Automated extraction of temporalized tumor evolution from oncology EMRs using natural language processing.

Area of Science:

• Oncology
• Medical Informatics
• Natural Language Processing

Background:

• Extracting time-sensitive oncology outcomes like tumor progression from electronic medical records (EMRs) is challenging.
• Unstructured EMR data hinders the accurate retrieval of clinical outcomes.
• Developing automated solutions is crucial for advancing cancer research.

Purpose of the Study:

• To evaluate a domain-adapted natural language processing (NLP) pipeline for extracting structured, temporally anchored clinical outcomes from EMRs.
• To assess the accuracy and efficiency of NLP in retrieving oncology-specific data.
• To enable the reconstruction of real-world endpoints for cancer patients.

Main Methods:

• A cohort of advanced or metastatic non-small-cell lung cancer (NSCLC) patients treated with targeted therapies was analyzed.
• A specialized NLP pipeline was employed to extract clinical outcomes from narrative EMR data.
• Extracted data were benchmarked against expert annotations and mapped to Observational Medical Outcome Partnership vocabularies.

Main Results:

• The NLP pipeline achieved an F1-score of 79.7% for tumor evolution concepts and 62.0% with temporality.
• Overall performance reached F1-scores of 76.5% for concept extraction and 63.7% with temporality.
• Automated data collection was 5.8 times faster than manual review, with median real-world progression-free survival (rwPFS) estimates aligning with benchmarks.

Conclusions:

• The developed NLP solution effectively extracts temporally structured tumor outcomes from EMRs.
• This approach supports the accurate reconstruction of real-world endpoints in oncology.
• The findings highlight the potential of NLP to streamline oncology research and clinical data analysis.