Evidence, read properly

Cells, Mice And People: Three Kinds Of Evidence On One Eye Label

A single supplement label can carry ingredients whose evidence ranges from large randomised human trials down to a study of cells in a dish, and the front of the bottle gives no hint which is which. The marketing language does not help either: a phrase like "shown to support retinal health" is equally available at every level. Here is how to tell them apart in about a minute an ingredient.

By TheyaVue Team · Last updated September 19, 2026 · 9 min read · Evidence, read properly · Reviewed by Dr. Jarrod Long, OD

The ladder, from a dish to a trial

Evidence in nutrition comes in levels, and the levels are not a matter of opinion. Each one answers a narrower question than the one above it and a more reliable one than the one below.

The reason this matters on a supplement label is that the language used to describe all five levels is identical. Studied, shown, supports, linked to, clinically researched: none of those phrases tells you which rung of the ladder you are on, and all of them are technically defensible at every rung.

So the phrase is useless and the underlying study is not. Five minutes with the reference list at the foot of a page will usually tell you more than an hour of reading the page itself.

Five levels and five phrases. Only the last one supports a benefit claim, and it is the level at which marketing language stops getting more impressive.

Cells in a dish

The bottom rung, and the one that produces the most impressive-sounding language relative to what it supports.

Quercetin is the example on the label this desk sells. A 2021 study applied quercetin to retinal pigment epithelium cells under oxidative stress and to a mouse model, and found the outer retinal barrier protected, apparently through increased autophagy. That is a coherent, interesting piece of work.

What it cannot tell you is whether swallowing quercetin does anything. A compound dropped directly onto cultured cells at a chosen concentration has bypassed every obstacle between a mouth and a retina: digestion, absorption, first-pass metabolism, protein binding, the blood-retinal barrier and the small matter of whether enough of it arrives.

Most compounds that work beautifully in a dish do nothing in a person, and that is not a scandal. It is the reason the other four rungs of the ladder exist.

Animals, and the special case of depletion

Animal work is one rung up and it answers a different question: does this matter in a whole organism?

Taurine is the interesting case here, because its animal evidence takes an unusual form. A review of the retinal case for taurine sets out what is known: taurine is one of the most abundant free amino acids in the retina, depleting it causes photoreceptor degeneration in animals, and people with mutations in the taurine transporter develop early retinal degeneration.

That is a strong case for taurine being necessary. It is not a case for taurine being useful in addition. The two are routinely confused and the confusion is worth naming, because it applies to a large share of the nutrients on any label.

Humans, wrong outcome

A rung that looks better than it is. The study is in people, it may even be randomised, and it measures something other than the thing being claimed.

Rutin is the example. A human supplementation study measured rutin's bioavailability and its antioxidant efficiency after ingestion, which is a genuinely useful piece of pharmacokinetics. It tells you the compound is absorbed and that some antioxidant activity survives the trip.

It does not tell you anything about an eye, because it did not look at one. A marketing page describing rutin as clinically studied would be accurate and completely uninformative.

Observational studies in people

Higher again, and the rung where correlation and causation get confused most often.

Lycopene is the case. A serum study found lycopene lower in people with macular degeneration while lutein and zeaxanthin were not different between the groups.

There are at least three explanations for a finding like that and only one of them is the one marketing reaches for. Low lycopene might contribute to the disease. The disease, or the lifestyle that accompanies it, might lower lycopene. Or lycopene might be a marker for a dietary pattern that is doing the work through something else entirely.

Observational data is how hypotheses get generated, and in nutrition it is often all there is. It is not a reason to discount an ingredient. It is a reason not to promote it past what the data says.

Randomised trials with the right outcome

The top rung, and the one that supports a claim.

Two ingredients on this label reach it. The year-long contrast-sensitivity trial randomised 59 healthy young adults to placebo or one of two amounts of macular carotenoid and measured contrast sensitivity and lateral inhibition at six and twelve months. Both improved, and the size of the improvement tracked the size of the pigment change.

The grape seed trial randomised 124 people with non-proliferative diabetic retinopathy to placebo, a licensed comparator or 150 mg a day of proanthocyanidin extract for up to twelve months, and found hard exudate severity improved more on the extract than on either of the others.

Both are in people, both are randomised, both have ocular outcomes and both name an amount. That last point is what makes them usable: a trial amount you can set against a label figure is the whole basis of comparison in this category.

Six ingredients from one label, ranked

Six ingredients from one blend, sorted by the strongest evidence that exists for each in an eye. The spread is four levels wide.

Four levels of evidence on one blend line, presented on a marketing tile as six equivalent names. That is not deceptive in any legal sense and it is exactly what the format produces.

It is also why a reference list at the foot of a page is worth more than the page above it. Five of those six rows can be established in ten minutes by anybody with the citations, and none of them can be established from the front of a bottle.

The one-minute version

1. Find the citation. If a page names an ingredient and cites nothing, that is the answer.
2. Read the title. It usually names the species, the design and the outcome.
3. Ask: in what? Cells, mice or people.
4. Ask: measuring what? A visual outcome, a biomarker, or something unrelated.
5. Ask: randomised? If not, expectation is in the result.
6. Ask: how much? And compare that with the label. No amount on the label means no comparison, and no comparison means the trial is not about the product.
7. Then decide how much weight the ingredient gets. Usually less than the front of the bottle implies, and occasionally more.

The bottom line

A label is a list of names, and names are equal on a label in a way they are not in the literature. On one ordinary eye formula the evidence spread across six ingredients runs from a year-long randomised trial with a visual outcome down to a study of cells in a dish.

Knowing which is which does not require a science degree. It requires the citation, the title, and five questions that take a minute each.

A page that publishes those citations is making itself checkable, which is the only form of credibility available to anybody who is paid when you buy something. A page that does not is asking to be trusted, and there is no reason to.

References