Abstract 6401: Predictive models for first-line immunotherapy based on spatial tumor immune microenvironment in patients with advanced non-small cell lung cancer (NSCLC)

Abstract

Abstract Background: Anti-programmed cell death protein 1/programmed cell death ligand 1 (anti-PD1/PDL1) based immunotherapy is the standard first-line treatment for advanced NSCLC patients without targetable driver mutations, but PDL1 immunohistochemistry (IHC) shows limited predictive value as companion diagnostics. Recent studies have elucidated a strong relationship between immunotherapy efficacy and tumor immune microenvironment (TIME) using single-cell multi-omics technology. In this study, we aim to develop predictive models for both objective response rate (ORR) and progression-free survival (PFS) using TIME features with machine learning in advanced NSCLC patients treated with first-line immunotherapy. Methods: A total of 210 patients with advanced NSCLC who received anti-PD-1/PDL-1 therapy at Shanghai Chest Hospital from Jan. 2020 to Jan. 2023 were included in this study. Among them, 110 cases who received pembrolizumab were randomly split into training set and validation set with a 2:1 ratio. The remaining patients were designated as independent validation cohort. Five multiplex immunohistochemistry (mIHC) panels were developed based on literature review and in-house dataset mining. mIHC tests were performed using baseline tumor samples. Immunotherapy efficacy was evaluated by ORR and PFS (RECIST v1.1). ORR and PFS models were developed using machine learning methods in training set with 5-fold cross-validation and further validated in validation set and independent validation cohort. Results: After data cleaning, 68 cases in training set and 34 cases in validation set were further analyzed. In terms of single immune- or spatial- based biomarker, positive rate of PD1+ cells in stroma and spatial relationship of PD1+/PDL1+ cell pairs exhibited the highest correlation with ORR. In addition, positive rate of PDL1+ tumor cells and nearest distance between FOXP3+ cells and tumor cells were highly correlated with PFS. Our ORR model achieved an AUC of 0.82 in the training set and 0.82 in the validation set. When applying optimal cutoff in our PFS model to stratify patients into low- and high- risk groups, low-risk group showed prolonged PFS in both training set (p < 0.0001; HR: 0.25, 95% CI [0.13 - 0.51]) and validation set (p = 0.00046; HR: 0.22, 95% CI [0.09 - 0.55]). Moreover, our ORR model outperformed PDL1 IHC (AUC: 0.81 vs 0.60; DeLong’s test p = 0.005) in these 102 patients. Independent validation of the models is still awaited. Conclusion: Using machine learning methods, we developed and validated ORR and PFS predictive models for first-line immunotherapy in advanced NSCLC patients based on mIHC technology. Our models displayed superior diagnostic performance compared to PDL1 IHC results. Citation Format: Chan Xiang, Jianjian Zheng, Sidong Chen, Zhanxian Shang, Shi Yang, Lianying Guo, Lei Jiao, Yang Wang, Yuchen Han. Predictive models for first-line immunotherapy based on spatial tumor immune microenvironment in patients with advanced non-small cell lung cancer (NSCLC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6401.