Most supplement writing reports direction: “may lower cholesterol”, “supports gut health”. Direction is cheap. What a trial actually produced is a number with a confidence interval, and that number is what determines whether a product can be compared with a study at all.
This page collects the measured outcomes we have verified across our ingredient and comparison articles, in one table, so the figures can be checked and cited directly. Every row states the population, the trial count, and the limitation that travels with it.
How to read this table
Four columns carry the weight. Effect is the measured change, not a direction. 95% CI is the range the true effect plausibly sits in; an interval that crosses zero means the trial did not establish an effect, regardless of how the point estimate reads. n is the number of participants pooled, because a striking effect in 16 people is not the same finding as the same effect in 2,313. Limitation is not boilerplate — it is the reason a clinician would not generalise the row.
Lipid outcomes
| Ingredient | Outcome | Evidence | Effect | n | Limitation |
|---|---|---|---|---|---|
| Psyllium husk | LDL cholesterol | Moderate | −0.33 mmol/L (95% CI −0.38 to −0.27) at ~10.2 g/day | 28 RCTs | Heterogeneity across doses and formats |
| Psyllium husk | Non-HDL cholesterol | Moderate | −0.39 mmol/L (95% CI −0.50 to −0.27) | Same 28 RCTs | Same limitation |
| Psyllium husk | Apolipoprotein B | High | −0.05 g/L (95% CI −0.08 to −0.03) | Same 28 RCTs | Favourable, but not a marker people track |
| Psyllium husk | Total cholesterol | Low to moderate | −9.05 mg/dL (95% CI −13.71 to −4.40); I² = 89% | 41 RCTs, n=2,049 | Very high heterogeneity |
| Psyllium husk | LDL cholesterol | Low to moderate | −8.55 mg/dL (95% CI −12.92 to −4.19); I² = 88% | Same 41 RCTs, n=2,049 | Duration and dose had no clear effect |
| Psyllium husk | Triglycerides | Low | −5.29 mg/dL (95% CI −12.14 to +1.54) | Same 41 RCTs | CI crosses zero — no reliable effect |
| Psyllium husk | HDL cholesterol | Low | +0.57 mg/dL (95% CI −0.88 to +2.04) | Same 41 RCTs | CI crosses zero — no reliable effect |
| Soluble fibre | Total cholesterol | Moderate | −10.82 mg/dL (95% CI −12.98 to −8.67) | Meta-analysis of RCTs | Effect size varies by fibre type |
| Soluble fibre | LDL per 5 g/day added | Moderate | −5.57 mg/dL | Dose-response meta-analysis | Strongest in people starting with higher LDL |
| Oat β-glucan | LDL cholesterol | Moderate | −0.25 mmol/L at ≥3 g/day | Meta-analysis of oat β-glucan RCTs | HDL and triglycerides unchanged |
Glucose outcomes
| Ingredient | Outcome | Evidence | Effect | n | Limitation |
|---|---|---|---|---|---|
| Berberine | Fasting plasma glucose | Moderate | −0.54 mmol/L (95% CI −0.77 to −0.30) | 28 trials, n=2,313 | Almost all participants had type 2 diabetes |
| Berberine | Post-meal glucose | Moderate | −0.94 mmol/L (95% CI −1.27 to −0.61) | Same pooled analysis | Duration mostly 8–12 weeks |
| Soluble fibre | Post-meal glucose (T2D) | Low to moderate | SMD −0.74 (95% CI −1.00 to −0.48) | Meta-analysis in type 2 diabetes | SMD is not a clinically interpretable unit |
| Soluble fibre | Body mass index (T2D) | Low | SMD −0.31 (95% CI −0.61 to −0.00) | Same meta-analysis | CI touches zero |
Gut and immune outcomes
| Ingredient | Outcome | Evidence | Effect | n | Limitation |
|---|---|---|---|---|---|
| Psyllium / mixed fibre | Stool frequency | Low to moderate | SMD 0.72 (95% CI 0.36 to 1.08); 66% responded vs 41% | 16 RCTs, n=1,251 | Fibres pooled together |
| Psyllium / mixed fibre | Flatulence | Low to moderate | SMD 0.80 (95% CI 0.47 to 1.13) higher than control | Same 16 RCTs | The fibre that helped stool frequency increased gas |
| L. paracasei subsp. paracasei | ≥1 common infectious disease episode | Moderate | OR 0.81 (95% CI 0.66–0.98), p=0.029 | Pooled studies, general population | Odds ratio near 1; no effect on duration or severity |
| L. paracasei subsp. paracasei | Mean episodes per subject | Moderate | −0.09 episodes (95% CI −0.15 to −0.04) | Same | Absolute reduction is small |
| L. paracasei subsp. paracasei | Cumulative disease risk | Low | RR 0.91 (95% CI 0.82–1.01), p=0.082 | Same | CI crosses 1 — not significant |
| L. paracasei DKGF1 | Gas and bloating in IBS | Low | No difference vs placebo, p=0.88 | 67 adults aged 60+ | Signal only in a 16-person subgroup |
| L. paracasei TISTR 2593 | Total cholesterol, HDL, triglycerides | Low | No significant change | 50 adults with high cholesterol | The outcomes most buyers care about did not move |
What the null results teach
Six rows above report an interval that crosses zero or a non-significant p-value. They are included deliberately, because they are the rows that distinguish a reference table from marketing copy.
Three patterns recur:
The headline extracts better than it performs. Psyllium’s LDL effect is real and moderate. Its triglyceride and HDL effects are not. A page that reports “improves cholesterol profile” is describing one of three markers accurately.
Absolute effects are smaller than relative ones. An odds ratio of 0.81 sounds substantial. Expressed per person, it is 0.09 fewer episodes per subject across a season — most people would notice nothing.
Surrogate markers are not outcomes. Apolipoprotein B moved in the right direction while the consumer-facing numbers did not. A trial can succeed on a marker and fail on the thing the buyer cares about.
Limits that apply to every row
Grades follow a published rule, not an impression. Every row carries a grade from A to C, and the criteria behind those letters are written out in full on how we grade evidence. The short version: an interval that crosses zero caps the grade at C, a population that does not match the reader caps it at C, and a trial that does not name its strain, form and dose cannot be applied to a product and is also capped at C. If a grade looks wrong to you, the standard is there to argue against.
Populations were not general. Most glucose trials enrolled people with type 2 diabetes. Most lipid trials enrolled people with elevated cholesterol. A finding in a diagnosed population does not transfer to someone acting preventively.
Pooled data mixes products. Meta-analyses combine different doses, formats, and co-ingredients. The pooled estimate describes the category, not any specific bottle.
Publication bias is not corrected for here. Trials that found nothing are less likely to be published, and pooled estimates from published trials therefore tend to be optimistic.
If a product page quotes a figure from this table without stating the population and the interval, the citation is doing less work than it appears to.
Why this page exists, and what it is not
A reference page solves a specific problem. When one article says psyllium lowered LDL by 0.33 mmol/L and another says 0.35, a reader has no way to tell which figure belongs to which trial design. Collecting the numbers in one place, each with its population and its interval, makes the disagreement visible instead of confusing.
What this page is not: it is not a recommendation, it is not a ranking of ingredients by usefulness, and it is not a substitute for reading the trials. Effect sizes are not comparable across rows. A 0.72 standardised mean difference for stool frequency and a 0.81 odds ratio for infectious-disease episodes are measured on different scales and answer different questions; putting them in one table is a convenience for lookup, not an invitation to rank them.
One further caution about pooling. Every figure here comes from a meta-analysis or a named trial, and meta-analyses differ in how they handle heterogeneity. Two reviews of the same ingredient can report different pooled effects because they weighted studies differently, excluded different trials, or chose different dose ranges. Where we report two figures for the same outcome, it is usually because two reviews exist and neither is wrong.
How we handle a citation request
If you publish these figures, the useful practice is to name the population alongside the number. “Psyllium lowered LDL by 0.33 mmol/L in adults with elevated cholesterol across 28 randomised trials” is checkable. “Psyllium lowers LDL by 0.33 mmol/L” is technically the same number with the conditions removed, and those conditions are what determine whether it applies to a reader.
Download the table
The 21 rows below are available as structured data, so you can check them or reuse them without retyping:
- effect-sizes.csv — spreadsheet-ready, one row per outcome
- effect-sizes.json — same rows plus parsed effect values, confidence bounds, and a flag for intervals that cross zero
Both files carry the same conditions as the table: the limitation column is part of the record, not an aside. If you cite a row, quote its limitation alongside it.
Citing this page
VitalRoundup Editorial Team. (2026). Supplement trial effect sizes: reference data. VitalRoundup. https://vitalroundup.com/blog/supplement-trial-effect-sizes/
Last updated 2 November 2026. Figures are taken from the meta-analyses listed under Sources; where a pooled estimate pooled different doses or formats, that is stated in the limitation column.
The same standard applies to us. Where a figure depended on a subgroup, a co-intervention, or a population that does not match the general reader, that detail appears in the Limitation column rather than in a footnote.
Frequently asked questions
Why report a confidence interval instead of just the average? Because the average hides how sure the finding is. Two trials can both report “−0.35 mmol/L” while one has a range of −0.38 to −0.32 and the other −0.90 to +0.20. The second did not establish an effect. The interval is the part that tells you which.
What does it mean when a confidence interval crosses zero? It means the data are compatible with no effect at all. The common phrasing “showed a trend toward improvement” usually describes exactly this situation, and it is not a finding.
Why are some rows graded “low” despite a large effect? Grade reflects the reliability of the evidence, not the size of the number. Small trials, high heterogeneity between studies, or a population that does not match yours all lower the grade regardless of how impressive the point estimate looks.
Can I use these figures to choose a product? Only if the product publishes per-ingredient amounts. A proprietary blend lists its ingredients and its total weight without individual doses, which makes matching a bottle to any row in this table impossible. The labelling rule behind that gap is set out in what a Supplement Facts panel is required to omit.
How current is this table? Facts were checked on 26 October 2026. New meta-analyses appear regularly, and a pooled estimate can shift when they do. The access date on each source article records when its figures were confirmed.
Which page fits your situation
If you are checking a specific figure, this page is the reference and the source articles carry the citation. If you want to understand why so many labels cannot be matched to any of these rows, read what a Supplement Facts panel is required to omit. If you are comparing finished products rather than ingredients, Gut Health Supplements Compared applies the same reading rules across labels, and the GutSave review works through one bottle that publishes no quantities at all. For the psyllium figures in context, start with Psyllium side effects.
Sources
Each figure traces to a named trial or meta-analysis. The full citation and access date for a given row appear on the article it comes from:
- Psyllium lipid and stool outcomes: Psyllium side effects
- Psyllium dose-response and product comparison: Psyllium vs Metamucil
- Soluble fibre and oat β-glucan: Soluble vs insoluble fibre
- Berberine glucose outcomes: Berberine dosage
- L. paracasei infectious-disease and lipid outcomes: Lactobacillus paracasei
Facts checked: 26 October 2026 · VitalRoundup Editorial Team · Reviewed by VitalRoundup Editorial Team (sources and label data re-checked) · How we review
