Truace
Subscribe

AI sorted 33.8% to 111.8% more breast cancer patients as low-risk in past data, a study found

Study title: Development and validation of a machine learning clinicogenomic model to improve prognostic stratification in ER-positive/HER2-negative early breast cancer

Researchers developed and externally validated a machine-learning-based clinicogenomic model for ER-positive/HER2-negative early breast cancer using four independent datasets totaling 5,132 patients. The random survival forest integrated a research-grade 21-gene recurrence score, a 14-gene immunoglobulin signature, and clinicopathologic features, and was tested against the standalone 21-gene score.

Read the study abstract

Background Several gene expression signatures (GESs) are used for risk stratification in estrogen receptor-positive/human epidermal growth factor receptor 2-negative (ER-positive/HER2-negative) early breast cancer. Recent integrative approaches combine tumour proliferation, estrogen receptor signalling, immune activity, and clinicopathologic features, yielding gains in prognostic accuracy. We therefore developed a machine-learning-based clinicogenomic prognostic model integrating a research-grade implementation…

Truvace record1 min read1 sourceEvidence: Moderate · 70/100 — Controlled or validated comparison · n = 5,132 · 1 independent sourcePeer-reviewedTRV-2026-1132 · cite or check it
Development and validation of a machine learning clinicogenomic model to improve prognostic stratification in ER-positive/HER2-negative early breast cancer
"Hard metal lung disease/Giant cell interstitial pneumonia (GIP) - Case 259" by Pulmonary Pathology is licensed under CC BY-SA 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/2.0/.

In brief

By the September 2026 publication date, retrospective validation showed the model identified 33.8% to 111.8% larger low-risk groups while maintaining survival estimates, and improved discrimination with concordance index gains of 0.035 to 0.058. The work remains retrospective and research-grade, so prospective clinical utility, treatment decision impact, and generalizability beyond the studied datasets remain uncertain.

Main points

  1. Retrospective integrative analysis included four independent datasets totaling n = 5132 systemically untreated or endocrine-treated patients
  2. One dataset used for development of random survival forest model, three datasets used for external validation
  3. Model incorporated research-grade implementation of 21-gene recurrence score, 14-gene immunoglobulin signature, and clinicopathologic features
  4. External validation showed relative increases of 33.8 to 111.8% in low-risk group size while maintaining similar or higher horizon-specific survival estimates

The gain

In three external retrospective datasets, an AI model combining gene scores and clinical features identified low-risk groups 33.8% to 111.8% larger than the standalone 21-gene score while keeping similar or higher survival estimates.

The full story

The study developed a random survival forest model using one dataset and validated it in three independent external datasets of systemically untreated or endocrine-treated patients. Performance was compared to a standalone research-grade 21-gene recurrence score using risk stratification, discrimination measured by concordance index and integrated area under time-dependent ROC, and prediction error via Cox-recalibrated integrated Brier score.

Explainability analyses showed time-dependent contributions: the 21-gene RS and clinicopathologic variables drove early prognostic performance while the immune component contributed a smaller but more stable effect over time. Risk separation measured by difference in restricted mean survival time was consistently greater for the RSF model.

Sources

  1. Peer-reviewedBreast Cancer Research2026-09-16

The debate