AI predicts breast cancer recurrence risk in patients better than genomics: Nature study
AI has been used to predict breast cancer recurrence risk and disease-free interval more accurately than the standard 21-gene assay, a study in Nature Communications suggests.
The AI test uses routine clinical data — such as age, cancer stage and hormone receptor status — and the patient’s digital pathology biopsy slides to produce a continuous risk score between 0 and 1.
Led by US researchers, the study said the predictions could be generated in less than 60 minutes rather than the 10-30 days wait involved in running the 21-gene assay, Oncotype DX.
The added that gathering the data costs around $1015, compared with the $5500 cost of the assay.
The AI test was developed based on a training cohort of 4659 breast cancer patients, of whom 16% had recurrence, and an evaluation cohort of 3502 patients (8% with recurrence).
Patients with stage IV disease or a prior history of breast cancer were excluded.
The concordance index (c-index) for predicting disease-free interval was 0.67 for the AI test and 0.61 for Oncotype DX, with the AI frequently giving a low-risk classification to patients judged as intermediate risk via Oncotype DX.
While the c-index confidence intervals overlapped, the researchers said the c-index “is known to have low statistical power and … often understates incremental improvements that are clinically significant”.
“Although these findings are based solely on prognostic associations and do not yet confirm predictive value for treatment benefit, strong risk stratification alone might provide an additional data point in therapy escalation or de-escalation decisions.”
For example, patients judged as low risk could potentially avoid adjuvant chemotherapy.
The AI also predicted disease-free interval reliably across cohorts, including age, menopausal status, nodal status, race, tumour size, oestrogen and HER2 receptor status and administered adjuvant therapy.
The researchers pointed out that the US National Comprehensive Cancer Network guidelines currently did not support any predictive tools for recurrence risk in triple-negative or HER2+ breast cancer.
Given the AI test’s reliability for both groups (c-index 0.71 for triple-negative and 0.67 for HER2+ breast cancer), it could become “a single tool to inform treatment decisions in all breast cancer patients”, they said.
“With more rigorous validation through clinical trials, it could become an essential tool in personalised cancer care.”
More information: Nature 2026; 20 May.
