Four million people is not a study. It is almost a census.

In April 2026, researchers published what may be the most comprehensive quantitative examination of how cardiorespiratory fitness relates to mental health outcomes ever attempted. The paper, led by Bruno Bizzozero-Peroni and colleagues and published in Nature Mental Health, pooled data from 27 separate prospective cohort studies spanning 9 countries, with individual participants followed for anywhere from four to twenty-nine years.

The scale is unusual. What the researchers found — and what they were careful to say they did not find — offers a more precise picture than the headlines about it tended to suggest.

How a Meta-Analysis of 27 Studies Reaches 4 Million People

The number that anchors this research — 4,007,638 participants — does not come from a single enrolled population. It comes from pooling the results of 27 independent prospective cohort studies, each of which tracked real people over time and measured both their cardiorespiratory fitness and whether they went on to develop a mental health condition or dementia.

That approach, called a systematic review and meta-analysis, is designed to surface patterns that individual studies are too small to detect reliably. A single study of 5,000 people may not observe enough dementia cases to draw firm conclusions about fitness and dementia risk. A pooled analysis of 4 million people, drawing from studies in nine countries including the United States, the United Kingdom, Sweden, Australia, and South Korea, can begin to answer questions that smaller samples cannot.

The studies were not identical in how they measured fitness. Most used metabolic equivalents of task, or METs — a standardised unit representing the ratio of energy burned during an activity to energy burned at rest. Many relied on maximal exercise testing, where participants exercise to exhaustion on a treadmill or bicycle while researchers measure peak oxygen uptake. Some used estimated fitness calculated from submaximal tests. The researchers included studies using either method, which made the pooled sample larger but also introduced variability in how fitness was defined across studies.

Participants ranged in age from 11 to 73, with a median follow-up of about twelve years. Roughly 30 percent were women. The researchers sorted participants into high, moderate, and low cardiorespiratory fitness groups, then compared the rates of developing mental health conditions and dementia across those groups.

The Numbers: What the Data Showed for Depression, Dementia, and Psychosis

The central finding of the Bizzozero-Peroni et al. (2026) meta-analysis, published in Nature Mental Health, is that people in the high cardiorespiratory fitness group had markedly lower rates of several conditions compared to people in the low fitness group over the follow-up period.

For depression, the hazard ratio comparing high to low fitness was 0.64 — meaning the high fitness group had about 36 percent lower risk of developing depression over follow-up than the low fitness group, after statistical adjustment for other factors (Nature Mental Health, 2026). For dementia, the hazard ratio was 0.61, corresponding to approximately 39 percent lower risk. For psychosis — a less commonly examined outcome in fitness research — the hazard ratio was 0.71, indicating about 29 percent lower risk.

The researchers also looked at what happened with each incremental increase in fitness. For every one-MET increase in cardiorespiratory fitness across the sample, they found approximately a 5 percent lower risk of depression and a 19 percent lower risk of dementia (Nature Mental Health, 2026).

Anxiety was a different story. The association between fitness and anxiety did not reach statistical significance in this meta-analysis. The researchers note this as a meaningful finding in itself — not a null result to be explained away, but a signal that the relationship between fitness and different types of mental health symptoms is not uniform across conditions.

These are associations measured across large populations. What they mean for understanding the biology of fitness and mental health — and what they do not mean for predicting any particular person’s risk — is a separate and important question.

Why Dementia Shows the Clearest Evidence

One of the more precise contributions of this meta-analysis is that it did not simply report associations — it graded the certainty of the evidence separately for each outcome. The grading system used, called GRADE, assesses the quality and consistency of the underlying studies rather than just the size of the effect.

By that measure, the dementia finding stands apart. The researchers rated the evidence linking cardiorespiratory fitness to dementia risk as “moderate” certainty — the strongest certainty grade of any outcome in the analysis (Nature Mental Health, 2026). The evidence for depression, anxiety, and psychosis was rated “very low” certainty.

This distinction matters. Moderate certainty in GRADE methodology means researchers are moderately confident that the true effect is close to the estimated one, and that further research is unlikely to change the direction of the finding substantially. Very low certainty means the true effect could be substantially different from the estimate — that limitations in the underlying studies make it hard to draw firm conclusions about the magnitude.

Several features of the dementia research likely explain its stronger evidence grade. Studies of dementia tend to use well-validated diagnostic criteria, be longer in follow-up, and have a more clearly defined biological question: cardiovascular health and brain vasculature are deeply connected, and there is substantial existing evidence that vascular risk factors influence dementia risk. The pathways are better characterised than they are for depression, where the relationship between fitness and symptoms involves more complex and still-contested mechanisms.

For a reader weighing this research, the practical implication is not simply that the dementia number was biggest — it is that researchers have more confidence in that particular signal than in the others.

What a Hazard Ratio Does and Does Not Tell You

The numbers in this meta-analysis — a 36 percent lower risk of depression, a 39 percent lower risk of dementia — are population statistics. They describe what happened across millions of people over years of observation. They do not describe what will happen to any particular individual.

A hazard ratio of 0.64 for depression, comparing the high fitness group to the low fitness group, means that across the entire pooled population, people in the high fitness group developed depression at 64 percent of the rate that people in the low fitness group did, over the same follow-up period. It does not mean that a person who improves their fitness has a 36 percent lower chance of depression. It does not account for the person’s specific history, their existing symptoms, their environment, or the many other variables that contribute to whether any individual develops depression.

This is not a limitation specific to this study. It applies to all epidemiological research of this kind, and the researchers are explicit about it. The study is designed to describe associations in populations, not to predict individual outcomes.

This distinction also carries a particular weight when thinking about mental health. Population statistics can tell us that groups of people with higher fitness have different mental health trajectories on average than groups of people with lower fitness. They cannot tell us whether someone already experiencing symptoms associated with depression or anxiety will benefit from changing their fitness level, how much, or over what timeframe. Those questions require different kinds of study designs — controlled trials, longitudinal tracking of individuals — and they remain more open than the population associations do.

What Cardiorespiratory Fitness Actually Means in Practice

Cardiorespiratory fitness — CRF — is a measure of how efficiently the heart, lungs, and circulatory system deliver oxygen to working muscles during sustained physical exertion. In research settings, it is typically expressed in METs, or metabolic equivalents of task: at rest, a person burns roughly 1 MET. Brisk walking burns approximately 3 to 4 METs. Running at a moderate pace is roughly 8 to 10 METs. Peak cardiorespiratory fitness — the maximum METs a person can sustain — is measured through a maximal exercise test, where the participant exercises to voluntary exhaustion while attached to instruments measuring oxygen uptake.

The difference between the high and low fitness groups in this meta-analysis was not a comparison between elite athletes and sedentary individuals. Most of the included studies defined “low fitness” as the bottom third or quartile of the sample and “high fitness” as the top third or quartile. In practical terms, across the kind of middle-aged adults who made up most of the sample, this often corresponds to a difference of a few METs in peak capacity — roughly the difference between someone who regularly walks for sustained periods and someone who does not.

The per-MET finding — approximately 5 percent lower depression risk and 19 percent lower dementia risk per 1-MET increment — is notable precisely because it suggests a gradient rather than a threshold. The signal does not appear to emerge only at some high fitness level; it tracks across the fitness range. Whether that gradient reflects a causal relationship or the influence of other correlated variables is not something this study design can resolve.

What This Study Cannot Answer

The researchers were careful in their discussion of what their findings do not establish, and those limitations deserve the same attention as the findings themselves.

The most fundamental is reverse causation. This is not a study that randomly assigned people to high or low fitness groups and then watched what happened. It observed people who already had varying fitness levels and tracked their outcomes. Depression, anxiety, and early cognitive changes can themselves reduce physical activity and impair fitness — meaning that a person developing symptoms may become less fit before a diagnosis, rather than becoming symptomatic because they were less fit. Studies that measure fitness at one point in time and then observe outcomes are particularly vulnerable to this problem. The researchers acknowledge it directly.

The participant pool also has notable gaps. Across the 27 studies, roughly 30 percent of participants were women — a significant imbalance given that depression, anxiety, and several other mental health conditions affect women at higher rates than men. The sample skewed toward middle-aged and older adults and was predominantly European and white, which limits how far the findings can extend to other populations, age groups, and cultural contexts. The authors note that data on outcomes like bipolar disorder, ADHD, and autism were too sparse to analyze.

The evidence certainty grades — “very low” for most mental health outcomes — reflect these limitations in the underlying studies rather than in this research team’s work. They are an honest accounting. A person reading this research should understand it as a large, carefully analyzed set of observations that identify real associations, while recognising that those associations are not yet supported by the kind of evidence that would allow confident causal conclusions. That is what well-conducted observational research does: it marks where certainty is earned and where it is not yet.