Abstract / Summary
Breast cancer is a leading cause of cancer-related mortality among women globally. Accurate identification of HER2 status, particularly the HER2-low category, is crucial for effective treatment strategies, given the emergence of HER2-targeted therapies for this subgroup. However, discrepancies often arise between core needle biopsy (CNB) and surgical resection tissue (SRT) regarding more refined HER2 status. This retrospective study included female patients with invasive breast cancer who underwent surgical resection between January 1, 2020, and December 31, 2024, at a single cancer center in China. We assessed the concordance of HER2 status, alongside other clinicopathological features on both CNB and SRT samples. We then developed a supervised binary modeling framework to predict SRT HER2 status (with metastasis) based on CNB data. Among 1,919 patients analyzed, the concordance rate for HER2 classification was 97.0% in a binary system but dropped to 75.9% when using a 3-category system, with discordance noted particularly between HER2-zero and HER2-low categories. Logistic regression identified significant factors impacting concordance. The predictive model utilizing CNB HER2 classification achieved a mean AUC of 0.824, indicating substantial predictive capability for SRT HER2-low status. Integrating CNB clinicopathological variables further improved prediction accuracy (mean AUC from 0.691 to 0.771) for SRT HER2-low with metastasis in clinically relevant regions. While CNB effectively determines binary HER2 status, establishing HER2-low categorization reveals discrepancies. Our findings highlight the importance of accurate preoperative HER2 assessment, supported by AI-based prediction models that enhance clinical decision-making in breast cancer management targeted toward HER2-low patients.