About a third of gastric and junctional adenocarcinomas are claudin 18.2 positive — but not a third are eligible

Design

Systematic review and random-effects meta-analysis of prevalence, with prespecified subgroup and sensitivity analyses

Population

12,173 patients with gastric or gastro-oesophageal junction adenocarcinoma across 22 predominantly retrospective cohorts

Primary outcome

Claudin 18.2 positivity at moderate-to-strong membranous staining in 75% or more of tumour cells

Effect

Pooled prevalence 33.99% (95% CI 30.13 to 38.07), I² 92.4%, 95% prediction interval about 18% to 55%

Twenty-two mostly retrospective cohorts totalling 12,173 patients were pooled to estimate how often gastric and gastro-oesophageal junction adenocarcinoma expresses claudin 18.2 at the threshold used in the zolbetuximab phase 3 trials: moderate-to-strong membranous staining in 75% or more of tumour cells. Pooled prevalence was 33.99% (95% CI 30.13 to 38.07). Between-study heterogeneity was 92.4% and the 95% prediction interval ran from roughly 18% to 55%. Antibody clone did not explain the spread — the 43-14A clone gave 32.79% against 41.74% for other clones (p = 0.281) — and neither did geography, with non-Asian cohorts at 37.85% and Asian at 32.10% (p = 0.169). Sensitivity analyses held and Egger's test found no publication bias (p = 0.56).

The authors are explicit about the misreading to avoid. These cohorts spanned different disease settings and were largely unselected for HER2, so the figure is a prevalence of a stain, not a proportion of patients who would be offered zolbetuximab. Anyone planning a service on 34% will overestimate demand.

What it is good for is testing policy. A biomarker present in a third of cases justifies reflex testing rather than testing on request, and the prediction interval is the warning that your own laboratory's positivity rate may land anywhere between one in five and one in two without anything being wrong. If your rate sits far outside that band, the assay or the scoring, not the population, is the first thing to check.

• Set up reflex claudin 18.2 testing on gastric and junctional adenocarcinoma rather than testing case by case.
• Do not quote 34% as the proportion eligible for zolbetuximab — HER2 status and disease setting cut into it.
• Score at the trial threshold: moderate-to-strong membranous staining in 75% or more of tumour cells.
• Audit your laboratory's positivity rate against the 18% to 55% prediction interval, not against the 34% point estimate.
• Record the antibody clone on the report — comparisons across laboratories are otherwise uninterpretable.

The statistics, in plain English

The confidence interval, 30.13 to 38.07, describes how precisely the average across studies is known; the prediction interval, roughly 18% to 55%, describes where a new study's result would be expected to fall. With heterogeneity at 92.4% those two answers are very far apart, and the second is the one that matters to a laboratory comparing itself with the literature. High heterogeneity that survives subgroup analysis by clone and by region means the variation is coming from something unmeasured — most likely differences in fixation, specimen type and scoring rather than in the tumours themselves. That is also why the authors call for standardisation rather than for more prevalence studies.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.