Use of artificial intelligence (AI)–powered spatial analysis to predict pathologic complete response (pCR) in HR+ HER2- breast cancer (BC) patients treated with neoadjuvant chemotherapy (NAC).

593 Background: In the KEYNOTE(KN)-756 study, adding pembrolizumab to NAC increased pCR in high-risk HR+ HER2- BC. There is an unmet need to discover who will benefit from adding immune checkpoint inhibitors (ICIs). This study aims to investigate the role of pCR in HR+ HER2- BC and to identify whether AI-powered TIL analysis could predict pCR in patients treated with NAC without ICIs. Methods: This is a single-center study conducted in Seoul National University Hospital, Korea. H&E whole-slide images (WSIs) of archival breast tumor tissues at diagnosis were analyzed by Lunit SCOPE IO, an AI-powered spatial TIL analyzer. Tumors with a high proportion of area with high intratumoral TIL were classified as immune-inflamed, those with a high proportion of area with low intratumoral but high stromal TIL as immune-excluded, and the remaining as immune-desert. Recurrence risk prediction AI model was trained and validated on independent 1,552 H&E WSIs to predict OncotypeDx score. Patients with histologic grade 3 and tumor size ≥2 cm with node-positive status, or tumor size ≥5 cm were classified as those eligible for KN-756 study. Results: A total of 425 BC patients who were treated with NAC without ICIs between January 2015 and October 2018 were included. The median age was 47 (range 24-80), 67.1% had stage III disease, 93.6% had node-positive disease, and 23.5% were eligible for KN-756 study. pCR was achieved in 57 (13.4%) patients and was higher in patients with whom were eligible for KN-756 study (20% vs 11.4%, p= 0.041). Patients who achieved pCR had better 5-year event free survival (EFS, 92.9% vs 77.7%, p= 0.010). AI-powered spatial analysis was performed in 340 patients. There were 125 (36.8%) with immune-desert tumor, 138 (40.6%) with immune-excluded tumor, and 77 (22.6%) with immune-inflamed tumor. Patients with inflamed tumor had higher pCR rate compared to those with immune-excluded or immune-desert in the whole population (Table) and in patients not eligible for KN-756 study (25.6% vs 11.9% vs 5.7%, p = 0.013). Patients with inflamed or excluded tumor had better EFS compared to desert group (80.9% vs 71.2%, p = 0.0275). In addition, recurrence prediction model showed that those who were predicted to have OncotypeDx score ≥ 26 had higher pCR rate (18.2% vs 4.3%, p= 0.002). Conclusions: Achieving pCR was associated with favorable EFS in HR+ HER2- patients treated with NAC. Patients with immunogenic tumor microenvironment had higher pCR rate and better EFS. Investigating the role of immune checkpoint inhibitors according to tumor microenvironment may be promising. Desert(N=125) Excluded(N=138) Inflamed(N=77) p-value pCR 8 (6.4%) 18 (13.0%) 23 (29.9%) <0.001 Stage III 85 (68.0%) 88 (63.8%) 51 (66.2%) 0.32 LN + 114 (91.2%) 132 (95.7%) 72 (93.5%) 0.34 HG 3 21 (16.8%) 31 (22.5%) 38 (49.4%) <0.001 KN-756 eligible 20 (16.0%) 29 (21.0%) 34 (44.2%) <0.001