Social connection and health

Social connection is one of the variables that predicts how long people live, and it predicts it at a size comparable to risk factors that have public health campaigns attached to them. In the most-cited pooled analysis, the increase in adjusted odds of dying over the studies' follow-up periods came out at 26% for loneliness, 29% for social isolation and 32% for living alone.

Three things are worth fixing in place before any of that gets interpreted. The evidence is observational, so none of it establishes that connection causes the difference. The three figures are three different exposures measured three different ways, and they do not behave alike. And a number of this kind describes risk across populations over years, which is a different object from anybody's own decade.

The phrase covers at least four measured things, and almost every confusing claim on the subject comes from collapsing them.

• Network structure. How many ties, how varied, whether there is a partner. Countable from records and questionnaires.
• Contact frequency. How often any of it actually happens.
• Living arrangement. Usually a single item: does this person live alone. The crudest measure in the set and, as it turns out, not the weakest.
• Loneliness. The subjective reading, measured with scales rather than counts. This is the felt gap described in the loneliness entry, and it is the only one of the four that asks the person.

The first three are structural and visible from outside. The fourth is subjective and the only one that tracks what the experience is like. They correlate, loosely. A well-connected person can score high on loneliness and a person with a very small network can score low, which is why studies that measure both usually find each predicts outcomes the other does not.

So "social connection is good for your health" is not one finding. It is a family of findings about different exposures, and when a headline gives one number it is reporting whichever of the four the underlying studies happened to use.

Holt-Lunstad, Smith, Baker, Harris and Stephenson pooled the mortality studies published up to 2014 and reported adjusted odds ratios separately for the three exposures: 26% increased odds of death for loneliness, 29% for social isolation, 32% for living alone. The adjustment matters, and so does what it cannot reach. All the pooled studies are observational, illness itself isolates people, and the review reported that participants' initial health status moved the findings. The effects were larger in samples with an average age under 65, which is not the direction most readers would guess.

Now the part that does not fit the tidy version of the argument, and it belongs here rather than in a footnote. Of the three exposures, the largest adjusted risk attached to living alone, which is the most purely structural measure in the set. The same review found no difference between objective and subjective measures of isolation. If the appealing formulation is that loneliness is not about whether anybody is in the room, this is the finding that argues with it.

The honest reading is that the two measures are answering different questions. The felt gap answers what a person is experiencing, which is what matters for recognising your own situation and for what might change it. Living alone answers what an arrangement does to a life across years: who notices an illness early, who is there after a fall, whether anyone has the household's health admin in view. Those two can diverge without either being wrong. A description of loneliness from the inside and a public health table of exposures are not competing accounts.

Valtorta and colleagues ran the same exercise on cardiovascular outcomes across 16 longitudinal datasets, which recorded 4,628 coronary events and 3,002 strokes over follow-ups of 3 to 21 years. Poor social relationships were associated with 29% higher risk of incident coronary heart disease and 32% higher risk of stroke. In 2025 the WHO Commission on Social Connection put global figures on the whole picture: around one in six people affected, and roughly 871,000 deaths a year associated with loneliness and isolation.

How much weight those figures can carry, and how to read an odds ratio without turning it into a prognosis, is worked through in the difference between lonely and alone.

Hawkley and Cacioppo's review collects the candidate mechanisms, and the honest summary is that the association is better established than any single explanation for it.

• Threat monitoring and stress physiology. Sustained loneliness is associated with heightened vigilance for social threat, higher vascular resistance and blood pressure, and altered cortisol patterns. The regulating effect that company has on the same systems is covered in co-regulation.
• Sleep. Lonely people in these studies do not sleep less so much as sleep worse, with more fragmentation and more night waking.
• Health behaviour. Less physical activity, worse adherence to treatment, later presentation to a doctor.
• Practical support. Someone who notices the symptom, drives you to the appointment, remembers what the specialist said. This pathway is unglamorous and probably does a great deal of the work, and it is the one that most plausibly explains why living alone performs the way it does.
• Inflammation and immune markers. Associations are reported and the picture is mixed.

Two cautions about a list like that. Most of these pathways are measured in the same observational designs as the outcomes, with the same reverse-causation problem: poor sleep and low activity isolate people too. And a mechanism list makes a finding feel explained. This one is a set of candidates, not a settled causal chain.

For most of the last century this was filed as a quality-of-life question rather than a health one, and the effect of that filing was that nobody counted it. The reason to state the size is to correct that, and the reason to state the limits in the same breath is that a frightened reader is not a better-informed one.

What these figures are not:

• Not a prognosis. They are population averages across years. They describe risk, not fate, and no individual month or decade appears in them.
• Not evidence that a quiet period is damaging you. Almost all the health evidence attaches to sustained states, not to a thin fortnight.
• Not a verdict on anybody's character. The circumstances that remove connection are mostly structural, which is precisely why they show up at population scale.
• Not an instruction to socialise more. Adding contact changes the structural exposures and does not reliably change the subjective one. That mismatch is the subject of why more friends doesn't fix loneliness.

What they are good for is proportion. Connection belongs on the same list as sleep and activity: something with measurable consequences, worth arranging deliberately rather than leaving to whatever a calendar happens to produce. The loneliness check reads the subjective exposure, which is the one a person can actually report on.

Nothing here is medical advice or a risk assessment for any individual. If loneliness has arrived alongside persistent low mood, lost interest or changed sleep, that combination is worth a conversation with a clinician, because the overlap with depression is substantial and the two are regularly mistaken for one another.