ECP 2023 Abstracts

S35 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 target population expands to include the HER2-low group, immuno- histochemistry-based assays’ power in distinguishing HER2-low from HER2-negative cases acquires substantial therapy-guiding relevance. Methods: 116 cases covering HER2 0 (n=39), 1+ (n=25), 2+ (n=26) and 3+ (n=26) groups were scored via HercepTest (poly) and PATHWAY 4B5 assays by 5 blinded pathologists. Amplification status of cases were further verified with Brightfield Dual in situ Hybridization (BDISH). Segregation of HER2-positive, HER2-low and HER2-negative cases was evaluated, as well as inter-assay and inter-observer concordance. Results: Identical IHC scores were achieved in both assays for 63.8% (74/116, Kappa=0.522, p<0.001) of cases. Among remaining, differ- ence was mainly due to higher HER2-status scores in HercepTest assay (37/42). Number of HER2-positive cases was 34 in HercepTest and 31 in PATHWAY 4B5. All 3 discordant cases were overscored (IHC 3+) by HercepTest, as they were 2+ according to PATHWAY 4B5 and further found to lack amplification in BDISH. Number of HER2-low (IHC 1+ and 2+/BDISH-) cases was also higher in HercepTest (57 vs 41). Inter-observer agreement was substantial in both assays, with Fleiss’ kappa 0.63 (p<0.001) in HercepTest and 0.62 (p<0.001) in PATHWAY 4B5 assay. Conclusion: Both assays have high sensitivity in detecting HER2 amplification, however more cases are allocated to HER2-positive and HER2-low group in HercepTest (poly). As clinical consequences in HER2-low group are not yet fully established, different segregation profiles among different IHC-based assays should be acknowledged. Probability of a case being overscored in one assay or underscored in another should be considered while therapeutic translations are made from pathology reports. OFP-09-007 Multi-reader study of a fully automated artificial intelligence solu- tion for HER2 IHC scoring in breast cancer biopsies S. Krishnamurthy*, S. Schnitt, A. Vincent-Salomon, R. Canas- Marques, E. Colon, K. Kantekure, M. Maklakovski, W. Finck, J. Thom- assin, Y. Globerson, L. Bien, M. Grinwald, M. Vecsler, C. Linhart, J. Sandbank *MD Anderson Cancer Center, USA Background & objectives: Recent proven efficacy of HER2-targeted therapy on HER2-low expressing tumours raise the need for accurate and reproducible HER2 scoring. We aimed to validate pathologists’ use of an artificial intelligence (AI) solution for interpreting HER2 scores in breast samples. Methods: Two-arm multi-reader study on 120 cases, containing H&E and complementary HER2 IHC, from four sites, compared HER2 scor- ing performance of 4 pathologists with an AI HER2 solution versus manual interpretation. Both arms were compared to rigorous ground truth (GT) established according to ASCO/CAP 2018 guidelines by five breast subspecialists, i.e. agreement of at least 4 out of 5 pathologists. Results: The AI solution demonstrated high accuracy for HER2 scor- ing of 92% on slides with a robust GT majority (N=92). Average inter- observer agreement among GT pathologists for across all HER2 scores was 72.8%; for each HER2 score, 0/ 1+/ 2+/ 3+, their agreement was 80.1%, 65.9%, 69.2% and 96.4%, respectively. Pathologists supported by AI showed significantly better performance with higher consistency (88.8% vs 81.9%), and accuracy (87.4% vs. 69.8%) for HER2 0/1+, while across all HER2 scores a trend of better performance was observed (83.7% vs. 75% for consistency, and 88% vs. 85% for accuracy). Conclusion: This study demonstrated the developed automated AI solution scored HER2 accurately according to 2018 ASCO/CAP guide- lines. Pathologists supported by AI showed improvements in HER2 IHC scoring consistency and accuracy, especially for HER2 0 and 1+. AI solutions could be used as decision-support tools for pathologists, enhancing the reproducibility and consistency of HER2 scoring, thus enabling optimal treatment pathways and better patient outcomes. OFP-09-008 Agreement among 4 observers and 4 reading modalities for HER2 Low designation of breast carcinoma S. Nofech-Mozes*, E. Olkhov-Mitsel, A. Plotkin, E. Slodkowska, F. Lu *Precision Diagnostics and Therapeutics Program, Sunnybrook Health Sciences Centre, Toronto, Department of Laboratory Medicine and Pathobiology, University of Toronto, Canada Background & objectives: Accurate designation of HER2 Low (HER2L) (IHC 1+ or 2+/ISH-) in breast carcinomas (BC) is critical due to recently approved antibody drug conjugate in the metastatic setting. We examined the agreement among 4 breast pathologists using different slide viewing modalities. Methods: Array with 51 cores enriched for HER2L, stained with 4B5, was independently scored using the 2018 ASCO/CAP guide- lines by 4 breast pathologists with 4 slide viewing modalities: SC1-Scanned slide. SC2- Scanned slide after consensus discussion (and washout). GSM- Glass slide using a microscope. GSTV- Glass slide projected to HDTV screen. Concordance between modalities, observers and assessments was evaluated. Results: Among 46 cores with BC adequate for scoring, there was complete agreement among raters and modalities in 39 (84.8%) cores. Seven (15.2%) cores had at least one discordant score, all HER2-0 vs HER2-1+ (κ = 0.792-0.894). The interrater agreement for each modal- ity was SC1 - κ = 0.860, p<0.001; SC2 - κ = 0.905, p<0.001; GSM - κ = 0.890, p<0.001 and GSTV - κ = 0.903, p<0.001. The agreement on HER2-0 improved from 0.844 on SC1 to 0.894 on SC2 (p<0.001). 6/7 discordant cores were scored as HER2-0 with GSM and HER2-1+ with scanned slides. Conclusion: Post analytical variables including interobserver, intrao- bserver and modality used for HER2 immunohistochemistry assess- ment are critical for the distinction between HER2-1+ and HER2-0 BC, particularly for cases bordering the 10% faint membranous stain- ing cutoff. We observed more HER2 1+ scores with scanned slides vs other modalities, raising a need for modality-specific validation as many pathology departments are transitioning to digital platforms for primary diagnosis. The effect of different scanners, focal points, white balance and computer monitors should be studied further. OFP-09-009 Upgrade rate and predictive factors of fibroepithelial lesions of the breast to phyllodes tumour R. Alaghehbandan*, T. Salisbury, M. Zokaei Nikoo, G. Sokhanran, S. Koonmee *Cleveland Clinic, USA Background & objectives: Fibroepithelial lesions (FELs) of the breast are heterogenous neoplasms that can present diagnostic challenges, particularly on core biopsies. The aim of this study was to determine the upgrade rate of FELs to phyllodes tumour on surgical follow-up and associated factors. Methods: This is a retrospective population-based study of FELs of the breast (fibroadenoma [FA], phyllodes tumour [PT]) on core biopsies over a three-year period (2017-2019) in Vancouver, BC, Canada. The data were obtained from the laboratory information system and patient’s chart. Core biopsy diagnoses were correlated with surgical pathology resection findings. Clinical and radiologic factors were compared to identify PT-related features. Results: Of 425 core biopsies of FELs (369 FA, 9 PT, and 47 indeter- minate), 91 underwent excisional biopsy, with an upgrade rate of 24.2% to PT. Of 44 core biopsies diagnosed/favoured as FA, 3 were PT on excisional biopsy (NPV=93.2%). Of 9 PT diagnosed (or favoured) on core biopsies, 7 were confirmed on excisional biopsy (PPV=77.8%). Mean age and mass size were significantly higher in the PT group (51