Acupuncture, an ancient practice rooted in traditional Chinese medicine, has been studied and argued over by scientists for decades. A few pieces of research stand out, both for what they found and, just as importantly, for what they don't tell us.
Chronic pain: a very large analysis
One of the most important studies is an individual patient data meta-analysis by the Acupuncture Trialists' Collaboration, led by Andrew Vickers and published in The Journal of Pain in 2018. Rather than simply combining published results, the team pooled the raw data from 39 trials involving 20,827 patients with chronic pain.
Acupuncture did better than both sham (fake) acupuncture and no acupuncture for each type of pain studied. The gap over sham was smaller than the gap over no acupuncture, which tells us that part of the benefit is likely to be placebo and the care that comes with treatment. Even so, the authors concluded that the drop in pain "cannot be explained solely in terms of placebo effects". The effects also largely lasted, with only around a 15% decrease in the treatment effect at one year.
UK research on cost and value
In 2017 a large NIHR-funded programme led by Hugh MacPherson reviewed the evidence for acupuncture in chronic pain. It found benefits beyond placebo for musculoskeletal pain, knee osteoarthritis and chronic headache, and estimated a cost per quality-adjusted life year below the threshold the NHS usually considers good value. As with the Vickers analysis, the effects were smaller against sham acupuncture than against usual care.
Migraine prevention
A 2016 Cochrane review looked at 22 trials with 4,985 participants. After treatment, headache frequency at least halved in 50% of people having real acupuncture, compared with 41% of those having sham. Where trials compared acupuncture with preventive drugs, the results were at least similar, and people having acupuncture reported side effects less often. The reviewers rated the evidence as moderate quality.
How this shows up in UK guidance
This body of research is reflected in NICE guidance. A course of up to 10 acupuncture sessions over 5 to 8 weeks can be considered to help prevent chronic tension-type headache, and to help prevent migraine where the usual preventive medicines haven't worked or suited. The chronic pain guideline (NG193) adds that a single course of acupuncture or dry needling can be considered for people aged 16 and over with chronic primary pain, provided certain conditions are met.
The UK advertising regulator, the ASA, accepts evidence for short-term relief of tension headache, migraine, chronic low back pain, neck pain and jaw (TMJ) pain, and temporary relief of knee osteoarthritis pain alongside other treatment. Those words "short-term" and "temporary" matter.
How might acupuncture work?
This is still an open question. One study that drew a lot of attention was published in Nature in 2021. Working with mice, the researchers found that gentle electroacupuncture at a point on the hind leg (known in acupuncture as ST36) activated a nerve pathway to the adrenal glands that reduced inflammation, while stimulating a point on the abdomen at the same intensity did not. They traced the effect to a particular group of sensory nerves found in deep tissues of the limbs.
It's a genuinely interesting result, because it hints at one reason the location of a point might matter. But this was research in mice, using electrical stimulation, and it doesn't show that the same thing happens in people having traditional acupuncture.
Where the evidence falls short
- It's hard to design a perfect placebo for acupuncture, and practitioners can't be "blinded".
- For many conditions the evidence is low quality or mixed, and in those areas I don't make claims.
- Benefits for pain are real in trials but often modest, and usually described as short-term.
I'd rather tell you that than overpromise. Traditional ideas like Qi and the Five Elements are part of how I practise, but they're a tradition, not something these studies set out to prove.
For a fuller round-up of the evidence, including where it's weakest, see my research page.