Confidence is earned.
Here’s how we read the evidence.
Two questions belong beside every claim: what did the research find, and how dependable is that answer?
They summarize selected published trials and review conclusions. They are not formal GRADE assessments, medical recommendations, percentages of effectiveness, or independently clinically reviewed scores. We have not conducted an exhaustive systematic review of every condition.
Updated September 24, 2026. Newly found abstracts enter the map without a numeric score until there is an adequate basis for interpretation.
The four-point confidence scale
High
Direct, substantial and consistent evidence supports the narrowly stated finding. Important limits still apply.
Moderate
The answer has a solid basis, but applicability, uncertainty or conflicting details limit confidence.
Low
A tentative answer. Small studies, inconsistent results or indirect measures could change the conclusion.
Very low
Too much uncertainty for a dependable estimate. A positive statistical signal can still have very low certainty.
Not yet scored is a separate status, not zero out of four. It can mean that a new paper awaits full appraisal, that only a record or abstract was available, or that relevant evidence is too limited to characterize. Read the page’s explanation.
What earns—or lowers—confidence?
01 / Bias: could the study design distort the answer?
We look for random allocation, adequate blinding where feasible, balanced follow-up, prespecified outcomes and transparent reporting. Funding is disclosed where available; it is a reason to inspect methods, not automatically dismiss results.
02 / Consistency: do credible studies tell a similar story?
Conflicting findings reduce confidence unless differences in formulation, population or other methods plausibly explain them. A small p-value in one subgroup does not prove that subgroup responds differently; an interaction test is needed.
03 / Directness: does the evidence answer this exact question?
We distinguish EPA from DHA, prescription drugs from retail supplements, food from capsules, and prevention from treatment. A blood marker or animal experiment is less direct than a patient-important clinical outcome.
04 / Precision: what effects remain compatible with the data?
We examine effect estimates and confidence intervals, not just whether p is below 0.05. Small samples or few events often leave room for both meaningful benefit and no benefit.
05 / Missing evidence: are we seeing the whole picture?
Small-study effects, unpublished null findings, selective reporting, multiple testing and overlapping reviews can make an effect look more certain than it is. Ten reviews may contain the same underlying trials.
These domains draw on established evidence-appraisal principles described in the Cochrane Handbook. Atlas scores are not mechanically calculated by adding domain points and are not presented as Cochrane or GRADE endorsements.
How a page gets its rating
- Define the population, intervention, comparator and exact outcome.
- Identify a relevant synthesis or large clinical trial and check what its published results actually support.
- Consider the five domains above; preserve a review’s reported certainty where it directly applies, and explain any Atlas adjustment.
- Publish the conclusion separately from confidence, with the sources, appraisal depth and reasons for the rating.
- Leave the page unscored when its source basis is inadequate. Reappraise when important new evidence changes the question.
What a score does not tell you
A 4/4 score is not a 100% chance of benefit. It is not a comparison of how well omega-3 works across diseases. Moderate confidence in no clear benefit can be more informative than low confidence in a promising effect.
A finding has a direction, too.
Benefit in a defined setting: a favorable finding within the named intervention and population. Possible benefit: promising, but unresolved. Mixed findings: results differ or depend on the outcome. No clear benefit demonstrated: studies do not establish benefit; read the interval and scope. Insufficient evidence: too little dependable information. Safety signal: a possible harm. Research area: investigated, but not yet appraised here.
Where the Atlas is today
This release expands the map to 49 questions. It uses a targeted search of PubMed-indexed records through Europe PMC, Cochrane summaries, selected landmark trials and professional sources, checked on September 24, 2026. It is not a preregistered systematic search. Page dates indicate editorial updates, not that every source is new.
Independent clinical review remains pending. New research discovery does not automatically change a score. A future review should include full-text risk-of-bias assessment, trial-registry checks, source reconciliation and a named qualified reviewer.