GI Oncology Daily Digest

CheckMate 648 randomised 970 patients with previously untreated advanced oesophageal squamous cell carcinoma to first-line nivolumab plus chemotherapy, nivolumab plus ipilimumab, or chemotherapy alone. At a minimum of five years' follow-up, both experimental arms retained their overall survival advantage in the PD-L1≥1% population — OS HR 0.62 (95% CI 0.48–0.79) for nivolumab plus chemotherapy and OS HR 0.62 (95% CI 0.48–0.80) for nivolumab plus ipilimumab — with an OS HR of 0.77 for both arms across all randomised patients. Progression-free survival tells a more divided story: nivolumab plus chemotherapy improved PFS in the PD-L1≥1% group (HR 0.67), while the chemotherapy-free nivolumab plus ipilimumab arm did not (PFS HR 1.03). No new safety signals emerged with the additional follow-up.

Mountain-02 (NCT06374901) randomised 136 patients with operable cT3–4aN+M0 gastric or gastro-oesophageal junction cancer to perioperative tislelizumab plus chemotherapy or chemotherapy alone, prospectively stratified by tumour-specific MHC class II (tsMHC-II) expression. In tsMHC-II-positive tumours, adding tislelizumab raised the major pathological response rate to 61.8% from 26.5% (p=0.003), meeting the primary endpoint; pathological complete response was 32.4% in tsMHC-II-positive versus 5.9% in tsMHC-II-negative tumours (p=0.006). No significant benefit was seen in the tsMHC-II-negative subgroup. The design is the point: rather than running an all-comers perioperative immunotherapy trial and mining for a biomarker afterwards, the investigators built the stratification into randomisation.

DENEB (NCT06342440) is a multicentre diagnostic accuracy study of a blood-based assay combining 19 cell-free and 20 exosomal microRNAs by RT-qPCR, with a locked machine-learning model. In the testing cohort the assay reached an AUC of 95% (95% CI 92–98) for colorectal cancer plus advanced adenoma versus controls, with 91% sensitivity for colorectal cancer at any stage and 92% for stage I–III disease. Advanced adenoma sensitivity was 81%; specificity was 85% for advanced neoplasia and 83% against a negative colonoscopy. The advanced adenoma figure is the headline, because blood-based colorectal screening has consistently performed well on established cancer and poorly on precancer — which is backwards for a test whose purpose is prevention. The outcome-enriched case-control design means spectrum bias is the obvious caveat, and a true screening-population cohort is the necessary next step.

This national observational study linked an English Lynch syndrome registry to NHS digital records to follow 4,732 mismatch repair gene carriers between 2010 and 2022, asking what colonoscopic surveillance actually delivers. Surveillance at a mean interval of three years or less (n=3,028) was associated with lower colorectal cancer-specific and all-cause mortality after multivariable adjustment — but with no accompanying reduction in total colorectal cancer incidence and no strong evidence of stage shift. More provocatively, surveillance at a mean interval of two years or less (n=1,569) was associated with a higher total colorectal cancer incidence, driven by more early-stage cancers with no corresponding fall in late-stage disease. The authors are appropriately careful: short follow-up, spectrum bias, overdiagnosis and selection bias are all live explanations in a non-randomised design. A mortality benefit without a stage shift is mechanistically awkward, and an incidence rise at the tightest interval is exactly the shape overdiagnosis makes.

NEO-Nal-IRI (NCT03483038) enrolled 45 patients with resectable or borderline-resectable pancreatic adenocarcinoma and gave neoadjuvant NALIRIFOX — liposomal irinotecan with 5-fluorouracil, leucovorin and oxaliplatin — before surgery. The primary endpoint, 30-day major postoperative complication rate, was 10% (95% CI 2.2–27%), meeting the pre-specified threshold (p=.012). R0 resection was achieved in 90% of patients who went to resection, and the radiographic objective response rate was 45% (95% CI 29–62%). This is a feasibility question answered cleanly rather than an efficacy result: the trial licenses the next study rather than the next prescription. But knowing that a NALIRIFOX backbone does not compromise the operation is genuinely useful before anyone designs a randomised neoadjuvant comparison.