← All chapters·Chapter 00

Who this is not for

Exclusion criteria, contraindications and stop signals — stated first, because for some readers the correct outcome of this text is to close it.

Most publications in this category put the disclaimer in six-point type at the bottom. This one opens with it, because the exclusions are load-bearing: a publication that applies to everyone is a publication that has not been checked against anything.

The population the evidence was measured in

A The synthesis this text is built on covers adult populations with at least one chronic condition. That is where the trials were run (Reyneke 2026, PMID 40657695; 30 reviews, 225 primary studies, 15 dietary patterns).

The precise statement matters more than the convenient one:

Not tested is not the same as disproved. There are very few trials in adults with no chronic condition, so no effect can be claimed for them either way.

There is also a mechanistic reason to expect a smaller effect where baseline markers are normal: if CRP already sits at the low end, there is little room for it to fall. The same direction shows up inside the data — in one subgroup analysis a vegetarian pattern reached significance only in patients with impaired kidney function (MD −3.91 mg/mL, 95% CI −5.23 … −2.60; via PMID 40657695).

The ten exclusions

This publication is not written for people who:

1. have a current or past eating disorder, or recognise disordered eating patterns in themselves. C Adults following vegetarian or vegan diets report orthorexia symptoms at roughly twice the rate of others (OR 1.99, 95% CI 1.21–3.25; Cohen d = 0.46; 26 cross-sectional studies, n = 23,783; Díaz-Goñi 2026, PMID 41230622). The authors flag high heterogeneity and low methodological quality, and the direction of causation is unresolved. The grade stays at C for that reason, and the exclusion stands anyway, because the cost of the two possible errors is not symmetric.
2. are pregnant or breastfeeding. No grade. No reviewed trial supports restricting energy or food groups in pregnancy, and the absence of evidence for safety is the whole basis of this line. It is a conservative position, marked as one.
3. have chronic kidney disease. Potassium, phosphate and protein need individual management. A generic instruction to eat more legumes, nuts and greens can be directly harmful here.
4. take insulin or sulfonylureas. Lowering glycaemic load measurably changes glucose handling in trials (Wu 2026, PMID 42370356), which is a hypoglycaemia risk if medication doses are not adjusted alongside.
5. take anticoagulant or immunosuppressive therapy and intend to add supplements. The reason is unproven safety, not proven harm. Those are different statements and this text keeps them apart.
6. are under 18. The evidence base is adult.
7. have an acute or undiagnosed illness — fever, unexplained weight loss, blood in stool, night sweats, joint swelling with redness and heat. Those are signs of an acute process, which is a different thing from chronic low-grade inflammation and not this text's subject.
8. want a weight-loss programme. This is not one, and see the first exclusion.
9. want a cure or a promise for a diagnosed disease. Claims that a food treats, prevents or reduces the risk of disease are prohibited for foods in the EU (Reg. 1924/2006), and the evidence would not support them here in any case.
10. have no chronic condition and expect a measurable result. See the section above. This is the exclusion readers find most surprising, and it is the one with the clearest citation behind it.

When this stops being the right tool

Presented as a list of moments to stop reading and talk to a clinician, not as a diagnostic instrument. Recognising the difference is the point.

A A hs-CRP result above 10 mg/L is not "very high risk". The AHA/CDC statement treats it as a probable acute event: repeat after about two weeks and look for an acute cause first (Pearson 2003, PMID 12551878).

Other stop signals: unexplained weight loss; fever or night sweats; joint swelling with redness and heat; blood in stool, vomiting or difficulty swallowing; fatigue that is getting worse rather than better; and — from the first exclusion — thoughts about food becoming intrusive, guilt attached to eating, or social avoidance because of food.

Supplement safety, where it has actually been checked

A Pooled data from 11 randomised trials (n = 120,643) found no increase in overall bleeding risk with omega-3 supplementation (RR 1.09, 95% CI 0.91–1.31, p = 0.34; Javaid 2024, PMID 38742535). This contradicts the warning almost every nutrition guide repeats.

A The risk that does exist tracks dose, not drug combination: high-dose purified EPA showed roughly a 50% relative increase and a 0.6% absolute increase in bleeding events (PMID 38742535). Risk was not significantly associated with background antiplatelet use (p = 0.056).

That is not a clearance to combine supplements with anticoagulants. It is a correction to a specific, widely repeated claim. Adding any supplement while on anticoagulant therapy is a conversation with a prescriber.

Curcumin: the one thing this text confirms, and why it is still not a recommendation

Chapter 5 grades bioavailable curcumin supplements at B — they lower CRP in 7 of 10 meta-analyses (PMID 38892660). That makes curcumin one of very few places where this publication says something worked. Which is exactly why its safety section had to be the most exact section in the book.

A The US National Institutes of Health reference database on drug-induced liver injury assigns turmeric its highest likelihood score, A: a well-documented cause of clinically apparent liver injury. Its own summary is that turmeric "appears to have become the most common cause of clinically apparent, herbal-related liver injury in the United States" (LiverTox, NIDDK/NIH, updated June 2025; Bookshelf NBK548561, PMID 31643876).

The clinical shape of that injury is consistent across every series. Latency is typically one to four months. Onset is insidious — fatigue, nausea, poor appetite — before dark urine and jaundice. ALT is often above 1,000 U/L. Most people recover within one to three months of stopping. LiverTox estimates incidence at 1 in 10,000 to 1 in 100,000 exposed, and estimates that fatal cases may occur in around 10% of those who reach jaundice, "particularly if the product is not discontinued promptly."

C In the ten adjudicated cases collected by the US Drug-Induced Liver Injury Network, five people were hospitalised and one died of acute liver failure. Seven of the ten carried the HLA-B*35:01 allele, two of them homozygous — an allele frequency of 0.450 against population controls of 0.056–0.069 (Halegoua-DeMarzio 2023, PMID 36252717). Independent regulators report the same pattern: Health Canada reviewed more than 60 international cases including three deaths, two attributed to turmeric; Australia's TGA logged 18 reports including one death; France's nutrivigilance programme recorded over 100 adverse events including 15 hepatitis cases (via PMID 42364655 and PMID 40773013).

The part that decides the question

The liver signal is concentrated in the same formulations that produce the B-grade result. The Italian pharmacovigilance analysis states it plainly: "in all cases, hepatotoxicity was associated with Curcuma longa formulations with high bioavailability and high dosage of curcumin/curcuminoids" (Lombardi 2021, PMID 32656820). Three of the seven DILIN products that were chemically assayed also contained piperine. Germany's federal risk institute reports hepatotoxicity with turmeric–piperine combinations even at doses below the regulatory intake limit (BfR opinion 040/2021, via PMID 40773013).

Plain turmeric in food carries a D grade for effect and is barely absorbed. The supplements that clear the bioavailability problem are the supplements that carry the liver signal. The benefit and the harm come out of the same bottle.

Two further findings close off the obvious workarounds. First, there is no safe dose line to stay under: the injury is idiosyncratic rather than dose-dependent, documented cases span 30 to 3,000 mg per day, and a review of 20 trials found abnormal liver tests in 5% of 526 supplement users with no dose relationship, all of it in studies running longer than a month (via PMID 42364655). Second, self-monitoring does not work: in an eleven-case pathology series, six of the eleven were asymptomatic and were found only because someone ordered blood tests (Papke 2025, PMID 39381846). "Stop if it feels wrong" is not a safety net here.

For completeness, the strongest argument in the other direction: the US Pharmacopeia's expert committee reviewed the same literature in 2026 and concluded that trials have not reported organ toxicity, that published cases frequently involve other drugs and supplements, and that the correct response is a cautionary label rather than removal (PMID 42364655). That reading is not unreasonable, and it is not decisive, because trials are too short and too small to detect a 1-in-10,000 event — an umbrella review of curcumin in osteoarthritis found "incomplete safety reporting across reviews" in its own included papers (PMID 42254374). Trials and pharmacovigilance are measuring different things.

Doses, for readers who want the numbers

The FAO/WHO expert committee set an acceptable daily intake for curcumin of 0–3 mg per kg of body weight — about 210 mg a day for a 70 kg adult, covering food and supplements together. That figure explicitly does not cover preparations containing an absorption enhancer such as piperine, a phospholipid complex or a micelle, and no equivalent limit exists for those (JECFA 2004, via PMID 40773013 and PMID 42364655).

The doses that moved inflammatory markers in trials were 1,000–3,000 mg a day — five to fifteen times that intake limit. Of 125 supplements surveyed across five countries, 36 (28.8%) recommended a dose above it, and 43 (34.4%) did not state their curcuminoid content at all (PMID 40773013).

Anticoagulants, and the difference from the omega-3 case

D What is actually documented in people is two spontaneous reports. A patient on warfarin with previously stable readings whose INR rose above 10 within weeks of starting a turmeric product (Medsafe New Zealand safety communication, 2018). A woman on the vitamin K antagonist fluindione whose INR went from 2–3 to 6.5 on turmeric infusions, resolving on withdrawal (Daveluy 2014, PMID 25230280). Everything else repeated in nutrition guides is extrapolation from platelet and clotting assays in glassware.

The one human study that tested a bioavailable curcumin formulation directly against warfarin, dabigatran and antiplatelet drugs found no change in INR or bleeding time over ten days (PMID 30070343) — but it reports no sample size, has no control group, tested one proprietary product, and an underpowered negative result is not evidence of absence.

That produces a conclusion that has to be stated carefully, because it is not the same as the omega-3 conclusion four paragraphs above. There, pooled data from 120,643 randomised patients actively refute the standard warning. Here, nobody has run that trial. The warning is unproven and also unrefuted, and those are different states.

Note too that the New Zealand regulator drew the same line this text draws elsewhere: "this warning does not apply to turmeric used in food."

Other interactions worth naming

B Tamoxifen. In breast cancer patients, curcumin with piperine cut exposure to the active metabolite endoxifen by 12.4% (95% CI −21.9 to −1.9, p = 0.02); the authors estimate this could push 20–40% of patients below the concentration threshold for effect (Hussaarts 2019, PMID 30909366). The direction is the surprising part — the drug gets weaker, not more dangerous.

B CYP enzymes, in a test tube versus in a person. Curcuminoids are potent inhibitors of several drug-metabolising enzymes at micromolar concentrations in vitro. In a randomised crossover trial, 4 g of curcuminoids plus 24 mg of piperine over two days produced no meaningful change in the handling of midazolam, flurbiprofen or paracetamol — because unconjugated curcuminoid concentrations in plasma stayed below the assay's detection limit (Volak 2013, PMID 22725836). The blanket warning that curcumin interferes with medication is stronger than the human data support. The interactions that do warrant care involve drugs with a narrow therapeutic window.

D Transplant medication, chemotherapy, iron. Single case reports exist for each: acute calcineurin-inhibitor toxicity in a liver transplant recipient consuming roughly 75 g of turmeric daily (PMID 28104136); one heavily confounded case of paclitaxel toxicity in a patient also taking a cyanotoxin-contaminated supplement (PMID 29857089); one case of iron deficiency anaemia resolving two weeks after stopping high-dose turmeric (PMID 30899609). Each is one case. The text says so rather than converting them into rules.

Where this text stops

B Curcumin should be avoided by people with any liver disease or history of liver problems; with gallstones or biliary obstruction — 40 mg of curcumin contracts the gallbladder by half within two hours (PMID 12495265, PMID 10102956), and the French agency advises against these supplements in bile duct disease; on vitamin K antagonists; on tamoxifen; on calcineurin inhibitors; and in pregnancy or breastfeeding, where there is no data at all.

On stopping before surgery: that advice is widely repeated and no clinical data supporting it was found. A surgeon should know about every supplement being taken regardless, which is the sound version of the same instruction.

And the conclusion this chapter exists to state:

A B grade describes what an intervention did to a marker in trials. It is not a recommendation to take it. For bioavailable curcumin those two things come apart, because the benefit is a surrogate marker and the harm is a hospital admission that cannot be predicted, cannot be screened for in advance, and in more than half of observed cases produced no symptoms to notice.

This publication grades curcumin at B and does not recommend it. Both of those statements are the output of the same method, and printing them next to each other is the point of the method.

Why the exclusions come first

A reader who is filtered out on page one has been told something true and useful at no cost. A reader who is not filtered out now knows the boundary the rest of the text operates inside. Neither of those is a legal formality.

The exclusions also cannot be operated as a form. This publication asks no health questions and stores no health data; the filter works as a reader's own decision against a stated list, which is both cleaner under GDPR and consistent with the rest of the method.