Nobody writes loneliness on a death certificate. It shows up in the paperwork as heart failure, or stroke, or a fall that went unnoticed for two days.

That gap between what kills people and what gets recorded is roughly where Julianne Holt-Lunstad went looking. A psychologist at Brigham Young University, she published a meta-analysis with two colleagues in PLoS Medicine in 2010, pooling 148 separate studies covering 308,849 people, each followed for an average of seven and a half years. Their question was blunt. Do your relationships change your odds of still being alive when the study ends?

They do. Participants with stronger social ties had roughly a 50 per cent better likelihood of survival across the follow-up window, and the effect held regardless of age, sex, cause of death, or how healthy people were at the starting line.

What made the paper travel, though, was the benchmarking. Holt-Lunstad’s team lined that effect up against the mortality risks public health already spends real money on. Social connection came out level with quitting smoking, and ahead of both obesity and physical inactivity.

One meta-analysis is one meta-analysis

So the same lab went back in 2015 with a sharper question, published in Perspectives on Psychological Science. This time they separated three things that get sloppily lumped together: loneliness (the subjective ache), social isolation (an objectively thin network), and simply living alone. Seventy studies, more than 3.4 million participants.

After controlling for the obvious confounds, social isolation was associated with a 29 per cent increased likelihood of death, loneliness 26 per cent, and living alone 32 per cent.

The detail that should stop you is what they did not find. Objective and subjective measures performed about the same. A person with a crowded calendar who feels unseen carried a comparable elevation in risk to someone whose phone genuinely never rings.

Both papers are pooled observational data, which means the causal arrow is not nailed down. Illness isolates people, and isolation is easier to measure than the slow decline that produced it. The researchers adjusted for baseline health, but statistical adjustment is not the same animal as a randomised trial, and nobody is going to randomise volunteers into three years of solitude to settle the question. The honest version of the smoking comparison is narrower than the headlines: two effect sizes landed in the same tier of that table.

What the body appears to be doing about it

I retired from an office job and found out how much invisible work the office had been doing.

Not friendship, which I had assumed was the point. Friction. The bloke by the lift who wanted to talk about the cricket, the woman on the third floor who noticed when I got a haircut, forty small transactions a day that I had spent thirty years finding mildly irritating. Somewhere around month four I worked out that I had gone eleven days without a conversation with anyone who was not reading from a script. Not sad, exactly. More like a room with the heating quietly turned off.

The mechanism proposed for this is not sentiment either. John Cacioppo, who spent decades on the neuroscience of loneliness at the University of Chicago, argued that the feeling evolved as an alarm, a nudge back toward the group when drifting to the perimeter was genuinely dangerous. In a review in Philosophical Transactions of the Royal Society B, Louise Hawkley and John Capitanio describe the follow-on effects when that alarm never switches off: implicit hypervigilance for social threat, more anxiety and hostility, attention that snags on the negative reading of any ambiguous interaction.

Which is the cruel part of the loop. Chronic loneliness tunes you to expect rejection, and expecting rejection makes you the sort of company people quietly stop seeking out.

Meanwhile the stress response keeps running. Elevated cortisol, low-grade inflammation, fragmented sleep, blood pressure that never quite settles. None of that produces a diagnosis at forty. Over twenty years it produces cardiovascular disease, and cardiovascular disease is what goes on the certificate.

From psychology journals to health ministries

Fifteen years of this evidence finally moved institutions. In June 2025 the World Health Organization’s Commission on Social Connection released its global report, and the WHO’s own summary put the figures at one in six people worldwide affected by loneliness, and an estimated 871,000 deaths a year linked to it. Around a hundred an hour, modelled from cohort data rather than a body count, and the reason the World Health Assembly passed its first resolution treating social connection as a public health matter a month earlier.

Some of what the commission found cuts against the usual assumptions. Summarising the report, the London School of Hygiene and Tropical Medicine noted that loneliness ran highest among teenagers rather than pensioners, with 17 to 21 per cent of people aged 13 to 29 reporting it. Rates in low-income countries came in around 24 per cent, roughly double the figure for wealthy ones.

Vivek Murthy, the former US surgeon general who co-chaired the commission, has been making the tobacco comparison in public for years, and he has been careful about what it is for. It is a lever, aimed at budgets.

The awkward thing about becoming a risk factor is that risk factors come with interventions, and the ones with the best evidence in the commission’s review are unglamorous. Cognitive behavioural therapy targets the threat-detection loop rather than the calendar, and social skills training in schools starts earlier still. The least glamorous fix of all is just building towns where walking somewhere puts you near other people by accident.

That last one is the piece I underestimated for three decades. The forty daily irritations were the infrastructure.

Editor’s note: This article was updated on September 14, 2026, to align its headline and wording more closely with the evidence and sources.