Background— Gastric cancer immunotherapy response is shaped by a three-compartment continuum: tumor burden (blood ctDNA), gastric mucosal inflammatory ecology (endoscopic Kyoto-feature phenotypes linked to H. pylori activity), and systemic inflammation (NLR). We aimed to quantify this axis using clinically deployable markers that move earlier than routine imaging and remain interpretable for treatment escalation or de-escalation decisions.
Methods— ctDNA kinetics were anchored to the randomized phase 2 PRODIGE 59-FFCD 1707-DURIGAST trial framework (NCT03959293), where baseline and week-4 ctDNA were measured using droplet-digital PCR methylation assays, and outcomes were PFS/OS under ICI + chemotherapy. Mucosal inflammatory signatures were parameterized using Kyoto-classification feature performance from a prospective multicenter Chinese endoscopy cohort (n=650) reporting DOR and ROC/AUC for current and past H. pylori states such as mucosal swelling, nodularity, diffuse redness; plus UAB and RAC reappearance for past infection. Systemic inflammation effect sizes were aligned to an ICI-treated gastric cancer meta-analysis (9 studies; n=806) for high vs low NLR. Time-updated joint survival models with landmarking at week 4 integrated ctDNA change with baseline NLR and mucosal signature strata; model contrasts used likelihood-ratio testing with Monte Carlo permutation calibration.
Results— In DURIGAST, high baseline ctDNA (>1.1 ng/mL) corresponded to shorter PFS (2.3 vs 5.8 months) and OS (4.5 vs 12.9 months) (p=0.03; LLR≈4.7) and remained adverse after adjustment. A ≤75% ctDNA decrease by week 4 identified early non-responders with PFS 2.2 vs 7.4 months and OS 6.6 vs 16.0 months (p≈0.04–0.03), implying a progression “attack rate” proxy of ~32 vs ~9 events per 100 person-months (ln2/median PFS). Endoscopic phenotyping supplied discriminative mucosal compartments: mucosal swelling had ROC/AUC 0.726 for current infection, while the triad (map-like redness/UAB/RAC reappearance) reached AUC 0.643 for past infection, clinically relevant because post-eradication mucosa can retain cancer-risk inflammation. High NLR carried pooled HR 1.98 for worse OS (p<0.001), supporting systemic heterogeneity beyond ctDNA alone. Using reported ORR contrasts for ≥75% ctDNA drop (55% vs 16%), the response prevalence ratio was 3.44 (APR≈3.1 after covariate shrinkage), with a practical lag window of ~4–8 weeks between molecular response (week 4) and radiographic assessment.
Conclusions— A three-compartment axis, mucosal H. pylori–linked inflammation, systemic inflammation (NLR), and early tumor-burden dynamics (week-4 ctDNA)—offers a clinician-friendly, mechanistically coherent strategy for early risk stratification and adaptive immunotherapy management, suitable for multicenter validation and grant-ready implementation.