OBJECTIVES: Significant differences in survival rates between Asian and non-Asian patients have been noted in prior studies. Imbalances in patient characteristics between trials could violate transitivity assumption in a network meta-analysis (NMA). Conventional meta-regression methods are data-intensive and population adjustment approaches like ML-NMR or MAIC require access to patient-level data. We conducted an NMA of overall survival (OS) using external data as priors to account for population-level differences across the included trials.

METHODS: Data were sourced from Yang 2024, an NMA of immune checkpoint inhibitors (ICIs) in advanced esophageal squamous cell carcinoma, including five clinical trials (ATTRACTION-3, ESCORT, KEYNOTE-181, ORIENT-2, RATIONALE-302) with varying proportion of Asian patients (39%-100%). We used a Bayesian NMA with meta-regression on the proportion of Asian patients, modeling the log-hazard ratio (HR) relative to the baseline treatment. To adjust for differences in Asian proportions, the regression coefficient was assigned an informative prior for Asian versus non-Asian patients, derived from a meta-analysis. To ensure the adjustments are valid, we explored the reported values of percent Asian patients across the studies to recover the original HRs.

RESULTS: All ICIs showed significantly longer OS versus chemotherapy. After accounting for Asian population differences , pembrolizumab showed highest reduction in the risk of mortality (HR=0.64, 36%), followed by tislelizumab (34%), camrelizumab (29%), sintilumab (29%), and nivolumab (24%). The adjustment to reported percent Asian values recovered the originally reported HRs in respective trials. Additionally, the adjusted HR for KEYNOTE-181 to 100% Asian population (0.64, 95%CrI 0.51-0.80) was comparable to the Asian subpopulation analysis from KEYNOTE-181 (0.63, 95%CI 0.50-0.80).

CONCLUSIONS: