What community does to your health

Most of what gets written about community and health is unfalsifiable. It asserts that belonging is good for you and leaves it there. The research literature is more specific than that, and more interesting, because it puts numbers on the effect and it also contains findings that complicate the story.

This post covers both.

What the data actually shows

The foundational work is a 2010 meta-analysis in PLoS Medicine that pooled 148 studies covering 308,849 people, average age 63.9, followed for an average of 7.5 years. Adults with stronger social relationships had a 50% greater likelihood of survival over the follow-up period (OR 1.50, 95% CI 1.42–1.59).1

The authors put that number in context deliberately. They calculated it as comparable in magnitude to quitting smoking, and larger than the mortality effect of obesity or physical inactivity. That comparison is the single most-quoted line in this literature, and it is worth knowing that it comes from the authors of the paper rather than from a press release.

How you measure it changes the answer

Inside that same meta-analysis is a finding that gets dropped from most summaries. The size of the effect depended heavily on what the study measured.

• Complex measures of social integration — how many kinds of relationship a person is embedded in, how often they participate — produced the strongest association: OR 1.91 (95% CI 1.63–2.23).
• Functional measures — whether someone felt they had support available — came in lower: OR 1.46 (95% CI 1.28–1.66).
• Living alone versus living with someone was the weakest, and was not statistically significant: OR 1.19 (95% CI 0.99–1.44).1

So the thing that tracks with mortality is not household composition. It is the number of distinct, active connections a person actually has. Living alone and being socially integrated are not in conflict, and the data says so directly.

A larger follow-up, with the confounders stripped out

The 2015 sequel pooled 70 prospective studies and 3,407,134 participants, and reported fully adjusted figures — controlled for age, sex, baseline health and other confounders. Those numbers are smaller than the 2010 headline, which is what you would expect and what you want to see: social isolation OR 1.29 (95% CI 1.06–1.56), loneliness OR 1.26 (95% CI 1.04–1.53), living alone OR 1.32 (95% CI 1.14–1.53).2

Loneliness and isolation were measured separately here, and both predicted mortality independently. They are not the same variable. A person can be objectively isolated without feeling lonely, and can feel lonely in a full house.

The finding worth sitting with is the age breakdown. The association was stronger in younger adults, not older ones: OR 1.57 under age 65, 1.25 between 65 and 75, and 1.14 above 75.2

It shows up in specific diseases

All-cause mortality is a blunt outcome. The disease-specific work is narrower and easier to interrogate.

A 2016 systematic review in Heart drew on 16 longitudinal datasets covering 4,628 coronary heart disease events and 3,002 stroke events, with follow-up running from 3 to 21 years. Poor social relationships were associated with a 29% increase in incident coronary heart disease (pooled RR 1.29, 95% CI 1.04–1.59, from 11 studies) and a 32% increase in stroke (RR 1.32, 95% CI 1.04–1.68, from 8 studies).3 The authors compared the size of that risk to anxiety and job strain — real, but not the largest thing on the list.

On cognition, a 2024 meta-analysis in Nature Mental Health pooled 21 samples and 608,561 people. Loneliness was associated with a 31% higher rate of all-cause dementia (HR 1.31, 95% CI 1.20–1.43) and a 15% higher rate of cognitive impairment (HR 1.15, 95% CI 1.11–1.19).4

Those confidence intervals are tight, which is what 600,000 people buys you. The authors are explicit that heterogeneity between studies was large, driven mostly by the fact that different research groups measure loneliness differently.

The part that argues against us

Here is where a gym has to be careful, because the obvious inference — that exercising in a group is therefore better for you than exercising alone — is not what the trial evidence shows.

A 2026 meta-analysis in Nature Human Behaviour pooled 71 studies and 523 effect sizes, covering roughly 22,000 to 31,600 participants, comparing group-based against individual physical activity interventions head to head. On physical activity itself the group advantage was small and not statistically significant (g = 0.086, 95% CI −0.061 to 0.233, p = 0.249). Equivalence testing supported the conclusion that the two approaches produce statistically equivalent change. Psychosocial outcomes (g = 0.292, p = 0.214) and health outcomes (g = 0.125, p = 0.096) were also non-significant. Only functional outcomes favoured groups, and only after outliers were removed (g = 0.164, p = 0.015).5

Put plainly: a class does not make the same workout work better. If anyone tells you otherwise, they are selling something.

The second inconvenient finding concerns fixing loneliness on purpose. A meta-analysis of loneliness interventions found that effects in randomised designs were meaningfully smaller than in single-group and non-randomised designs — the usual signature of weak methodology inflating results. Among the randomised trials, the interventions that worked best were not the ones that simply increased social contact, but the ones that addressed how people interpret social situations.6

Putting people in a room together is not, by itself, an intervention.

The honest caveats

Nearly all of the mortality and disease evidence above is observational. That matters in three specific ways.

Reverse causation is a real problem here, more than in most of this literature. People who are becoming ill withdraw from social life before a diagnosis arrives, sometimes by years. Some of the association between isolation and death is that mechanism running backwards. The better studies exclude early deaths and adjust for baseline health, which is why the 2015 adjusted figures are lower than the 2010 unadjusted ones.

Confounding is hard to eliminate. Social integration correlates with income, education, mobility, and access to care. Statistical adjustment reduces that overlap; it does not remove it.

The decisive trial cannot be run. You cannot randomly assign several thousand people to thirty years of rich social connection, or to the absence of it. Nobody will ever produce the evidence that would settle this the way a drug trial settles a drug. What exists instead is a large, consistent body of cohort data pointing the same direction, plus mechanistic work on stress physiology, with the honest gap left visible.

What this looks like in practice

Given all of the above, the defensible version is narrower than the slogan, and still worth acting on.

The measure that carried the strongest association was social integration: the count of active, recurring connections in a person’s life. Not a feeling. Not a household arrangement. A number of relationships that are actually in use.

A gym brings the same people through the door several times a week. A class has a fixed time, a roster and someone who notices when you aren’t there. That is not a health intervention in the clinical sense, and the trial evidence says it will not make your training more effective than the identical session done alone. What it does is make you more likely to still be doing it in five years, and it adds a recurring connection to the count that the mortality data is actually tracking.

That is the honest claim. It is smaller than “community is medicine,” and unlike that phrase, it is supported.

The bottom line

Social connection predicts mortality, cardiovascular disease and dementia across millions of people and decades of follow-up, at effect sizes comparable to well-established risk factors. The strongest signal comes from how many active relationships you have, not from who you live with. The risk is not confined to the elderly — it is larger, proportionally, under 65.

And exercising in a group will not, on its own, make you fitter than exercising alone. It will make you more likely to keep going, and it will put more people in your week. On this evidence, that is the part that counts.

If you want somewhere to put those hours, the class schedule is published two weeks out and The Gym runs intro sessions.