Natural Language Processing framework for identifying abdominal aortic aneurysm repairs using unstructured electronic...

Area of Science:

• Medical Informatics
• Natural Language Processing
• Health Data Science

Background:

• Patient identification for national registries often relies on manual methods or inaccurate clinical codes, leading to data gaps.
• Electronic Health Records (EHRs) contain unstructured clinical notes that are challenging to analyze manually.
• Automated methods are needed to improve the efficiency and accuracy of patient identification for medical registries.

Purpose of the Study:

• To develop and evaluate Natural Language Processing (NLP) models for identifying and classifying abdominal aortic aneurysm (AAA) repairs from unstructured EHRs.
• To demonstrate the feasibility of using NLP for automated patient identification in vascular surgery registries.
• To compare the performance of different NLP models in accurately extracting AAA repair information.

Main Methods:

• A multi-tiered NLP approach was used on the MIMIC-IV-Note dataset to identify vascular patients, AAA repairs, and classify repair types (primary vs. revision).
• Four NLP models were trained and evaluated: scispaCy, BERT-base, Bio-clinicalBERT, and a scispaCy/Bio-clinicalBERT ensemble.
• Model performance was assessed using accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• The scispaCy model offered the fastest training and inference times.
• High accuracy (0.97) was achieved for identifying vascular patients by scispaCy and ensemble models.
• Models demonstrated excellent performance in identifying AAA repairs (accuracy 0.99, AUC 1.00), with Bio-clinicalBERT and the ensemble model excelling in classifying repair types (AUC 1.00, accuracy 0.98).

Conclusions:

• NLP models can accurately and efficiently identify and classify abdominal aortic aneurysm (AAA) repair cases from unstructured EHR data.
• Automated patient identification using NLP holds significant potential for improving vascular surgery registries and other medical data collection efforts.
• This approach can reduce administrative overhead and enhance the completeness and accuracy of registry data.