Why Acupuncture Research Is Hard to Judge

Every condition page on this site carries a short note that says acupuncture trials are hard to run fairly. This article explains what I mean by that, why it matters to you as a patient, and how I read the research so that I do not oversell it. I link every study to its PubMed record, so you can check my work.

The short version: acupuncture has been tested in thousands of people, and the best pooled data show it helps chronic pain more than no treatment and a little more than imitation needling. But the tests have built-in problems that make every number a bit softer than it looks. Knowing those problems makes you a better judge of any treatment, mine included.

In a good drug trial, people are assigned by chance to the real pill or a dummy pill that looks identical. That is a randomized trial with a placebo. Nobody in the room knows who got which until the study ends. Hiding the answer is called blinding, and it keeps hope and expectation from tilting the result.

A needle is different. The person placing it knows whether it is a real needle or an imitation. In 13 pain trials that compared real acupuncture, imitation acupuncture and no acupuncture, none blinded the clinicians (Madsen 2009). A 2020 Cochrane review of low back pain trials made the same point: most trials were at high risk of bias because the acupuncturist could not be blinded (Mu 2020).

Patients can often guess, too. Across 54 trials, about 61 in 100 participants stayed fully blinded, which is decent but far from perfect (Moroz 2013). A 2024 review of 64 trials found that people given real acupuncture mostly guessed they had the real thing, and the authors warned that the imitation needling was not convincing in some trials (Liu 2024).

Why this matters: when the person giving the treatment believes in it, enthusiasm can leak into the way patients answer questions. When patients guess they got the real thing, hope can lower their reported pain. Neither is fraud. Both are human, and both can nudge a result.

To fix the blinding problem, researchers invented imitation acupuncture. There are two main kinds. One uses a blunt needle that presses on the skin but does not enter it. The other uses very shallow needling away from the usual points. The goal is a treatment that looks and feels real but does nothing.

It turns out that doing nothing is hard to fake.

So when a trial finds real acupuncture only slightly ahead of imitation, two things can be true at once. The needling may add only a little. Or the imitation may have been doing part of the real work. The trial alone cannot tell you which.

Some trials give up on blinding altogether. They compare acupuncture with usual care, physiotherapy, rest or a waiting list. These are called pragmatic trials, and everyone knows which group they are in. They answer a different, very practical question: if I add acupuncture to what I would do anyway, do I do better?

A German trial gives an example. It randomized 3,093 people with chronic low back pain to acupuncture plus routine care, or routine care alone. After three months, back function had improved by 12.1 points in the acupuncture group and 2.7 points in the routine care group (Witt 2006).

That is a big gap, and it is a real result for real patients. But hope counts here. In four trials with 864 people, those who expected more benefit had about twice the odds of a good response (Linde 2007). A result against usual care can therefore include some expectation along with the needle.

Blinding is only the start. Here are the other problems I watch for, and what I do about each.

Smart researchers look at the same trials and disagree. Here are both readings, fairly stated.

Many of my condition pages show a GRADE rating. GRADE is a system researchers use to say how much to trust a body of evidence. It runs from high to very low, and a low rating is not a failure. It is a note to expect surprises.

You do not have to settle the research debate to make a good choice. Here is the plan I give patients.

A word on safety. A review of 21 prospective studies found serious side effects in about 1 in 10,000 patients, and any side effect, usually minor, in about 9 in 100 (Baumler 2021). Definitions varied between studies, so treat the numbers as a rough guide.