• Gastric Cancer

We commend Kang et al1 for their phase 2 study evaluating neoadjuvant durvalumab plus docetaxel, oxaliplatin, and S-1 (DOS) in Asian patients with resectable locally advanced gastric cancer. The reported pathologic complete response (pCR) rate of 30% and manageable safety profile are promising. However, several critical limitations in study design, statistical methodology, and clinical interpretation warrant further discussion to contextualize these findings.

The study emphasizes pCR as a primary endpoint, correlating it with favorable progression-free survival trends. However, the phase 3 KEYNOTE-585 trial demonstrated that pembrolizumab plus chemotherapy significantly improved pCR rates but failed to enhance event-free survival,2 highlighting the limitations of pCR as a surrogate in gastric cancer. Kang et al’s survival analysis (median follow-up: 21.8 months) is premature for assessing long-term outcomes, particularly in a cohort with 64% stage III tumors. Notably, the PRODIGY trial, which this study references, reported a 10% pCR rate with neoadjuvant DOS alone but achieved a 5-year overall survival (OS) of 58%—similar to the current study’s 3-year OS of 88%.3 This suggests that pCR may not reliably predict survival in gastric cancer, necessitating caution in framing pCR as a definitive efficacy marker.

The exploratory arm evaluated durvalumab plus tremelimumab in deficient mismatch repair (dMMR) tumors but reported a surprisingly low pCR rate (22.2%), contrasting with the 58.6–60% pCR rates observed in the GERCOR NEONIPIGA and INFINITY trials using dual checkpoint inhibitors without chemotherapy.4 The authors attribute this discrepancy to advanced cT4 stage in their cohort but overlook critical factors such as tumor mutation burden (TMB) and programmed death-ligand 1 (PD-L1) heterogeneity. For example, the INFINITY trial stratified patients by TMB and PD-L1 expression, showing pCR rates of 89% in TMB-high tumors.4 Kang et al did not analyze TMB in the main arm or correlate it with pCR, despite its established role in predicting immunotherapy response.⁵ A more granular analysis of dMMR tumor biology (eg, Lynch syndrome vs sporadic dMMR) is also missing, which could explain differential responses.

The exploratory arm’s sample size (n=9) and statistical assumptions are problematic. The authors applied Fleming’s single-stage design assuming a pCR rate of 60% (vs 12.5% in historical controls), yet enrolled only nine patients due to slow accrual. With two pCRs observed, the study claims the regimen is “lower than expected” but fails to acknowledge that the 95% CI (0–49.7%) overlaps with historical data. This undermines the conclusion that dual ICIs underperform in dMMR tumors. A Bayesian adaptive design or pooled analysis with similar trials (eg, NEONIPIGA) would have provided more robust insights.

Recommendations for future studies

Dose optimization: Conduct a phase Ib PK/PD study to define the optimal durvalumab dose and schedule with DOS.

Biomarker-driven design: Incorporate ctDNA monitoring, TMB, and spatial transcriptomics to refine patient stratification.

Dual ICI reassessment: Re-evaluate dual ICIs in dMMR tumors using adaptive designs enriched for TMB-high or PD-L1-high subgroups.

Surrogate endpoint validation: Partner with international consortia to establish pCR as a validated surrogate through pooled meta-analyses.

While this study advances perioperative immunotherapy in Asian gastric cancer, addressing these limitations is crucial before phase 3 validation.