The Smoking Comparison: Where Does It Actually Come From

The 2023 U.S. Surgeon General advisory on loneliness contains a line that has since travelled further than almost any other sentence in the field: the mortality risk associated with social disconnection is comparable to smoking up to 15 cigarettes a day. It appears in newspaper coverage, conference slides, agency fact sheets, and no small number of opinion pieces. It is, by a wide margin, the most quoted single statistic in loneliness policy writing.

It is worth asking what the claim is actually based on, and whether anyone has tried to reproduce it.

Where the number comes from

The Surgeon General’s advisory does not present the smoking comparison as a new finding. It is a translation of relative risk estimates already in the literature into a unit ordinary readers can picture. The most cited source behind that kind of comparison is Julianne Holt-Lunstad’s 2010 meta-analysis in PLoS Medicine, which pooled 148 studies covering 308,849 participants and found that stronger social relationships were associated with a 50% increased likelihood of survival over the follow-up periods studied. That paper explicitly framed the effect size as comparable to other well-established mortality risk factors, which is the same rhetorical move the 2023 advisory makes with cigarettes specifically.

Holt-Lunstad’s 2015 follow-up in Perspectives on Psychological Science, a meta-analysis addressing isolation, loneliness, and living alone as distinct predictors, reported more modest figures: an odds ratio of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, each for early mortality, with effects holding after adjusting for baseline health. These are real and consistently estimated effects. They are not, however, expressed in cigarettes. An odds ratio of 1.29 and “equivalent to smoking 15 cigarettes a day” are two different kinds of statement, produced by two different kinds of calculation, and the path from one to the other is not spelled out in any of the sources that repeat it.

What “comparable to” is doing

The phrase comparable to smoking is not the same claim as caused by smoking-equivalent biological damage. It is a statement about relative magnitude of mortality risk, translated across two literatures that use different exposure metrics, different reference populations, and different follow-up windows. Smoking research typically expresses risk in terms of pack-years and cause-specific mortality, most often lung cancer and cardiovascular disease, tracked over decades in cohorts assembled explicitly to study tobacco exposure. The loneliness and isolation literature works from observational cohorts assembled for other purposes, with social connection measured by instruments that vary from a single item (“do you feel lonely”) to the full 20-item UCLA Loneliness Scale, and with all-cause mortality as the endpoint rather than a specific disease pathway.

Holt-Lunstad’s own 2021 review in the American Journal of Lifestyle Medicine argues for treating social connection as a modifiable risk factor belonging in the same preventive category as diet, exercise, and smoking. That is a normative argument about where connection should sit in clinical priority-setting. It is a different thing from a quantitative claim that a specific level of loneliness carries the same hazard ratio as a specific number of cigarettes. The 2021 review does not, on its own terms, attempt that second calculation.

What the cardiovascular literature actually says

The American Heart Association’s 2022 scientific statement on social isolation and cardiovascular and brain health is the closest thing to an independent check on the mortality comparison, and it is worth reading precisely because it does not repeat the cigarette figure. It reports that social isolation and loneliness are associated with roughly a 30% increased risk of heart attack, stroke, or death from either, breaking out 29% for heart attack and death from heart disease and 32% for stroke. These numbers come from a different evidence synthesis than Holt-Lunstad’s mortality meta-analyses, focused on cardiovascular endpoints rather than all-cause mortality, and they land in a similar but not identical range: roughly a 30% increase, not a 50% increase in survival likelihood, not an odds ratio near 1.3 restated in cigarettes.

More tellingly, the AHA statement explicitly identifies the absence of intervention evidence as the central gap in the field. A body of work confident enough to hand journalists a precise cigarette-equivalent figure has not, according to its own most rigorous synthesis, produced the trial evidence needed to say that intervening on isolation changes the outcome at all. That is a strange position for a widely repeated causal-sounding comparison to be in.

A different number altogether, from the same year

The World Health Organization’s 2025 Commission on Social Connection report offers a third framing that does not map onto either of the above. It estimates that loneliness is linked to roughly 871,000 deaths annually worldwide, or about 100 an hour, and that one in six people globally is affected by loneliness, with the highest rates among young people and in low-income countries. This is a population-attributable-fraction estimate: a statement about aggregate deaths across a global population, built from prevalence figures multiplied by relative risk. It is not a per-person hazard comparison and cannot be converted into a cigarettes-per-day figure without additional assumptions that the report does not supply. Three organisations, three framings, three units of measurement, all cited in support of the same underlying intuition that loneliness is dangerous. None of the three replicates either of the others.

So has it been replicated

Not in the sense that matters. No study identified in the current literature takes the Surgeon General’s specific comparison — 15 cigarettes a day — as a hypothesis and tests it against new data using a comparable exposure metric. What exists instead is a chain of citation: an odds ratio from one meta-analysis, translated into a mortality-risk comparison in a policy document, then repeated as though the comparison itself had been independently measured. A 2024 review of the epidemic framing of loneliness, published in the wake of the advisory, examines how the language of contagion and epidemic came to attach to loneliness research, and that review is a useful reminder that framing decisions of this kind tend to precede, rather than follow from, a settled evidence base.

This does not mean the underlying concern is wrong. The direction of the effect — that social disconnection predicts worse survival, independent of baseline health — appears across enough methodologically distinct studies, using enough different instruments, that it is one of the more durable findings in this literature. A 2023 review in BMC Public Health mapping the state of loneliness and social isolation research flags inconsistent measurement as the field’s central barrier to comparison, which is exactly the problem visible in the smoking claim: the number survives not because it has been reproduced, but because no one has had to reproduce it. It travels well, and travelling well is not the same test as replication.

What would settle it

A genuine test would require a single cohort, followed prospectively, with both loneliness and smoking status measured using instruments validated against each other, and a mortality model that expresses both exposures in the same hazard units — years of life lost, ideally, rather than odds ratios converted after the fact into a consumer product. Until that study exists, the honest version of the claim is narrower than the one in circulation: social disconnection is associated with mortality risk in the same broad range as several established risk factors, the size of that risk depends heavily on which instrument and which endpoint a given study used, and the specific cigarette-equivalent figure is a communication device, not a separately measured finding.