Hiring in life sciences? Share your open positions with our professional community. Read more Close

Advertisement

Identifying Distant Breast Cancer Recurrence: A Proof-of-Concept Comparison of Discrete-Field Algorithm and Large Language Model Performance and Cost.

Created on 17 Nov 2026

Authors

Megan A Mullins, Lesi He, Chahat Rana, Omer Abdrabbuh, Abril Carrillo, Malcolm Su, Pranathi Pilla, Yingzi Zhang, Francisco Villalobos, Shifa Kanjwal, Wenqi Shi, Arthur S Hong, Jason B Fleming, Timothy P Hogan, Wendy Chapman

Published in

JCO clinical cancer informatics. Volume 10. Issue 4. Pages e2600103. Epub Nov 16, 2026.

Abstract

Reliably identifying distant cancer recurrence and determining its timing remain longstanding challenges that limit metastatic cancer research. We characterized the performance of a large language model (LLM)-based approach versus gold standard chart review and a validated discrete-field algorithm for identifying distant metastatic breast cancer (mBC) recurrence.
Using electronic health record data and notes from January 2010 to June 2023, we applied the Ritzwoller et al discrete-field algorithm and an LLM using GPT-5-mini for identifying distant mBC recurrence. Both were assessed against manual chart review. To optimize computation, the LLM processed clinical notes containing metasta by year in ascending temporal order. Once a patient was classified as having a distant recurrence, subsequent notes were excluded from further processing. We assessed the performance of case identification and timing of recurrence.
Our cohort included 429 women initially diagnosed with stage I to III breast cancer. Compared with manual chart review, GPT-5-mini achieved a sensitivity of 1.00 (95% CI, 0.96 to 1.00), specificity of 0.94 (95% CI, 0.91 to 0.96), PPV of 0.80 (95% CI, 0.72 to 0.87), and NPV of 1.00 (95% CI, 0.99 to 1.00). The algorithm had a sensitivity of 0.82 (95% CI, 0.73 to 0.89), specificity of 0.98 (95% CI, 0.96 to 0.99), PPV of 0.91 (95% CI, 0.83 to 0.96), and NPV of 0.95 (95% CI, 0.93 to 0.97). For recurrence timing, 69.0% of GPT-5-mini recurrence dates were within 1 month of the gold-standard date compared with only 10% of algorithm estimates.
In this single-institution proof-of-concept study, LLM performance is well-suited for more inclusive case capture than manual review or complex algorithm application with minimal instruction and could be further improved with disease-specific knowledge.

PMID:
43038486
Bibliographic data and abstract were imported from PubMed on 17 Nov 2026.

Read full publication at:
Please sign in to see all details.

Advertisement

Stats

  • Community rating n/a 0 votes
  • Reviewers' rating n/a 0 votes
  • Your rating

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

  • Recommendations n/a n/a positive of 0 vote(s)
  • Views 6
  • Comments 0

Recommended by

  • No recommendations yet.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.

Advertisement