The breast cancer prognostic step is moving from tissue-burning to tissue-reading

The standard prognostic step for early-stage breast cancer is shifting from genomic assays that burn tissue and take weeks to AI that reads slides a pathologist already examines and returns a recurrence-risk score in days.

A Nature Communications paper, 3,500+ patients across 15 populations in seven countries, provides the evidence: the AI test reads the same slides a pathologist would already examine, adds stage, age, and hormone-receptor status, and returns that same recurrence-risk estimate in days — while matching the standard 21-gene genomic recurrence test at how well it separated patients who recurred from those who didn't (C-index 0.67 versus 0.61). The workflow advantage — days instead of weeks, no tissue consumed, no sample gone before a second opinion — plus comparable accuracy makes this a category-level shift in how prognostic testing is performed.

Lab economics is where this lands first: pathology departments earn revenue per assay, and an H&E-slide-based test collapses that line. Patients gain faster treatment conversations and tissue preserved for whatever test comes next. Triple-negative breast cancer, where no genomic assay is standard, hit a C-index of 0.71. The missing baseline is the win.

The paper does not replace the oncologist. A recurrence-risk estimate is a probability, not a treatment. Clinicians sit on top of any number a model produces, and that is how it should stay.