Top 5 Papers
#1
Source: JAMA Oncology | Authors: Samaille T, Henriques J, Tougeron D, et al.; André T (GERCOR, French multicentre) | Published: October 15, 2026
Score: 10/20 — JAMA Oncology (base 7) + Phase II (+2) + biomarker-guided dMMR/MSI-H selection (+1) = 10. No survival-benefit bonus: the trial is single-arm, and the authors themselves state there is no clear evidence of improved survival over published surgery-alone series in this population. No colleague-engagement bonus: one GI oncology KOL post was reported by the X sweep but could not be confirmed at source. Tier-1 safety-net paper.
NEONIPIGA treated 32 patients with resectable (cT2–T4, any N, M0) dMMR/MSI-H gastric or gastro-oesophageal junction adenocarcinoma with six cycles of neoadjuvant nivolumab plus two doses of ipilimumab, then surgery and nine cycles of adjuvant nivolumab. The primary endpoint, pCR, was reported earlier: 17 of 29 resected patients (59%). This paper adds the long-term data at a median follow-up of 48.3 months. Four-year OS was 84.1% (95% CI 65.8–93.0%) in the intention-to-treat population, and four-year EFS was 83.5% (95% CI 64.8–92.8%) in the 31 surgery-eligible patients. Only one patient relapsed, with brain metastases. All three patients who did not have surgery (two declined) had a clinical complete response on immunotherapy alone. Grade 3/4 treatment-related adverse events occurred in 19% during neoadjuvant and 17% during adjuvant therapy, with no new late toxicity. The authors are careful: with no comparator arm, they see no clear survival gain over historical surgery-alone outcomes, and they frame the result as groundwork for randomised organ-preservation trials.
Post angle: Four years out, NEONIPIGA has one relapse in 32 patients with resectable MSI-H gastric/GEJ cancer, and every patient who skipped surgery is in clinical complete response. The authors won't claim a survival win without a control arm. The question that matters now is who still needs a gastrectomy. #GIOnc #GastricCancer #MSI #Immunotherapy
#2
Source: Cancer | Authors: Kim RD, Naleid N, Bubie A, Zhang N, Clemens K (Moffitt Cancer Center; Guardant Health) | Published: October 15, 2026
Score: 7/20 — Cancer (base 5; Other = 5, matching the 2026-10-10 edition) + retrospective/real-world (+1) + biomarker-guided treatment selection, ctDNA-detected IDH1 (+1) = 7. No survival-benefit bonus: the OS difference is an unadjusted real-world association that runs against the intervention, not a demonstrated benefit. Ranked above the two other 7s because it bears directly on first-line treatment choice.
TOPAZ-1 and KEYNOTE-966 made a checkpoint inhibitor plus gemcitabine–cisplatin (GemCis) the first-line standard for advanced biliary tract cancer, but some data suggest IDH1-mutant tumours derive less from immunotherapy. This study linked Guardant360 ctDNA results to claims data in the InfinityAI Data Library and identified 416 patients with IDH1-mutant advanced cholangiocarcinoma between April 2019 and February 2025; 155 received GemCis alone and 113 GemCis plus an ICI. Real-world time to next treatment (HR 0.75, p = .13) and time to treatment discontinuation (HR 1.12, p = .40) did not differ. Real-world OS, however, was 16.0 months (95% CI 13.6–19.9) with GemCis plus ICI versus 30.0 months (95% CI 24.1–41.0) with GemCis alone (reported HR 0.40, p < .001). Within the 1,150-patient GemCis-plus-ICI cohort, IDH1-mutant patients had numerically shorter OS than everyone else (HR 0.75, p = .06). These are Kaplan–Meier and log-rank comparisons in claims data, with no adjustment for who was offered immunotherapy, and the treatment eras differ. The authors call for prospective studies and recommend ctDNA profiling before first-line therapy starts.
Post angle: A 14-month real-world OS gap, running against immunotherapy, in IDH1-mutant cholangiocarcinoma. It's unadjusted claims data, so it can't change the standard. But it's a strong argument for having the ctDNA result back before cycle 1. #GIOnc #Cholangiocarcinoma #BTC #IDH1 #LiquidBiopsy
#3
Source: Med (Cell Press) | Authors: Fuchs HS, Xie P, James SA, et al.; Middleton MR, Lu X (Ludwig Institute for Cancer Research, University of Oxford) | Published: October 14, 2026
Score: 7/20 — Med (base 5; Other = 5, matching the 2026-06-13, 06-30 and 10-15 editions) + correlative analysis of a single-arm Phase II (+1, one step below the +2 an ancillary biomarker analysis of a randomised Phase III earns) + biomarker-guided immunotherapy stratification (+1) = 7. No survival-benefit bonus: the abstract reports improved outcomes in the operable cohort but no comparator and no effect size.
LUD2015-005 is a UK Phase II trial that treated 73 patients with oesophageal cancer, 38 advanced or inoperable and 35 operable. Each patient received four weeks of a checkpoint inhibitor alone (durvalumab and tremelimumab, supplied by AstraZeneca) before it was combined with standard chemotherapy or chemoradiotherapy. Paired, longitudinal biopsies of tumour and adjacent normal tissue underwent whole-genome, methylome, single-cell and bulk RNA sequencing. The group previously showed that tumour monocyte content predicts OS with immunochemotherapy in advanced disease. Here they report clinical outcomes in the 35 operable patients, describe improved overall outcomes with neoadjuvant ICI-based therapy, and link tumour monocytes to immune-microenvironment remodelling and tumour-reactive CD8 T cells. They define five DNA-methylation subtypes with distinct immune microenvironments, plus a five-gene monocyte signature, CFISH (CXCL8, FCN1, IFI30, SLC11A1, HLA-DRA), which they report as highly predictive of OS in oesophageal cancer and beyond. The abstract gives no effect sizes. Treat it as hypothesis-generating until the signature is tested in a randomised cohort.
Post angle: Most immunotherapy biomarkers in oesophageal cancer start with PD-L1 and T cells. This one starts with monocytes. A five-gene myeloid signature plus methylation subtypes, from a trial with an ICI-only lead-in window, is the kind of design that can separate a drug effect from a chemo effect. #GIOnc #EsophagealCancer #Immunotherapy #Biomarkers
#4
Source: Nature Communications | Authors: Jin M-J, Xu Y-W, He X-X, et al.; Li E-M, Xu L-Y (Shantou University Medical College, multicentre) | Published: October 15, 2026
Score: 7/20 — Nature Communications (base 6) + retrospective multicentre case–control (+1) = 7. No precision bonus: a diagnostic panel, not a treatment-selection biomarker. No survival-benefit bonus.
Phage immunoprecipitation sequencing (PhIP-seq) profiles antibodies against a very large library of peptides at once. This group applied it to 1,039 plasma samples from a multicentre Chinese cohort to look for circulating autoantibodies that separate oesophageal squamous-cell carcinoma (ESCC) from healthy controls. Machine learning selected a panel of core antigenic peptides, including autoantibodies against KMT2C and INPP4B, and built an XGBoost diagnostic model. AUC was 0.912 (95% CI 0.878–0.946) in training and 0.782 (0.696–0.868) in internal validation. In the two independent external cohorts it was 0.809 (0.762–0.856) and 0.729 (0.570–0.889). The step-down from training to external validation is the honest part of the paper, and the wide interval in the second external cohort shows how small it was. The authors also report preliminary signal for monitoring treatment response, and say that needs independent prospective validation.
Post angle: A blood test for ESCC built from autoantibodies rather than tumour DNA. External AUCs of 0.81 and 0.73 are not yet a screening test. But autoantibodies amplify a tiny tumour signal, and that's the problem early detection in this disease has always had. #GIOnc #EsophagealCancer #EarlyDetection #LiquidBiopsy
#5
Source: Annals of Surgery | Authors: Mack CE, Klotz R, Klaiber U, et al.; Büchler MW, Hackert T (Heidelberg) | Published: October 15, 2026
Score: 7/20 — Annals of Surgery (base 5; Other = 5, matching the 2026-10-11 edition) + Phase II RCT (+2) = 7. No impact bonus: a negative trial, stopped for futility, on a surgical-complication endpoint. Ranked last of the three 7s because it is a perioperative rather than an oncological question.
Postoperative pancreatic fistula follows up to 30% of distal pancreatectomies. One proposed fix is to relax the sphincter of Oddi before surgery, so pancreatic juice drains into the duodenum instead of out through the stump. This monocentre, open-label Phase II RCT randomised patients scheduled for elective distal pancreatectomy, for any indication, to endoscopic injection of 100 units of botulinum toxin into the sphincter 3–10 days before surgery, or to no intervention. The combined primary endpoint was clinically relevant (grade B/C) fistula or death within 30 days. Fifty patients were enrolled between March 2020 and August 2023. Recruitment was stopped for futility: even the most favourable outcome in the remaining 10 planned patients could not have shown benefit. Every injection succeeded, and there was no post-procedure pancreatitis. Grade B/C fistula occurred in 28.6% of the botulinum arm versus 15% of controls (P = 0.294), and one patient in the botulinum arm died of fistula-related post-pancreatectomy haemorrhage.
Post angle: A well-run negative trial worth knowing about. Botulinum toxin into the sphincter of Oddi before distal pancreatectomy was safe and technically easy, and it did nothing for fistula: 28.6% vs 15%, stopped for futility. Don't build it into pathways. #GIOnc #PancreaticCancer #HPB #SurgicalOncology
Additional Papers of Interest
-
JAMA Network Open — unplanned post hoc analysis of two German Phase II TNT trials (346 of 385 patients with baseline questionnaires, 185 still reporting at 5 years). Watch-and-wait was associated with clinically meaningful advantages over TME in fecal incontinence (−21.2 points, 95% CI −30.0 to −12.3), stool frequency (−19.8), social, physical and role functioning, and global health status (+9.2, 95% CI 3.2–15.2). Non-randomised, with questionnaire attrition to 48% by year 5, so selection into watch-and-wait cannot be excluded.
-
Modern Pathology — trained on 644 resected PDACs and tested in six retrospective cohorts plus a prospectively accrued external cohort (n = 155), where high-risk patients had adjusted HR 4.45 (95% CI 2.28–8.70) for death. It added 0.015–0.047 to the C-index, but the locked cutoff failed in one external cohort, labelling 96.2% of patients high-risk, which is a transportability warning.
-
Diseases of the Colon & Rectum — single-centre retrospective pilot, 76 patients. A weighted point score built from mucosal whitening, telangiectasia, scarring, ulceration, necrosis and degree of mass regression predicted pCR with AUC 0.85 (95% CI 0.76–0.93) and inter-rater weighted κ 0.9. Needs external validation before it informs watch-and-wait selection.
-
Cancer Research — in CTNNB1-mutant HCC, β-catenin and ARID2 jointly drive TDO2–kynurenine–AhR signalling in myeloid-derived suppressor cells. Arid2 loss or pharmacological AhR inhibition re-sensitised immune-cold tumours to anti-PD-1 in mouse models, and CTNNB1/ARID2 co-mutation tracked with better ICB response in human cohorts. A mechanistic route into the β-catenin immune-exclusion problem.
-
Cell Genomics — an ensemble model trained on 11,188 experimentally validated neoantigens reached AUC 0.904 on held-out data and 0.82 in pancreatic cancer. In 118 PDACs with resolved phylogenies, subclonal neoantigen load was associated with worse PFS (p < 0.001) and OS (p = 0.038). Hypothesis-generating, and relevant to the personalised mRNA vaccine programmes in PDAC.
Back to all digests