Limited predictive value of PD-L1 TPS for pathological response in NSCLC is not attributable to interobserver...

PD-L1 tumor proportion score (TPS) is used to guide immunotherapy in non-small cell lung cancer (NSCLC), yet its ability to predict pathological response in the neoadjuvant setting remains limited.

Thirty pathologists from 11 countries independently assessed PD-L1 TPS in pre-treatment biopsies and residual viable tumor (RVT) in matched resection specimens after neoadjuvant chemoimmunotherapy in 30 digitized cases from the ReGraDE (regression grading in Germany) study. Interobserver agreement was evaluated using intraclass correlation coefficients (ICC) and Fleiss' kappa. Associations between TPS and RVT were analyzed using Pearson correlation, and correlations derived from single-rater and averaged TPS were compared using the Williams test.

Interobserver agreement was moderate for both TPS and RVT (ICC = 0.74, 95% CI 0.60-0.87, and ICC = 0.74, 95% CI 0.62-0.85, respectively) and became near-perfect when mean scores per case were calculated across observers (ICC = 0.99, for both). However, TPS remained poorly correlated with RVT for both single-rater assessment (r = -0.17, p = 0.38) and averaged TPS (r = -0.16, p = 0.39) with no significant difference between these approaches (p = 0.96). Differences in interpretation thresholds were observed particularly in borderline cases around the 1% TPS cut-off and in distinguishing 0% from minimal RVT, but did not account for the lack of association between TPS and RVT. Systematic differences were observed depending on individual professional experience, particularly in borderline cases.

Interobserver variability does not explain the limited predictive value of PD-L1 TPS in the neoadjuvant setting, suggesting an intrinsic limitation of the biomarker.