Core Needle Biopsy Falls Short for HER2-Low Breast Cancer Classification, Study Finds

Core needle biopsy remains highly dependable for deciding whether a breast cancer is HER2-positive or HER2-negative, but its reliability collapses once the increasingly important HER2-low category is introduced, according to a large retrospective study from a single cancer center in China. The research, published in Diagnostic and Prognostic Research, analyzed 1,919 women with invasive breast cancer who underwent surgery at the Cancer Hospital of Chongqing University between January 1, 2020, and December 31, 2024, and found that while agreement between biopsy and surgical specimens reached 97.0 percent under the traditional binary classification, it fell to 75.9 percent when HER2-zero, HER2-low and HER2-positive were treated as separate categories, and dropped further to 63 percent under a four-category system that distinguishes IHC 0, IHC 1+, IHC 2+ with negative FISH, and IHC 2+ with positive FISH or IHC 3+.

The stakes of this discrepancy are considerable. HER2-low, defined as IHC 1+ or IHC 2+ with negative FISH testing, has become a therapeutically meaningful category because antibody-drug conjugates such as trastuzumab deruxtecan have demonstrated efficacy in HER2-low disease. The phase III DESTINY-Breast04 trial showed that trastuzumab deruxtecan significantly improved progression-free and overall survival in HER2-low metastatic breast cancer compared with physician’s choice chemotherapy. A tumor misclassified as HER2-zero on a biopsy could therefore deny a patient an effective therapy, while the reverse error could lead to overtreatment. The study’s authors argue that these numbers underscore the difficulty of robustly identifying true HER2-low disease on core needle biopsy alone using existing approaches.

The discordance was not random in direction. Among patients whose biopsies showed HER2-low status, 279 cases, representing 26.0 percent of all biopsy HER2-low tumors, were later classified as HER2-zero on the surgical resection specimen. Conversely, 125 patients, or 29.3 percent of all biopsy HER2-zero cases, converted to HER2-low on the surgical sample. Under the four-category system, the largest discordant group appeared among biopsy IHC 1+ tumors, where 234 cases, 36.7 percent of all biopsy IHC 1+ tumors, were reclassified as IHC 0 on the resection specimen. The study also found that 129 biopsy IHC 0 patients, 30.2 percent of that group, were classified as non-IHC 0 based on surgical tissue. These bidirectional shifts, particularly at the boundary between IHC 0 and IHC 1+, mirror quality assurance studies showing that weak, incomplete membrane staining is especially difficult to score consistently.

HER2 was not the only biomarker affected. Concordance rates between biopsy and surgical specimens were 95.4 percent for estrogen receptor, 91.9 percent for progesterone receptor, and 78 percent for Ki-67. The largest Ki-67 discordance involved 275 biopsy-negative patients, 28.3 percent of all biopsy Ki-67-negative cases, who were classified as Ki-67-positive on the surgical specimen, suggesting that core needle biopsy tends to underestimate proliferation in a subset of patients. In multivariate analysis, Ki-67-positive patients had more than 100 percent higher odds of being FISH-positive compared with Ki-67-negative patients regardless of sample type, while progesterone receptor positivity lowered the chance of FISH positivity only for biopsy samples.

The researchers also identified clinicopathological factors associated with HER2 concordance, indicating that discordance reflects underlying tumor biology rather than pure chance. Under the three-category system, patients whose biopsies showed more advanced clinical tumor stage had significantly higher odds of concordance: moving from cT1 to cT2-4 increased the odds of concordance by 58 percent. The presence of lymph node invasion was associated with 25 percent lower odds of being HER2-concordant. The authors suggest that larger tumors may be more comprehensively sampled by biopsy, reducing the effect of spatial heterogeneity on HER2 assessment, whereas lymph node invasion is often associated with greater genomic instability and clonal diversity, which may contribute to differences in HER2 status between the biopsy site and other tumor regions.

Under the four-category system, the pattern differed. Progesterone receptor positivity on the biopsy was associated with 35 percent lower odds of HER2 concordance, while Ki-67 positivity on the surgical specimen was associated with 31 percent higher odds. The authors note that HER2-low is enriched in hormone receptor-positive disease and displays distinct molecular signatures compared with both HER2-zero and HER2-positive cancers, and that such tumors may show more subtle or regionally variable HER2 expression, making precise category assignment on small biopsy samples more challenging. Highly proliferative tumors, by contrast, may exhibit more uniform HER2 expression across tumor areas.

Beyond documenting discordance, the study developed supervised machine learning models to predict the surgical HER2-low status using only preoperative biopsy data. Models based solely on biopsy clinicopathological features achieved only modest discrimination, with a mean area under the receiver operating characteristic curve of 0.615, better than random guessing but limited. Models using only the biopsy HER2 classification results performed substantially better, with a mean AUC of 0.824, indicating that the HER2 immunohistochemistry and FISH data themselves remain the key determinants of the surgical HER2-low classification. Adding all clinicopathological variables to the HER2 classification produced only a small gain, with a mean AUC of 0.831, and more flexible approaches such as random forest models and neural networks, as well as interaction variables and integrated risk scores, did not improve accuracy.

The final selected model was a simple decision tree built from two engineered biopsy HER2 variables, one three-category and one four-category, which can be translated into the probability of surgical HER2-low status for each biopsy category. The estimated probabilities of surgical HER2-low were 0.29 for biopsy IHC 0, 0.62 for IHC 1+, 0.88 for IHC 2+ with negative FISH, and 0.07 for HER2-positive tumors. In practice, this means biopsy IHC 2+/FISH-negative tumors can be considered highly likely to remain HER2-low on the surgical specimen, whereas IHC 0 or IHC 1+ tumors may warrant confirmatory testing when antibody-drug conjugate eligibility is under consideration. The authors suggest that IHC 0 and IHC 1+ results from biopsies should be interpreted cautiously, and that surgical confirmation or expert adjudication should be considered when treatment decisions hinge on that distinction.

The team also built a model for a clinically relevant composite outcome: patients whose surgical specimens showed both HER2-low status and metastasis in lymph node, lymphovascular or perineural regions. In this setting, biopsy clinicopathological variables alone and biopsy HER2 classification alone performed comparably, with mean AUCs of 0.692 and 0.719 respectively, but combining both types of features in a Lasso-regularized model improved discrimination to a mean AUC of 0.771. The final risk score included biopsy HER2 classification in binary and four-category forms, estrogen and progesterone receptor status, and clinical T and N stage. More advanced nodal status and higher hormone receptor expression were associated with higher risk of the composite outcome, while biopsy HER2 positivity was strongly protective, consistent with the observation that classic HER2-positive tumors are rarely in the surgical HER2-low group. All variables in the risk score are routinely collected in clinical practice, making the model implementable without additional testing or cost.

The authors acknowledge several limitations. The study is retrospective and single-center, conducted in a Chinese population, and external validation in other geographic and institutional settings is needed before the models can be generalized, particularly given documented inter-laboratory variability in HER2-low classification. Although discordant cases were re-reviewed by a blinded experienced breast pathologist, residual inter-observer variability in HER2 scoring, especially at the IHC 0/1+ and 1+/2+ boundaries, is inevitable and was not adjusted for. Follow-up duration and the number of death events were limited, precluding robust conclusions about the prognostic impact of HER2 concordance, and the biopsy histologic grade variable had a high proportion of missing data that prevented its use in predictive modeling. The authors call for multi-center, prospective validation of both the concordance findings and the prediction models, and suggest that embedding such models into digital pathology systems or electronic medical records could eventually enable automated estimation of surgical HER2-low probability at the time of biopsy reporting, though they emphasize that real-world application requires external validation, prospective testing, and careful evaluation of workflow impact first.

Subject of Research: Concordance of HER2-low status between core needle biopsy and surgical resection specimens in breast cancer and machine learning prediction of surgical HER2-low status

Article Title: Accurate assessment of surgical HER2-low status based on core needle biopsy data of breast cancer

Article References: Li, W., Chen, Z., Li, J., Ding, T., Lei, H., Chen, R., & Li, J. (2026). Accurate assessment of surgical HER2-low status based on core needle biopsy data of breast cancer. Diagnostic and Prognostic Research, 10(1), Article 28. https://doi.org/10.1186/s41512-026-00241-y

Image Credits: AI Generated

DOI: 10.1186/s41512-026-00241-y

Keywords: breast cancer, HER2-low, core needle biopsy, surgical resection, HER2 concordance, immunohistochemistry, FISH, machine learning, prediction model, antibody-drug conjugate, lymph node metastasis, retrospective study