How Big Is the Effect, Really

A study can report a statistically significant association between loneliness and heart disease with a sample large enough to detect almost anything. The question that matters for policy is different: how large is the effect, and is it large enough to build a health strategy around. The loneliness literature answers the first question constantly and the second one rarely. Lining up the actual effect sizes, rather than the p-values attached to them, produces a more useful and less flattering picture.

The mortality numbers, side by side

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 is a large effect by the standards of population health, comparable to the difference between smokers and non-smokers.

Her 2015 follow-up in Perspectives on Psychological Science, which separated the construct into components, reported smaller odds ratios: 1.29 for social isolation, 1.26 for loneliness, 1.32 for living alone. These are still real effects, and they held after adjusting for baseline health status. But an odds ratio of 1.3 is a different order of magnitude from a 50% survival advantage. The two papers are frequently cited interchangeably as evidence that “loneliness is as deadly as smoking,” when in fact the 2010 estimate is considerably larger than the 2015 one, and the two are not measuring the same construct. The 2010 analysis pooled a broader category of social relationship measures; the 2015 analysis isolated loneliness, isolation, and living alone as distinct exposures. Reporting only the larger number, as the Surgeon General’s 2023 advisory does when it draws the smoking comparison, is not wrong, but it obscures that the field has since produced a more conservative, decomposed estimate.

Cardiovascular risk: consistent, moderate, not decisive

The American Heart Association’s 2022 scientific statement on social isolation and cardiovascular and brain health gives some of the most precise figures available. Isolation and loneliness combined were associated with roughly a 30% increased risk of heart attack, stroke, or death from either. Broken down, the statement reports a 29% increased risk of heart attack or death from heart disease and a 32% increased risk of stroke. These are moderate effect sizes: real, replicated across the studies the AHA reviewed, and clinically relevant at a population level, but nowhere near the magnitude implied by the “as deadly as smoking” framing that circulates in press coverage of the same body of evidence. A 30% relative increase in risk, applied to a large population, still produces a meaningful number of excess events. It does not mean loneliness is doing to the cardiovascular system what fifteen cigarettes a day does. The AHA statement itself is unusually candid about this, stating plainly that the absence of intervention evidence is the central gap in the field — an acknowledgment that association research has outpaced anything that tells clinicians what to do about it.

The global estimate, and why it is hard to compare

The WHO Commission on Social Connection’s 2025 report puts the global toll at an estimated 871,000 deaths annually attributable to loneliness, or roughly 100 an hour. This figure is not directly comparable to the Holt-Lunstad or AHA odds ratios, because it is an attributable-burden estimate built by applying relative-risk figures to population prevalence data, not a fresh effect size from a new study. It is useful for conveying scale to a policy audience. It is not a number that tells a clinician or a program designer how much benefit to expect from a given intervention, and citing it as though it does invites the same confusion that plagues the mortality literature generally: large, headline-ready aggregate numbers standing in for effect sizes that are, individually, moderate.

Where the intervention evidence actually lands

This is where effect size matters most, because it is where decisions about resource allocation get made, and where the evidence is thinnest.

The HEAL-HOA trial, a randomized controlled trial of volunteering and prosocial engagement among lonely older adults in Hong Kong published in the Lancet Healthy Longevity in 2024, is notable simply for being a randomized trial in a field otherwise dominated by uncontrolled programme evaluations. Its value lies less in the size of the effect than in the design that produced it — a rare instance where a comparison group makes the estimate trustworthy rather than merely statistically significant.

The 2025 befriending trial in aged care, published in Clinical Gerontologist, gives an unusually specific number: a reduction of 2.39 points on the UCLA Loneliness Scale at eight weeks and 2.71 points at sixteen weeks, relative to a control group. Whether that is a large effect depends on the scale’s range and clinical meaning, which the trial itself frames as modest but durable — the effect grew rather than decayed between eight and sixteen weeks, which is more informative than the raw magnitude. Read next to HEAL-HOA, the pattern suggests structured psychological or prosocial engagement outperforms simple social contact, though neither trial is large enough to settle that ranking definitively.

By contrast, the two systematic reviews of social prescribing, from 2021, report outcomes almost entirely in terms of direction rather than magnitude: increases in self-esteem and self-confidence, positive impacts reported in nine of nine included studies. Neither review can report a pooled effect size, because the underlying trials use different instruments and different comparators, a problem the 2023 BMC Public Health review of the field identifies explicitly as inconsistent measurement across studies. This is not a minor caveat. Social prescribing has become a default policy response to loneliness in several health systems in part because it is popular and low-risk, not because trials have quantified how much loneliness it removes, on what instrument, compared to what alternative.

What the size of an effect should change

None of this means loneliness research is thin. The link between social disconnection and mortality has been replicated with large samples using consistent methods for over a decade. But treating every statistically significant finding as equivalent obscures a real hierarchy: the isolation-mortality odds ratios around 1.3, the cardiovascular relative risks around 1.3, and the population-attributable death estimates in the hundreds of thousands are all defensible numbers, and none of them is the same claim as “loneliness kills like smoking.” That claim rests on the largest and least decomposed of the available estimates.

A better evidence base would report intervention effect sizes on a single standardized instrument, with a control arm, the way the two 2024-2025 randomized trials do, rather than continuing to add uncontrolled programme evaluations whose only reportable finding is that something, on average, improved. Until more of those trials exist, the honest summary is narrower than the headlines: social disconnection is associated with moderately elevated risk across several serious health outcomes, the size of that elevation is smaller than popular framing suggests, and evidence on how much any specific intervention reduces it remains scarce enough that effect size, where it exists at all, should be stated rather than assumed.