GI Oncology Weekly — Issue 3

This week’s calls — 5 readouts · 3 to act on · ~8 min read. Tap any row to jump.

TORCH TACE plus selective ablation of residual tumour more than doubles PFS in BCLC-B HCC Practice-shaping FIND ctDNA methylation-guided surveillance doubles curative-intent salvage after resected CRC Practice-shaping ARISTOTLE Adding irinotecan to preoperative chemoradiotherapy fails, at a real toxicity cost Practice-shaping COMMIT Chemo/bevacizumab added to atezolizumab improves PFS in first-line dMMR mCRC — at double the toxicity Worth watching PRODIGE-24/CCTG PA6 (genomic ancillary) Adjuvant mFOLFIRINOX benefit was confined to KRAS-mutant tumours; HRR status predicted nothing On the radar

This week · 5 readouts

The Verdicts

Every trial that matters this week, distilled to a single call — act on it, watch it, or keep it on the radar.

Practice-shaping Liver · intermediate-stage HCC

TORCH

A positive phase 3 in unresectable intermediate-stage hepatocellular carcinoma: going back after chemoembolization and thermally ablating whatever viable tumour remains more than doubled progression-free survival and halved the hazard of death, with essentially no added grade 3–4 toxicity. The principle is simple and hard to argue with — residual viable tumour after TACE is what kills these patients.

Go deeper

This open-label phase 3 randomized trial enrolled 241 patients with BCLC stage B unresectable hepatocellular carcinoma across two Chinese centres and compared transarterial chemoembolization followed by selective radiofrequency ablation of residual viable tumour against TACE alone. Median progression-free survival was 17.7 versus 7.3 months (hazard ratio 0.47, 95% CI 0.34–0.65, p<0.001), and median overall survival was 88.6 versus 35.1 months (hazard ratio 0.50, 95% CI 0.34–0.73, p<0.001). Grade 3–4 treatment-related adverse events were comparable at 23.2% versus 18.3%, so the benefit does not appear to be bought with meaningful added toxicity. Two caveats deserve stating plainly. This is a two-centre trial in a highly selected BCLC-B population treated by operators skilled in both modalities, and the 88.6-month median overall survival in the experimental arm is exceptionally long for intermediate-stage disease — a number that will need reproduction in a Western, more aetiologically mixed cohort before it anchors practice. The direction of effect, though, is consistent with the principle that residual viable tumour after chemoembolization is the thing that kills these patients. Full daily entry →

Survival over time · schematic — PFS*

TACE + selective thermal ablationTACE alone

· Median PFS 17.7 vs 7.3 mo (HR 0.47, 0.34–0.65; p<0.001)· Median OS 88.6 vs 35.1 mo (HR 0.50, 0.34–0.73; p<0.001)· Grade 3–4 TRAE 23.2% vs 18.3%· n=241

HR 0.47 ↓ 53% risk of death 95% CI 0.34–0.65 · Δ +10.4 mo median

JAMA Oncology · phase 3 randomized, two centres, n=241 Debate on X →

Practice-shaping Colorectal · post-resection surveillance

FIND

The first randomized phase 3 to show that a surveillance biomarker changes what you can offer at relapse. ctDNA-guided monitoring did not find more recurrences than standard CT follow-up — it found them roughly four months earlier, when twice as many patients were still candidates for curative-intent metastasis-directed therapy. Overall survival is not yet answered, and that caveat is real.

· Curative-intent salvage 48.1% vs 23.6% (RR 2.03; p=.008)· Recurrence detected 3.9 mo earlier (9.5 vs 13.4 mo; p<.001)· Recurrence rate 18.0% vs 18.6% (p=.919)· Liver/lung-only curative resection 42.3% vs 18.2% (p=.002)· n=584

Recurrences reaching curative-intent metastasis-directed therapy · 0–100% scale

48.1%ctDNA methylation-guided surveillance

23.6%Standard CT-based surveillance

050100%

Go deeper

The FIND trial (NCT05904665) randomized 584 patients with resected nonmetastatic colorectal cancer to ctDNA methylation-guided dynamic surveillance versus standard CT-based monitoring; a positive ctDNA triggered immediate imaging, and two consecutive negatives returned the patient to routine intervals. At a median follow-up of 23.3 months recurrence rates were essentially identical (18.0% vs 18.6%, p=.919) — but what happened at recurrence was not. The ctDNA-guided arm reached curative-intent metastasis-directed therapy in 48.1% versus 23.6% of recurrences (relative risk 2.03, p=.008), and detected recurrence a median 3.9 months earlier (time to clinical recurrence 9.5 vs 13.4 months, p<.001). Among recurrences confined to liver and/or lung, curative resection rates were 42.3% versus 18.2% (p=.002), with more favourable hepatic anatomy: ≤3 lesions in 75.0% vs 28.6% (p=.005), ≤3 cm in 90.0% vs 57.1% (p=.033), and unilobar disease in 80.0% vs 28.6% (p=.002). The honest caveat is that this is a resectability endpoint, not a survival endpoint — whether the extra salvage surgery translates into overall survival remains unproven. Full daily entry →

Journal of Clinical Oncology · randomized phase 3, n=584 Debate on X →

Practice-shaping Rectal · high-risk LARC

ARISTOTLE

A definitively negative phase 3 — and negative trials this clean are practice-shaping. Seven years of follow-up in 589 UK patients show adding weekly irinotecan to capecitabine chemoradiotherapy does nothing for disease-free survival while nearly doubling grade ≥3 toxicity and degrading delivery of the radiotherapy and capecitabine themselves. The authors' conclusion is blunt: don't do it.

· 36-mo DFS 68% vs 67% (HR 0.91, 0.68–1.23; p=0.54)· Deaths 31% vs 32%· Grade ≥3 AEs 78% vs 52%· Full 45 Gy delivered 75% vs 89%· Median follow-up 78 mo· n=589

Grade 3 or worse adverse events · 0–100% scale

78%Irinotecan + capecitabine CRT

52%Capecitabine CRT (standard of care)

050100%

Go deeper

ARISTOTLE (EudraCT 2008-005782-59) randomized 589 patients with MRI-defined locally advanced rectal cancer threatening or involving the resection margin to preoperative 45 Gy in 25 fractions with either capecitabine alone or reduced-dose capecitabine plus weekly intravenous irinotecan 60 mg/m² in weeks 1–4; 564 were analysed in the modified intention-to-treat population. After a median follow-up of 78 months the trial is unambiguously negative: 36-month disease-free survival was 68% (95% CI 63–73) with irinotecan versus 67% (61–72) with standard of care, hazard ratio 0.91 (95% CI 0.68–1.23, p=0.54), and deaths were near-identical at 31% versus 32%. The cost side is where the trial speaks loudest. Grade 3 or worse events occurred in 78% of the irinotecan group versus 52% of controls, with diarrhoea at 14% versus 4%, lymphopenia at 66% versus 35%, and neutropenia at 10% versus 1%. That toxicity fed straight back into deliverability — only 75% of irinotecan patients received the full 45 Gy versus 89% of controls, and only 68% received at least 90% of planned capecitabine versus 89%. Five deaths were attributed to protocol treatment, three of them in the irinotecan arm. The authors' conclusion is appropriately blunt: irinotecan should not be combined with radiotherapy plus capecitabine in this setting. Full daily entry →

Lancet Oncology · phase 3 randomized, 75 UK sites, n=589 (Cancer Research UK)

Worth watching Colorectal · first-line dMMR/MSI-H

COMMIT

The mature publication of NRG-GI004/SWOG-S1610 confirms the triplet beats single-agent atezolizumab on progression-free survival and response rate in first-line dMMR/MSI-H metastatic colorectal cancer. The open question is not whether it works but who needs it: single-agent checkpoint blockade already produces durable disease control in roughly half these patients, and grade ≥3 events nearly doubled with the combination.

· PFS HR 0.439 (0.23–0.84; p=.0103)· ORR 86.1% vs 46%· 12-mo disease control 64.7% vs 32.4%· Grade ≥3 AEs 34 vs 18 patients· Median follow-up 46 mo· n=102

Objective response rate · 0–100% scale

86.1%mFOLFOX6/bevacizumab + atezolizumab

46%Atezolizumab monotherapy

050100%

Go deeper

COMMIT (NRG-GI004/SWOG-S1610) is now fully published — the mature peer-reviewed paper behind the ASCO GI topline this digest covered on April 10. This three-arm open-label phase III randomized first-line dMMR/MSI-H mCRC patients 1:1:1 to mFOLFOX6/bevacizumab, atezolizumab monotherapy, or the triplet; the chemotherapy-alone arm was closed after 20 patients once KEYNOTE-177 reported, and the trial continued as atezolizumab versus triplet with a revised target of 100 patients. From November 2017 to March 2025, 102 patients enrolled (FFX/bev n=20, atezolizumab n=41, triplet n=41). At a median follow-up of 46 months, PFS favoured the triplet over atezolizumab monotherapy (hazard ratio 0.439, 95% CI 0.23 to 0.84; p=.0103, below the prespecified critical value of 0.0152). Objective response rate was 86.1% versus 46%, and 12-month disease control was 64.7% versus 32.4%. Grade 3 or higher adverse events of any attribution occurred in 52 patients — 18 in the atezolizumab arm and 34 in the combination arm. The rationale was that VEGF inhibition plus chemotherapy might synergize with PD-L1 blockade in the roughly half of dMMR patients who progress within 12 months on single-agent PD-1 therapy. Full daily entry →

Journal of Clinical Oncology · NRG-GI004/SWOG-S1610, phase 3, n=102 Debate on X →

On the radar Pancreas · resected PDAC, adjuvant

PRODIGE-24/CCTG PA6 (genomic ancillary)

The molecular readout of the trial that set our adjuvant pancreatic standard. In 317 sequenced tumours, the mFOLFIRINOX advantage over gemcitabine was confined to KRAS-mutant disease, with a formal interaction test of p=0.010 — while homologous recombination repair status, the biomarker most people expected to matter, predicted nothing. The authors are explicit that this does not change practice. It does raise a question worth a prospective answer.

· mFFX vs gemcitabine in KRAS-mutant: DFS sHR 0.60 (0.45–0.79; p<0.001)· No benefit in KRAS-wild-type (interaction p=0.010)· PurIST classical vs basal-like within mFFX: sHR 0.48 (0.31–0.77)· HRR status not predictive· 317/350 tumours sequenced

Go deeper

This is the molecular ancillary study of PRODIGE-24/CCTG PA6, the phase III trial that made adjuvant mFOLFIRINOX (mFFX) the standard after resection of pancreatic ductal adenocarcinoma. Tumour DNA sequencing succeeded in 317 of 350 tumours (168 mFFX, 149 gemcitabine), with transcriptomic subtyping by the PurIST classifier, four PDAC driver genes, 24 homologous-recombination-repair (HRR) genes, and single-base-substitution signatures. Two findings stand out. First, PurIST subtype was prognostic within the mFFX arm — classical tumours had markedly better disease-free survival than basal-like tumours (stratified HR 0.48, 95% CI 0.31–0.77). Second, and more provocatively, the mFFX benefit over gemcitabine was confined to KRAS-mutant tumours (sHR 0.60, 95% CI 0.45–0.79, p<0.001), with no benefit detected in KRAS-wild-type tumours and a formal interaction test of p=0.010. HRR gene status was not predictive, and the mFFX advantage held across SBS-signature-positive and -negative subgroups. The authors are careful: these results do not support a change in adjuvant strategy, mFFX remains standard, and the KRAS-wild-type observation is hypothesis-generating. Full daily entry →

Journal of Clinical Oncology · translational ancillary of a phase 3 trial, n=317 sequenced

* Survival curves are illustrative — modeled from each trial’s reported median and hazard ratio (assuming exponential survival), not the published Kaplan–Meier curves.

The take

The take — our standards of care are averages

Three of this week's five trials ask the same question from different angles: we know the treatment works on average, but do we know whose average it is?

PRODIGE-24's genomic ancillary is the sharpest version. Adjuvant mFOLFIRINOX is what we give essentially every fit patient after pancreatic resection — and the molecular data say its advantage over gemcitabine lived in the KRAS-mutant tumours, with a formal interaction of p=0.010 and no benefit detectable in KRAS-wild-type disease. That subgroup is roughly a tenth of cases and increasingly looks biologically distinct. A trial waiting to be written, not a footnote. The quieter surprise is the null: HRR status predicted nothing.

FIND is the optimistic version. It doesn't find more recurrences — rates were identical — it finds them about four months earlier, when twice as many patients were still operable with curative intent. Keep saying out loud that overall survival is still pending.

ARISTOTLE is the version where the answer is 'nobody'. No DFS signal across 589 patients and 78 months, while grade ≥3 toxicity went from 52% to 78% and fewer patients completed their radiotherapy and capecitabine. Intensification that degrades delivery of the backbone is the failure mode this trial documents cleanly.

TORCH — ablate what TACE left; PFS 17.7 vs 7.3 mo, no meaningful added toxicity. Hold the 88.6-mo OS loosely until reproduced outside two expert centres.
COMMIT — the triplet works in first-line dMMR mCRC, at roughly double the grade ≥3 events. Which patients need it is the live question.
Thin week, honestly reported. Tuesday's daily ran four papers, not five. Fewer verified papers beats padding.

GI Oncology Weekly · curated by Dr. Allan Pereira, Moffitt Cancer Center. Educational summaries for practising oncologists — not medical advice; verify against primary sources before acting.

All weekly issues