a data source information feature, depending on registrar needs.

Conclusions:

• CRCP demonstrates the effectiveness of NLP in accurately detecting reportable cancer cases from clinical notes.
• Implementing NLP features, including rule-based filters and machine learning, significantly improves precision, recall, and speed.
• This automated approach supports efficient and accurate cancer case identification for registries.