Tumour-infiltrating lymphocytes (TILs) are prognostic and predictive biomarkers in breast cancer. High pre-treatment TILs are associated with a favourable prognosis and improved response to systemic therapies. However, the role of TILs in mediating response to radiotherapy in breast cancer remains underexplored, with scarce clinical evidence to date. In this Review, we present an overview of current evidence and potential mechanisms, highlight opportunities for integrating TILs into radiation oncology trials and clinical practice, and call for standardised TIL reporting to accelerate biomarker-driven individualisation of radiotherapy in breast cancer.

Abstract Ductal carcinoma in situ is a non-obligate precursor lesion of breast cancer. Often detected by mammography, most cases are managed through surgical and/or radiotherapy approaches. Today, it is not possible to predict which patients will progress to invasive disease. Here, we evaluate high-depth whole-genome sequenced ductal carcinoma in situ, enriched for high-grade clinical lesions, to understand whether deep WGS could reveal biological insights and/or personalized therapeutic vulnerabilities that may be targetable. We find genomic locations that are likely susceptible to producing the initiating lesion for structural variations prone to subsequent evolution, termed SHOREs. We additionally highlight individualized therapeutic potential that would otherwise not be appreciable without whole genome sequencing. We posit that holistic whole genome sequencing profiling could offer a more precise stratification approach, discerning higher-risk cases for prospective clinical studies on personalized therapies, from truly low-risk cases suitable for active monitoring.

ABSTRACT Background Population-scale molecular profiling integrated into routine healthcare could accelerate biomarker discovery, validation, and implementation, but the feasibility and sustainability of such an approach have rarely been demonstrated prospectively. The Sweden Cancerome Analysis Network - Breast (SCAN-B) Initiative was established to integrate prospective molecular profiling with population-based breast cancer care and create an infrastructure for translating molecular discoveries into clinical practice ( ClinicalTrials.gov identifier NCT02306096 ). Methods We evaluated the first 10 full calendar years of SCAN-B, encompassing patients with primary invasive breast cancer enrolled between August 30, 2010 and December 31, 2020. Enrollment and biospecimen collection were compared with all eligible breast cancer diagnoses in participating hospitals to assess population coverage and representativeness. Clinicopathological characteristics, treatments, recurrence-free survival, overall survival, RNA-sequencing-based molecular subtypes and risk-of-recurrence, and somatic mutations were evaluated. We additionally report the translation of SCAN-B molecular profiling from the research setting into routine clinical diagnostics. Results Among 16,381 estimated eligible breast cancer diagnoses, 13,940 patients (85.1%) were prospectively enrolled across participating Swedish hospitals. Baseline blood samples were obtained from 98.4% of enrolled patients and tumor specimens from 71.1%; 9,323 tumors (94.0% of submitted tumor specimens) underwent RNA-sequencing. The enrolled cohort was broadly representative of the underlying breast cancer population across major clinicopathological characteristics. Integration of longitudinal clinical data with molecular profiling enabled characterization of real-world treatment patterns, long-term outcomes, molecular subtypes, risk-of-recurrence, and the somatic mutational landscape in this population-based cohort. Building on prospective real-time RNA-sequencing and subsequent development and validation of single-sample molecular subtype and risk-of-recurrence predictors, the SCAN-B workflow was transferred into routine clinical molecular diagnostics in Skåne and Blekinge in 2021. Through January 2026, more than 3,000 patients had received clinical RNA-sequencing-based molecular subtype and risk-of-recurrence reports, while prospective SCAN-B enrollment and transfer of samples and molecular data into the research infrastructure continued. Patient enrollment continues prospectively, with over 23,000 patients accrued as of January 2026. Conclusions A prospective, population-based molecular profiling program can be integrated into routine breast cancer care at scale while maintaining high population coverage and representativeness. Over more than a decade, SCAN-B progressed from prospective biosampling and molecular profiling through biomarker development and validation to implementation of RNA sequencing-based testing in routine healthcare. This model establishes a continuous framework linking population-based molecular research, biomarker discovery and validation, and clinical implementation, and provides a strategy for integrating precision oncology research with routine cancer care. Trial registration ClinicalTrials.gov identifier NCT02306096

Persistent xcirculating tumor DNA (ctDNA) during neoadjuvant treatment (NAT) of early breast cancer (EBC) indicates high-risk disease. Similarly, detection of ctDNA post-resection indicates molecular residual disease (MRD) and impending relapse. For ctDNA to be integrated into EBC management, accessible and scalable diagnostics are required. Here we apply an ultrasensitive, personalized tumor-informed approach to ctDNA evaluation predicated on analyses of structural variants (SVs) using a novel digital PCR (dPCR) multiplex SV technology. 136 patients eligible for NAT (29.4 For 136 patients with early breast cancer, 1497 plasma samples were collected during neoadjuvant therapy and post-operative adjuvant therapy follow-up (median 6.5 years) and analyzed for ctDNA using an ultrasensitive tumor-informed structural variant-based personalized multiplex dPCR approach. For 136 patients with early breast cancer, 1497 plasma samples were collected during neoadjuvant therapy and post-operative adjuvant therapy follow-up (median 6.5 years) and analyzed for ctDNA using an ultrasensitive tumor-informed structural variant-based personalized multiplex dPCR approach.

In the era of immune checkpoint inhibitors for cancers, the need for prognostic biomarkers to identify patients most likely to achieve a durable response has become increasingly more relevant. Tumour-infiltrating lymphocytes (TILs) have gained significant interest, as they can be evaluated using standard haematoxylin and eosin-stained slides, making it a widely accessible and cost-effective biomarker. In addition to their practicality, TILs provide prognostic insights into the interplay between the immune system and tumour cells. While the morphological assessment of TILs has been standardised in breast cancer, comprehensive guidelines for their evaluation in gastro-oesophageal carcinomas (GEC) are still lacking. This narrative review examines the current literature on the composition, clinical implications and therapeutic utility of TILs in GEC. These insights are used to propose a framework with recommendations for standardised evaluation and reporting of TILs in GEC, while also highlighting pitfalls specific to GEC pathology. These recommendations serve as a vital first step towards the widespread use and validation of TILs as a biomarker.

External quality assessment (EQA) schemes for pathology are essential, yet large/international programmes do not assess morphology-based biomarkers or address local/regional needs. This study outlines bottom-up initiated, flexible Swedish Digital Pathology EQA rounds for breast pathology, and presents results from the 2021 and 2023 rounds. Six breast carcinoma cases were selected for each EQA round by the Swedish Breast Pathology Expert Group (KVAST Breast). Whole tissue slides stained with HE, IHC, and ISH were anonymized, digitized, and uploaded to the digital EQA platform. Biomarkers were selected based on national registry data analysis and pathologist and clinician feedback. The 2021 round assessed Nottingham grade (NHG), oestrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2), while the 2023 round focused on NHG, HER2-low, and global Ki67. Twenty-seven pathology departments participated. From 2021 to 2023, the variability of NHG assessment on digital slides improved from moderate to substantial (kappa 0.50; 95

3057 Background: Persistent circulating tumor DNA (ctDNA) detection during neoadjuvant treatment (NAT) of early-breast cancer (EBC) indicates high-risk disease. Following surgical resection, ctDNA-positivity indicates molecular residual disease (MRD) and heralds occult metastatic disease relapse. To incorporate ctDNA into EBC management, scalable and widely accessible diagnostic methods are necessary. Here we apply an ultrasensitive, personalized tumor-informed approach to ctDNA analysis leveraging structural variant (SV) detection using a novel multiplex digital PCR (dPCR) technology. Methods: 116 patients with stage I-III EBC (31.0% TNBC, 43.1% HR+/HER2- and 24.1% HER2+) and eligible for NAT were recruited through the prospective SCAN-B study (NCT02306096, substudy NeoCircle) between December 2014 and March 2019 and have been analyzed for ctDNA. Whole genome sequencing was performed on tumor material and personalized multiplex dPCR assays tracking up to 16 SVs were used for ctDNA monitoring. Plasma samples were collected at baseline, during NAT, pre- and post-surgery and at 6-monthly intervals during follow up. Results: High baseline detection was observed across all stages and subtypes (90.5% overall), and ctDNA-positivity at end-of-NAT (end-NAT) was a significant predictor of eventual disease relapse and death (relapse-free interval, RFI, hazard ratio, HR, 3.7, 95% CI 1.4-9.7; overall survival HR 7.7, 95% CI 2.2-26.6). A significant association was observed between end-NAT ctDNA clearance and pathological complete response (pCR), whereas non-pCR by itself was not a significant predictor of relapse or death in this cohort. At one or more post-operative timepoints, MRD+ was detected in 10 patients who experienced distant recurrence, with lead times up to 4 years (median 13.9 months, range 1.8-47.7 months). Similarly, ctDNA was detected in 3 of 4 patients with local recurrences and 1 of 2 patients with CNS-only recurrences. For 2 patients without presentation of clinical recurrence to date, ctDNA was detected post-operatively, with subsequent clearance during follow-up. Post-operative MRD associated with poor RFI (HR 45.5, 95% CI 13.0-159.8) and OS (HR 15.3, 95% CI 4.5-52.9). Conclusions: In this analysis of 116 patients from a prospective study in patients with EBC receiving NAT, we monitored ctDNA using an ultrasensitive tumor-informed dPCR assay tracking patient-specific SVs. ctDNA detection post-NAT and prior to surgery was associated with high-risk of disease relapse and death, outperforming pCR. Moreover, post-operative ctDNA detection was also significantly associated with disease relapse and death, with long lead-times over standard-of-care clinical assessments. These findings further validate the feasibility of SVs as an MRD analyte and support the clinical use of this approach in EBC.

PURPOSE:There is uncertainty whether estrogen receptor (ER)-low tumors with 1% to 10% IHC staining of nuclei represent a distinct molecular biological entity of breast cancer, posing significant challenges for their clinical management and the development of novel therapies. We aimed to elucidate ER-low tumor biology. EXPERIMENTAL DESIGN:We analyzed primary breast tumors included in the Swedish population-based Sweden Cancerome Analysis Network-Breast (SCAN-B) cohort, 2% (n = 174) of which were classified as ER-low. Transcriptional patterns, tumor inflammatory infiltration, and prognosis were compared between ER-low versus ER-negative (ER-neg; 0%) and ER-positive (ER-pos; >10%) tumors. RESULTS:The transcriptomes of ER-low and ER-neg tumors remarkably overlapped, displaying predominantly nonluminal PAM50 subtypes and downregulated ER signaling. All triple-negative breast cancer (TNBC) molecular subtypes were represented within ER-low/HER2-negative breast cancer. Unsupervised clustering algorithms failed to segregate ER-low/HER2-negative from TNBC tumors, and only two genes showed significant differential expression above a 1.5-fold difference between the groups. However, borderline ER-low tumors (with exactly 10% ER) were mostly assigned labels associated with luminal disease biology, suggesting possible endocrine responsiveness. Lymphocyte infiltration was comparable between ER-low and ER-neg but was significantly higher relative to ER-pos tumors. Within ER-low/HER2-negative disease, hormone receptor positivity and low/intermediate PAM50 risk of recurrence score inferred from RNA sequencing data and lymphocyte fraction ≥30% were respectively associated with a better prognosis. CONCLUSIONS:ER-low/HER2-negative is not a distinct breast cancer molecular biological entity but an integral part of TNBC, deserving similar treatments. Nonetheless, a few borderline cases with moderately active ER signaling can potentially respond to endocrine therapies. Hormone receptor-related signatures and tumor-infiltrating lymphocytes may stratify ER-low/HER2-negative tumors according to the risk of recurrence. The true benefit of endocrine therapies in ER-low breast cancer requires prospective investigation.

Introduction: Tumor-infiltrating lymphocytes (TILs) have become a significant biomarker during recent years, showcasing its predictive and prognostic potential for early and metastatic triple-negative breast cancer (TNBC). However, pathologist-read stromal TILs (sTILs) remain a semi-quantitative biomarker, susceptible to inter-observer variability. With the surge in Artificial Intelligence (AI) research, various automated approaches have been proposed to score TILs with the promise to overcome the limitations of manual assessment. However, there is a lack of studies comparing different AI models in both analytical and clinical validity with respect to mimicking the challenges of clinical practice. Methods: In this study, we aimed to investigate the variability among ten AI-based TILs scoring models (seven own-developed machine learning models in QuPath –KNN, Random Forest, Neural Network– and three pre-trained deep learning models –HoverNet Graham et al. Medical Image Analysis 2019, CellViT Hörst et al. Medical Image Analysis 2024, Abousamra et al. Frontiers in Oncology 2022–) with respect to their analytical and clinical validity on internal and external validation sets. The development cohort consisted of diagnostic tissue slides of 79 women with surgically resected primary invasive TNBC tumors diagnosed between 2012 and 2016 from the Yale School of Medicine. An independent prospective set comprising of 215 TNBC patients from Sweden diagnosed between 2010 and 2015, with 4 years median follow-up, was used for assessing the models’ clinical validity. The gold standard of this study regards manual sTILs scoring from two expert pathologists. Results: Moderate correlation in analytical validity (Internal validation set: Spearman’s r= 0.72-0.84, p<0.001; External validation set: Spearman’s r=0.63-0.73, p<0.001) is demonstrated across AI methodologies and training strategies. Training on progressively increasing number of samples improved the correlation with sTILs in internal (10 patients:r=0.79, 20:r=0.81, 30:r=0.82, 40:r=0.84, 50:r=0.83, p<0.001) but not in the external validation sets (10:r=0.70, 20:r=0.68, 30:r=0.70, 40:r=0.68, 50:r=0.73, p<0.001). HoverNet & CellViT achieved the second highest correlation with sTILs in the internal validation set (r=0.83, p<0.001) but second and third to worst in the external validation set (r=0.67 & r=0.64, p<0.001). Variabilities in the distribution of TILs scores were identified across models. Interestingly, eight out of ten models (KNN, RF, NN and HoverNet), even less extensively trained ones, showed statistically significant prognostic potential, with similar and overlapping hazard ratios (HR) in the external validation cohort (Cox regression based on IDFS-endpoint and dichotomized TILs scores at 10%, HRadjusted=0.38-0.50, p<0.047). For reference, manual sTILs demonstrated a HRadjusted=0.43 (p=0.003). Conclusion: Most AI TIL methods demonstrated similar and statistically significant clinical validity, which we believe may be attributed to the intrinsic robustness of TILs as a biomarker. The analytical discrepancies between the AI models should not be overlooked; rather, we believe that there is a need for a large and diverse clinical benchmark dataset to be used for independent model validation ensuring the comparability and reliability of AI tools before integration into the clinical practice. Citation Format: Nikolaos Tsiknakis, Joan Martinez Vidal, Johan Staaf, Ana Bosch, Anna Ehinger, Emma Nimeus, Roberto Salgado, Yalai Bai, David L. Rimm, Johan Hartman, Balazs Acs. Comparison of analytical and prognostic performance among various Artificial Intelligence models for Tumor Infiltrating Lymphocytes scoring in Triple Negative Breast Cancer: An independent validation on a prospective cohort [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P4-03-16.

The assessment of residual lymphovascular invasion (LVI) in breast cancer patients undergoing neoadjuvant therapy may be a critical factor influencing prognosis and treatment decisions. However, there is a notable discrepancy between the RCB, UICC/AJCC, and ICCR guidelines regarding how LVI should be evaluated and reported in this context. ICCR recommends including LVI in the invasive tumor size for neoadjuvant treated patients with only residual LVI affecting the Residual Cancer Burden (RCB) score. AJCC suggests that LVI should not be evaluated as invasive cancer. However, they do not recommend that such cases are considered as complete response. The RCB method does not address the LVI question at all. This editorial aims to explore the implications of these differing recommendations, highlighting the challenges in clinical practice. Even though there is limited evidence in the literature on this subject, leaving this discrepancy unaddressed leads to high variability in the staging of neoadjuvant-treated breast cancer patients among pathologists. This, in turn, may cause confusion in the clinical decision-making for these patients. The recommendation of the Swedish Breast Pathology Expert Group (KVAST breast) based on current evidence, is to report LVI as a separate prognostic biomarker in neoadjuvant setting and reporting it separately from the RCB treatment response criteria. For breast cancer patients with only LVI as residual disease in the breast without any lymph node metastasis after NACT, the Swedish Breast Pathology Expert Group recommends the following staging: RCB-0, pPR, ypT0, ypN0, L1.

Immune checkpoint inhibitors are now a part of the treatment arsenal for triple-negative breast cancer (TNBC) but refinement of PD-L1 as a prognostic and predictive biomarker in TNBC is a clinical priority. We aimed to evaluate the relevance of novel PD-L1 immunohistochemical (IHC) thresholds with regards to prognostic value, PD-L1 gene expression and TNBC molecular subtypes. We evaluated PD-L1 on a tissue microarray with the SP142 (immune cell score; IC) and 22C3 (combined positive score; CPS) IHC assays and scored abundance of TILs on H&E sections in a population-based cohort of 237 early-stage TNBC patients diagnosed from 2010 to 2015. The majority of the patients were treated with adjuvant chemotherapy and none received checkpoint inhibitors. Survival analysis was performed (overall survival, invasive disease-free survival, distant relapse-free interval) and RNA sequencing data employed for molecular profiling. As expected, PD-L1 positivity (IC≥1%; CPS ≥1) was significantly associated with better prognosis compared to zero PD-L1 expression. Importantly however, also patients with intermediate expression (IC >0%, <1%; CPS >0, <1) showed a trend towards improved outcome for all the endpoints. Tumors with intermediate IHC expression also had intermediate PD-L1 (CD274) gene expression (mRNA). Patients low in TILs (<30%) and PD-L1 tended to have the poorest prognosis. PD-L1 positive tumors clustered significantly more often as Immunomodulatory-high and Basal-Like 1-high TNBC subtypes. PD-L1-zero tumors were more prevalently of the Luminal-Androgen-Receptor-high, Mesenchymal-high and Mesenchymal Stem-like-high molecular subtypes. PD-L1-intermediate tumors categorized with neither PD-L1-positive nor PD-L1-zero tumors on the hierarchical clustering level, forming thus a unique subgroup. With both SP142 and 22C3, we identified an intermediate IHC PD-L1 group within TNBCs that was supported on the molecular level. Any PD-L1 IHC expression, even though it is <1, tended to have positive prognostic impact. We suggest that the generally accepted threshold of PD-L1 IHC positivity in TNBC should be investigated further.

568 Background: Persistent circulating cell-free tumor DNA (ctDNA) detection during neoadjuvant treatment of early-breast cancer (EBC) indicates high-risk disease. Detection of ctDNA post-resection of EBC (molecular residual disease, MRD) indicates occult metastatic disease and impending disease relapse. For ctDNA to be integrated into EBC management, accessible and scalable diagnostic tools are required. Here, we apply a highly sensitive, personalized tumor-informed approach to ctDNA evaluation predicated on analyses of structural variants (SVs) using a novel digital PCR (dPCR) SV technology. Methods: 170 patients with EBC and eligible for neoadjuvant therapy (NAT) were recruited through the prospective SCAN-B study (NCT02306096) between Dec 2014 and Mar 2019 (25.8% TNBC, 47.1% HR+/HER2- and 24.1% HER2+). Interim results are presented for the first 46 consecutive patients (comprising 567 plasma samples) where minimum QC criteria were met (10% tumor content and 10x sequencing depth). Whole genome sequencing (WGS) was performed on tumor material and personalized multiplex dPCR assays designed tracking up to 8 SVs for use in ctDNA analyses. Plasma samples were collected at baseline, during NAT, pre- and post-surgery and at 6-monthly intervals during follow up. Clinical characteristics and recurrence outcomes were recorded. Results: ctDNA was detected at one or more timepoints prior to surgery in 43/46 (93%) patients across all breast cancer subtypes, at a median variant allele frequency (VAF) of 0.14% (range 0.0002% - 27.6%). ctDNA levels remained detectable at the end of NAT in 24% patients (11/46); 5/11 (45%) ctDNA positive patients experienced disease relapse versus 1/35 (3%) ctDNA negative patients (P=0.002, Fisher’s exact test). At one or more post-operative timepoints, ctDNA was detected in 6/6 (100%) patients who experienced clinical recurrence, with lead times up to 52 months (median 11.8 months, range 3.5 to 52 months). In 40 patients without presentation of clinical recurrence to date, ctDNA was undetectable across 285/287 (99.3%) plasma timepoints collected post-surgery. Post-operative detection of ctDNA associated with poor overall survival (OS) compared to absence of ctDNA (log-rank P<0.0001). Conclusions: In this interim analysis of an ongoing prospective study in patients with EBC receiving NAT, we analyzed plasma for ctDNA using a novel tumor-informed dPCR assay tracking patient-specific SVs. ctDNA detection post-neoadjuvant therapy, prior to surgery associated with high-risk of disease relapse. Postoperative ctDNA detection was observed in 100% patients with clinical recurrence, and associated with poorer OS and long lead-times. Our data demonstrate the feasibility of SVs as an MRD analyte and provide evidence for high levels of clinical sensitivity achievable with this approach in EBC. Clinical trial information: NCT02306096 .

Abstract Background: Estrogen receptor (ER) expression is currently the most consequential biomarker for prognostic and therapeutic decision making in clinical management of early breast cancer. Endocrine therapy (ET) is an important part of standard of care in most cases with ER-positive disease, and an effective therapy for patients with endocrine-responsive tumors, decreasing the risk of recurrence and improving the rate of survival. Unfortunately, ET comes with frequently bothersome side-effects and tend to decrease quality of life. The threshold to define ER positivity as a marker of endocrine responsiveness is non-uniform between international and local guidelines. Among ER low positive tumors (ERlow ,1-10% expression) endocrine responsiveness is uncertain. Swedish national guidelines recommend a cut-off of ≥10% for ER positivity and ET prescriptions, thus patients with ERlow tumors are often exempted from adjuvant ET. This study explores the clinicopathological characteristics, global transcriptional complexity and clinical outcome of ERlow tumors to better understand their biology and response to therapy. Methods: 9138 patients diagnosed with early breast cancer between 2010 – 2021 in Sweden with available clinicopathological data and RNA sequencing data were included. Patients were classified according to ER expression: ERneg (< 1%; n=897), ERlow (1-10%, n=158) and ERhigh ( >10%, n=8083). Adjuvant ET was provided to 3.6%, 16.8% and 91.5% of patients with ERneg, ERlow, and ERhigh tumors, respectively. Clinicopathological characteristics, overall survival (OS) and global transcriptional profiles of ERlow tumors were compared with ERneg and ERhigh tumors, respectively. Results: Generally, ERlow and ERneg tumor pathological characteristics were more similar to each other but were significantly distinct compared to ERhigh tumors. However, among patients with HER2-negative disease only, significant differences related to tumor biology persisted between ERlow compared with ERneg tumors; ERlow was enriched with tumors displaying a lobular histology, NHG grades 1&2, PgRlow/high expression, low proliferation and Luminal & HER2-enriched molecular subtypes (p< 0.05 for all comparisons). Moreover, multivariable survival analyses revealed that the risk of dying was significantly lower for patients with ERlow tumors compared to patients with ERneg (p=0.005) or ERhigh tumors (p=0.011) in this cohort. The distribution of clinicopathological characteristics and overall survival between ERlow compared with ERneg or ERhigh tumors were similar among patients with HER2-positive tumors only. Global transcriptional comparisons identified only 42 genes to be differentially expressed between ERlow vs ERneg tumors (FDR< 0.05); but gene set enrichment analyses failed to identify any significantly enriched cancer-related processes/pathways within the queried databases. Conclusions: Identification of optimal therapies for all subsets of breast cancer is necessary in the pursuit of personalized medicine. These results confirm that ERneg and ERlow tumors are pathologically and transcriptionally distinct from ERhigh tumors. Although subtle differences exists in the underlying biology of ERlow compared to ERneg tumors, similar therapeutic management excluding ET for patients with ERlow and ERneg disease is recommended in the Swedish context. Omission of adjuvant ET did not seem to compromise overall survival for ERlow relative to ERneg or ERhigh tumors, urging the need for prospective studies investigating the true benefit of ET for ERlow tumors. Our results also raise reasonable clinical thoughts on the benefits of new treatment strategies such as cyclin-dependent kinase 4/6 inhibitors, immunotherapy and antibody-drug conjugates for patients with ERlow expressing tumors. Citation Format: Siker Kimbung, Srinivas Veerla, Anna Ehinger, Johan Vallon-Christersson, Martin Malmberg, Niklas Loman. Clinicopathological and global transcriptional complexity of estrogen receptor low positive breast cancers in a contemporary Swedish prospective population-based cohort [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO1-16-01.

Background Pathologist-read tumor-infiltrating lymphocytes (TILs) have showcased their predictive and prognostic potential for early and metastatic triple-negative breast cancer (TNBC) but it is still subject to variability. Artificial intelligence (AI) is a promising approach toward eliminating variability and objectively automating TILs assessment. However, demonstrating robust analytical and prognostic validity is the key challenge currently preventing their integration into clinical workflows. Methods We evaluated the impact of ten AI models on TILs scoring, emphasizing their distinctions in TILs analytical and prognostic validity. Several AI-based TILs scoring models (seven developed and three previously validated AI models) were tested in a retrospective analytical cohort and in an independent prospective cohort to compare prognostic validation against invasive disease-free survival endpoint with 4 years median follow-up. The development and analytical validity set consisted of diagnostic tissue slides of 79 women with surgically resected primary invasive TNBC tumors diagnosed between 2012 and 2016 from the Yale School of Medicine. An independent set comprising of 215 TNBC patients from Sweden diagnosed between 2010 and 2015, was used for testing prognostic validity. Findings A significant difference in analytical validity (Spearman's r = 0.63-0.73, p < 0.001) is highlighted across AI methodologies and training strategies. Interestingly, the prognostic performance of digital TILs is demonstrated for eight out of ten AI models, even less extensively trained ones, with similar and overlapping hazard ratios (HR) in the external validation cohort (Cox regression analysis based on IDFS-endpoint, HR = 0.40-0.47; p < 0.004). Interpretation The demonstrated prognostic validity for most of the AI TIL models can be attributed to the intrinsic robustness of host anti-tumor immunity (measured by TILs) as a biomarker. However, the discrepancies between AI models should not be overlooked; rather, we believe that there is a critical need for an accessible, large, multi-centric dataset that will serve as a benchmark ensuring the comparability and reliability of different AI tools in clinical implementation.

This work advances and demonstrates the utility of a reporting framework for collecting and evaluating annotations of medical images used for training and testing artificial intelligence (AI) models in assisting detection and diagnosis. AI has unique reporting requirements, as shown by the AI extensions to the CONSORT (Consolidated Standards of Reporting Trials) and SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) checklists and the proposed AI extensions to the STARD (Standards for Reporting Diagnostic Accuracy) and TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) checklists. AI for detection and/or diagnostic image analysis requires complete, reproducible, and transparent reporting of the annotations and metadata used in training and testing datasets. Prior work by Wahab et al. proposed an annotation workflow and quality checklist for computational pathology annotations. In this manuscript, we operationalize this workflow into an evaluable quality checklist that applies to any reader-interpreted medical images, and we demonstrate its use for an annotation effort in digital pathology. We refer to this quality framework as CLEARR-AI: The Collection and Evaluation of Annotations for Reproducible Reporting of Artificial Intelligence.

A growing body of research supports stromal tumour-infiltrating lymphocyte (TIL) density in breast cancer to be a robust prognostic and predicive biomarker. The gold standard for stromal TIL density quantitation in breast cancer is pathologist visual assessment using haematoxylin and eosin-stained slides. Artificial intelligence/machine-learning algorithms are in development to automate the stromal TIL scoring process, and must be validated against a reference standard such as pathologist visual assessment. Visual TIL assessment may suffer from significant interobserver variability. To improve interobserver agreement, regulatory science experts at the US Food and Drug Administration partnered with academic pathologists internationally to create a freely available online continuing medical education (CME) course to train pathologists in assessing breast cancer stromal TILs using an interactive format with expert commentary. Here we describe and provide a user guide to this CME course, whose content was designed to improve pathologist accuracy in scoring breast cancer TILs. We also suggest subsequent steps to translate knowledge into clinical practice with proficiency testing.

HER2/ERBB2 evaluation is necessary for treatment decision-making in breast cancer (BC), however current methods have limitations and considerable variability exists. DNA copy number (CN) evaluation by droplet digital PCR (ddPCR) has complementary advantages for HER2/ERBB2 diagnostics. In this study, we developed a single-reaction multiplex ddPCR assay for determination of ERBB2 CN in reference to two control regions, CEP17 and a copy-number-stable region of chr. 2p13.1, validated CN estimations to clinical in situ hybridization (ISH) HER2 status, and investigated the association of ERBB2 CN with clinical outcomes. 909 primary BC tissues were evaluated and the area under the curve for concordance to HER2 status was 0.93 and 0.96 for ERBB2 CN using either CEP17 or 2p13.1 as reference, respectively. The accuracy of ddPCR ERBB2 CN was 93.7% and 94.1% in the training and validation groups, respectively. Positive and negative predictive value for the classic HER2 amplification and non-amplification groups was 97.2% and 94.8%, respectively. An identified biological “ultrahigh” ERBB2 ddPCR CN group had significantly worse survival within patients treated with adjuvant trastuzumab for both recurrence-free survival (hazard ratio, HR: 3.3; 95% CI 1.1–9.6; p = 0.031, multivariable Cox regression) and overall survival (HR: 3.6; 95% CI 1.1–12.6; p = 0.041). For validation using RNA-seq data as a surrogate, in a population-based SCAN-B cohort (NCT02306096) of 682 consecutive patients receiving adjuvant trastuzumab, the ultrahigh-ERBB2 mRNA group had significantly worse survival. Multiplex ddPCR is useful for ERBB2 CN estimation and ultrahigh ERBB2 may be a predictive factor for decreased long-term survival after trastuzumab treatment.

Background Immune checkpoint inhibitors are now a part of the treatment arsenal for triple-negative breast cancer (TNBC) but refinement of PD-L1 as a prognostic and predictive biomarker is a clinical priority. We aimed to evaluate the relevance of novel PD-L1 immunohistochemical (IHC) thresholds in TNBC with regards to PD-L1 gene expression, prognostic value, tumor infiltrating lymphocytes (TILs) and TNBC molecular subtypes. Material & Methods We evaluated PD-L1 on a tissue microarray with the SP142 (immune cell (IC) score) and the 22C3 (combined positive score; CPS) IHC assays and evaluated abundance of TILs in a population-based cohort of 237 early-stage TNBC patients. Survival analysis was performed and RNA sequencing data employed for molecular profiling. Results As expected, PD-L1 positivity (IC ≥1% and/or CPS ≥1) was significantly associated with better prognosis compared to zero PD-L1 expression. Importantly however, also patients with intermediate expression (IC >0%, <1%; CPS >0, <1) showed a trend towards improved outcome. Tumors with intermediate PD-L1 IHC expression also had intermediate PD-L1 (CD274) gene expression (mRNA). Patients that were both low in TILs (<30%) and PD-L1 (IC <1%; CPS <1), tended to have the poorest prognosis. PD-L1 positive tumors clustered significantly more often as Immunomodulatory-high and Basal-Like 1-high TNBC molecular subtypes and were enriched in immune response and cell cycle/proliferation signaling pathways. PD-L1-zero tumors on the other hand were enriched in cell growth, differentiation and metastatic potential pathways and clustered more prevalently as Luminal-Androgen-Receptor-high and Mesenchymal-high. PD-L1-intermediate tumors categorized with neither PD-L1-positive nor PD-L1-zero tumors on the hierarchical clustering level, consigning them as a unique subgroup. Conclusion With both SP142 and 22C3, we identified an intermediate IHC PD-L1 group within TNBCs that was supported on the molecular level. Any PD-L1 IHC expression, even though it is <1, tended to have positive prognostic impact. We suggest that the generally accepted threshold of PD-L1 IHC positivity in TNBC should be investigated further. Trial Registration The Swedish Cancerome Analysis Network – Breast (SCAN-B) study was retrospectively registered 2nd Dec 2014 at ClinicalTrials.gov; ID NCT02306096.

Cox regression analysis of TIL variables for OS in all available cohorts

Background Immunohistochemical (IHC) PD-L1 expression is commonly employed as predictive biomarker for checkpoint inhibitors in triple-negative breast cancer (TNBC). However, IHC evaluation methods are non-uniform and further studies are needed to optimize clinical utility. Methods We compared the concordance, prognostic value and gene expression between PD-L1 IHC expression by SP142 immune cell (IC) score and 22C3 combined positive score (CPS; companion IHC diagnostic assays for atezolizumab and pembrolizumab, respectively) in a population-based cohort of 232 early-stage TNBC patients. Results The expression rates of PD-L1 for SP142 IC ≥ 1%, 22C3 CPS ≥ 10, 22C3 CPS ≥ 1 and 22C3 IC ≥ 1% were 50.9%, 27.2%, 53.9% and 41.8%, respectively. The analytical concordance (kappa values) between SP142 IC+ and these three different 22C3 scorings were 73.7% (0.48, weak agreement), 81.5% (0.63) and 86.6% (0.73), respectively. The SP142 assay was better at identifying 22C3 positive tumors than the 22C3 assay was at detecting SP142 positive tumors. PD-L1 ( CD274 ) gene expression (mRNA) showed a strong positive association with all two-categorical IHC scorings of the PD-L1 expression, irrespective of antibody and cut-off (Spearman Rho ranged from 0.59 to 0.62; all p -values < 0.001). PD-L1 IHC positivity and abundance of tumor infiltrating lymphocytes were of positive prognostic value in univariable regression analyses in patients treated with (neo)adjuvant chemotherapy, where it was strongest for 22C3 CPS ≥ 10 and distant relapse-free interval (HR = 0.18, p = 0.019). However, PD-L1 status was not independently prognostic when adjusting for abundance of tumor infiltrating lymphocytes in multivariable analyses. Conclusion Our findings support that the SP142 and 22C3 IHC assays, with their respective clinically applied scoring algorithms, are not analytically equivalent where they identify partially non-overlapping subpopulations of TNBC patients and cannot be substituted with one another regarding PD-L1 detection. Trial registration The Swedish Cancerome Analysis Network - Breast (SCAN-B) study, retrospectively registered 2nd Dec 2014 at ClinicalTrials.gov; ID NCT02306096.