A blind retrospective analysis of a novel predictive marker to ICB response in NSCLC, calculated directly from histopathological slides.

• DOI: 10.1200/jco.2024.42.16_suppl.2665
• Publication Date: 2024-06-04T19:28:39Z
• Authors: Omer Tirosh; Leon Gugel; Gal Dinstag; Yaron Kinar; Anna Elia; Michael Tabi; Tzivia Gottlieb; Eli Pikarsky; Aron Popovtzer; Ranit Aharonov; Johnathan Arnon
• Keywords: Predictive marker

2665 Background: Immune checkpoint blockers (ICB), and primarily PD-1/PD-L1 inhibitors, are in the forefront of contemporary clinical oncology have become an integral part treatment many malignancies, including non-small cell lung cancer (NSCLC). Nevertheless, tumor response to ICB varies widely. Predictive markers commonly used distinguish patients likely respond ICB, such as PD-L1 expression mutational burden (TMB) limited predictive value, which calls for development practical more accurate tests. We present results a blind retrospective analysis novel marker NSCLC, relying solely on histopathological slides. Methods: obtained high resolution Hematoxylin Eosin (H&E) slides from tumor-tissue samples 50 cases metastatic NSCLC treated with first-line PD-1 inhibitors. retrospectively applied our ENLIGHT-DeepPT (ENLIGHT-DP short) pipeline generate, blinded manner, individual score inhibition each slide. ENLIGHT-DP is composed two main steps: (i) predict mRNA directly H&E slide using DeepPT, digital-pathology based algorithm; (ii) use these values input ENLIGHT, transcriptome-based precision platform, generates that predicts targeted therapies (based 10-gene signature this case). then unblinded outcome (RECIST1.1), evaluated ENLIGHT-DP’s performance vs. standard markers. Results: cohort, had overall rate 68% (34 50), ROC AUC = 0.69 (p 0.01, one-sided permutation test). Using predefined threshold binary classification derived independent data, all 15 were predicted by indeed responded (100% PPV, 44% sensitivity). In comparison, predicting according > 1% achieves PPV 62% sensitivity, while 50% 65% 38% i.e, both thresholds exhibit no power (PPV <= baseline rate). Patients TMB (>10) 82% 26% showing lower benefit than ENLIGHT-DP. was particularly good at stratifying < (18 patients, 0.8, p 0.03). Conclusions: demonstrates accessible slides, outperforming also able identify responders within 1%, whom usually considered ineffective. Importantly, approach does not require training prior outcomes, can therefore be generalized drugs data unavailable or scarce.