ESMO 2026: Eight Solid-Tumour CAR-Ts, and Four Reasons the Percentages Do Not Compare

The 2026 congress of the European Society for Medical Oncology, held in October, was the first major meeting after the approval of satri-cel — the Claudin18.2 CAR-T cleared by China's NMPA in June 2026 for advanced gastric and gastro-oesophageal junction adenocarcinoma. That approval changed the register of the conversation. For a decade, solid-tumour CAR-T was a category of promising preclinical work with occasional dramatic case reports. After June, there was a product with a label, and everything else on the programme had to be positioned relative to it.

Eight CAR-T programmes presented data in solid tumours. They covered gastric, hepatocellular, colorectal, renal, small-cell lung and large-cell neuroendocrine cancers. The ESMO summary described T cell-based therapies as showing strong and durable response signals in solid tumours, which is a fair description of the aggregate and a misleading one for any individual row.

What follows is the scorecard, and then the more useful part: an explanation of why the numbers in it should not be ranked against each other.

The Scorecard

Read the table as a description of what each programme reported in its own population. It is not a league table, and the columns are deliberately not ordered by response rate.

Reason One: Half of These Patients Were Not Comparable at Entry

The hepatocellular cohort had a median of four prior lines of therapy, with 94% classified BCLC stage C. The colorectal cohort was treated after at least four lines in the high-dose group. The adjuvant gastric cohort, by contrast, had no measurable disease at all — these were patients who had just had a curative gastrectomy and a D2 lymph node dissection, selected because their pathology put them at high risk of recurrence.

Those are not variations on a theme. They are different clinical questions. A patient with no visible tumour is being asked whether the therapy can clear microscopic residual disease; a patient with four prior lines is being asked whether the therapy can shrink a tumour that has already defeated everything else. Response rate is a meaningful measure for the second question and largely meaningless for the first, which is why the gastric programme reported disease-free survival instead.

Reason Two: The Denominators Are Single Digits

Seven patients, nine patients, ten patients, twelve, fourteen, thirty-six. In a cohort of seven, one patient changing category moves the response rate by fourteen percentage points. A 57.1% response rate in fourteen patients and a 44.4% response rate in thirty-six are, statistically, compatible with each other.

This is not a criticism of the investigators. Early-phase trials are designed to find signals and doses, and a small cohort is the correct size for that purpose. It is a criticism of the way those numbers travel. When a percentage with a denominator of seven is printed next to one with a denominator of thirty-six and no confidence interval, readers supply a comparison that the data did not make.

Reason Three: The Dose Cohort Being Reported Is Often the Best One

Several programmes reported both an overall figure and a figure for a selected dose level. IM96 reported 42% overall and 50% in the high-dose cohort. The hepatocellular programme reported 65.4% twelve-month survival overall and 75.6% at the recommended phase 2 dose. LB2102 reported 28.6% at dose level three and above.

Both numbers are legitimate. The overall figure is what happened in everyone treated; the dose-level figure is an estimate of what the eventual labelled product might do. The problem is that headlines and patient enquiries almost always quote the second and not the first. When you are evaluating a programme for yourself or a family member, the number to ask for is the overall figure plus the reason the selected dose was selected — because the selected dose is also the one that will be used on you, and its safety profile is not automatically the one seen at lower doses.

Reason Four: Target Expression Thresholds Differ

The gastric indication has an explicit biomarker definition: Claudin18.2 positive, HER2 negative. The renal programme enrolled CD70-expressing clear-cell histology. The neuroendocrine programme relied on DLL3, which is close to ubiquitous in small-cell lung cancer and therefore less of a filter. The colorectal programme targets GCC, which is expressed across the adenoma-to-carcinoma sequence.

A therapy tested in patients selected for high antigen density cannot be compared with one tested in an unselected population, because the selection is doing part of the work. This is one of the reasons the Claudin18.2 story moved faster than the GPC3 story: a stricter biomarker produces a cleaner signal, and a cleaner signal produces a faster approval.

What the Designs Have in Common

Four of the eight programmes do not primarily modify the part of the receptor that binds the antigen. They modify what the cell does after binding, and the modifications are all attempts to survive the tumour microenvironment.

The hepatocellular and neuroendocrine programmes both use a dominant-negative TGF-beta receptor II. The receptor binds TGF-beta in the microenvironment but cannot transmit its suppressive signal, so the cell absorbs the inhibition without obeying it. The renal programme uses a co-stimulation platform intended to activate host immune cells alongside the CAR-T cells themselves. The colorectal peritoneal programme polarises cells toward Th9 and Tc9 phenotypes, on the argument that these persist longer and exhaust less. And the colorectal liver programme engineers antigen-independent activation, so cells can proliferate without waiting for a target to present itself at sufficient density.

The convergence is the point. A few years ago the assumption was that finding a tumour-specific antigen was the bottleneck. The 2026 data suggest the bottleneck has moved: antigens are available, and the work is now about making a cell that can function in a tumour that is actively hostile to it. That is a harder problem with less obvious solutions, and it explains why the results cluster around forty per cent rather than eighty.

Two Results That Redefine What Success Looks Like

The renal programme reported something that does not fit the usual framing. In four patients, tumour regression continued after CAR copies in peripheral blood had fallen below the limit of detection. If that holds up, it means the infused cells did more than kill directly — they changed the immune environment in a way that outlasted their own presence. The mechanism proposed is the co-stimulation platform activating endogenous immunity, and the clinical implication is that persistence may not be the only route to durability.

The two-day manufacturing result points in a different direction. A transposon-based process completed cell engineering in two days, produced an 85.7% response rate in seven relapsed B-cell patients, and achieved durable remissions at doses well below those used by conventional products. The team's hypothesis is that rapid processing preserves naive and stem-cell memory subsets that a three-week culture degrades. If that reproduces, the constraint on CAR-T access stops being biology and becomes logistics — and logistics is a solvable problem.

What a Patient in China Can Actually Reach

Of the eight programmes, exactly one corresponds to a product you can be prescribed today: satri-cel, approved in June 2026 for Claudin18.2-positive, HER2-negative advanced gastric and gastro-oesophageal junction adenocarcinoma after at least two prior lines of therapy. Everything else is a trial.

For hepatocellular carcinoma, the GPC3 programme is in trials in China. For colorectal cancer, the GCC and CEACAM5 programmes are in early-phase trials. For small-cell lung cancer, the DLL3 programme is in dose escalation. None of these is commercially available, and all of them have eligibility rules that will exclude most patients who ask.

The honest framing for an international patient is this: if your diagnosis matches the approved gastric label, China is one of very few places in the world where a solid-tumour CAR-T exists as a product rather than a protocol. If it does not, the realistic question is whether a trial exists that your histology and prior treatment history fit, and whether entering it is worth the disruption — a question that has a different answer for each of the eight rows above.