Abstract 171P
Background
Disorder in collagen fiber architecture, characterized by alterations in collagen organization, plays a critical role in tumor progression. In this study, we developed a collagen-based computational pathology biomarker and evaluated its association with the benefits of adjuvant treatment in a clinical trial cohort.
Methods
The study utilized H&E-stained Whole Slide Images (WSIs) of DCIS (n = 755) from the UK/ANZ DCIS randomized controlled trial, which included four treatment allocations: no adjuvant treatment, radiotherapy, tamoxifen, radiotherapy+tamoxifen. Two-thirds of ipsilateral breast events (IBE) from each treatment group were used for training, with each case (patient with IBE) matched to two controls by age (+/-7 years) and treatment. The collagen score was evaluable in 713 of 755 samples, including 270 of 288 in training set. For the development of the collagen-tamoxifen score (CTS), 102 samples (IBE=35) out of 270 formed the training set (TrS) and 140 samples (IBE=20) formed the validation set (VaS). Machine learning models were employed to extract a series of features relating to collagen arrangement from WSIs. A logistic regression model was trained using these features and generated a continuous score (CTS). A 66th percentile risk score threshold was applied on CTS in TrS to stratify patients into low or high-risk groups in TrS and VaS.
Results
In TrS+VaS, 194 of 713 (27%) patients were classified as CTS-high. Among 481 patients with ER (clonal) status data, a weak negative correlation (spearman rho -0.19, p<0.0001) was seen between ER (clonal) and CTS. Specifically, 61% (109/178) of ER-negative DCIS were CTS-low, while 21% (63/303) of ER-positive DCIS were CTS-high. In TrS and VaS, CTS-high DCIS had a greater than 3-fold risk of ipsilateral breast event (IBE) as compared with CTS-low DCIS [TrS: Hazard Ratio (HR) = 4.54; 95% Confidence Interval (95%CI), 2.27-9.06, p<0.0001; VaS: HR = 3.46; 95%CI, 1.41-8.48, p=0.0067).
Conclusions
Disorder in collagen architecture is associated with tamoxifen resistance in DCIS patients. Our computational pathology based collagen-tamoxifen score has a role, independent of ER-status, in predicting tamoxifen benefit in DCIS.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.
Resources from the same session
173P - Unveiling a novel EpCAM-CD24+ circulating cells with unidentified origin associated with breast cancer distant metastasis
Presenter: Evgeniya Grigoryeva
Session: Poster session 08
174P - Prognostic value of the immune and metabolic profile in the response to neoadjuvant treatment with ICIs in triple-negative breast cancer patients (TNBC)
Presenter: Lucía Serrano García
Session: Poster session 08
175P - Utility of artificial intelligence (AI) in Ki67 scoring of a breast cancer (BC) patient population
Presenter: Xavier Pichon
Session: Poster session 08
176P - ERBB2 amplifications across sex, race, and cancer types
Presenter: Marc Machaalani
Session: Poster session 08
177P - HER2 testing in multiple solid tumors: Concordance between 3 scoring algorithms
Presenter: Wentao Yang
Session: Poster session 08
178P - PD-L1 expression in ER-low versus triple-negative (TN) advanced breast cancer (aBC), and according to phenotypic evolution from primary to recurrent disease
Presenter: Federica Miglietta
Session: Poster session 08
179P - Multimodal deep learning integrating MRI and molecular profiles for predicting outcomes in triple-negative breast cancer
Presenter: Seong Hwan Park
Session: Poster session 08
181P - Molecular characterization and immune microenvironment analysis of MSI-H patients with or without MMR gene mutations
Presenter: Mengxi Ge
Session: Poster session 08
182P - Multi-modal artificial intelligence outperforms image-based approaches for mutation prediction from H&E tissue images in colorectal cancer
Presenter: Marc Päpper
Session: Poster session 08