GI Oncology Daily Digest

The first randomised trial to ask whether responders to first-line immunotherapy in advanced HCC should go to surgery. 489 treatment-naive patients with macrovascular invasion and no extrahepatic spread received three cycles of atezolizumab plus bevacizumab; the 201 who achieved a response or stable disease and were judged resectable were randomised 1:1 to liver resection or continued maintenance. Time to treatment failure was 20.4 versus 11.8 months (HR 0.60, 95% CI 0.39-0.91; p=0.015) at a median follow-up of 18.4 months. The cost is real and belongs in the same sentence as the benefit: grade 3-4 treatment-related adverse events ran 39% in the surgery arm versus 21% with maintenance, with two treatment-related deaths after resection (abnormal liver function and liver failure). Two further caveats before this changes anyone's practice — the endpoint is time to treatment failure rather than survival, and all 24 sites are in China, so generalisability to populations with predominantly non-HBV aetiology is untested. A companion Lancet editorial in the same issue asks whether this opens a new era for conversion surgery; the honest answer today is that it opens the question.

A 760-patient, 166-centre, 28-country trial that failed its primary endpoint and is more instructive for it. Patients with stage II-IVA unresectable oesophageal squamous cell carcinoma who had completed definitive chemoradiotherapy were randomised 1:1:1 to atezolizumab plus the anti-TIGIT antibody tiragolumab, atezolizumab plus placebo, or placebo. The primary comparison — the doublet versus placebo — missed: investigator-assessed PFS HR 0.82 (95% CI 0.65-1.03; p=0.0947) and OS HR 0.91 (95% CI 0.70-1.18; p=0.4772). Underneath it, atezolizumab alone versus placebo produced OS HR 0.69 (95% CI 0.52-0.91; p=0.0085) and PFS HR 0.74 (95% CI 0.58-0.93). Because the testing hierarchy failed at the first gate, those monotherapy numbers are descriptive and cannot support a licence or a guideline on their own. Read plainly, the trial says two things at once: adding TIGIT blockade bought nothing, and single-agent PD-L1 blockade after chemoradiation looks genuinely active in a setting with few options. The second finding needs its own confirmatory trial, and this one cannot substitute for it.

A reformulation trial that turned into a survival trial, which is not how these usually go. 536 patients with unresectable, recurrent or metastatic gastric cancer progressing after fluoropyrimidine-based first-line therapy were randomised 1:1 to paclitaxel oral solution (200 mg/m² twice daily on days 1, 8 and 15 of a 28-day cycle) or standard paclitaxel injection (175 mg/m² on day 1 of a 21-day cycle), with dual primary endpoints of blinded independent-review PFS and OS. The oral formulation was non-inferior on PFS — 3.02 versus 2.89 months, HR 0.894 (95% CI 0.719-1.112; P=0.311) — and superior on overall survival, 9.13 versus 6.54 months, HR 0.770 (95.5% CI 0.635-0.934; P=0.006). That combination deserves scrutiny rather than a headline: a 2.6-month OS gain without a PFS difference is not the usual shape of a cytotoxic benefit, and the most plausible explanations are tolerability and time on treatment rather than greater tumour control. The safety data point the same way, with less peripheral neuropathy, fewer hypersensitivity reactions, less alopecia and fewer musculoskeletal disorders; treatment-related fatal events were rare in both arms (1.5% versus 1.1%). Two limits belong beside the result: every site is in China, and the sponsor, Haihe Biopharma, has three co-authors on the paper. Worth watching for whether regulators outside China accept an OS-only win in a formulation comparison.

A small-area analysis of 3,110 US counties covering a mean annual population of 306 million from 2000 to 2019, with estimates corrected for the well-documented misclassification of race and ethnicity on death certificates. Nationally the story is good: colorectal cancer mortality in adults 45 and older fell from 82.5 to 54.4 per 100,000 in men and from 58.9 to 39.7 per 100,000 in women. The county-level story is not. Among American Indian and Alaska Native men aged 45 and over, county rates ranged from 13.6 to 213.5 per 100,000 — a roughly 16-fold spread within one country. In 275 counties at least one population group had mortality above 100 per 100,000, including 182 counties for Black men and 92 for American Indian and Alaska Native men. And in adults under 45, mortality was stable or rising throughout. Two limits worth stating: the data end in 2019, so nothing here reflects pandemic-era screening disruption, and an ecological analysis identifies where the gaps are without explaining why.

331 plasma samples from 36 patients with locally advanced oesophageal cancer, drawn before and after surgery, run head-to-head on two MAESTRO-based ultrasensitive tumour-informed assays. Nine of the 14 patients who recurred and had post-operative plasma available were ctDNA-positive on their very first post-operative sample, and all 14 were positive in at least one post-operative sample before radiological recurrence. The number that matters for assay design: 59% of all ctDNA-positive samples sat below the limit of detection of first-generation assays, which is a direct measurement of what earlier platforms were missing rather than an inference. Oncodetect V2 tracked a median 2.1-fold more mutations per patient, up to 5,000, with 98.9% positive percent agreement against MAESTRO-Pool and a longer lead time to recurrence. Two things to hold onto: n=36 is small, and this is a detection study — nobody's treatment was changed by the result. It tells us we can see residual disease earlier in oesophageal cancer; it does not yet tell us what to do when we do.