Development and validation of venous thromboembolism-bidirectional encoder representations from transformers...

Area of Science:

• Medical Informatics
• Computational Biology
• Clinical Research

Background:

• Accurate and timely phenotyping of venous thromboembolism (VTE) is crucial for longitudinal studies.
• A validated natural language processing (NLP) tool for VTE detection in real-world patient populations is currently lacking.

Purpose of the Study:

• To develop and externally validate a novel NLP platform, NLPMed, to facilitate thrombosis research.
• To create and validate VTE-BERT, a specialized NLP model for detecting acute VTE events and their anatomical locations longitudinally.

Main Methods:

• A novel NLP platform, NLPMed, was developed for data preprocessing, annotation, and model finetuning.
• Bio_ClinicalBERT was finetuned using clinical notes from patients with cancer to create VTE-BERT.
• The VTE-BERT model underwent internal and external validation in patient cohorts from two healthcare institutions.

Main Results:

• VTE-BERT achieved high performance during training, reaching 95% precision and 98% recall.
• Internal validation demonstrated 95% precision and 91% recall.
• External validation in an independent dataset yielded 85% precision and 92% recall.

Conclusions:

• An efficient NLP model, VTE-BERT, was successfully trained and externally validated for longitudinal VTE event detection.
• The adoption of VTE-BERT is expected to accelerate thrombosis research by improving large-scale VTE detection.
• This NLP tool can reduce the time and cost associated with manual chart review in big data epidemiological studies.