TimeX: Phenotype Onset Extraction from Clinical Narratives

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Clinical Data Science

Background:

• Accurate disease phenotype onset estimation is crucial for diagnosis and clinical decisions.
• Electronic Health Record (EHR) data offers potential but faces challenges in precise onset determination.
• Current methods using EHR timestamps or conventional NLP lack scalability and struggle with temporal nuances.

Purpose of the Study:

• To introduce TimeX, an open-source pipeline for extracting phenotype onset from clinical narratives.
• To leverage Llama-3.1 and instruction-based prompting for enhanced temporal information extraction.
• To improve the accuracy and scalability of phenotype onset estimation from EHR data.

Main Methods:

• Developed TimeX, a modular pipeline including family history filtering, phenotype extraction, negation handling, and temporal extraction.
• Utilized Llama-3.1 with instruction-based prompting for clinical narrative analysis.
• Validated TimeX on 102 manually annotated clinical notes, comparing against five baseline methods.

Main Results:

• TimeX achieved an average accuracy of 81.24% in timestamp extraction, outperforming baselines by at least 14.86%.
• Case studies on rare diseases demonstrated that narrative-derived onset is more precise than documentation timestamps.
• The pipeline shows significant improvements in accuracy and scalability for phenotype onset extraction.

Conclusions:

• TimeX offers an accurate and scalable solution for phenotype onset extraction from clinical narratives.
• This approach enhances disease trajectory characterization and supports timely diagnosis.
• The findings highlight the potential of advanced NLP models in clinical data analysis.