Oral vs injectable semaglutide: nausea, vomiting, diarrhoea and constipation in trials
How often people in each drug’s own clinical-trial arms reported common stomach side effects, pooled from results posted to ClinicalTrials.gov, as held in our corpus on 9 October 2026.
In plain words
- Nausea: reported by about 1 in 8 people (12%) in oral semaglutide trial arms, and about 1 in 3 (31.3%) in injectable semaglutide trial arms. The ranges overlap, so on this evidence the two can’t be told apart for nausea.
- Vomiting: reported by about 1 in 8 people (12.4%) in oral semaglutide trial arms, and about 1 in 6 (15.9%) in injectable semaglutide trial arms. The ranges overlap, so on this evidence the two can’t be told apart for vomiting.
- Diarrhoea: reported by about 1 in 7 people (14.7%) in oral semaglutide trial arms, and about 1 in 5 (21%) in injectable semaglutide trial arms. The ranges overlap, so on this evidence the two can’t be told apart for diarrhoea.
- Constipation: reported by about 1 in 9 people (11.3%) in oral semaglutide trial arms, and about 1 in 6 (16.6%) in injectable semaglutide trial arms. The ranges overlap, so on this evidence the two can’t be told apart for constipation.
For every effect where both have a published rate, the ranges overlap. These trials do not show that either is gentler on the stomach than the other. A trial rate describes a group of people, not what will happen to you.
Not medical advice — talk to your doctor before starting, stopping or switching any medicine.
Read this before comparing
- Not head-to-head. Each figure pools that drug’s own trial arms. The comparison is between two separate sets of trials, not one trial that randomly assigned people to one or the other.
- Not dose- or population-matched. Each figure pools every dose and every population its trials enrolled — obesity, type 2 diabetes and heart-outcome trials together.
- Registry reporting thresholds. Trials list common side effects only above a set threshold, so a trial where an effect was rarer posts no row. That pushes less common figures up.
- Not placebo-adjusted. Some people on placebo report these effects too; the rates here are as posted, before subtracting anything.
The numbers in detail
Pooled rate of each effect in the drug’s own trial arms. 95% CI is the range the true figure probably sits in; it includes a between-trial heterogeneity term. Distinct trials counts separate studies; arm-level rates counts the arms pooled. A row with fewer than 3 distinct trials for an effect shows “insufficient data” and no rate. Where one trial carries more than half of a figure’s pooling weight, it is named.
| Effect | Semaglutide (oral) | Semaglutide (injectable) |
|---|---|---|
| Nausea | 12% about 1 in 8 95% CI 1–60% 12 distinct trials · 23 arm-level rates One trial carries 62.5% of the weight: NCT03914326 | 31.3% about 1 in 3 95% CI 12–59% 15 distinct trials · 22 arm-level rates |
| Vomiting | 12.4% about 1 in 8 95% CI 4–32% 9 distinct trials · 18 arm-level rates | 15.9% about 1 in 6 95% CI 7–34% 15 distinct trials · 22 arm-level rates |
| Diarrhoea | 14.7% about 1 in 7 95% CI 6–33% 10 distinct trials · 21 arm-level rates | 21% about 1 in 5 95% CI 9–42% 15 distinct trials · 22 arm-level rates |
| Constipation | 11.3% about 1 in 9 95% CI 3–39% 8 distinct trials · 17 arm-level rates | 16.6% about 1 in 6 95% CI 5–45% 14 distinct trials · 20 arm-level rates |
Source: benchmark-sample.json, computed by scripts/build-benchmark-sample.mjs on 9 October 2026. Every trial arm behind every figure is listed in that file and on the benchmark sample.
Need more than these effects?
For protocol design, briefing documents and investor decks: the full benchmark covers every effect we track for each approved incretin drug in our corpus (an effect with fewer than 3 trials is marked “insufficient data”, never filled), split by dose where registry arm labels state one, with the row-level source list behind every figure.
Only need one drug? A single-drug pack is also available.
Method and limits
How each figure is computed. Registry-posted results only: ClinicalTrials.gov adverse-event tables for each drug's own trial arms, as stored in the Magistra corpus. Each cell pools the arm-level rates for one drug and one effect with the same function the site publishes its estimates with: rates from one trial collapse to one entry (the plain mean of that trial's arms), entries are weighted by sample size and extraction confidence, and the 95% interval adds a between-trial heterogeneity term on the log-odds scale. A cell with fewer than 3 distinct trials publishes no rate.
- statedRates counts arm-level rate rows, not independent participants: a crossover or multi-period study can post the same people in more than one arm.
- Registries list non-serious adverse events only above each trial's reporting threshold (usually 5%). A trial whose rate for a term fell below that threshold posts no row and is absent from the cell, which biases low-frequency cells upward.
- Each cell pools every dose and population its registry arms describe (obesity, type 2 diabetes and cardiovascular-outcomes trials together). Semaglutide is split into oral and injectable rows; every other drug is one row across its doses. They are not dose-matched or population-matched comparisons.
- obesityOnly narrows a cell to trials whose registry conditions are tagged obesity/overweight (some, such as STEP 2 and SCALE Diabetes, enrol people with type 2 diabetes as well), with the same function and the same 3-trial floor. It is printed beside the all-population figure, never instead of it.
- Pooling weights follow sample size, so one very large trial can carry most of a cell. Where a single trial carries more than half of a cell's pooling weight, the cell names it.
- Trial durations and adverse-event collection windows differ between trials; rates are as-posted, not annualised.
- Rates are as posted to the registry; we do not adjust them for placebo.
- Rates are not drug-attributed: the rule governing ClinicalTrials.gov results reporting (42 CFR 11.10) defines an adverse event as an occurrence "temporally associated with the subject's participation in the research, whether or not considered related" to it.
- Rates come from each registry's non-serious ("Other") adverse-event table, except for a term a trial posts only in its serious table. A participant whose event was recorded as serious is not counted in the non-serious table, and the two tables can share participants, so they cannot simply be added: every cell is a floor on all-severity incidence.
The full method is on the methodology page. This is a sourced summary of the public record, not a substitute for the prescribing information, and not medical advice.