A 12-Day Study and REWIND's 7-Year One Both Feed Our Pooled GLP-1 Nausea Rate — Here's How Much Each Counts

Two of the studies behind our pooled nausea rate watched their participants for very different lengths of time. NCT05936138, a Phase 1 single-dose study of orforglipron, followed people for 12 days and posted a 16.7% nausea rate in one arm — one of its six participants. REWIND, dulaglutide's cardiovascular-outcomes trial, followed 9,901 people for up to 7 years and posted 14.9% in its 4,943-person dulaglutide arm. Our pooled nausea figure averages rows from both — weighted by each study's own sample size, never told which window it came from.

The window nobody states

Every trial posts its own adverse-event counting window — how long it watched people before reporting a rate — and ClinicalTrials.gov calls it the `timeFrame`. Nothing in a stated rate carries that window along with it: 20% from a 12-week arm and 20% from a 240-week arm read identically once pooled. We added the field to our own rows (`extractedWeeks`) and, as of 3 October 2026, across the 1,734 eligible rows behind our 15 tracked effects, 1,727 come from a ClinicalTrials.gov record. Of those, 1,671 actually state a counting window — the other 56 are CT.gov-sourced rows whose record carries no adverse-event `timeFrame` at all, an honest null rather than a zero. The shortest stated window is 1.7 weeks (about 12 days); the longest is 364 weeks (about 7 years). 755 of those 1,671 window-stating registry rows — 45.2% — come from trials of 26 weeks or less, almost all from seven Phase 1 studies of orforglipron — formulation, dose-escalation, drug-interaction and special-population studies.

That is a lot of rows from short studies. It is not a lot of weight. Our pooling (described in full in what a pooled rate is made of) weights each study by its own stated sample size, not by its window — but in this corpus the short-window studies are overwhelmingly small early-phase trials, so weight concentrates in the long trials anyway. Computed from the same `windowMix` function that runs on magistra.health, against the repo-committed corpus on 3 October 2026:

On every one of these seven effects, rows from trials of 26 weeks or less carry 1.1–2.4% of the pooled weight despite being 43–52% of the row count for that effect, and the median stated rate in those short trials sits well below the median in trials over two years — for reduced appetite (shown above) and two more we don't table here, abdominal pain and fatigue, the short-window median is literally 0%, not merely low: in a short Phase 1 arm of a few dozen people, many report no events for a given term at all. Nausea's own median counting window, across the distinct studies behind its eligible rows, is 57 weeks.

What this changes, and what it doesn't

Nothing here is a correction — no pooled rate moves, and no row is withheld. It is a disclosure our methodology page has carried since 2 October 2026 (v5.40): "A pooled clinical rate combines trials of every indication... and each trial counts an event at least once within its own window, which runs from days to years... read a pooled rate as a share over roughly one to several years." This article is the worked version of that sentence for one effect, with the two trials that sit at its extremes named.

The practical reading rule: a GLP-1 side-effect rate pooled from trial registries is, by weight, mostly a rate over a year or more of use, not a rate at any single dose or duration — a 12-day arm can add a row, but it will almost never move the number. If you are looking for what a specific short period shows, read the study, not the pool; the live endpoint names every source with its URL and its own stated window is one `fetch` away for anyone who wants to re-weight it differently.

A related, separate selection effect — which non-serious terms a trial lists at all — is covered per effect in the headache and abdominal pain pieces: a trial only lists a term in its "other events" table if it reached 5% in some arm, so low-rate effects are structurally under-represented in exactly the same base this article describes. The two effects are independent axes of the same pooled number.

Reproduce it

Every figure above was computed on 3 October 2026 using `classifyRatePoint`, `windowMix` and `medianWindowWeeks` — the exact functions `/api/data` calls in production — first against the repo-committed corpus and then re-run against the live production data store the same evening, with identical results; the trial windows, enrolment and arm sizes were re-read from ClinicalTrials.gov the same day. The corpus grows, so re-check the live endpoint (`rateBase.clinical.windowMix`) before citing a number from this article as current.

The full dataset — 15 effects, every source with its URL and its own counting window where the registry states one, CC BY 4.0 — is at magistra.health/en/data-api.