PD-L1 expression scoring using artificial intelligence algorithms in non-small cell lung cancer harbouring ROS1...

PD-L1 expression in ROS1-positive non-small cell lung carcinoma (NSCLC) patients remains unclear regarding its possible clinical-biological role.

A retrospective, descriptive study of 24 ROS1-positive NSCLC patients (2013-2023) from five Spanish hospitals were evaluated. Manual (clone SP263) PD-L1 tumour proportion score (TPS) was also subjected to two artificial intelligence (AI) testing algorithms. Clinical and molecular data categorized according to PD-L1 levels by double independent review, were retrospectively collected and, potential correlations were analysed.

The median age was 60 years (46-85), female (54.1%), adenocarcinoma (87.5%), and former/never smokers (83.3%). Predominant ROS1 fusion partner (FP) was CD74 (41.6%). Most patients (79.1%) received crizotinib at some point during their treatment. PDL1 TPS 1-49% (41.7%), ≥ 50% (33.3%), and < 1% (25%). Statistically significant association between the ROS1 FP category and PD-L1 TPS ≥ 50% (p = 0.008), specifically with EZR (mean manual TPS: 62.1) was found. No statistically significant associations for clinical categorical variables and PD-L1 TPS were observed. Although no significant difference among PD-L1 TPS categories in terms of overall survival was detected (p = 0.36), a shorter overall survival was noted in patients with TPS ≥ 50% and EZR. No significant differences were found between the manual PD-L1 TPS evaluation and the use of PD-L1-based AI algorithms.

PD-L1 expression may have a biological significance in ROS1-positive NSCLC, particularly when associated with the EZR FP and PD-L1 TPS ≥ 50% which determines a lower overall survival. No significant differences were found between manual PD-L1 TPS evaluation and the use of AI.