When the Authors Corrected Their Own Headline: Reading a 2026 Meta-Analysis of Biofield Therapies Honestly

Key Takeaways

A 2026 meta-analysis pooled 28 randomised trials of biofield therapies, with Reiki as one of four practices included, and reported moderate benefits for pain and for symptoms associated with mental disorders.
Once the same authors corrected for statistical outliers and for studies that were never published, their pain estimate fell from 0.72 to 0.16 and their mental-symptom estimate from 0.66 to 0.27.
The steadiest result in the paper was also the smallest: depression, at 0.39, where the included trials agreed with one another almost perfectly.

Key Terms Explained

靈氣 (Reiki) / Universal Energy — the practice of placing the hands and letting life settle, passed down in the Usui lineage.
手当て (Te-ate) / Placing the Hands — the ordinary Japanese word for care, literally "putting the hand on."
師範 (Shihan) / Master Teacher — the level at which one is able to teach and pass on the practice.

A Number I Would Have Liked to Keep

When a review of this kind appears, the first number is always the one that travels. In this case it was pleasant reading: a moderate benefit for pain, a moderate benefit for anxiety, depression and stress. If I had stopped there, I could have written a short and cheerful post about it.

I read the rest instead. By the end of the paper, the same authors had adjusted their own headline down to roughly a fifth of where it started. That is the part worth talking about, and it is the part that almost never gets quoted.

I trained as an engineer before I trained as a Shihan. Engineers are taught to distrust the first reading of an instrument. Not because instruments lie, but because the first reading usually includes the things you forgot to account for. A meta-analysis is an instrument like any other.

What the Review Actually Did

The paper is called Efficacy of Biofield Therapies in Alleviating Pain and Reducing Symptoms Associated with Mental Disorders: A Systematic Review and Meta-Analysis, by Rick Sá, Robison José Quitério and Godofredo Pignataro Neto. It appeared in OBM Integrative and Complementary Medicine, volume 11, issue 2, on 2 June 2026. It follows PRISMA 2020 and was registered in advance with PROSPERO.

Four practices, not one

The first thing to be clear about: this is not a review of Reiki. It pools four practices — Reiki, Therapeutic Touch, Healing Touch and External Qigong Treatment — under the single label of biofield therapies.

Those four share a family resemblance, but they are not the same thing, and they are not taught the same way. When someone tells you "the research shows Reiki works," and the research they mean is this one, they are quoting a number that also contains three practices you may never have encountered.

Twenty-eight trials, two thousand people

Twenty-eight randomised controlled trials were included, with 2,013 participants in total. These were split into two groups: fourteen trials with 918 people looking at pain, and twenty-two trials with 1,581 people looking at symptoms associated with mental disorders — anxiety, depression, stress and aggression.

The numbers before and after

The primary model reported a Hedges' g of 0.72 for pain, with a confidence interval running from 0.27 to 1.16. For the mental-symptom group it was 0.66, from 0.37 to 0.94. In the language of effect sizes, those are moderate results.

Then the authors ran a sensitivity model, removing statistical outliers. Pain fell to 0.27, and the mental-symptom group to 0.42. The confidence intervals narrowed considerably, which is what you want, but the estimate itself dropped by more than half for pain.

Finally they tested for publication bias — the tendency for studies that found nothing to remain unpublished. They found significant asymmetry. For the mental-symptom group, the trim-and-fill procedure imputed seven additional studies that ought to exist statistically but do not appear in the literature. Adjusted for that, pain came out at 0.16 and mental symptoms at 0.27.

So the honest range from a single paper is not "0.72." It is something like "between 0.16 and 0.72, depending on how much you correct for the papers nobody published."

Where the results held steadiest

One detail is worth keeping. In the subgroup analysis, depression came out at 0.39 with an I² of 0.0 per cent — meaning the trials agreed with one another almost perfectly. Anxiety was 0.32 with low variation. Stress showed the largest effect at 0.68 but also the widest disagreement between trials, at 73.1 per cent.

The pattern is the opposite of the exciting one. The biggest headline number is the least consistent, and the most consistent finding is a modest one.

What the authors say against themselves

The paper is candid about its own limits, and this is the reason I trust it more rather than less.

A quarter of the included trials — seven of them — were rated as high risk of bias, and eighteen more were rated as having "some concerns." Seventy-nine per cent were at high risk in the domain covering deviations from the intended intervention. The reason is structural and unavoidable: you cannot blind a practitioner to whether they are placing their hands with intent or pretending to.

Heterogeneity was substantial. For pain, I² reached 88.2 per cent before the sensitivity analysis. Several subgroups rested on only five to eight trials. Some standard deviations had to be estimated rather than reported.

They also note that no adverse effects were documented across the included trials.

Why a Smaller Number Does Not Change My Morning

Here is where I have to be careful, because there is an easy move available to me and I do not want to make it.

The easy move is to say the corrected numbers do not matter, because I know what I have seen. That is not an argument. It is a way of ending one.

What I will say is narrower. A pooled effect size across four practices, twenty-eight trials and many different settings is not a statement about your Tuesday morning. It is a statement about an average, and averages are built by flattening exactly the things that make a session what it is — who is sitting with you, whether you slept, whether you wanted to be there at all.

A common misunderstanding is that the harder you concentrate, or the longer you sit, the more it works. In my experience it is the opposite. Relaxation matters most, and even a short session is enough. My own five honest minutes in the morning have done more for me than any hour I forced.

None of that is a claim about illness, and I want to be plain: Reiki is not medical care, and nothing in this review changes that. If you are unwell, see a doctor. What Reiki has offered the people I work with is something quieter — feeling settled, feeling lighter, sleeping better than they did.

What turned my own half-belief into conviction was never a mystical experience. It was my life slowly turning for the better over many years. That is a sample size of one, and I know exactly what that is worth as evidence. I am not offering it as proof of anything. I am saying that a person can hold a modest research finding and a personal conviction at the same time, without either one having to defeat the other.

FAQ

Q: Does this study prove that Reiki works? A: No, and the authors do not claim that. It reports a pooled average across four different practices, and that average shrinks substantially once you correct for publication bias. It is one input among many, not a verdict.

Q: Why did the effect size change so much within one paper? A: Because three different models were run. The primary model gave 0.72 for pain, removing outliers gave 0.27, and adjusting for missing unpublished studies gave 0.16. Each model answers a slightly different question, and reporting only the first one would be misleading.

Q: What was the most reliable finding? A: The depression subgroup, at 0.39, where the trials agreed with each other almost completely — an I² of zero per cent. It is a modest number, but it is the steadiest one in the paper.

Key Insights to Remember

The first number out of a meta-analysis is rarely the final one. In this paper the same authors walked their own pain estimate from 0.72 down to 0.16 as they corrected for outliers and for the studies that were never published. Anyone quoting only the first figure is not quoting the paper; they are quoting its opening sentence.
Consistency and size point in different directions here. Stress showed the largest effect and the widest disagreement between trials, while depression showed a smaller effect and near-perfect agreement. If you want to know what the evidence actually supports, follow the agreement rather than the headline.
A pooled average is a poor description of any single session, in either direction. It cannot tell you that your practice is worthless, and it cannot promise you that it will help. It can only tell you what happens to a large group of people on average, under conditions that no one's real Tuesday morning resembles.