Social Relationships and Mortality Risk: The 148-Study Meta-Analysis Behind “Loneliness Is Like Smoking”
By Noam Shemla · The Studies
TL;DR Holt-Lunstad, Smith and Layton pooled 148 studies that followed 308,849 people for an average of seven and a half years. People with stronger social relationships had 50% higher odds of being alive at the end of follow-up. The link was as strong as some of medicine’s best-known risk factors, including moderate smoking, and stronger than physical inactivity or obesity. It held across ages, sexes, health status and causes of death.
- Paper: Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social relationships and mortality risk: a meta-analytic review. PLoS Medicine, 7(7), e1000316.
- Studies: 148 prospective studies, searched from 1900 to 2007, of how people’s social relationships related to whether they later died.
- People: 308,849 participants, average age 63.9 at the start, half women, mostly from North America and Europe.
- Follow-up: 7.5 years on average (from 3 months to 58 years); 29% of participants died during their study’s follow-up.
- Measures: Structural (marriage, networks, living alone, integration), functional (support received or perceived, loneliness), or both.
- Method: Random-effects meta-analysis of log odds ratios, with metaregression to test what changed the size of the link.
This is the paper usually behind the claim that loneliness is as bad for you as smoking. Julianne Holt-Lunstad, Timothy Smith and Bradley Layton set out to measure, across every suitable study they could find, how strongly people's social relationships predict whether they die.1
What a meta-analysis does
Each of the 148 studies measured people's social relationships at the start, then counted who had died some years later. The authors converted each study's result into the same unit, the natural log of an odds ratio, and averaged them, giving more weight to more precise studies. Because the studies differed in who they followed and how they measured relationships, they used a random-effects model, which assumes the true effect varies from study to study and estimates both its average and its spread.
The headline result
Across the 148 studies the pooled odds ratio was 1.50 (95% CI 1.42 to 1.59). People with stronger social relationships had 50% higher odds of surviving the follow-up period. Individual studies ranged from 0.77 to 6.50, and they disagreed more than chance alone would explain, which is why the authors went on to look for what made the effect larger or smaller.
The funnel plot below draws every one of the 148 studies. Precise studies sit near the top; imprecise ones spread out lower down. If small studies with disappointing results had gone unpublished, the lower left of the funnel would look empty.
Which relationships matter most
Not every way of measuring relationships predicted survival equally well. Complex measures of social integration, which combine things like marriage, the size of someone's network and how much they take part in social life, showed the strongest link. A single yes-or-no question about living alone showed the weakest, and its interval includes no effect at all.
The metaregression tested whether the link depended on who was studied. It did not depend on age, sex, initial health, cause of death, length of follow-up or region. What did change it was how relationships were measured, and whether the estimate was adjusted for other variables: adjusted estimates were smaller than raw ones.
What changed the size of the link, for structural measures (random-effects metaregression)
| Predictor | B | SE | p |
|---|---|---|---|
| Average age of participants | −0.002 | 0.002 | .49 |
| All-female sample (vs mixed) | 0.038 | 0.066 | .57 |
| All-male sample (vs mixed) | 0.049 | 0.068 | .48 |
| Pre-existing illness (vs community) | −0.103 | 0.085 | .23 |
| Cardiovascular deaths (vs all-cause) | 0.081 | 0.161 | .61 |
| Cancer deaths (vs all-cause) | −0.208 | 0.139 | .13 |
| Length of follow-up, per year | −0.003 | 0.005 | .54 |
| Measured as living alone | [object Object] | 0.106 | .013 |
| Measured as marital status | −0.097 | 0.074 | .19 |
| Measured as social isolation | −0.144 | 0.178 | .42 |
| Measured as social networks | −0.050 | 0.071 | .48 |
| Complex measures of integration | [object Object] | 0.095 | .007 |
| Statistically adjusted estimate | [object Object] | 0.058 | .01 |
As strong as smoking?
The paper's most quoted figure puts social relationships beside other risks, using effect sizes from other meta-analyses. Social relationships sit alongside smoking fewer than 15 cigarettes a day, and above physical inactivity and obesity.
What 50% higher odds does and does not mean
An odds ratio of 1.5 is not the same as 50% more people surviving. Odds compare survivors with non-survivors, so the translation to people depends on how many survived in the first place.
If half of the people with weaker ties are still alive at some point, an odds ratio of 1.5 means about 60% of those with stronger ties are: ten more people in every hundred, not fifty. The authors' own summary makes the same point with a smaller sample: by the time half of a hypothetical 100 people have died, there are five more survivors among those with stronger relationships.
Has it held up?
The authors extended the work in 2015 with a meta-analysis focused on the absence of relationships. Counting only studies that adjusted for possible confounds, social isolation was associated with 29% higher odds of death, loneliness with 26% and living alone with 32%. Objective isolation and the feeling of loneliness mattered about equally, and the links were, if anything, stronger in people under 65.2
What this study cannot tell you
- Cause. Every study was observational. Sick people may lose social contact, and some other factor, such as income or personality, may drive both. Adjusted estimates were smaller than raw ones, which is exactly what confounding would produce.
- Which relationships to build. The strongest measures were the broadest, which suggests it is the overall web of relationships, not any one of them, that tracks with survival.
- The mechanism. The authors discuss two, relationships buffering stress and relationships encouraging healthier behaviour, but the meta-analysis cannot choose between them.
- The comparison with smoking is approximate. It sets effect sizes from different meta-analyses, with different designs, side by side.
What a speaker can take from it
Relationships are built mostly in conversation, and this is the strongest evidence there is that they matter for more than mood. The measures that predicted survival best were not about one close tie but about a life woven into many: people seen, groups belonged to, roles held.
Every one of those starts with someone speaking to someone else. The skills that make a talk land, attention to the listener, the nerve to start, saying the thing clearly, are the same ones that turn acquaintances into the kind of network this paper measured.
This article summarises an open-access paper. The pooled estimates, subgroup results and metaregression are the authors’ own, from the text and Tables 3 and 4. The funnel plot and our reanalysis use the 148 values in Table 1. The comparison figure is measured from the paper’s Figure 6 and is approximate.
Measured
- Mortality during follow-up in 148 prospective studies
- Social relationships as each study measured them, from marital status to complex integration scales
Inferred, not measured
- Any causal effect of relationships on survival: all the studies are observational
- The equivalence with smoking: a comparison across separate meta-analyses
References
- Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social relationships and mortality risk: a meta-analytic review. PLoS Medicine, 7(7), e1000316. https://doi.org/10.1371/journal.pmed.1000316
- Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T., & Stephenson, D. (2015). Loneliness and social isolation as risk factors for mortality: a meta-analytic review. Perspectives on Psychological Science, 10(2), 227–237. https://doi.org/10.1177/1745691614568352