Has the '15 Cigarettes a Day' Claim Ever Been Replicated?

The 2023 U.S. Surgeon General advisory on loneliness stated that the mortality risk of social disconnection is “comparable to smoking up to 15 cigarettes a day.” The line has since appeared in newsroom copy, public health presentations, and legislative testimony, almost always without a citation attached to the number itself. It reads as an empirical finding: someone measured loneliness’s health cost and found it equivalent to a specific, familiar hazard. It is worth asking where that comparison actually comes from, and whether anyone has tested it since.

One meta-analysis, one analogy

The comparison traces to 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 study follow-up periods. Holt-Lunstad’s team then noted that this effect size was of a similar magnitude to the association between smoking and mortality documented elsewhere in the epidemiological literature. That is the entire empirical basis for the analogy: a comparison of two independently estimated effect sizes, not a single study that measured both loneliness and cigarette consumption in the same cohort and compared them directly.

Cacioppo’s 2008 book on loneliness had already framed chronic loneliness as a biological stressor with measurable physiological cost, which gave the comparison intuitive plausibility before it had a specific number attached. Once Holt-Lunstad supplied the number, it stuck. Vivek Murthy’s 2020 book, written before he returned to the Surgeon General’s office, used similar language to argue loneliness belongs in the same category as other major public health risks. The 2023 advisory then restated the comparison as settled fact, without commissioning or citing a study that tested it directly.

The numbers that followed do not match

If the smoking comparison had been replicated, later meta-analyses would be expected to produce similar effect sizes using comparable methods. They do not.

Holt-Lunstad’s own 2015 follow-up in Perspectives on Psychological Science, using a different pool of studies, reported an odds ratio of 1.29 for social isolation and 1.26 for loneliness as predictors of early mortality, with living alone at 1.32. These are meaningfully smaller effects than the 50% survival advantage reported in 2010, and they are odds ratios rather than the relative risk framing used originally. The two figures are not measuring the same thing in the same way, and neither has been translated into a cigarettes-per-day equivalent by an independent analysis.

The American Heart Association’s 2022 scientific statement, covering cardiovascular and brain health specifically, reported roughly a 29% increased risk of heart attack or death from heart disease and a 32% increased risk of stroke associated with isolation and loneliness. That statement explicitly names the absence of intervention evidence as the field’s central gap, and it makes no attempt to restate its findings in smoking-equivalent terms. The WHO Commission on Social Connection’s 2025 report uses a different metric again: an estimated 871,000 deaths annually attributable to loneliness worldwide, roughly 100 an hour. That is a population-attributable-fraction estimate, not a personal relative-risk comparison, and it cannot be converted into the Surgeon General’s cigarette framing without additional assumptions that the report does not supply.

So the field now has at least four distinct quantitative claims — a 50% survival advantage, odds ratios near 1.3, a roughly 30% cardiovascular risk increase, and an 871,000-death global toll — circulating under the shared heading of “loneliness is as dangerous as smoking.” None of these four figures was derived by comparing loneliness directly against smoking within a single dataset. The 15-cigarettes line is the only one of the four phrased that way, and it has not been updated or retested since 2010.

Why this matters beyond pedantry

This is not an argument that loneliness lacks a real mortality association. The 2015 meta-analysis’s finding that social deficits were more predictive of death in samples averaging under 65 years old, after adjustment for existing health status, is a robust and separately interesting result that does not depend on any smoking comparison at all. The concern here is narrower: a single fifteen-year-old analogy has circulated as though it were a repeatedly confirmed empirical constant, when in fact each subsequent major report has produced a different number using a different method, and none has revisited the original comparison directly.

The 2023 BMC Public Health review of the state of loneliness and social isolation research names inconsistent measurement as a structural barrier to comparing findings across studies. The smoking comparison is a case study in exactly that problem, compounded by the fact that reports tend to quote the memorable line rather than the underlying odds ratio it was drawn from. A number that survives fifteen years of restatement without independent testing has not been replicated; it has been repeated. Those are different things, and public health communication tends to reward the second while implying the first.

What a real test would look like

A study designed to test the comparison directly would need to estimate hazard ratios for loneliness or isolation and for smoking status within the same cohort, using the same follow-up period, adjustment set, and outcome definition, then compare the resulting dose-response curves. No source reviewed here does this. The AHA statement’s identification of the missing intervention evidence points toward a related gap: even if the relative risks were comparable in magnitude, that would not establish that reducing loneliness produces a mortality benefit of the kind that smoking cessation reliably does. Until such a study exists, the appropriate use of the smoking comparison is as a rhetorical device for conveying that the observed effect sizes are large by epidemiological standards — not as a citation-worthy finding in its own right.