Has the Cigarette Comparison Ever Been Replicated?

The 2023 Surgeon General advisory on loneliness contains a single sentence that has since appeared in hundreds of news articles, conference slides, and policy documents: the mortality risk of social disconnection is comparable to smoking up to 15 cigarettes a day. It is the most quoted line in the entire loneliness literature. It is worth asking where it comes from, and whether it has been replicated in the sense a clinical claim usually needs to be.

The answer is that it traces to one source, has never been directly replicated using the same method, and the comparison itself is doing more rhetorical work than the underlying data supports.

Where the number originates

The figure comes from 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 and colleagues then noted, as one comparison among several, that this effect size was of similar magnitude to well-established mortality risk factors — the paper mentions smoking as one such comparator, alongside obesity and physical inactivity. It was a way of communicating scale to a public health audience already fluent in smoking-attributable risk, not a claim that loneliness and smoking operate through the same biological pathway or carry an equivalent dose-response curve.

The “15 cigarettes a day” framing specifically is the version the Surgeon General’s office adopted in 2023, restating Holt-Lunstad’s comparison for a general audience. It is a paraphrase of a paraphrase: a rhetorical device from a 2010 paper, restated thirteen years later as if it were a discrete, freshly measured finding.

What the 2015 follow-up actually changed

Holt-Lunstad returned to the question in a 2015 meta-analysis in Perspectives on Psychological Science, this time separating isolation, loneliness, and living alone as distinct predictors rather than treating “social relationships” as one bundled variable. That paper reported odds ratios of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, each for early mortality, with effects surviving adjustment for health status. Notably, the deficits were more predictive of death in samples averaging under 65 than in older samples — a finding that cuts against the popular assumption that loneliness is chiefly an old-age problem.

This is a genuine advance on the 2010 paper. It is not a replication of the cigarette comparison. The 2015 study does not repeat or test the smoking analogy at all; it refines the underlying mortality estimates and, in doing so, produces effect sizes that are smaller and more differentiated than the single pooled 50% figure from 2010. An odds ratio of 1.29 is a real and policy-relevant effect. It is also a considerably less dramatic number than “increased likelihood of survival by 50%,” and neither figure was designed to be measured on the same scale as smoking-attributable mortality, which is typically expressed in relative risk terms derived from a completely different body of longitudinal smoking cohort research.

The comparison has not been tested against smoking data directly

No study identified in the current literature takes smoking mortality data and loneliness mortality data and runs them through a common model to produce a validated equivalence — the kind of analysis that would be needed to say, with the same confidence as an epidemiologist saying “one pack-year is associated with X% increased lung cancer risk,” that a given duration or intensity of loneliness is associated with a specific, comparable increment of mortality risk. What exists instead is a chain of citation: the 2010 paper makes an illustrative comparison, subsequent reviews and advisories repeat it, and the number’s apparent precision — 15 cigarettes, not “a moderate amount of smoking” — has hardened through repetition rather than through additional measurement.

Holt-Lunstad’s 2021 review in the American Journal of Lifestyle Medicine makes a version of this point explicitly, arguing that social connection belongs alongside diet, exercise, and smoking in preventive health frameworks. That is a claim about where loneliness should sit in the hierarchy of modifiable risk factors clinicians attend to. It is a different claim from asserting a specific quantitative equivalence between loneliness and a stated number of cigarettes, and the review does not attempt to defend the equivalence at that level of specificity.

What the American Heart Association added, and did not add

The 2022 American Heart Association scientific statement is the most careful treatment of cardiovascular-specific risk in the literature to date. It reports that social isolation and loneliness are associated with roughly a 29% increased risk of heart attack and death from heart disease, and a 32% increased risk of stroke. These are AHA’s own pooled estimates, not restatements of the Holt-Lunstad smoking comparison, and they cover a narrower outcome set — cardiovascular events specifically, rather than all-cause mortality.

What the AHA statement does that most citations of the cigarette line do not is name the actual gap in the evidence: the absence of intervention evidence. The statement is explicit that association has been established across many observational cohorts, but that no comparable body of trial evidence exists showing that reducing isolation or loneliness reduces cardiovascular events. That is the replication question in its most useful form. Not “has someone else found a similar association,” which several dozen cohort studies have, in various forms, but “has anyone shown that intervening on the exposure changes the outcome,” which almost nothing in the literature has attempted for mortality-scale endpoints.

Why this matters for how the figure gets used

None of this means the underlying concern is unfounded. The association between social disconnection and mortality is one of the more consistently replicated findings in the health literature — it appears across the 2010 meta-analysis, the 2015 meta-analysis, and the 2022 AHA statement, using different pooling methods, different outcome definitions, and different subsets of cohort studies, and the direction of the effect does not flip. A 2023 review in BMC Public Health mapping the state of loneliness and isolation research identifies inconsistent measurement as the field’s central methodological problem, but even that review does not dispute the basic direction of the mortality association.

The problem is narrower and more specific than “is loneliness bad for you.” It is that a single illustrative comparison from a 2010 paper has been treated, through thirteen years of citation, as though it were itself an independently measured and replicated finding, when in fact the studies that followed it — including Holt-Lunstad’s own 2015 work — moved away from the pooled 50% figure toward more differentiated, and generally smaller, effect sizes for isolation and loneliness considered separately. The cigarette framing survives not because it has been re-derived, but because it is vivid and porous enough to travel from an academic journal into a Surgeon General advisory into a news article without anyone checking whether the comparison was ever meant to bear that much weight.

What would settle it

A genuine test of the equivalence claim would require a single analytic framework applied to both exposures — comparable cohort data on smoking intensity and duration, and comparable data on loneliness or isolation intensity and duration, modeled with the same covariates and the same mortality endpoint, ideally in the same population. That study does not currently exist. Absent it, the honest version of the claim is the one Holt-Lunstad’s data actually support: chronic social disconnection is associated with mortality risk of a magnitude that public health researchers consider comparable in scale to other major modifiable risk factors, established across multiple independent meta-analyses using different methods. That is a strong and useful finding on its own terms. It is not the same as saying the equivalence has been measured, and the repetition of a specific cigarette count across a decade and a half of secondary citation should not be mistaken for replication.