The evidence on herbal medicine

What it is

What controlled trials have and have not shown about herbal preparations taken as preparations — as distinct from isolated compounds, which are covered under when herbalism became pharmacology.

The distinction matters and is usually blurred. That artemisinin cures malaria is not evidence that a herbal tea does anything.

Where it came from

Herbal medicine largely sat outside the trial system until the 1990s, when the growth of the supplement market and the founding of dedicated research bodies — the US Office of Alternative Medicine in 1991, later NCCIH — produced funding for proper studies for the first time.

Germany's Commission E, which from 1978 assessed several hundred herbal drugs for the German regulator, was the first systematic attempt anywhere and remains widely cited. Its monographs were based substantially on traditional use and expert consensus rather than trials, which is both why they were possible and why they are weaker evidence than they are often presented as being.

How it is used

Selectively, in both directions. Advocates cite the positive trials; critics cite the negative ones; and the actual picture is legible if you look at it whole.

What we can and cannot say

We can say a small number have reasonable support.

• St John's Wort for mild to moderate depression is the strongest case. Cochrane reviews have found it comparable to standard antidepressants for mild-to-moderate presentations with fewer side effects. The catch is enormous and is covered under interactions — it is pharmacologically active enough to disable other medicines, which is itself confirmation that it does something.
• Peppermint oil for irritable bowel syndrome — decent trial support, and it appears in some clinical guidance.
• Ginger for nausea, particularly in pregnancy and post-operatively — modest but real.
• Senna as a laxative, which is uncontroversial and works by a well-understood mechanism.
• Psyllium for constipation and cholesterol — good evidence, and it is a fibre rather than a pharmacologically interesting plant.
• Horse chestnut for chronic venous insufficiency, and butterbur for migraine prophylaxis, both with reasonable trials — butterbur carrying a real hepatotoxicity problem alongside.

We can say the more famous ones have mostly failed.

• Echinacea for colds. Extensively studied, and the large well-controlled trials find little or nothing. This is the most-studied herb with the least to show for it.
• Ginkgo for dementia and cognitive decline. The GEM study followed over 3,000 older adults for around six years and found no reduction in dementia incidence. Later trials agree.
• Saw palmetto for prostate symptoms. Early positive results did not survive the larger, better-blinded trials, including a well-run dose-escalation study finding no benefit over placebo.
• Milk thistle for liver disease. Persistently popular, persistently unconvincing at trial.

We can name the methodological problem that is specific to this field, and it is the most useful part of this article.

A drug trial tests a defined molecule. A herbal trial tests a preparation — and two preparations of the same species can differ in active content by an order of magnitude depending on subspecies, growing conditions, harvest time, which part of the plant was used, how it was dried, and what solvent extracted it. Echinacea trials have used three different species and several different plant parts. When such a body of work comes out mixed, there is genuinely no way to tell whether the herb does nothing or whether half the trials tested something inert. That ambiguity is real, it is not a debating point, and it is why this literature does not resolve the way pharmaceutical literature does.

Blinding is also harder. Many of these preparations taste unmistakable, and a participant who can identify their allocation is not blinded.

Publication bias runs both ways here, unusually. Manufacturer-funded studies tilt positive. There is also a documented tendency for trials run by researchers unsympathetic to the field to select doses and preparations that practitioners say are wrong — a complaint that is sometimes special pleading and sometimes correct.

And we can say the traditional-use argument does not do the work asked of it. "Used for thousands of years" establishes that a plant was used, and nothing about whether it worked. Bloodletting was used for thousands of years across multiple unconnected civilisations. So was mercury. Widespread persistent use is evidence about human beings, not about pharmacology.

We cannot summarise this field in a verdict, and any source that does is selling something. What can be said is that a handful of preparations have real support, the best-known ones largely do not, most have never been tested at all, and the field has a preparation-variability problem that makes it much harder to settle than it looks.

Further reading

• The Cochrane Library, which has reviews on most of the herbs named here and is free to search.
• NCCIH's herb-by-herb summaries, which are unusually blunt about negative findings.
• Edzard Ernst's work — hostile to the field, methodologically careful, and worth reading against the Commission E monographs.