GI Oncology Weekly — Issue 10

This week’s calls — 5 readouts · 2 to act on · ~8 min read. Tap any row to jump.

EPoS II Five-year surveillance is noninferior to three after high-risk adenoma removal, in 10,799 randomised patients Practice-shaping KEYNOTE-859 At 4.5 years the first-line benefit holds, and the CPS ≥10 tail is where it earns its place Practice-shaping PANXEON 86.8% sensitivity for stage I–II pancreatic cancer in 1,785 people — and a 15.6% false-positive rate where it would actually be used Worth watching Dutch Barrett cohort 90% of patients treated for Barrett-related cancer had no prior endoscopy, and 70% of them were already stage II or beyond Worth watching PEGASUS Missed its feasibility endpoint while confirming an emphatic prognostic split — the gap the whole field sits in 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 Colorectal · post-polypectomy surveillance

EPoS II

The strongest randomised evidence yet that we are scoping post-polypectomy patients more often than we need to. Guidelines say three years; this says five holds, with colorectal cancer incidence of 0.77% versus 0.82%. The honest caveat is that this is a prespecified interim analysis at 5.5 years of a trial whose actual primary endpoint is 10-year cumulative incidence — so it is directionally convincing rather than finished, and worth saying that way to a patient.

· 5-year CRC incidence 0.77% vs 0.82%, difference −0.05 percentage points· upper boundary of the one-sided 99.12% CI 0.68, inside the prespecified margin of 0.7· 10,799 randomised across 8 European countries· 5 CRC deaths in total (3 vs 2)

Go deeper

Across eight European countries, 10,799 patients with high-risk adenomas were randomised to a first surveillance colonoscopy at 5 years (n=5398) or at 3 years (n=5401), with the 3-year group also scoped at 5 years. The 5-year cumulative incidence of colorectal cancer was 0.77% with less-frequent surveillance versus 0.82% with more-frequent surveillance — a difference of −0.05 percentage points, and the upper boundary of the one-sided 99.12% CI was 0.68, inside the prespecified noninferiority margin of 0.7 percentage points. Cancer stage distribution at diagnosis did not appear to differ substantially between groups, and 5 patients died of colorectal cancer in total (3, or 0.06%, in the 5-year group; 2, or 0.04%, in the 3-year group). The critical caveat is that this is a prespecified interim analysis at 5.5 years of a trial whose primary endpoint is 10-year cumulative incidence, with inverse probability weighting used for non-participation at the 5-year scope. Full daily entry →

New England Journal of Medicine · randomised noninferiority trial, prespecified interim analysis

Practice-shaping Gastric / GEJ · first-line immunotherapy

KEYNOTE-859

Long-term follow-up of an established first-line standard, and the tail is the news rather than the medians. Absolute median gains stay modest, but 48-month overall survival was 26% versus 10% in the CPS ≥10 group with a plateau emerging. The framing worth adopting is the authors' own: PD-L1 expression enriches for benefit rather than selecting who may receive treatment.

· Median OS 12.9 vs 11.5 mo in the ITT population (HR 0.78, 95% CI 0.70–0.86)· CPS ≥10: 15.8 vs 11.8 mo (HR 0.64, 0.53–0.77)· 48-month OS 14% vs 8% ITT and 26% vs 10% at CPS ≥10· grade 3–4 TRAEs 58.3% vs 49.3%· median follow-up 54.8 mo

HR 0.78 ↓ 22% risk of death 95% CI 0.70–0.86 · Δ +1.4 mo median 48-month OS, CPS ≥10 26% vs 10%

Survival over time · schematic — OS*

Pembrolizumab + chemotherapyPlacebo + chemotherapy

Go deeper

With median study follow-up now 54.8 months (Q1–Q3 46.8–62.1), KEYNOTE-859's 1579 randomised patients show a first-line benefit that has held. Median OS was 12.9 versus 11.5 months in the ITT population (HR 0.78, 95% CI 0.70–0.86), 13.0 versus 11.4 months at CPS ≥1 (HR 0.74, 0.66–0.84), and 15.8 versus 11.8 months at CPS ≥10 (HR 0.64, 0.53–0.77). The absolute medians are modest, but the tail is where this trial earns its place: 48-month OS was 14% versus 8% in the ITT population and 26% versus 10% at CPS ≥10, with the Kaplan–Meier curves separating through 48 months and a modest plateau emerging in the CPS ≥10 group. Grade 3 or 4 treatment-related adverse events occurred in 58.3% versus 49.3%, and grade 5 treatment-related events were lower with pembrolizumab (1.0% versus 2.0%). Full daily entry →

ESMO Open · randomised phase 3, 4.5-year follow-up analysis

Worth watching Pancreatic · blood-based early detection

PANXEON

The most interesting early-detection dataset of the week, and the number to argue about is not the sensitivity. A 15.6% false-positive rate in high-risk controls matters far more than the 3.2% in low-risk ones, because high-risk surveillance is precisely the population you would deploy this in. Promising, prospective, international — and explicitly not yet ready, which the authors say themselves.

· Sensitivity 86.8% for stage I–II PDAC with CA19-9 added· miRNA signature alone AUC 88.6%, sensitivity 83.8% early-stage· false-positive rate 3.2% in low-risk vs 15.6% in high-risk controls· 1,785 individuals across 4 countries

False-positive rate by control population · 0–100% scale

3.2%Low-risk controls

15.6%High-risk controls

050100%

Go deeper

An international, multicentre, observational, prospective biomarker study across four countries enrolled 1,785 individuals with and without PDAC to develop and validate a 10-microRNA blood signature, then combined it with CA19-9 into a composite score called PANXEON. The miRNA signature alone reached an AUC of 88.6% in the testing cohort with 83.8% sensitivity for early-stage PDAC, and showed minimal cross-reactivity with other GI cancers. Adding CA19-9 lifted sensitivity to 86.8% for stage I–II disease, at a false-positive rate of 3.2% in low-risk controls but 15.6% in high-risk controls. PANXEON also flagged high-grade dysplasia in 64.3% of individuals with high-risk pancreatic cysts, and in a small serial cohort of 19 people the signature fell during neoadjuvant chemotherapy and after surgery and rose again before recurrence. The authors pair the assay with a Markov cost-effectiveness model against annual MRCP/EUS surveillance, and are explicit that larger prospective studies are still required. Full daily entry →

Nature Medicine · prospective multicentre biomarker development and validation study

Worth watching Oesophageal · Barrett surveillance and case-finding

Dutch Barrett cohort

Read this next to EPoS II and the week tells one story. We have strong randomised evidence about how often to scope the people already inside a surveillance programme, and a national cohort showing that nine out of ten patients who need curative treatment were never in one. Surveillance itself worked where it was applied — only 11% of surveilled patients had advanced neoplasia missed. The bottleneck is not the interval. It is entry.

· 90% presented de novo with no prior endoscopy (95% CI 89–91; 8,025/8,914)· stage II or beyond in 70% of de novo vs 11% of surveilled patients, p<0.001· 97% of those needing non-endoscopic therapy were de novo· advanced neoplasia missed in only 11% under surveillance

Stage II or beyond at diagnosis · 0–100% scale

70%Presented de novo

11%Under surveillance

050100%

Go deeper

Among 8,914 Dutch patients treated with curative intent for Barrett-related dysplasia or oesophageal adenocarcinoma between 2008 and 2018, 90% (95% CI 89–91; 8,025/8,914) presented de novo — with no prior endoscopy of any kind. Stage at diagnosis diverged sharply: 70% of the de novo group presented at stage II or beyond, against 11% of those already under surveillance (p<0.001). Even among patients whose disease was still amenable to endoscopic therapy, 65% (95% CI 62–67) had never been scoped, and 97% (95% CI 97–98) of those who ultimately required non-endoscopic treatment were de novo presentations. Critically, surveillance itself performed well where it was applied: only 11% (96/889) of patients under surveillance had advanced neoplasia missed. Full daily entry →

Gut · nationwide retrospective registry cohort, 2008–2018

On the radar Colorectal · ctDNA-guided adjuvant therapy

PEGASUS

The instructive negative of the week. Post-surgical ctDNA separated who relapses with real force — 3-year disease-free survival 58% versus 83% — while the trial missed the feasibility endpoint it set for acting on that information. Knowing who relapses is settled. Knowing what to do about it is not, and no amount of prognostic power substitutes for that.

· 3-year DFS 58% (ctDNA-positive) vs 83% (ctDNA-negative), HR 2.71, 95% CI 1.35–5.45, P=0.0036· primary endpoint missed: 2-year relapse-free rate 88% (90% CI 81–93) vs a prespecified ≥92% threshold· 26% ctDNA-positive at the post-surgery landmark· 135 patients enrolled

Go deeper

PEGASUS enrolled 135 patients with resected MSS T4N0 or stage III colon cancer and used post-surgical ctDNA to direct adjuvant therapy, de-escalating ctDNA-negative patients and escalating ctDNA-positive ones. At the post-surgery landmark 26% (n=35) were ctDNA-positive. The prognostic separation was striking: 3-year DFS was 58% in ctDNA-positive versus 83% in ctDNA-negative patients (HR 2.71, 95% CI 1.35–5.45, P=0.0036). The primary endpoint was not met — the observed 2-year relapse-free rate was 88% (90% CI 81–93) against a prespecified threshold of ≥92%, with only 100 of a planned 134 ctDNA-negative cases accrued. Against matched controls from TOSCA, DFS was similar with substantially less neurotoxicity, and relapse tracked with consensus molecular subtype 4, stroma-rich biology. Full daily entry →

Nature Cancer · phase 2, ctDNA-guided adaptive adjuvant trial

* 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

Two papers this week point in opposite directions and are both right, which is the most useful thing that happened.

EPoS II is a genuinely good trial telling us we can stretch post-polypectomy surveillance from three years to five without paying for it in cancer incidence. Four days later a Dutch national cohort reported that 90% of patients treated for Barrett-related cancer had never had an endoscopy at all before the one that found their disease — and that 70% of them were already stage II or beyond, against 11% of the patients who were being surveilled.

Put plainly: our randomised evidence is concentrated on optimising the experience of the minority already inside a surveillance pathway, while the mortality sits overwhelmingly with the majority who never enter one. Both of those facts can be acted on, but only one of them is where the deaths are.

The same asymmetry shows up in the early-detection work. PANXEON reports 86.8% sensitivity for stage I–II pancreatic cancer, which is the number that will travel. The number that decides whether it is deployable is the 15.6% false-positive rate in high-risk controls — because high-risk surveillance is the only setting anyone would actually use it in. A test's headline performance and its performance where it would be used are different questions, and this week they diverged in both diseases.

And PEGASUS is the reminder that sits under all of it:

Detecting disease earlier and knowing what to do about it are separate problems.
ctDNA's prognostic power is no longer in question — 3-year DFS 58% versus 83% settles that.
Acting on it safely is still unproven, and PEGASUS missed the endpoint it set to test exactly that.

A good week for knowing who is in trouble. A thinner one for knowing what to do next.

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