Each health claim in the evidence review on our ingredient encyclopaedia entries carries a grade: strong, moderate, limited, insufficient, or mixed. This page explains what the grades mean, what we read before we assign one, who does the work, and — as plainly as we can put it — where the method does not yet reach.
What CEGA is
CEGA — Camden Evidence Grading & Appraisal — is our internal method for weighing the published evidence behind a health claim before we grade it. It is not a certification, and we do not describe it as one: it is the process our editorial team follows, and this page is that process made public rather than kept as an internal document only we can see. It is a self-assessment by Camden Medicals, a retailer that sells some of the ingredients it grades, not an independent audit.
The name covers two acts. Grading is placing a claim on the grade scale below. Appraisal is the structured risk-of-bias assessment of an individual study that a claim rests on — the working that a grade has to be able to show, not just assert.
A grade describes how strong the published evidence is. It is never a statement that a product works, and it is not an endorsement by us or by anyone else.
The five grades
Every graded claim sits at one of five grades:
- Strong — multiple high-quality randomised trials, a positive Cochrane review, or support in NICE guidance.
- Moderate — supportive randomised trials, even where the results are not fully consistent, or guideline support that does not come from NICE.
- Limited — early evidence, or a single positive trial.
- Insufficient — no good-quality human trials. This is not the same as “disproven”: it means the question has not been answered well yet.
- Mixed — good-quality studies conflict strongly with one another: some show benefit, others show none or show harm.
Two ceilings are built in. A single small trial cannot earn more than a moderate grade, and evidence from laboratory or animal studies alone cannot support more than insufficient.
A grade at the low end of the scale is not a defect in the page — it is the page doing its job. We do not hide a limited or insufficient grade, soften it visually, or leave it out. Where the evidence does not support a strong claim, the page says so.
A grade is not a licence to make a claim
What we may lawfully say about a nutrient is a separate question from how strong its evidence is. UK law answers it, through the Great Britain Nutrition and Health Claims Register and the rules on medicines, and we keep the two apart on purpose. A grade can never widen what we are allowed to say, and what we are allowed to say can never raise a grade. Tradition, online popularity, price, and whether we sell the ingredient cannot raise a grade either. Where nothing may lawfully be claimed, the entry sets out the facts and the evidence and makes no claim.
What we read before we grade: authority first
Most of the claims in our encyclopaedia rest on regulators and national guidance rather than on individual studies, and we treat those sources differently from papers:
- Law and regulators (the Great Britain Nutrition and Health Claims Register, the MHRA, the Food Standards Agency) — read in full and quoted precisely, with the version and date recorded. They govern what may be said; we do not grade them.
- UK clinical guidance (NICE, NICE CKS, the BNF, the NHS) — read in full, summarised faithfully in our own words, and cited to the section it comes from. We summarise this guidance; we do not critique it.
- International reviews and public bodies (Cochrane, the World Health Organization, EFSA, the US National Institutes of Health Office of Dietary Supplements) — read, summarised, and dated.
- Individual studies — critically appraised from the full text, using the instruments below. Where a claim rests on studies that no guidance covers, this appraisal is what stands behind the grade.
Where a newer study points beyond what a regulator permits, the regulator’s wording still governs. New evidence can prompt us to lower a grade or revisit an entry; it cannot lift a claim above what the register allows.
How a study is appraised: the instrument routing table
Not every kind of study is checked the same way. A randomised drug trial and an animal study are read for different failure modes, so each is routed to the risk-of-bias instrument built for its design — never a single generic checklist applied everywhere:
| Study design | Instrument used |
|---|---|
| Randomised controlled trial | RoB 2 |
| Non-randomised intervention study | ROBINS-I |
| Observational exposure study | ROBINS-E |
| Diagnostic accuracy study | QUADAS-2 |
| Prediction model | PROBAST |
| Animal / mechanistic / preclinical study | SYRCLE |
| Systematic review | AMSTAR 2 |
| Qualitative research | CERQual |
Two common tools are deliberately not used as the recorded verdict. The Jadad scale is not used as a primary risk-of-bias tool for trials — it is older and coarser than RoB 2, and we do not fall back to it. CASP is used only as an early screening aid, never as the recorded risk-of-bias verdict.
The full-text rule
An appraisal has to be grounded in the paper’s full text. Reading an abstract and recording a verdict from it does not count as an appraisal on this site: it is recorded as a screen, which helps decide what to read next, and a grade of strong or moderate may not rest on it. An abstract can misrepresent, oversimplify, or omit the limitations a full read would surface. Acquiring a citation is not the same act as analysing it, and we do not let the first stand in for the second.
Each full-text appraisal records where and when the paper was retrieved and a fingerprint of the text that was read, so that someone else can obtain the same paper and repeat the reading.
Who does the appraisal, and what is not claimed
Appraisals are drafted under this method by Camden Medicals’ editorial process. Almost all of those drafted so far were produced by AI systems working to the method: routing the right instrument, recording a verdict for each domain that instrument defines, and quoting the specific passage that justifies any verdict lower than “low risk”.
A draft is a first pass. The step that lets an appraisal carry full editorial weight is an independent second reading of the full text by a different rater, with disagreements recorded and an agreement figure reported. That step is being introduced and is not yet operating, which is why the evidence-review panel on each entry says that verification is in progress.
Independence is checked, not assumed: an appraisal is not recorded as independent if the appraiser is, or cannot be shown not to be, an author of the study it appraises.
We publish under an institutional byline and do not name a reviewing clinician. No grade on this site carries a clinician’s sign-off, and nothing here should be read as one.
What an appraisal record contains
Each appraised study gets one record, keyed to its PubMed ID, holding:
- The study design and the question it addresses, which together route the instrument
- A verdict for each risk-of-bias domain that instrument defines
- For every verdict that would justify downgrading the evidence, a verbatim quote from the paper and where in the paper it appears — a verdict without one is treated as a defect, not filled in from memory or inference
- An overall verdict, driven by the weakest domain rather than an average
- Who appraised it, when, and — where known — who wrote the paper, for the independence check above
Known limitations
We would rather this section were shorter. It isn’t yet, and we are not going to write around that.
- Depth of appraisal. Most of the grades in our encyclopaedia rest on regulators and national guidance, which we read and summarise rather than appraise. Where a grade rests on individual studies, only a minority of those studies have so far been appraised from their full text; most of the appraisal records on file were drafted from abstracts and count as screening only. We are re-reading the papers behind the ingredients in products we currently sell first. Until a study has been appraised from its full text, treat the grade that rests on it as provisional.
- No second reading yet. No appraisal has so far had the independent second reading described above, and no grade carries a clinician’s sign-off. Until the second-reading step operates, treat an appraisal as a well-structured first pass, not a finished editorial judgement.
- No agreement figure. Because no study has been appraised independently twice, we cannot yet report an inter-rater agreement figure — a standard reliability check for this kind of work, which we intend to add.
- Claim-level linkage is partial. Ideally every individual claim on a page links to the exact appraisal that supports it. Today that precise linkage exists for a minority of appraisals; the rest link at the level of the ingredient entry rather than the individual sentence.
These limits do not loosen the rules that cap a grade: nothing on this page can raise one, and where a re-read shows that a grade is not supported, we lower it or move it onto a stronger source. What the limits do mean is that the depth of work behind any single grade varies, and the grade alone does not tell you which. We would rather you knew that than assume otherwise.
Corrections and independence
If you find a grade that looks wrong, an appraisal record that misquotes its source, or a study routed to the wrong instrument for its design, please contact us with the page and what you would like us to look at. We mark significant corrections on the affected page.
Camden Medicals sells some of the ingredients graded in the encyclopaedia, and we declare that interest. This method, and this page describing it, are owned by Camden Medicals’ editorial team; no supplier or advertiser has input into a grade, and we do not adjust a grade to favour our own products. See our editorial policy for how that independence is maintained more generally.
This page describes the method as it stands on 8 October 2026, and we revise it when the method changes. It is information, not medical advice. Talk to your pharmacist or GP before starting any supplement, especially if you take prescribed medicines, are pregnant or breastfeeding, or have an existing condition.