Top 5 Papers
#1
Source: Journal of Clinical Oncology | Authors: Ikeda M, Nakachi K, Konishi M, … Ueno M — Japan Clinical Oncology Group | Published: September 23, 2026
Score: 14/20 — Journal of Clinical Oncology (8) + Phase III RCT (+3) + survival benefit, mature 5-year OS (+2) + colleague engagement (+1: one GI oncology KOL posted it within 72h) = 14
The prespecified final 5-year analysis of JCOG's ASCOT trial confirms adjuvant S-1 as a substantive intervention after curative resection for biliary tract cancer. Among 440 patients randomised (S-1 218, observation 222) with a median follow-up of 61.5 months, 5-year overall survival was 64% with S-1 versus 52% with observation — HR 0.72 (95% CI 0.55–0.95), one-sided p=.01 — corresponding to a median OS of 8.2 versus 5.9 years. Relapse-free survival favoured S-1 less decisively: the stratified hazard ratio was 0.76 (95% CI 0.59–0.98), while the unstratified estimate of 0.80 (95% CI 0.62–1.03) crosses unity. Treatment effects were generally consistent across prespecified subgroups. A significant 12-point absolute OS gain at five years is a large effect in a disease whose adjuvant evidence base has been thin; the open question is how well it generalises outside Japan, where S-1 pharmacokinetics differ.
Post angle: Mature 5-year data are where adjuvant trials either hold up or quietly regress to the mean — ASCOT held. Lead with the 64% vs 52% and the 8.2 vs 5.9 year median OS, then raise the generalisability question honestly. #GIOnc #BiliaryTractCancer #Cholangiocarcinoma
#2
Source: U.S. Food and Drug Administration | Authors: Elevar Therapeutics — pivotal REFOCUS trial (NCT04526106) | Published: September 23, 2026
Score: 13/20 — FDA regulatory action (8) + registrational Phase 1b/2 (+2) + biomarker-guided/precision, FGFR2 fusion or rearrangement selected (+1) + colleague engagement (+2: two GI oncology KOLs each posted it within 72h) = 13. No new-standard-of-care bonus applied: this is the third FGFR2-selective inhibitor approved in this setting, not a first-in-class entry.
The FDA approved lirafugratinib on September 23, 2026 for adults with previously treated unresectable, locally advanced or metastatic cholangiocarcinoma harbouring an FGFR2 gene fusion or other rearrangement. Approval rests on the REFOCUS trial's efficacy population of 116 FGFR-inhibitor-naïve patients with prior chemotherapy or chemoimmunotherapy: confirmed ORR was 46% (95% CI 36–55) with a median duration of response of 11.8 months (95% CI 7.5–13.0). The agent is dosed at 70 mg orally once daily and carried breakthrough therapy, priority review and orphan drug designations. This digest covered the same drug's Priority Review on April 18, 2026 with a PDUFA date of September 27 — the approval landed four days early. Importantly, this is not a first-in-class event: pemigatinib and futibatinib are already approved in this setting and infigratinib was withdrawn, so the practical question has shifted from whether to inhibit FGFR2 in cholangiocarcinoma to how to sequence the agents that do, and what to offer after acquired FGFR2 resistance emerges.
Post angle: Resist the 'breakthrough' framing — this is the third FGFR2 inhibitor into the same 2L niche, and that is the story. Sequencing and post-progression resistance are where the unanswered questions now live. #GIOnc #Cholangiocarcinoma #FGFR2 #PrecisionMedicine
#3
Source: Nature Medicine | Authors: Lee CK, Chung T, Kim IH, … Choi HJ — Korean Cancer Study Group (KCSG-HB23-05) | Published: September 22, 2026
Score: 12/20 — Nature Medicine (9) + Phase II (+2) + biomarker-guided/precision, HER2-positive selection (+1) = 12. No survival-benefit bonus: single-arm design with no randomised comparator.
HERBOT, a multi-institutional phase 1b/2 trial from the Korean Cancer Study Group, added trastuzumab and nivolumab to gemcitabine–cisplatin as first-line therapy in 40 patients with HER2-positive unresectable biliary tract cancer. The trial met its primary endpoint with an objective response rate of 55% (95% CI 38.5–70.7), comprising one complete and 21 partial responses, and a disease control rate of 95% (95% CI 83.5–99.4). Median duration of response was 12.6 months (95% CI 5.7–NR) and median progression-free survival 10.6 months (95% CI 7.8–17.4); at a median follow-up of 17.0 months median overall survival was not reached, and two patients (5.0%) proceeded to curative-intent conversion surgery. Toxicity was not trivial — grade ≥3 neutropenia in 57.5%, anaemia in 30.0% and thrombocytopenia in 22.5%. A preplanned AI whole-slide analysis found that a higher proportion of HER2 3+ tumour cells was associated with greater benefit, which is arguably the more generalisable finding: quantifying HER2 heterogeneity, rather than scoring it dichotomously, may be what selects these patients properly.
Post angle: Single-arm and n=40, so hold the enthusiasm — but a 55% ORR with a 12.6-month DOR earns a randomised trial. The whole-slide AI quantification of HER2 3+ cell fraction is the part worth watching. #GIOnc #BiliaryTractCancer #HER2
#4
Source: Nature Medicine | Authors: Zhou J, Guo G, Yao J, … Wang Q — Sun Yat-sen University / DAMO Academy | Published: September 22, 2026
Score: 11/20 — Nature Medicine (9) + large multicentre external and prospective validation, real-world tier (+1) + biomarker-guided/precision risk stratification (+1) = 11
EAGLE is a deep-learning model that detects esophageal cancer on noncontrast CT scans performed for other indications, trained on 6,813 patients at two centres and validated across 12 centres in three countries totalling 80,612 patients. In external opportunistic-screening cohorts spanning eight centres (n=11,466) it achieved 90.0% sensitivity for cancer at 98.5% specificity, and real-world calibration across three centres (n=35,402) cut false positives by 72.7% while preserving sensitivity. Prospective hospital validation in 17,446 patients yielded a positive predictive value of 42.2%, and real-world low-dose CT screening in 10,959 people reached 99.94% specificity. Performance against paired endoscopy (n=702) was 78.4% for stage I cancer but only 65.0% for precancerous lesions, and the opportunistic cohorts showed 52.5% sensitivity for precancer — this is a tool that finds cancer rather than dysplasia. The design insight is that the scans already exist; the constraint that will decide deployment is whether endoscopy capacity can absorb referrals at a 42.2% PPV.
Post angle: Opportunistic screening on CTs that were already ordered is the most deployable form of AI screening there is. Lead with 90.0% sensitivity at 98.5% specificity, then make the honest point: 52.5% for precancer means this is cancer detection, not prevention. #GIOnc #EsophagealCancer #AI #CancerScreening
#5
Source: Lancet Gastroenterology & Hepatology | Authors: Yaqub S, Valberg M, Bjørnbeth BA, … Waage A — 14 centres, Norway/Sweden/Denmark | Published: September 23, 2026
Score: 10/20 — Lancet Gastroenterology & Hepatology (7) + Phase III RCT (+3) = 10. No clinical-impact bonus: the primary endpoint was not met and the overall survival signal favours placebo. The scoring table has no category for a practice-relevant harm finding, so the score understates this trial's clinical importance — placement in the Top 5 reflects that.
ASAC randomised 466 patients (428 in the full analysis set: aspirin 217, placebo 211) to aspirin 160 mg daily or placebo for up to three years after curative-intent treatment of colorectal cancer liver metastases, across 14 centres in Norway, Sweden and Denmark. The primary endpoint was not met: disease-free survival HR was 1.08 (95.44% CI 0.85–1.38, p=0.58) across 269 events, with median DFS of 1.13 years on aspirin versus 1.40 years on placebo. More concerning, 36-month overall survival was 76% with aspirin versus 85% with placebo, HR for death 1.64 (95% CI 1.01–2.65, p=0.045), and serious adverse events occurred in 17/217 (8%) on aspirin versus 4/211 (2%) on placebo, including excess myocardial infarction, epistaxis, duodenal ulcer and cerebral haemorrhage. The OS result is a nominally significant secondary endpoint in a trial that missed its primary, so it should be read as a safety signal rather than proof of harm. Read against ALASCCA — which this digest led with on April 21 and which halved recurrence with the same 160 mg daily dose — the discriminating variable is patient selection: ALASCCA enrolled PI3K-pathway-altered localized disease, ASAC an unselected population after liver-metastasis resection.
Post angle: The headline is not 'aspirin doesn't work in colorectal cancer' — ALASCCA showed it does, in PI3K-altered localized disease. ASAC shows what happens when you drop the biomarker and move to resected liver metastases. Same dose, opposite result. #GIOnc #CRC #ColorectalCancer
Additional Papers of Interest
-
Gut — multi-omics across 156 patients (RNA-seq, spatial transcriptomics and metabolomics, lipidomics, serum proteomics) defines an SH-HCC index and a 12-lipid diagnostic model, with FABP4 functionally validated in Sleeping Beauty mouse models; drug-response analysis infers reduced sensitivity to sorafenib and IO-based regimens while TACE may remain effective.
-
Nature Communications — single-cell RNA, TCR and whole-exome sequencing of 52 patients nested in a phase 3 neoadjuvant trial: responders show clonal enrichment of cytotoxic CX3CR1⁺CD8⁺ clonotypes and reduced dominance of exhausted CXCL13⁺CD8⁺ states, with mouse and peripheral-blood validation.
-
JAMA Internal Medicine — retrospective US Veterans Affairs national cohort (2008–2020) of adults with imaging-confirmed steatotic liver disease and no viral hepatitis, comparing ALBI, aMAP, APRI, BARD, FIB-4, NFS, SAFE and two further scores head-to-head to inform surveillance decisions.
-
JCO Precision Oncology — single-arm phase II in 28 heavily pretreated patients (82.1% with ≥2 prior lines) met its primary endpoint with a 16-week PFS rate of 35.7% (95% CI 17.8–53.2); mPFS 3.3 months, mOS 4.6 months, DCR 50%, at only 4.8 months median follow-up.
-
European Journal of Cancer — extended follow-up of the 52-patient FLOT + nivolumab arm shows mPFS 7.0 and mOS 14.6 months; every prespecified biomarker subgroup comparison (CPS, NLR, MLR) is non-significant despite wide median separations, so these remain hypothesis-generating.
-
ESMO Gastrointestinal Oncology — 86 consecutive patients across Spain, Slovenia and Italy: 45.3% were diagnosed via the emergency department versus 46.5% through primary care, with a median symptom-onset-to-diagnosis interval of 2.56 months.
Back to all digests