What the Effect Sizes Actually Say About Loneliness and Mortality

Every major report on loneliness now states that its health effects are comparable to smoking. The U.S. Surgeon General’s 2023 advisory made the comparison directly, putting the mortality risk of social disconnection on par with smoking up to 15 cigarettes a day. The comparison is now repeated so often that it has become the number people remember. It is also the number least likely to survive contact with the underlying statistics, because “comparable to smoking” collapses several different effect sizes, measured in different ways, across different outcomes, into one memorable sentence.

This is worth unpacking, because the field’s own effect sizes are more interesting, and more limited, than the headline suggests.

Two meta-analyses, two different numbers

Julianne Holt-Lunstad’s 2010 meta-analysis, drawing on 148 studies and 308,849 participants, reported that stronger social relationships were associated with a 50% increased likelihood of survival. That is a striking number, and it is the one most frequently quoted. But “50% increased likelihood of survival” is a relative measure of a composite outcome across a highly heterogeneous set of studies, measuring social relationships in different ways: marital status, contact frequency, integration into networks, and more. It says relatively little about any one form of disconnection.

Her 2015 meta-analysis in Perspectives on Psychological Science is more precise, and precisely because it is more precise, its numbers look smaller. It separates the exposure into three constructs and reports an odds ratio for each: social isolation at 1.29, loneliness at 1.26, and living alone at 1.32, for early mortality. These are not small effects in epidemiological terms — they sit in a range that public health takes seriously — but they are also not the kind of number that “comparable to smoking” evokes for most readers, where the relative risk of lung cancer from smoking runs many times higher than 1.3.

The reconciliation is straightforward once stated: the two Holt-Lunstad papers are answering different questions. The 2010 paper asks how much social connection generally protects against death, using a broad and varied definition of connection, and gets a large number. The 2015 paper isolates specific, narrower constructs and gets more moderate but more interpretable numbers. Neither number is wrong. They are not the same number, and citing one without noting which is being used tends to make the field’s evidence look either more dramatic or more equivocal than it actually is.

What the American Heart Association found, and how it differs

The American Heart Association’s 2022 scientific statement gives a third kind of figure. Social isolation and loneliness were associated with roughly a 30% increased risk of heart attack, stroke, or death from either — specifically, a 29% increased risk of heart attack and death from heart disease, and a 32% increased risk of stroke. These are disease-specific outcomes rather than all-cause mortality, and the effect sizes land in a similar range to Holt-Lunstad’s 2015 odds ratios, which is itself worth noting: a cardiovascular-focused statement from a different research tradition arrives at a number close to the general mortality odds ratio from psychological science. Convergence across method and discipline is a stronger form of evidence than any single large number, even a dramatic one.

The AHA statement is also unusually candid about a gap the “smoking” comparison tends to paper over: it explicitly identifies the absence of intervention evidence as the central research gap. Association is well established. Whether treating loneliness or isolation actually lowers cardiovascular risk, the way smoking cessation lowers lung cancer risk, has not been demonstrated with anything like the same rigor. The smoking comparison imports causal confidence that the isolation literature has not yet earned through trials.

Why the smoking comparison is the wrong unit

The Surgeon General’s comparison is not fabricated — it draws a real analogy between relative risk magnitudes in the mortality literature — but it invites a category error. Smoking’s mortality risk is dose-dependent, biologically mechanistic, and supported by decades of intervention trials showing that cessation changes outcomes. The loneliness literature, even at its most rigorous, is still mostly observational. Odds ratios of 1.26 to 1.32 are real, replicated across a large pooled sample, and worth taking seriously in a population health sense — a small relative risk applied across tens of millions of people produces a large absolute burden. But treating an odds ratio of 1.3 as if it carries the same evidentiary weight as smoking’s well-established dose-response curve overstates what a single relative-risk number can support.

The 2024 CDC surveillance report illustrates a further wrinkle: it was published in June 2024 but is built on 2022 survey data, and it documents associations between loneliness, lack of social and emotional support, and downstream conditions including heart disease, stroke, dementia, type 2 diabetes, depression, and anxiety, alongside premature mortality. These are the same kinds of associational claims, restated as a federal surveillance finding rather than a meta-analysis, and they carry the same caveat: association across a wide set of outcomes, not a demonstrated causal pathway for any one of them.

The National Academies’ more cautious framing

The National Academies’ 2020 consensus report on older adults is worth returning to precisely because it does not lean on a single headline effect size. It reports that roughly a quarter of adults 65 and older are socially isolated, and it calls on the health care system to assess isolation and loneliness routinely, but its central argument is about screening infrastructure rather than about the size of the mortality effect. That is a more defensible use of the evidence: build a case for measurement and intervention on the consistency of the association across many studies, rather than on any one dramatic ratio.

The 2023 review in BMC Public Health on the state of loneliness and social isolation research makes the underlying problem explicit. It maps the field’s methodological gaps and identifies inconsistent measurement as a persistent barrier to comparing findings across studies. Different instruments, different follow-up periods, different covariates adjusted for, different outcome definitions — all of this variation is baked into any attempt to average effect sizes across the literature. A meta-analytic odds ratio of 1.3 is a genuine, defensible summary of dozens of underlying studies. It is not a single, clean biological fact in the way a blood pressure reading is.

What would make the comparison more honest

A more defensible statement than “loneliness is as deadly as smoking” would specify: which outcome, which exposure definition, which effect measure, and over what time horizon. It would also distinguish, as the AHA statement does, between what is established about association and what remains untested about intervention. Readers reasonably want a single number they can repeat. The literature, read carefully, offers several numbers, clustered in a plausible but modest range, converging across disciplines, and still missing the trial evidence that would let anyone say with confidence that reducing loneliness reduces death in the way that quitting smoking does. That is a less quotable finding. It is also the more accurate one.