Background: HER2-low expression has gained clinical relevance in breast cancer (BC) due to the availability of anti-HER2 antibodyedrug conjugates for patients with HER2-low metastatic BC. The well-reported instability of HER2-low status during disease evolution highlights the need to identify patients with HER2-0 primary BC who may develop a HER2-low phenotype at relapse. In response to the urgency of maximizing treatment access, we utilized artificial intelligence to predict this occurrence. Patients and methods: We included a large multicentric retrospective cohort of patients with BC who underwent tissue resampling at relapse. The dataset was preprocessed to address relevant issues such as missing data, feature abundance, and target class imbalance. We then trained two models: one focused on explainability [Extreme Gradient Boosting (XGBoost)] and another aimed at performance (an ensemble of XGBoost and support vector machine). Results: A total of 1200 patients were included in this study. Among 386 patients with HER2-0 primary BC and matched HER2 status at relapse, 42.5% (n = 157) converted to a HER2-low phenotype. The explainable model achieved a balanced accuracy of 58%, with a sensitivity of 53% and a specificity of 64%. The most important variables for this model were primary BC phenotype [mean Shapley value (SHAP) 0.540], primary BC histological type (SHAP 0.101), grade (SHAP 0.182), and sites of relapse (SHAP 0.008-0.213). The ensemble model had a balanced accuracy of 64%, with a sensitivity of 75% and a specificity of 53%. Conclusions: This work represents one of the first proof-of-concept applications of machine learning models to predict a highly relevant phenomenon for drug access in modern BC oncology. Starting with an explainable model and subsequently integrating it with an ensemble approach enabled us to enhance performance while maintaining transparency, explainability, and intelligibility.

We and others reported that HER2-low expression is unstable during disease evolution. The availability of anti-HER2 antibody-drug conjugates (ADCs) for HER2-low metastatic BC (MBC) patients (pts) commands attention to the identification of those with HER2-0 primary BC who may acquire HER2-low phenotype at relapse. Driven by the urgency of maximizing treatment access, we developed an AI-based model to predict this phenomenon. We included a large multicentric retrospective cohort of pts with matched HER2 status on primary and MBC samples. All the variables in the dataset were used to build the model, including, among others: primary and MBC phenotype, timing and site of relapse and of relapse biopsy, treatments for primary BC. The dataset underwent preprocessing to address missingness: we applied Multiple Imputation by Chained Equations algorithm to observations without missing values in key relapse-related features. Features were tested for collinearity through the calculation of the Variance Inflation Factor, and one-hot-encoded due to the high representation of categorical variables. In the final dataset, pts were randomly assigned to training and test set. The training set (n=561) was subjected to random under-sampling to mitigate the target class imbalance. To prevent lack of explainability, a Generalized Linear Model was then fitted (10-fold repeated cross-validation to prevent overfitting). Among 749 pts (final dataset), 296 had HER2-0 primary BC, of which 109 (37%) gained HER2-low expression at relapse. The model was able to predict this switch with a 74% accuracy. Sensitivity and specificity were 74%. Hormone receptor expression and timing of relapse biopsy (≤ vs >2 years from relapse) were the variables associated with the highest importance for the AI model (p<0.05). Our AI model, based on clinicopathological features, showed promising accuracy in predicting the conversion from HER2-0 primary BC to HER2-low phenotype at relapse. This model may help identifying pts with HER2-0 primary BC for whom a relapse biopsy should be prioritized to maximize treatment access to anti-HER2 ADCs for HER2-low BC.

Although 1% is the recommended cutoff for defining triple-negative breast cancer (TNBC), growing evidence suggests that 10% cutoff may better recapitulate TNBC. Conversion to TNBC at relapse is associated with poor survival. We primarily aim to assess the prognostic impact of phenotypic conversion to estrogen receptor (ER)-low BC in patients experiencing relapse. Relapsing BC patients from two Institutions were included. Patients were categorized in: TNBC (ER=0%, HER2-0/low), ER-low (ER=1-9%, HER2-0/low), Luminal (ER=10-100%, HER2-0/low), HER2+. Overall survival (OS) and post-relapse survival (PRS) were adopted as endpoints. 877 patients were included. The proportion of ER-low tumors was 3.2% on primary BC and 2.9% on relapse. When assessing the prognostic impact of primary BC phenotype, TNBC and ER-low retained a similar and significantly poorer prognostic impact than Luminal and HER2+ BC. In detail, median OS [mos] was: TNBC 68.6, ER-low 47.6, Luminal 125.4, HER2+ 121.4, p<0.001). PRS analysis described the same phenomenon (p<0.001) Superimposable findings were observed when considering the prognostic impact of tumor phenotype at relapse (OS, p<0.001; PRS, p<0.001). At relapse, 6.4% of TNBC, 2.5% of Luminal and 1.9% of HER2+ primary BC cases switched to ER-low phenotype (overall conversion rate to ER-low BC: 2.8%). Among Luminal BC patients, those converting to ER-low at relapse showed the worst outcome, with poorer survival than those maintaining Luminal BC or converting to either TNBC or HER2+. In detail, median OS [mos] was: concordant Luminal 134.7, conversion to ER-low 54.4, conversion to TNBC 80.4, conversion to HER2+ 129.9, p<0.001; median PRS [mos] was: concordant Luminal 51.4, conversion to ER-low 12.4, conversion to TNBC 29.6, conversion to HER2+ 44.4, p<0.001. ER-low BC was associated with unfavorable prognosis, similar to TNBC and significantly poorer than Luminal and HER2+. Luminal BC patients converting to ER-low phenotype experienced the worst survival rates, even worse than those converting to TNBC, possibly due to the limited access to TNBC treatment algorithms. Our study supports the assimilation of ER-low BC to TNBC.

Approximately a half of breast tumors traditionally classified as HER2-neg exhibit HER2-low expression (IHC 1+ or 2+ and ISH neg.). We recently described a high instability of HER2-low expression from primary breast cancer (BC) to relapse (Miglietta F et al., ESMO Breast Cancer 2021). Aim of this study is to track the evolution of HER2-low expression from primary BC to residual disease (RD) after neoadjuvant treatment. Patients undergoing neoadjuvant treatment with available baseline tumor tissue and matched samples of RD (in case of no pCR) were included. HER2 expression was evaluated according to ASCO/CAP recommendations in place at the time of diagnosis. Cases diagnosed between 2007 and 2013 were reviewed to comply with the 10% cutoff of IHC for HER2-positivity. HER2-neg cases were further classified as HER2-0 or HER2-low (IHC 1+ or 2+ and ISH neg.). 447 patients were included. Primary BC phenotype was: HR-pos/HER2-neg 23%, triple-negative (TN) 35%, HER2-pos 42%. HER2-low cases were 56% of the HER2-neg cohort and were significantly enriched in the HR-pos/HER2-neg vs TN subgroup (69% vs 47%, p=0.001). In patients failing to achieve pCR after neoadjuvant treatment (n=292), the overall rate of HER2 expression discordance was 27%, mostly driven by cases converting either from HER2-0 primary BC to HER2-low RD (9%) or from HER2-low primary BC to HER2-0 RD (15%; Table). Overall, 36% of non-pCR patients had a HER2-low expression on RD, including 12% of patients with TN and 24% of patients with HR-pos/HER2-neg disease. Among HR-pos/HER2-neg patients with HER2-low expression on RD, 23% had an estimated high risk of relapse according to the residual proliferative cancer burden (RPCB class 3).Table: 212PPrimary tumorResidual diseaseHER2-0HER2-lowHER2-posTotaln%n%n%n%HER2-05117269007726HER2-low431570241<111439HER2-pos0083933210135Total9432104369432292100 Open table in a new tab HER2-low expression showed high instability from primary BC to RD after neoadjuvant treatment. HER2-low expression on RD may guide personalized adjuvant treatment for high-risk patients in the context of clinical trials with novel anti-HER2 antibody-drug conjugates.

About a half of breast cancers traditionally classified as HER2-negative show a low HER2 expression (IHC 1+ or IHC 2+ and ISH negative) that can be targeted by new antibody-drug conjugates. There is no data on the evolution of HER2-low status from primary tumor to relapse. Patients with matched primary and relapsed breast cancer samples from two Institutions (IOV-IRCCS Padova and Treviso Hospital) were included. HER2 was evaluated according to ASCO/CAP recommendations in place at the time of diagnosis. Cases diagnosed between 2007 and 2013 were reviewed by IHC to comply with the cut-off of >10% cells staining for HER2 positivity. Moreover, 100 random samples were reviewed by a blinded pathologist: agreement with the original report was 80%. HER2-neg cases were sub-classified as HER2-low (IHC 1+, or IHC 2+ and ISH not amplified), or HER2-0 (IHC 0). 575 patients were included. Primary tumor phenotype was: 59% luminal-like (HR+/HER2-neg), 25% HER2-pos, 16% triple-negative. The proportion of HER2-low cases was 34% on the primary tumor and 38% on the relapse samples. Among HER2-neg cases, HER2-low status was more frequent in Luminal-like vs triple-negative tumors (47% vs 41% on primary tumor samples, p=0.268; 54% vs 40% on relapse samples, p=0.006). The overall rate of HER2 discordance was 38% (Table), mostly represented by HER2-0 switching to HER2-low (15%) and HER2-low switching to HER2-0 (14%). A minority (9%) of cases lost or acquired HER2-positivity. Among patients with a primary HER2-neg tumor, the rate of HER2 discordance was higher in luminal-like vs triple-negative cases (45% vs 35% p=0.080). This difference was mostly driven by cases switching from HER2-0 to HER2-low: 40% of luminal-like/HER2-0 vs 24% of triple-negative/HER2-0 patients (p=0.088).Table: 4MO_PRRelapseHER2-0HER2-lowHER2-positiveTotalPrimary tumourn%n%n%n%HER2-013423%8515%132%23240%HER2-low7814%10919%92%19634%HER2-positive61%234%11820%14726%Total21738%21838%14024%575100% Open table in a new tab HER2-low expression is highly unstable during disease evolution. Relapse biopsy in case of a primary HER2-0 tumor may open new opportunities for treatment in a relevant proportion of patients.