A Deep Learning-Enabled Workflow to Estimate Real-World Progression-Free Survival in Patients With Metastatic Breast...

Area of Science:

• Oncology
• Medical Informatics
• Natural Language Processing

Background:

• Progression-free survival (PFS) is critical in cancer drug research, with real-world PFS (rwPFS) from clinical notes serving as a key indicator.
• Traditional methods like RECIST are impractical for real-world data, and manual abstraction is resource-intensive.
• Natural Language Processing (NLP) offers a promising solution for accelerating the extraction of tumor progression data from real-world sources.

Purpose of the Study:

• To configure a healthcare NLP framework for transforming unstructured clinical notes and radiology reports into structured progression events.
• To study real-world progression-free survival (rwPFS) in metastatic breast cancer (mBC) cohorts using this NLP framework.

Main Methods:

• Developed and validated a semiautomated workflow using deidentified electronic health records from the Nference nSights platform.
• Validated the workflow on 316 patients with hormone receptor-positive, HER2-negative mBC treated with palbociclib and letrozole.
• Defined outcome events (progression, therapy change) and censoring events (death, loss to follow-up, study end) for rwPFS computation.

Main Results:

• The NLP engine achieved 98.2% accuracy at the sentence level and 88% accuracy at the patient level for capturing progression within ±30 days.
• Median rwPFS for the cohort (N=316) was 20 months, closely aligning with manual curation (25 months) and showing high accuracy in external validation.
• Subanalysis integrating radiology reports and clinical notes provided rwPFS estimates of 30 and 23 months, respectively; patient-reported outcomes (PHQ-8) showed associations with disease progression.

Conclusions:

• The developed workflow enables rapid and reliable determination of rwPFS in mBC patients receiving combination therapy.
• Further validation across diverse external datasets and other cancer types is recommended to ensure broader applicability and generalizability.