Model Predicts Breast Cancer Residual Burden
An interpretable machine learning model was developed to predict residual cancer burden after neoadjuvant therapy in HER2-positive breast cancer.
Updated on Nov. 19, 2026 in Cancer

Researchers developed an interpretable gradient boosting machine model to predict residual cancer burden after neoadjuvant therapy in HER2-positive breast cancer. The model was built from a retrospective study of 692 patients treated between November 2018 and May 2025 at a tertiary center.
Why it matters
Pre-treatment prediction of residual cancer burden could help clinicians stratify patients by risk before therapy begins. The researchers note the model's current predictive performance does not support stand-alone clinical decision-making.
The gradient boosting machine reached an AUC of 0.796 in the independent test set across 692 patients, selected from ten algorithms using nine routinely available predictors identified by LASSO. Prospective validation and performance in external cohorts remain unreported.
The details
SHAP analysis identified HER2 status, hormone-receptor status and clinical nodal stage as the strongest contributors to predictions. The researchers also implemented a Shiny web application of the model for research and educational use, and in a propensity score-matched comparison found no significant difference in achieving favorable residual cancer burden between trastuzumab-pertuzumab and trastuzumab-pyrotinib regimens.
Timeline
Patients in the retrospective cohort were treated between November 2018 and May 2025.
Health Landscape
The study extends residual cancer burden assessment, a post-surgical response measure used in HER2-positive breast cancer, by attempting to forecast that outcome before neoadjuvant therapy begins. Its performance and the decision to use it in practice will depend on validation beyond this single-center retrospective cohort.
For patients with HER2-positive breast cancer, a validated pre-treatment risk tool could eventually inform discussions about expected response to neoadjuvant therapy. The model is not yet validated for that use, and current performance does not support it as a stand-alone decision aid.
The takeaway
Pre-treatment prediction tools for breast cancer response are an emerging research area, but interpretable models still require prospective and multi-center validation before influencing treatment choices. Patients should continue to rely on their oncology team for regimen decisions.
Further reading
Explore more on Cancer research and treatment developments.
Source note: This article includes information reported by Nature.



