PD-L1 alone versus a five-variable genomic-immunotherapy score for predicting response to immune-checkpoint...

Immune checkpoint inhibitors (ICIs) have transformed the landscape of cancer care, yet reliable predictive biomarkers remain elusive. PD-L1 expression, the most widely used IHC-based biomarker, is limited by inter-observer variability and tumor-type-specific cut-offs. We hypothesized that a composite score integrating PD-L1 with tumor and blood TMB, MSI and MMR would improve prediction over PD-L1 alone in a real-world multi-tumor cohort.

We retrospectively enrolled consecutive solid-tumor patients who had undergone paired tissue-plasma NGS (425-gene), PD-L1 IHC, and at least one cycle of anti-PD-1/PD-L1 therapy between April 2021 and June 2023. Multivariable logistic regression, ROC-AUC, calibration, and decision-curve analysis compared PD-L1 alone vs a 5-variable composite (PD-L1, tTMB, bTMB, MSI, MMR).

Of 286 patients with complete molecular data, 102 formed the efficacy-evaluable cohort (ORR 47.1%). On multivariable analysis, only PD-L1 remained independently associated with response (OR 4.78, 95% CI 1.90-12.01, P = 0.001). tTMB, bTMB, MSI‑H, and MMR were not significant (all P > 0.1). PD-L1-positive vs PD-L1-negative: 55.9% vs 29.4% ORR (χ²=6.38, P = 0.012). The composite showed comparable performance to PD-L1 alone (AUC 0.698 vs 0.657; DeLong test, P = 0.42), with no net benefit on decision-curve analysis. Findings were robust across subgroups.

In this multi-tumor cohort, PD-L1 IHC was independently associated with ICI response, but its predictive effect varied significantly across tumor types (P_interaction=0.049), strongest in lung cancer and weaker in gastrointestinal malignancies. These findings underscore the need for tumor‑type-specific PD-L1 interpretation and do not support a generalized pan-cancer conclusion.