Why PD-L1 expression varies between studies of lung cancer: results from a Bayesian meta-analysis

PD-L1 expression is an important biomarker for the management of non-small cell lung cancer (NSCLC) but has been highly heterogeneous across studies. We developed a statistical model to reconcile conflicting estimates of PD-L1 prevalence by accounting for between-study variation in test sensitivity, specimen age, and laboratory count. In doing so, we obtained refined estimates for PD-L1 expression prevalence and identified differences by histological subtype, mutational status, and stage. Across 92 studies published between 2015 and 2023, the detectability of PD-L1 declined with increasing specimen age while the consistency of detection rates was greater for studies incorporating data from a higher number of laboratories. Using the 22C3 antibody as a benchmark, we predicted that 58.3% (95% CrI 49.8-66.1%) and 27.0% (95% CrI 21.2-33.1%) of NSCLC will have PD-L1 tumour proportion scores at the ≥ 1% and ≥ 50% threshold. PD-L1 expression was lower in EGFR-mutated NSCLC and higher in NSCLC with ALK, KRAS, MET, ROS1, and RET alterations. PD-L1 expression was more common with later-stage disease. Overall, this work highlights the continuing challenge of consistency in PD-L1 testing. Although the underlying prevalence of PD-L1 expression varies in the lung cancer population based on tumour-related factors, controllable differences in testing parameters also account for variations in PD-L1 prevalence.