How to read a study about breathing

How many people took part?

With few people, chance carries a lot of weight: two or three who respond differently can change the result. The famous finding that humming multiplies nitric oxide in the nose by fifteen comes from ten healthy people. That's enough to show something happens. It isn't enough to know how much it happens in people at large.

Look at the size of each group too, not just the total. The 2023 Stanford trial on cyclic sighing had 108 participants, spread across four practices. Twenty-one did box breathing. The authors caution that the study wasn't sized to compare the exercises with one another.

Compared with what?

If a group breathes for five minutes a day and feels better a month later, you don't know it was the breathing. It could be the pause, the attention, the passing of time or the wish to feel better. That's why the comparison group matters.

The Stanford trial had no group that went without a practice. In-the-moment anxiety dropped in all four groups, meditation included, and nobody can say how much it would have dropped with nothing at all. By contrast, a 2023 trial of 4-7-8 breathing after surgery did include a group that got routine care only.

A study that measures one group before and after, with no comparison, is the weakest starting point. A 2018 review of bhramari found six studies and not one randomized trial.

Did they know what they were doing?

In drug trials, neither the person nor the evaluator knows whether the pill is the real one. That's called blinding, and it keeps expectations from shifting the result. With breathing it can't be done: you know perfectly well whether you're sighing or holding your breath, and you usually know what's expected of you.

That doesn't invalidate the studies, but it puts a ceiling on them. A 2023 meta-analysis looked at twelve randomized trials of breathwork and stress, with 785 adults. When it rated risk of bias, none came out low: three were high risk and nine raised some concerns.

Who measures, and what?

Almost everything of interest, such as mood, anxiety or stress, is measured with questionnaires the person fills out. They're valid tools, and they're sensitive to expectations. The 2023 meta-analysis acknowledges it: all of its outcomes were self-reports, with no objective measures.

Readings from the body help, but they don't always point the same way. At Stanford, positive mood went up while heart rate variability and sleep, tracked with a wrist strap, didn't change. In a 2025 study of 84 students the opposite happened: breathing at six a minute raised heart rate variability, and mood didn't budge.

And a body reading isn't a benefit. A pulse that dips for a moment, or more gas coming out of the nose, is a data point. What it means for your life is a different question.

How long did it last?

Many studies measure what happens during the session or right after. A 2022 study of 43 young adults recorded heart rate and blood pressure immediately after 4-7-8 breathing. The 2025 study measured a single session of each pattern.

The Stanford trial lasted 28 days and had no follow-up. Almost nothing published says what happens over months or years of practice, or how long an effect lasts once you stop.

What never got published, and who paid?

Studies with eye-catching results get published more easily than studies that find nothing. That's called publication bias, and it makes the published record look better than it is. The 2023 meta-analysis went looking for it: its statistical test didn't detect any, though the plot showed some asymmetry the authors attribute to small studies.

One defense is preregistration: writing down in a public registry what will be measured before starting, so nobody picks what turned out well afterward. The Stanford trial was registered retrospectively, and its authors say so.

Check the disclosures too. In that same trial, the company behind the wrist straps donated them, and one of the authors later became an advisor to it. It's disclosed, which is the right thing. It doesn't make the result false. It's one more thing to weigh as you read.

Questions to take with you

• How many people were in each group, and who were they? Healthy students at one university aren't everybody.
• Was there a comparison group, and what did that group do?
• Is the result a questionnaire, a body reading, or both?
• Was it measured in the moment, after weeks, or after months?
• How big was the change, and not just whether it was significant? At Stanford, the difference was less than one point on a scale of 10 to 50.
• Is it one study or a review of several? What do the authors themselves say about the limits?
• Is the headline about a practice, or is it promising a treatment? The second is a decision for you and a health professional, not for an article.