Two headlines land on the same morning. One says walnuts lower cholesterol. The other says walnuts may protect the colon. Both come from real studies, but they do not weigh the same: the first pools 26 trials with 1,059 people, and the second comes from mice.
That gap is what this post is about. A good nutrition study fits its design to its question, is big and long enough to see what it claims, reports absolute as well as relative numbers, and says who paid. Even then, one study is a clue, not a verdict.
We will use the five evidence labels on our own pages, then one cautionary tale: beta-carotene, a pigment that looked like a cancer fighter in diet studies and then raised lung cancer in smokers when it came in a pill.
What do the five evidence labels mean?
Every study card on NutriScrolls carries one of five labels. The rules are on our sources page [1]: we weigh the type of study, its size, whether different studies agree, how long it ran and whether a regulator such as the FDA or EFSA approved the claim. If industry paid for a study and the paper says so, the card says so too.
The chart counts every card on our published food, nutrient and phytonutrient pages. The top rung is crowded with very little.
Drawn to scale: each bar is that label’s share of all 6,810 study cards on the 286 published food, nutrient and phytonutrient pages, counted on 25 Oct 2026 [5]. Label meanings: NutriScrolls sources page [1]. Data: NutriScrolls pages.
About 3 in 100 cards are Strong, because Strong asks a lot: pooled randomized trials that agree, or a claim a regulator has approved. Limited is the biggest group, which is honest. Many nutrition questions have been tested once, in a small group, for a few weeks.
To see the ladder in use, take one food. Our walnuts page [2] has a study on every rung. The questions differ, so read each label as a rating of that study, not of walnuts.
Walnut-enriched diets lowered LDL cholesterol by 5.51 mg/dL (3.73%) compared with control diets, with no change in body weight
Meta-analysis of 26 controlled trials with 1,059 people. Total cholesterol fell 6.99 mg/dL and triglycerides 4.69 mg/dL more than on control diets. Weight and blood pressure did not change significantly.
Two years of walnuts did not change cognitive decline in healthy older adults
Randomized trial of 708 adults aged 63 to 79: a diet with 30 to 60 g of walnuts a day against no walnuts. The global cognition score changed by -0.072 with walnuts and -0.086 without (P = 0.491). A post hoc look at the Barcelona site alone found a difference, which the authors call encouraging but inconclusive.
Eating 56 g of walnuts a day for 8 weeks improved blood vessel widening by 1.4%, against 0.3% on the control diet
Randomized crossover trial of 46 overweight adults with a large waist. Each ate a walnut diet and a walnut-free diet for 8 weeks, with 4 weeks between. Blood vessel widening is a marker, not a heart attack. A small, short trial.
Women who ate walnuts 2 or more times a week had a 33% lower risk of type 2 diabetes, or 24% lower after adjusting for body mass index
Two cohorts of 58,063 and 79,893 US women (the Nurses’ Health Studies) followed for 10 years, with 5,930 new cases. Walnut intake came from food questionnaires every 4 years. A link, not proof of cause. One author discloses funding support from the California Walnut Commission.
Male mice on a Western-style diet with 7% walnuts developed 2.3 times fewer colon tumors after a cancer-causing chemical
Mouse study in two experiments. In the first, a diet with 9.4% walnuts cut tumor numbers 1.3-fold. Higher walnut amounts lacked the effect, particularly in female mice. No people were studied.
Same food, five different questions. The big pooled analysis says walnuts lower LDL. The good trial on thinking found nothing, and a null result counts. The mouse study says nothing yet about your colon. A Strong label on cholesterol does not make walnuts a memory food, and a Limited label is a reason to wait for more, not to look away.
Why do diet studies and pill trials disagree?
The case for beta-carotene looked solid on paper. It is the orange pigment in carrots and sweet potatoes, and the body turns it into vitamin A. The opening lines of two big trials say why they were run: studies of eating habits had linked diets rich in carotenoid-containing fruits and vegetables, and high blood levels of beta-carotene, to lower risks of lung cancer (ATBC Study Group, 1994) and of cancer and heart disease (Hennekens et al., 1996).
A 2008 review pooled 25 prospective observational studies and six pill trials on lung cancer (Gallicchio et al.). People who ate the most carotenoids in total had a 21% lower risk than people who ate the least (relative risk 0.79). For beta-carotene eaten on its own the figure was 0.92, with a range of 0.83 to 1.01 that cannot rule out no difference. The six pill trials pooled to 1.10, also unclear. Then add the two biggest trials.
Drawn to scale. Dot = estimate, line = 95% range; left of the dashed mark (1.0) is less lung cancer. Diet rows: pooled prospective studies, highest vs lowest intake (Gallicchio et al., 2008). Pill rows: six trials pooled (same review); ATBC and CARET are the trials’ reported changes (18% = 1.18), our conversion.
Both trials were randomized, double-blind and placebo-controlled, in people at high risk of lung cancer. Even a single large trial like these is Moderate on our beta-carotene page, because Strong asks for pooled trials that agree.
20 mg of beta-carotene a day for 5 to 8 years raised lung cancer by 18% in Finnish male smokers
Randomized, double-blind trial of 29,133 male smokers aged 50 to 69, with vitamin E tested alongside. Among 876 new lung cancers, the beta-carotene group had 18% more (range 3% to 36%). Total deaths were 8% higher (range 1% to 16%), mostly from lung cancer and heart disease.
30 mg of beta-carotene plus vitamin A a day raised lung cancer by 28% in smokers and asbestos workers, and the trial stopped early
Randomized, double-blind trial of 18,314 smokers, former smokers and asbestos-exposed workers. Deaths from lung cancer were 46% higher (relative risk 1.46, range 1.07 to 2.00) and deaths from any cause 17% higher. The trial was stopped 21 months earlier than planned.
Not every trial found harm. In the Physicians’ Health Study, 22,071 male doctors took 50 mg of beta-carotene every other day for 12 years, and cancer, heart disease and death rates matched the placebo group: 1,273 cancers against 1,293 (relative risk 0.98), with 11% of the men current smokers at the start (Hennekens et al., 1996). Across these three trials, the pill did not lower cancer or heart disease.
Dose is part of the story, though not all of it. Against food, a 20 mg pill holds as much beta-carotene as 1.4 medium sweet potatoes or about four medium carrots (our arithmetic on USDA values [4]). Pills and plates differ in how the body absorbs the pigment, and our beta-carotene page [3] has the numbers.
Drawn to scale. Pill doses: ATBC and CARET abstracts. Food: USDA FoodData Central via NutriScrolls, per household serving [4]. Pills and food are absorbed differently, so equal milligrams are not equal effects.
Now the twist. Years later, researchers went back to the ATBC men and looked at beta-carotene in their blood, not in a pill.
Men in the top fifth for blood beta-carotene had a 36% lower risk of dying over 31 years than men in the bottom fifth
Analysis of 29,103 men from the ATBC cohort, with 23,796 deaths in 31 years. Hazard ratios for death from any cause were 0.81, 0.71, 0.69 and 0.64 for the second to fifth fifths against the first. Adjusting for major risk factors, and for fruit and vegetable intake, did not remove the link. An association, not proof of cause.
Same men, two designs, opposite pictures: a randomized pill raised lung cancer, while a high blood level tracked longer life. Both can be true because they answer different questions. A blood level mostly reflects what a person eats and how they live. The authors of the 2008 review warned that carotenoid measures may be a marker of a healthier lifestyle, or of leftover confounding by smoking. Randomization breaks that tie: chance, not habits, decides who gets the pill.
None of this says carrots are risky. EFSA’s 2024 opinion found no indication that beta-carotene from the background diet is linked to adverse health effects, and it advised smokers to avoid supplements containing it (EFSA NDA Panel, 2024). If you smoke or have worked with asbestos, ask your doctor before taking any beta-carotene pill. Diet studies pointed researchers to the right question, and only the trials could answer it.
Does a 49% lower risk mean what it sounds like?
Headlines like relative numbers because they sound big. Whether they are big depends on how much risk there was to cut, and that is the number the headline tends to leave out. Here is a trial where beta-carotene did good.
Weekly beta-carotene at a nutritional dose cut pregnancy-related deaths by 49% among women in rural Nepal
Double-blind trial in 270 wards covering 44,646 married women and 22,189 pregnancies. Women took a weekly placebo, vitamin A or 42 mg of beta-carotene for over 3.5 years. Deaths per 100,000 pregnancies were 704 on placebo and 361 on beta-carotene (relative risk 0.51, range 0.30 to 0.86).
Drawn to scale. Top bars: deaths per 100,000 pregnancies. Lower bars: the same deaths as a share of all 100,000 pregnancies, shown as thin slivers. Data: West et al., 1999; the gap is our arithmetic.
Both views are true. The relative drop is 49%. The absolute drop is 343 deaths per 100,000 pregnancies, or 0.34 percentage points, about one death avoided for every 290 pregnancies (our arithmetic). In a rural, undernourished population where 704 in 100,000 pregnancies ended in death, that is a real gain. For someone starting from a much lower risk, the same relative cut would be a much smaller absolute one.
The trick works in reverse. In CARET, 388 of the 18,314 people (2.1%) were diagnosed with lung cancer during follow-up, both groups together. When a headline says “28% higher risk”, ask: 28% of what?
Look at the range too. ATBC reported 18% more lung cancer, with a range of 3% to 36%, so the data fit a small rise or a large one. A relative-risk range that crosses 1.0, like the 0.83 to 1.01 for beta-carotene in the diet, cannot rule out no effect at all.
How big, how long, and in whom?
Size and length decide what a study can see. ATBC followed 29,133 men for five to eight years and counted 876 lung cancers, enough events to see the 18% rise. A trial of 46 people for 8 weeks can only catch quick changes in a marker, such as how far an artery widens, not heart attacks.
Drawn to scale; the 46-person bar is less than one pixel wide and is drawn as a hairline. Durations are in the labels. Data: abstracts of ATBC, Hennekens, CARET, Guasch-Ferré, WAHA and Katz.
A marker is not an outcome. The walnut analysis measured LDL, a risk factor, and the crossover trial measured artery widening. Neither counted heart attacks. That does not make them useless, but it is why our cards say what was measured.
Then ask who took part. The same pigment gave three kinds of answer: harm in male smokers (ATBC) and in smokers and asbestos workers (CARET), no effect in physicians, 11% of them smokers (Hennekens), and 49% fewer pregnancy-related deaths among undernourished women in Nepal, whose authors limit their conclusion to rural, undernourished populations in south Asia. A result is a result about the people studied.
Who paid for the study?
Money does not make a study wrong, and a funder’s silence does not make it right. It is a reason to look closer. The walnut and diabetes card above carries a disclosure in PubMed: one author has received funding support from the California Walnut Commission (Pan et al., 2013). That does not decide whether the result is true. It tells you to check whether independent studies point the same way.
Studies of studies show why checking pays off. A review of 17 systematic reviews on sugary drinks and weight gain found that among those disclosing a financial tie to the food industry, 5 of 6 conclusions said the evidence was insufficient to link the drinks to weight gain; among reviews without ties, 10 of 12 said the drinks could be a risk factor (Bes-Rastrollo et al., 2013). Reviews with ties were five times as likely to reach the no-link conclusion (relative risk 5.0, range 1.3 to 19.3). It is a small sample, and the authors say they cannot rule out publication bias.
Our cards name a funder or an author’s ties when the paper states them. For example, our protein maxxing post notes a dairy council link on one meta-analysis card. Look for the same line in anything you read.
Which questions should you ask of the next headline?
The whole post fits in seven red flags. Each one comes with the question that tests it.
Seven red flags and the question that tests each. Not every flag means a study is wrong; each is a reason to look closer.
And the same seven as questions, in the order you would ask them:
- What kind of study is it? A randomized trial can show cause. A cohort study shows a link, and cells or mice show a possibility. Look for the label.
- Which number are you being shown? Ask for the absolute figures and the range, not only the percentage.
- How many people, for how long? Small and short studies catch markers, not outcomes.
- Who took part? Smokers, rural mothers and doctors are not the same audience as you.
- What was measured? LDL and artery widening are markers. Deaths and diagnoses are outcomes.
- Who paid, and who wrote it? A funder is a reason to look for independent studies, not a verdict.
- Do other studies agree? One study is a clue. Quality matters inside a pooled analysis too: in a Cochrane review of 78 antioxidant trials, deaths barely changed overall (relative risk 1.02, range 0.98 to 1.05), but the 56 trials at low risk of bias found 4% more deaths (1.04, range 1.01 to 1.07) (Bjelakovic et al., 2012).
You will not need all seven every time. Two or three, asked out loud, sort most headlines into the ones worth a second look and the ones that can wait for the next study.
Frequently asked questions
What is the best kind of nutrition study?
A meta-analysis of randomized trials that agree is the strongest design, because randomization sorts out habits and pooling smooths out chance. Even so, check what was measured, who took part and who paid. Our Strong label asks for such a meta-analysis or an approved regulatory claim, and only about 3 in 100 of our study cards reach it.
Why do nutrition studies seem to contradict each other?
They often ask different questions of different people. Studies of eating habits linked more beta-carotene to less lung cancer, while randomized pill trials in smokers found 18% and 28% more. Doses, forms, groups and length differ too. Contradictions usually shrink once you compare design, size and who was studied.
What is the difference between relative and absolute risk?
Relative risk compares two groups, such as a 49% lower risk of death. Absolute risk gives the actual rates: in one trial, pregnancy-related deaths fell from 704 to 361 per 100,000 pregnancies, a gap of 0.34 percentage points. Ask for both, plus the range, because a big relative change on a small base can be a small absolute one.
Does industry funding mean a study is wrong?
No, it means check more. A review of reviews on sugary drinks found that those disclosing food industry ties were five times as likely to conclude there was no positive link (relative risk 5.0, range 1.3 to 19.3), though the sample was small. Look for disclosures, and for independent studies that agree.
Can I trust a study in mice?
Treat it as a lead, not an answer. In one walnut study, male mice on a Western-style diet with 7% walnuts developed 2.3 times fewer colon tumors, yet higher walnut amounts lacked that effect, and no people were studied. Mouse results tell researchers where to look. Trials in people tell you whether it works for you.
Sources
- NutriScrolls. Sources: how we rate evidence (the five evidence labels). nutriscrolls.com/sources/#how-we-rate-evidence. Accessed 24 Oct 2026.
- NutriScrolls. Walnuts: study cards and evidence labels. nutriscrolls.com/food/nuts-and-seeds/walnuts/. Accessed 24 Oct 2026.
- NutriScrolls. Beta-carotene: study cards, foods and safety. nutriscrolls.com/phytonutrients/carotenoids/beta-carotene/. Accessed 24 Oct 2026.
- U.S. Department of Agriculture, Agricultural Research Service. FoodData Central, entries used on NutriScrolls: sweet potato (FDC 168484), carrot (FDC 170393), cantaloupe (FDC 169092), spinach (FDC 168462). fdc.nal.usda.gov. Accessed 24 Oct 2026.
- NutriScrolls. Study cards counted by evidence label across published food, nutrient and phytonutrient pages, 25 Oct 2026 (site database).
- Guasch-Ferré M, Li J, Hu FB, Salas-Salvadó J, Tobias DK. Effects of walnut consumption on blood lipids and other cardiovascular risk factors: an updated meta-analysis and systematic review of controlled trials. Am J Clin Nutr. 2018;108(1):174-187. doi:10.1093/ajcn/nqy091
- Sala-Vila A, Valls-Pedret C, Rajaram S, et al. Effect of a 2-year diet intervention with walnuts on cognitive decline. The Walnuts And Healthy Aging (WAHA) study: a randomized controlled trial. Am J Clin Nutr. 2020;111(3):590-600. doi:10.1093/ajcn/nqz328
- Katz DL, Davidhi A, Ma Y, Kavak Y, Bifulco L, Njike VY. Effects of walnuts on endothelial function in overweight adults with visceral obesity: a randomized, controlled, crossover trial. J Am Coll Nutr. 2012;31(6):415-423. doi:10.1080/07315724.2012.10720468
- Pan A, Sun Q, Manson JE, Willett WC, Hu FB. Walnut consumption is associated with lower risk of type 2 diabetes in women. J Nutr. 2013;143(4):512-518. doi:10.3945/jn.112.172171
- Nakanishi M, Chen Y, Qendro V, et al. Effects of walnut consumption on colon carcinogenesis and microbial community structure. Cancer Prev Res (Phila). 2016;9(8):692-703. doi:10.1158/1940-6207.CAPR-16-0026
- The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group. The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med. 1994;330(15):1029-1035. doi:10.1056/NEJM199404143301501
- Omenn GS, Goodman GE, Thornquist MD, et al. Effects of a combination of beta carotene and vitamin A on lung cancer and cardiovascular disease. N Engl J Med. 1996;334(18):1150-1155. doi:10.1056/NEJM199605023341802
- Hennekens CH, Buring JE, Manson JE, et al. Lack of effect of long-term supplementation with beta carotene on the incidence of malignant neoplasms and cardiovascular disease. N Engl J Med. 1996;334(18):1145-1149. doi:10.1056/NEJM199605023341801
- Gallicchio L, Boyd K, Matanoski G, et al. Carotenoids and the risk of developing lung cancer: a systematic review. Am J Clin Nutr. 2008;88(2):372-383. doi:10.1093/ajcn/88.2.372
- Huang J, Weinstein SJ, Yu K, Männistö S, Albanes D. Serum beta carotene and overall and cause-specific mortality. Circ Res. 2018;123(12):1339-1349. doi:10.1161/CIRCRESAHA.118.313409
- EFSA Panel on Nutrition, Novel Foods and Food Allergens (NDA). Scientific opinion on the tolerable upper intake level for preformed vitamin A and β-carotene. EFSA J. 2024;22(6):e8814. doi:10.2903/j.efsa.2024.8814
- West KP Jr, Katz J, Khatry SK, et al. Double blind, cluster randomised trial of low dose supplementation with vitamin A or beta carotene on mortality related to pregnancy in Nepal. BMJ. 1999;318(7183):570-575. doi:10.1136/bmj.318.7183.570
- Bes-Rastrollo M, Schulze MB, Ruiz-Canela M, Martinez-Gonzalez MA. Financial conflicts of interest and reporting bias regarding the association between sugar-sweetened beverages and weight gain: a systematic review of systematic reviews. PLoS Med. 2013;10(12):e1001578. doi:10.1371/journal.pmed.1001578
- Bjelakovic G, Nikolova D, Gluud LL, Simonetti RG, Gluud C. Antioxidant supplements for prevention of mortality in healthy participants and patients with various diseases. Cochrane Database Syst Rev. 2012;(3):CD007176. doi:10.1002/14651858.CD007176.pub2
Not medical advice
This article is general information, not a diagnosis or treatment. Talk to a doctor or registered dietitian before changing your diet, especially if you are pregnant, take medication or have a health condition. Read our medical disclaimer.
