Top 5 Papers
#1
Source: U.S. Food and Drug Administration | Authors: U.S. Food and Drug Administration (Genentech, Inc.); efficacy from the ATOMIC/ML39057 trial (NCT02912559) | Published: October 8, 2026
Score: 17/20 — FDA regulatory action (base 8) + registrational randomised Phase III evidence, ATOMIC (+3) + new standard of care in the adjuvant setting (+3) + survival benefit (+2, granted on disease-free survival, the adjuvant-setting time-to-event endpoint; the FDA notice makes no OS statement) + biomarker-guided / precision, dMMR-selected (+1) = 17. No colleague-engagement bonus: the X sweep for this edition ran unattended and the one GI oncology KOL post it surfaced carried no outbound link and was not click-confirmed.
On October 8, 2026, the FDA approved atezolizumab (Tecentriq) in combination with a fluoropyrimidine and oxaliplatin as adjuvant treatment for adult and pediatric patients two years and older with resected stage III mismatch repair–deficient (dMMR) colon cancer; the subcutaneous formulation, atezolizumab and hyaluronidase-tqjs (Tecentriq Hybreza), was approved alongside it for patients 12 and older weighing at least 40 kg. Efficacy came from ATOMIC/ML39057 (NCT02912559), a multicentre, randomised, open-label, active-controlled trial in 711 adults and one pediatric patient after complete resection, randomised 1:1 to atezolizumab plus mFOLFOX6 for 12 cycles followed by six months of atezolizumab monotherapy, or mFOLFOX6 alone for 12 cycles. Investigator-assessed disease-free survival was the major efficacy outcome: HR 0.50 (95% CI 0.35–0.73; p = 0.0001), with median DFS not reached in either arm. The FDA notice gives no overall survival data, so the approval rests on halving the risk of recurrence or death. The recommended adult dose is 840 mg every two weeks, 1200 mg every three weeks, or 1680 mg every four weeks with chemotherapy for six months, then single-agent atezolizumab for six months. For practice, this makes dMMR testing on every resected stage III colon cancer a treatment decision rather than a prognostic footnote.
Post angle: Halving recurrence risk in resected stage III dMMR colon cancer is now an on-label adjuvant option. DFS HR 0.50, OS not yet reported — and every resected stage III colon cancer now needs an MMR result before adjuvant therapy starts. #GIOnc #CRC #ColonCancer #dMMR #Immunotherapy
#2
Source: JAMA Surgery | Authors: Fujiya K, Tokunaga M, et al. (Yoshikawa T, senior author); Stomach Cancer Study Group, Japan Clinical Oncology Group | Published: October 7, 2026
Score: 8/20 — Base 7 (JAMA Surgery) + 1 (retrospective multicentre cohort) = 8. No survival-benefit bonus: treatment was not randomised, and the study's two headline comparisons point in opposite directions (upfront gastrectomy beats upfront chemotherapy overall, while chemotherapy-then-gastrectomy beats gastrectomy-then-chemotherapy among multimodal patients), so there is no single strategy whose benefit the data establish.
Gastric cancer with positive peritoneal lavage cytology (CY1) and/or localized peritoneal metastasis (P1a) sits on the line between curable and palliative disease, and whether to operate first or treat first remains unresolved. This retrospective study from 48 institutions of the JCOG Stomach Cancer Study Group included 1,181 patients treated from 2016 to 2022 (median age 72): 838 (71%) had upfront gastrectomy and 343 (29%) upfront chemotherapy, the latter split into chemotherapy alone (205), gastrectomy after conversion to P0CY0 (92) and non-conversion gastrectomy (46). Overall 5-year OS was 25.0% (95% CI 22.3–28.0%), median 24.3 months. Upfront gastrectomy had longer OS than upfront chemotherapy (5-year 27.6% vs 19.0%; HR 0.80, 95% CI 0.69–0.93; P = .004), yet among multimodal strategies chemotherapy followed by gastrectomy had higher 5-year OS than gastrectomy followed by postoperative chemotherapy (39.9% vs 31.4%; P = .01). Gastrectomy after conversion to P0CY0 had the best 5-year OS at 50.5%, against 17.2% without conversion; gastrectomy alone (10.3%) and chemotherapy alone (3.2%) did poorly. The apparent contradiction is the signature of selection: the upfront-chemotherapy group includes patients who never reached surgery, and the authors' own conclusion is that better preoperative chemotherapy is what this population needs.
Post angle: The number that matters is 50.5%: five-year OS when chemotherapy clears the peritoneum before gastrectomy. Surgery-first 'wins' overall only because the chemo-first group includes everyone who never made it to the OR. Treat first, then operate on responders. #GIOnc #GastricCancer #SurgicalOncology #PeritonealMetastasis
#3
Source: Nature Communications | Authors: Pomej K, Frick A, Scheiner B, et al. (Pinter M, senior author) | Published: Published online October 7, 2026, ahead of PubMed indexing
Score: 8/20 — Base 6 (Nature Communications) + 2 (Phase II) = 8. No survival-benefit bonus: single-arm, 12 patients, with safety as the primary endpoint and no comparator.
Resistance to atezolizumab plus bevacizumab is now the central problem in advanced HCC, and the gut microbiome is one of the few levers proposed for restoring checkpoint sensitivity. In FAB-HCC (NCT05750030), a single-centre, single-arm, hypothesis-generating phase II pilot, 12 patients who had progressed on atezolizumab–bevacizumab received a single colonoscopic FMT — from a donor with HCC who had responded to anti-PD-1 therapy, or from a healthy donor — and were re-challenged with atezolizumab–bevacizumab. The primary endpoint was safety: grade 3–4 treatment-related events were hypertension (n = 5) and proteinuria (n = 3), attributed to bevacizumab, and diarrhoea (n = 1), attributed to FMT, with no serious treatment-related events. The objective response rate was 33% (4 of 12) and median PFS 6.9 months (95% CI 3.0–10.8). Recipients' microbiomes shifted toward their donors', healthy-donor FMT appeared to engraft better, and circulating PD-1+CD8+ T cells fell after FMT plus immunotherapy. The authors call the findings hypothesis-generating and ask for adequately powered trials — the right framing for 12 patients with no control arm, in a setting where re-challenge alone is an untested comparator.
Post angle: Four responses in 12 patients who had already progressed on atezo-bev, after a single FMT. Safe, intriguing, and uncontrolled — the trial that matters is FMT vs no FMT on the same re-challenge. Published online October 7, ahead of PubMed. #GIOnc #HCC #LiverCancer #Microbiome #Immunotherapy
#4
Source: Nature Communications | Authors: Zhao B, Gong Z, Xiao W, et al. (Ju S, corresponding author) | Published: September 8, 2026; indexed in PubMed October 2026
Score: 8/20 — Base 6 (Nature Communications) + 1 (retrospective multi-institutional model development and validation) + 1 (biomarker-guided / precision: an imaging signature used to select patients for neoadjuvant therapy) = 8. No survival-benefit bonus: the neoadjuvant-therapy OS difference is an observational, non-randomised comparison.
Early liver metastasis after resection is one of the main reasons upfront surgery fails in pancreatic ductal adenocarcinoma, and identifying those patients before the operation is exactly the selection problem neoadjuvant therapy is meant to solve. This multi-institutional study built a Mamba-based deep-learning model that integrates CT features from both the primary tumour and the liver, and evaluated it in 1,063 PDAC patients, with AUCs of 0.806–0.890 for early liver metastases. Model-defined high-risk patients had shorter PFS (HR 1.93) and OS (HR 1.89), both p < 0.001. In the high-risk group, neoadjuvant therapy was associated with an OS benefit (17.4 vs 34.1 months; p < 0.001) that persisted after propensity score matching (p = 0.004), while low-risk patients showed no survival difference; radiotranscriptomic analyses linked the high-risk signature to more aggressive biology. The neoadjuvant finding is retrospective and non-randomised, so it is a hypothesis for patient selection rather than evidence of benefit — but it is the kind of selection tool a prospective neoadjuvant trial could stratify on.
Post angle: If you could see liver micrometastases on the baseline CT, you'd never operate first. This model claims to get close: AUC up to 0.89 across 1,063 PDAC patients, with the neoadjuvant survival signal confined to the high-risk group. Retrospective — a stratification factor for the next trial, not a decision tool yet. #GIOnc #PDAC #PancreaticCancer #AI
Additional Papers of Interest
-
Annals of Oncology (Letter) — the CIRCULATE-Japan investigators test the core assumption behind treat-on-molecular-result trials: that resected colorectal cancer passes through an "MRD-positive, scan-negative" window long enough to enrol in. They quantify how often and how long that window occurs, and how many patients were eligible for and enrolled in ALTAIR, the phase III trial of trifluridine/tipiracil vs placebo in ctDNA-positive CRC that this digest led with on June 9 (primary DFS endpoint not met). The figures sit behind the paywall and are not reproduced here.
-
The Lancet (Correspondence) — a letter challenging ASTRUM-006, the perioperative serplulimab gastric trial this digest led with on June 4: the arms differ in both the adjuvant agent and its duration, unblinding at surgery leaves the investigator-assessed EFS open to bias, OS is immature, and MMR/MSI/EBV status was not collected. In their reply the investigators agree the design cannot isolate the adjuvant contribution or show that adjuvant chemotherapy can be omitted, and say blinded central review was consistent with the primary analysis.
-
The Lancet (Correspondence) — three letters on the long-term NordICC report and a reply from the NordICC Study Group. The authors restate a 45% per-protocol reduction in colorectal cancer incidence, call an incidence reduction without a mortality reduction biologically implausible and likely a precision problem with few deaths, and caution against over-reading the Norway–Poland difference, which may reflect participation of 61% vs 33%.
-
Cell Reports Medicine — prospective study of 70 ESCC patients and 183 serial plasma samples. Methylation-based cancer-detection positivity fell from 74% before to 31% after neoadjuvant therapy; it was 33% in non-pCR vs 0% in pCR patients before surgery (p = 0.028); a post-treatment model reached AUC 0.764 for pCR; and post-treatment negativity tracked with longer DFS (HR 0.33, p = 0.035), reproduced in a 30-patient expansion cohort. A candidate tool for organ-preservation decisions, not yet validated.
-
Nature Communications (published online October 6, ahead of PubMed indexing) — sequencing of 431 diffuse gastric cancers found most RHOA mutations without CDH1 alterations; in engineered gastric organoids, RhoA Y42C with Kras G12D and Trp53 loss was sufficient to drive diffuse gastric cancer, with AREG-mediated T-cell exhaustion and marked sensitivity to PD-1 blockade in models. Preclinical; in human samples the PD-1 link is a correlation.
-
Clinical Cancer Research — preclinical. Exatecan-based CDH17 ADCs outperformed MMAE-based ones in P-gp-expressing colorectal cancer models, and forced P-gp expression caused resistance to the MMAE but not the exatecan ADC. In mice expressing human CDH17, clearance was about four-fold faster than in wild-type mice or cynomolgus monkeys — a target-mediated sink, and a warning about on-target, off-tumour toxicity given CDH17's normal gut expression.
-
Digestive and Liver Disease (trial protocol, no results) — a French open-label phase III (NCT06501482) randomising 254 patients with resected metachronous colorectal liver metastases who responded to 2–8 cycles of preoperative FOLFIRI to FOLFIRI reintroduction, up to 12 perioperative cycles in total, vs no postoperative treatment. Primary endpoint 3-year DFS. A response-tailored adjuvant question worth watching.
Back to all digests