FINANCE

AI Outshines Old Tools in Detecting Distant Breast Cancer Recurrence

Single-institutionTue Nov 17 2026

The researchers wanted to see if a large language model could find when breast cancer returned far from the original tumor. They pulled electronic health records and notes from 2010 through 2023. A group of 429 women with early‑stage breast cancer was the focus. The team ran two detection methods: a GPT‑5‑mini based LLM and a known discrete‑field algorithm. Both were checked against manual chart reviews to see how well they worked.

The LLM performed almost perfectly. It caught every true recurrence (sensitivity 100%) and missed almost no healthy cases (specificity 94%). Its positive predictive value was 80% and its negative predictive value was 100%. The older algorithm was a bit weaker. It identified 82% of true recurrences and correctly ruled out 98% of non‑cases. Its PPV was 91% and NPV 95%. When it came to timing, the LLM got the date right within a month for about 69% of patients, while the algorithm did so for only 10%.

Because the LLM needed little guidance, it could capture more cases than a human reviewer or a complex algorithm. The study suggests that AI tools could become a practical way to monitor patients for distant spread. Adding disease‑specific knowledge might push performance even higher. The findings point to a future where routine AI checks help doctors spot late recurrences faster.

This proof‑of‑concept was done at a single hospital, so results may vary elsewhere. The researchers still see strong potential for LLM‑based detection in real‑world settings. More work will be needed to validate the approach across different centers and cancer types.

actions