How a Statistic Becomes a Headline: Tracing the Smoking Comparison

The 2023 U.S. Surgeon General’s advisory on loneliness opens with a comparison that has since traveled further than almost any other line in the document: the mortality risk of social disconnection, it states, is comparable to smoking up to 15 cigarettes a day. The figure has appeared in subsequent coverage, in public health commentary, and in the framing the World Health Organization used when it launched its Commission on Social Connection in November 2023. It is now doing a great deal of rhetorical work. It is worth asking where it came from, and whether it can bear the weight it is being asked to carry.

The number behind the number

The advisory, issued under Vivek Murthy’s office in May 2023, does not derive the 15-cigarette figure from new data. It is a public-facing translation of a body of meta-analytic work stretching back over a decade, most centrally Julianne Holt-Lunstad’s two large meta-analyses.

The first, published in PLoS Medicine in 2010, 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. The paper’s own framing compared this effect size to other established mortality risk factors, smoking among them, on the grounds that a 50% difference in survival likelihood is roughly the same order of magnitude as the differences attributable to well-studied behavioral risks.

The second, published in Perspectives on Psychological Science in 2015, refined the picture with separate estimates: an odds ratio of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, each predicting early mortality. Notably, the effects held after adjusting for baseline health status, and were more predictive of death in samples averaging under 65 than in older cohorts — a detail that complicates any simple story about isolation being mainly a problem of old age.

Neither paper reports a cigarette-equivalent. What they report are relative risk estimates, generated by pooling heterogeneous studies with different populations, different follow-up windows, and different ways of operationalizing “isolation” or “loneliness.” The comparison to smoking is a rhetorical device used to communicate the size of an effect to a non-specialist audience, not a like-for-like unit conversion. Smoking’s mortality risk is itself a range depending on which studies and which endpoints are used; converting a hazard ratio for social disconnection into “15 cigarettes a day” requires choosing a specific smoking-risk benchmark and treating the underlying relationship as if it scaled the same way. That is a defensible communication choice. It is not a finding.

Where the number picked up precision it didn’t earn

By the time Holt-Lunstad’s 2021 review in the American Journal of Lifestyle Medicine argued for treating social connection as a modifiable risk factor alongside diet, exercise, and smoking, the comparison had become a settled talking point in the prevention literature. That review’s contribution was conceptual: positioning connection as something a health system could intervene on, the way it already intervenes on smoking cessation or nutrition. It is a reasonable analogy for motivating investment. It does not claim that a specific dose of loneliness maps onto a specific dose of tobacco.

The 2022 American Heart Association scientific statement, led by Crystal W. Cene on behalf of several AHA councils, took a more cautious position with harder outcomes. It reported 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 — numbers of a similar magnitude to the earlier mortality meta-analyses, but derived from cardiovascular endpoints specifically rather than all-cause mortality. Crucially, the statement is explicit about what remains unknown: it names the absence of intervention evidence as the central research gap in the field. Association is well established across dozens of large studies. Whether reducing isolation or loneliness actually lowers cardiovascular risk, as opposed to merely correlating with lower risk, has not been demonstrated in trials. The AHA’s own newsroom summary of the statement, released the same day, kept that caveat intact for general audiences, even while leading with the more attention-grabbing relative-risk figures.

The Surgeon General’s advisory, issued the following year, is where the smoking comparison became a headline claim rather than a supporting detail. It sits in the advisory’s opening framing, ahead of the six-pillar National Strategy to Advance Social Connection that follows. Once a number appears at the top of a federal advisory, it stops being read as an illustrative analogy and starts being read as an established fact with a specific numeric value — a status the underlying meta-analyses were never built to support.

The propagation problem

This is not a story about any single document getting it wrong. Each step in the chain — the 2010 and 2015 meta-analyses, the 2021 prevention review, the 2022 AHA statement, the 2023 advisory — is defensible on its own terms, and each is more careful than the next step’s public reception of it. The 2024 review in PMC on the framing of loneliness as an “epidemic” makes a related point: the epidemic label itself compresses a set of heterogeneous findings, collected with different instruments across different populations, into language that implies a single, quantifiable, contagious phenomenon. The smoking comparison does the same compression for mortality risk. It takes a range of relative-risk estimates from studies using different definitions of isolation and loneliness, and different statistical adjustments, and renders it as one clean, comparable number.

By the time the WHO Commission on Social Connection launched in November 2023, co-chaired by Murthy and Chido Mpemba, loneliness was being framed globally as a public health priority on the strength of this kind of comparison — reasonably, given the consistency of the underlying association literature, but with the smoking figure functioning as shorthand for a body of evidence rather than as a finding anyone actually measured.

What the comparison obscures

Three things get lost when the 15-cigarette figure stands in for the underlying research.

First, the difference between isolation and loneliness. The 2015 meta-analysis reported separate, non-identical odds ratios for isolation, loneliness, and living alone. A single mortality comparison flattens three distinct constructs — one structural, one subjective, one about household composition — into a single “social disconnection” figure.

Second, the difference between association and intervention evidence. The AHA statement’s own gap analysis is unambiguous: no one has shown, in a controlled trial, that reducing isolation reduces cardiovascular risk by a measurable amount. The observational literature is large and consistent. The causal chain from intervention to outcome is not yet established.

Third, the fact that smoking’s mortality risk is itself a familiar, well-quantified benchmark precisely because it has been studied with a kind of longitudinal, dose-response rigor that the social connection literature does not yet have. Comparing an emerging body of association research to a mature body of dose-response research lends the newer finding a false precision.

None of this undermines the underlying claim that social disconnection is associated with meaningfully worse health outcomes. The consistency of that finding across the 2010 meta-analysis, the 2015 meta-analysis, and the 2022 AHA statement is real and worth taking seriously. What would strengthen the case is not a better analogy but a different kind of study: a randomized trial of an intervention that measurably changes isolation or loneliness, followed long enough to observe hard outcomes like cardiovascular events or mortality. Until that exists, the honest version of the claim is the one in the meta-analyses — a relative risk in the same range as other established risk factors — not the more quotable version that has since traveled from a research paper into a federal advisory and from there into the public record as settled fact.