HERIZON-GEA-01 Study Design, Geographic Distribution, and PD-L1 Testing Considerations

In this episode, Dr. Wainberg, Dr. Elimova, and Dr. Shah walk through the HERIZON-GEA-01 design and work through a cross-trial discrepancy in PD-L1–negative rates. Dr. Wainberg lays out the structure: approximately 900 patients with previously untreated HER2-positive metastatic gastroesophageal adenocarcinoma (GEA), randomized to one of three arms: control (trastuzumab plus chemotherapy), doublet (zanidatamab plus chemotherapy), or triplet (zanidatamab plus tislelizumab plus chemotherapy). Two contextual points matter for interpretation: pembrolizumab was not yet standard when the trial was designed, and tislelizumab is a different programmed death-1 (PD-1) inhibitor from pembrolizumab.

Dr. Elimova fills in the design details. The trial required central HER2 confirmation, which excluded about 25% of locally tested patients, consistent with earlier experience and an important design strength. Stratification was by geographic region, HER2 status, and Eastern Cooperative Oncology Group (ECOG) performance status; PD-L1 was not a stratification factor because its predictive role was not yet established. Geographically, ex-Asian patients made up about 50% overall, with roughly one-third from Europe and Canada. United States sites did not enroll — KEYNOTE-811 was competing for enrollment at the time.

The discrepancy worth dwelling on: PD-L1 tumor area positivity (TAP) was tested retrospectively in HERIZON-GEA-01, and about 35% of patients were TAP <1 compared with roughly 15% CPS <1 in KEYNOTE-811. Dr. Shah's leading explanation is assay stability over time in archival tissue, not a true population-level difference in PD-L1 biology.

In the next episode, “HERIZON-GEA-01 Efficacy Results and Interpretation of Duration of Response,” Dr. Shameem summarizes the headline efficacy data and Dr. Shah shares a hypothesis about the duration-of-response signal.