Has the Smoking Comparison Ever Been Replicated?

The U.S. Surgeon General’s 2023 advisory on loneliness states that the mortality risk associated with social disconnection is comparable to smoking up to 15 cigarettes a day. The line has since appeared in conference talks, op-eds, and policy documents as a settled fact about the size of the loneliness problem. It is a striking number, and it does real rhetorical work: it converts an abstract social phenomenon into something as concrete as a pack of cigarettes. The question this note asks is narrower and less comfortable. Has anyone actually replicated it?

The short answer is no, not in the sense that word usually carries in epidemiology. No study has taken a single population, measured loneliness and smoking in the same people, followed them for the same period, and reported that the two hazards land on the same point. What exists instead is a comparison of comparisons, built once, and repeated many times since without anyone rebuilding it.

Where the number actually comes from

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 follow-up periods studied. That paper’s contribution was to put social relationships into the same units — odds ratios for mortality — as other well-studied risk factors, and to note that the resulting effect size sat in the same range as things like physical inactivity and, yes, smoking. That is a legitimate way to communicate the magnitude of an effect. It is not the same as measuring both risk factors head-to-head in one dataset.

The specific “15 cigarettes a day” framing appears to descend from this kind of cross-literature comparison rather than from a study designed to produce that exact figure. The Surgeon General’s advisory cites it as established, but the advisory is a synthesis document, not a primary study, and it does not present new data pitting loneliness against smoking prevalence in a single cohort. It is, in effect, restating a comparison that Holt-Lunstad’s group had already made using separately estimated risk factors from separate bodies of research: one literature on relationships and mortality, another on smoking and mortality, aligned by their respective odds ratios.

This matters because the two literatures do not share methods. Smoking-mortality studies typically have well-defined exposure measures — pack-years, cigarettes per day, biomarkers of tobacco exposure — collected with reasonable consistency across decades of research. Loneliness and isolation research, by contrast, uses a patchwork of instruments: the UCLA Loneliness Scale, single-item self-report questions, structural measures of network size, and composite isolation indices that combine living arrangement, contact frequency, and group membership. A 2023 review in BMC Public Health flags inconsistent measurement as one of the central barriers to comparing findings across the loneliness literature. Comparing an odds ratio from that literature to an odds ratio from the comparatively standardized smoking literature is a comparison across two different levels of measurement precision, not just two different exposures.

What the more careful numbers say

Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science, which followed the 2010 paper, reported more specific figures: an odds ratio of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, as predictors of early mortality. These are meaningful effects, and the paper notes they held up after adjusting for existing health status, with social deficits proving more predictive of death in samples averaging under 65 years old. But an odds ratio around 1.3 is a different order of magnitude from the roughly threefold or greater mortality risk typically associated with heavy, sustained smoking. The “15 cigarettes a day” framing borrows the intuitive force of a familiar hazard to describe a set of odds ratios that, read directly, sit closer to moderate obesity or physical inactivity than to a pack-a-day habit.

The American Heart Association’s 2022 scientific statement offers a useful check from a different literature. It reports that social isolation and loneliness carry roughly a 29% increased risk of heart attack and death from heart disease and a 32% increased risk of stroke — figures in the same neighborhood as Holt-Lunstad’s 1.3 range, and notably not framed by the AHA in cigarette-equivalent terms. More important for this note, the AHA statement explicitly names the absence of intervention evidence as the central gap in the field: no trial has shown that reducing isolation or loneliness reduces cardiovascular events. That is a direct, if implicit, answer to the replication question. The people closest to this evidence, writing in 2022, were not treating the size of the effect as settled enough to anchor policy-grade comparisons to a specific number of cigarettes. They were flagging that even the association, let alone a causal pathway, needs more work.

Isolation and loneliness are not interchangeable inputs

A second problem sits underneath the first. The cigarette comparison is typically invoked for “loneliness” broadly, but the meta-analyses it draws on report separate figures for isolation (a structural property: how many social ties a person has, how often they see people) and loneliness (a subjective state: whether a person feels connected regardless of tie count). Holt-Lunstad’s 2015 paper treats these as distinct predictors with different odds ratios, precisely because they are not interchangeable. Public repetitions of the smoking comparison rarely preserve that distinction; the number gets attached to “loneliness” as a catch-all, when the underlying evidence base includes isolation, loneliness, and living alone as three separately measured, separately estimated exposures with different odds ratios. Collapsing them into one figure and then comparing that figure to smoking compounds the original cross-literature comparison problem with a second layer of aggregation.

What would actually settle this

A comparison sturdy enough to bear the weight this one has been given would need a few things the current literature does not have. First, a single prospective cohort with directly measured smoking history and a validated, consistent loneliness or isolation instrument, followed for mortality over the same period, so the two hazard ratios come from the same population under the same conditions rather than being imported from separate literatures. Second, that comparison would need to be run in more than one cohort, in more than one country, before anyone should call it replicated in the ordinary sense. Third, given the AHA’s own point about missing intervention evidence, a trial in which reducing isolation changed measured mortality or cardiovascular outcomes would do more to justify treating loneliness as a modifiable risk factor on par with smoking than any cross-literature comparison of odds ratios, however carefully done.

None of this means the underlying concern is overstated. A 50% increase in survival likelihood, or odds ratios in the 1.26 to 1.32 range holding after adjustment for existing health status, are not trivial findings, and they recur across a genuinely large evidence base. The issue is narrower: a specific, quotable comparison to a specific number of cigarettes has taken on the authority of a replicated empirical result when it is better described as an illustrative translation, made once, that has since been repeated as though someone checked it again. Nobody appears to have.