THC and pain: how big the effect really is

The number that matters

Wang and colleagues pooled the randomised trials of medical cannabis and cannabinoids in chronic pain for the BMJ. Their headline finding is that non-inhaled medical cannabis probably results in a small increase in the proportion of people who reach at least the minimally important difference in pain, a risk difference of about 10 per cent.

Ten per cent is a real effect and it is a small one. Put the other way round: for every ten people who take it for chronic pain, about one gets an improvement they would notice and nine do not.

What that threshold actually is

It is the smallest change a patient can actually perceive as better — conventionally about one centimetre on a ten-centimetre pain scale. It is a deliberately modest bar, which is worth holding in mind: the 10 per cent above is the proportion clearing the lowest meaningful hurdle, not the proportion getting substantial relief.

The harms, in the same units

Effect	Extra people per 1,000, against placebo
Meaningful pain improvement	about 100
Dizziness	90 in short trials, up to 280 in longer ones
Drowsiness	50
Nausea	50
Vomiting	30
Impaired attention	30
Cognitive impairment	20

Reading down that column is the honest version of this topic. The benefit and the dizziness are about the same size in trials under three months, and in trials of three months or more the dizziness is nearly three times as common as the benefit.

Sleep and function improved too

The same meta-analysis rated improvements in sleep quality and physical functioning at high certainty. Those are not nothing for somebody living with chronic pain, and they are a fairer description of what people who find it useful are usually describing.

The new Cochrane review

In January 2026 Cochrane published an updated review of cannabis-based medicines for chronic neuropathic pain. It is the most recent word on this page and it is not an encouraging one.

• THC-dominant formulations: no clear benefit for 50 per cent pain relief or for meaningful clinical improvement, and possibly more nervous-system adverse effects, at low certainty.
• Balanced THC and CBD: no clear benefit for 50 per cent pain relief, with modest increases in clinical improvement ratings the authors judged to lack clinical relevance.
• CBD-dominant: no clear benefit, at very low certainty.

Why it differs from the 2017 grading

Every competitor page still quotes the 2017 National Academies report, which graded the evidence for chronic pain in adults as substantial — its highest tier. That grading was accurate when it was written and it covered chronic pain broadly.

The Cochrane review is narrower, newer and harder: one pain type, a 50 per cent relief threshold rather than a minimal one, and a formal certainty rating on every outcome. The two are not really in conflict. A small average effect that fails a demanding threshold is exactly what both are describing.

Neuropathic pain is the hard case

Nerve pain is where cannabinoids have been studied most and where conventional options disappoint most often, which is why it keeps being tested. It is also where the newest and most rigorous review came back negative, and both of those things are true at once.

The trials used pharmacy products

Inhaled is a separate question

The BMJ finding is specifically about non-inhaled cannabis. Inhaled products reach the blood far faster and have been studied much less in chronic pain, so the numbers on this page should not be read across to them.

Opioid sparing, carefully

The hope that cannabis reduces opioid doses is reasonable and it is not settled. Observational work is encouraging and randomised work is thin, and observational designs are exactly where this kind of question goes wrong. Anyone on opioids should be having this conversation with the prescriber rather than acting on it.

Acute pain is not this

The evidence here is about chronic pain, months of it. Some trials in acute and post-operative pain have found cannabinoids no better than placebo and occasionally worse, and none of the figures above apply.

Tolerance, and what it costs

The dose that works tends to rise with steady use, and the adverse effects above are dose-related. A rising dose in a compound whose commonest harm is dizziness is a fall risk, particularly in older people.

Dizziness is the practical problem

Driving, in one line

Impaired attention and cognitive impairment are both measurable harms in the randomised evidence, and an oral dose lasts six to eight hours. Those two facts settle the question without needing a third.

What a label does not tell you

• A THC percentage is not a dose. How much arrives depends on the amount used and the method.
• Strain names carry no analgesic information, and nothing on a jar has been tested against pain.
• Topicals are a different question again, with far less evidence than the oral preparations these numbers come from.

What counts as chronic pain here

Pain lasting three months or more. The BMJ review covered both chronic non-cancer pain, which was the bulk of its trials, and cancer-related pain, and the summary figures above pool them. Anything shorter-lived than that is a different question with different evidence.

Why an average hides the people

The reason this page quotes a proportion rather than an average change is that the two say different things. An average improvement of half a point can be produced by everybody improving slightly, or by one person in ten improving a great deal and the rest not at all.

The second is closer to what the responder analyses suggest, and it is the more useful picture: a minority get something worth having, and there is no way to know in advance who they are.

Questions worth asking a prescriber

• What am I comparing this against, and what has that one shown?
• What would tell us in six weeks that it is not working?
• What does it do to the rest of my prescriptions?
• What happens to my driving?

How to read any page on this topic

Ask what the comparison was, whether the product was a measured pharmaceutical or something bought, how long the trial ran, and what proportion of people reached the threshold rather than what the average change was. Most of the confident writing on cannabis and pain fails at least two of those.

Where these numbers come from

The sources below are the meta-analysis, the current Cochrane review and the National Academies report, in that order of recency. Every figure quoted on this page is in one of them, and none of them is behind a paywall in abstract.