How Large Is the Loneliness Effect, Actually

The claim that circulates most widely in this field is a comparison: loneliness is as bad for health as smoking. The U.S. Surgeon General’s 2023 advisory put it precisely, stating that the mortality risk associated with social disconnection is comparable to smoking up to 15 cigarettes a day. That is a striking number, and it gets repeated far more often than the studies that produced it. Worth asking what “comparable” actually means here, and whether the effect sizes across the literature agree with each other closely enough to support a single soundbite.

They do not agree closely. They point in the same direction, but the magnitude varies by nearly a factor of two depending on which outcome was measured, which population was sampled, and whether isolation, loneliness, or living alone was the variable.

Two Holt-Lunstad papers, two different numbers

Julianne Holt-Lunstad’s 2010 meta-analysis in PLoS Medicine, pooling 148 studies and 308,849 participants, found that stronger social relationships were associated with a 50% increased likelihood of survival. That figure is the source of most “comparable to smoking” claims, because the paper frames the effect as similar in magnitude to established mortality risk factors.

Her 2015 meta-analysis in Perspectives on Psychological Science, examining social isolation, loneliness, and living alone separately, produced smaller odds ratios: 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone. Those are meaningfully different numbers from a 50% survival advantage. An odds ratio of roughly 1.3 for early mortality is a real effect, well above chance, and it held up after adjusting for baseline health status. But it is not the same magnitude as the 2010 figure, and the two are sometimes cited interchangeably as if they were.

Part of the discrepancy is what was measured. The 2010 analysis pooled a broader construct of social relationship strength across many different measurement approaches. The 2015 analysis disaggregated the construct into isolation, loneliness, and living alone as three separate predictors, each with its own effect size. Disaggregating a variable tends to shrink the apparent effect of any one component, because some of the pooled signal in a composite measure comes from the combination.

There is a further wrinkle in the 2015 paper worth stating plainly: the social deficits were more predictive of death in samples averaging under 65 than in older samples. That cuts against an intuitive assumption that isolation matters more as people age and lose social contacts. The data say the opposite, at least for mortality risk specifically. This is the kind of finding that gets dropped when a study is cited only for its headline number.

What the American Heart Association’s statement adds

The AHA’s 2022 scientific statement, led by Crystal W. Cene on behalf of several AHA councils, offers a third set of figures, this time specific to cardiovascular outcomes. Social isolation and loneliness carried roughly a 30% increased risk of heart attack, stroke, or death from either — specifically 29% for heart attack and/or death from heart disease, and 32% for stroke. These numbers sit closer to the 2015 mortality odds ratios than to the 2010 survival figure, which is what one would expect: cardiovascular events are a narrower outcome than all-cause mortality, and narrower outcomes tend to produce more modest, more specific effect sizes than broad composite ones.

The AHA statement also reports something the mortality meta-analyses do not: isolation and loneliness predict worse prognosis in people who already have coronary heart disease or a prior stroke, including recurrent stroke. That is a different claim from mortality risk in a general population. It suggests the effect operates at more than one point in a disease course — on the likelihood of a first cardiovascular event, and separately on recovery and recurrence risk once disease is established. Most public discussion of the loneliness-health link only addresses the first.

The AHA statement is candid about the biggest gap in this literature: it explicitly identifies the absence of intervention evidence as the central research problem. Every effect size discussed above comes from observational data. None of it comes from a trial in which isolation or loneliness was reduced and cardiovascular outcomes were then measured. That distinction matters more than the exact size of any odds ratio.

Association, not intervention

This is where effect size and causal interpretation start to pull apart. An odds ratio of 1.3 from an observational cohort tells us that people who are more isolated die earlier, on average, than people who are not, after adjusting for measured confounders. It does not tell us that reducing isolation in a given population would lower mortality by a corresponding amount. The AHA statement says as much directly. Holt-Lunstad’s 2021 review in the American Journal of Lifestyle Medicine argues that social connection should be treated as a modifiable protective factor, in the same preventive category as diet, exercise, and smoking cessation — but the argument for treating it that way rests on the strength and consistency of the association, not on trial evidence showing that intervening on it changes outcomes at a comparable magnitude.

The National Academies’ 2020 consensus report on older adults reaches a similar position from the clinical side: it calls for the health care system to routinely assess isolation and loneliness, but the case for assessment is built on the association with poor health outcomes, not on demonstrated evidence that assessment plus intervention improves those outcomes at scale. Assessment is a reasonable first step precisely because the intervention evidence is thin.

Why the smoking comparison survives anyway

None of this means the smoking comparison is wrong to use. An all-cause mortality effect in the range of the 2010 meta-analysis is genuinely large by the standards of epidemiology, and it is fair to describe it as being in the same range as other well-established risk factors. The problem is using one number — the largest, most striking one — as a stand-in for a literature that actually produces a range: roughly 1.3 for specific mortality odds ratios by isolation type, up to 1.5 for the broader survival-likelihood framing, and around 1.3 for specific cardiovascular endpoints. These are not contradictory findings. They are what happens when different studies ask different questions of different populations and report the answer on different scales. A meta-analysis of “social relationship strength” and a meta-analysis of “loneliness” isolated from other social variables will not produce the same number, even measuring related phenomena in overlapping literatures.

The 2023 review in BMC Public Health on the state of loneliness and isolation research names this directly: inconsistent measurement is identified as a barrier to comparing findings across studies. That is the methodological reality underneath every effect-size comparison in this piece. The instruments differ, the outcome definitions differ, and the populations differ. Effect sizes from this literature should be read as a consistent direction of risk with real but variable magnitude, not as a single number that any one study or advisory can be trusted to represent on its own.

What would settle the magnitude question

A study designed to resolve this would need to do two things current work does not. First, it would measure isolation and loneliness with the same instrument across cardiovascular, all-cause mortality, and general health outcomes in the same cohort, so that effect sizes could be compared on a common scale rather than reconstructed across separate meta-analyses with different pooling methods. Second, and more consequentially, it would need to be interventional — a trial that reduces isolation or loneliness in a defined population and follows health outcomes forward, rather than another observational cohort measuring association after the fact. The AHA statement says this evidence does not yet exist. Until it does, every effect size in this literature describes a correlation of real and consistent size, not a lever anyone has shown can be pulled.