Abstract 4329: Combined extracellular vesicle PD-L1 and radiomics as predictors of response in patients with advanced lung cancer undergoing immunotherapy

Background: Immune-checkpoint inhibitors (ICIs) have improved the clinical benefit of only a subset of patients with advanced non-small cell lung cancer (NSCLC), since tissue PD-L1 has showed low predictive accuracy. We have previously showed that extracellular vesicle (EV) PD-L1 identified durable responders to ICIs and that the radiomics analysis of baseline CT scans improved the EV PD-L1 predictive model. Here, we aim to validate this combination model to identify non-responders to ICIs. Methods: We enrolled 3 cohorts of patients, cohort (1) included 27 patients receiving ICIs, cohort (2) 17 on ICIs + chemotherapy (ChT), and cohort (3) with 13 undergoing ChT. Baseline and 3rd cycle plasma samples were collected, EV were isolated by ultracentrifugation, and EV PD-L1 dynamics was analyzed by immunoblot. 800 radiomics features were evaluated in baseline CT scans and a set of 6 features identified in cohort 1, were analyzed in cohort 2 and 3. Early and durable responses were assessed in CT scans after the 3rd and the 6-8 treatment cycle, respectively, using RECIST v1.1. Results: Increased expression of EV PD-L1 during treatment showed good predictive value at identify non-responders to ICIs and was not associated with ChT response. The addition of radiomics features improved the predictive model for early and durable response to ICIs, while similar AUC was observed in particular for durable response in cohort 2. Moreover, these biomarkers outperformed tissue PD-L1 and were unspecific for response to Docetaxel (Table 1). No predictive models could be evaluated for early response to ChT. Conclusions: We present for the first time, validated evidence of the combination of EV PD-L1 and radiomics as a highly accurate minimally invasive biomarker for predicting response to ICIs. This combinatorial approach outperformed the current standard-of-care tissue PD-L1 and is a promising tool for the stratification of patients to receive ICIs. Table 1. Predictive models for early and durable response: A: Predictive models for early RECIST response (3 treatment cycles) Cohort: 1: ICIs (n=27) 2: ICIs + Docetaxel (n=17) 3: Docetaxel (n=13) Biomarker: AUC Sensitivity Specificity AUC Sensitivity Specificity AUC Sensitivity Specificity ΔEV PD-L1 + Radiomics 73.3% 83.3% 73.3% 75.0% 100.0% 75.0% NA ΔEV PD-L1 61.7% 50.0% 73.3% 71.9% 100.0% 56.3% NA Radiomics 67.8% 83.3% 53.3% 90.6% 100.0% 87.5% NA Tissue PD-L1 ≥1% 57.5% 75.0% 40.0% 62.5% 100.0% 25.0% NA B: Predictive models for du