What Happened When 'Social Relationships' Split Into Three Things

In 2010, Julianne Holt-Lunstad and colleagues published a meta-analysis in PLoS Medicine covering 148 studies and 308,849 participants. Its headline finding, still the most quoted statistic in this field, was that stronger social relationships were associated with a 50% increased likelihood of survival. That figure treated “social relationships” as one thing: a composite pooling however each original study happened to measure connection, whether that was network size, contact frequency, marital status, perceived support, or loneliness.

Five years later, in Perspectives on Psychological Science, Holt-Lunstad’s team revisited the question and split the composite apart. The 2015 paper reported separate odds ratios for social isolation (1.29), loneliness (1.26), and living alone (1.32), each for early mortality. The effects held after adjusting for health status, and were more predictive of death in samples averaging under 65 than in older samples. But none of the three individual figures comes close to the 50% figure that made the 2010 paper famous.

This is not a contradiction. It is what happens when a pooled construct gets disaggregated, and it is worth tracing what that disaggregation changed downstream.

Why the numbers moved

The 2010 estimate was not wrong; it was answering a different question. Pooling studies that measured isolation, loneliness, network diversity, and social support all under one umbrella produces an estimate of “having weaker social relationships in some measured sense,” which will tend to look larger than any single component, partly because the composite absorbs studies using whichever measure showed the strongest effect in that particular sample. The 2015 paper deliberately asked a narrower question for each construct: holding constant that you are specifically measuring isolation, or specifically measuring loneliness, or specifically measuring living arrangement, what is the mortality association. Narrower questions produce smaller, more defensible numbers.

The consequence for anyone citing this literature is that “social connection is associated with a 50% higher chance of survival” and “loneliness carries a 26% higher mortality risk” are both accurate statements from the same research group, describing different things. Using them interchangeably, which happens constantly in secondary reporting and advocacy material, overstates the case for whichever specific factor a document happens to be about.

The distinction became institutional

The 2015 split was not an academic footnote. It set the template that subsequent authoritative bodies adopted.

The National Academies of Sciences, Engineering, and Medicine’s 2020 consensus report on older adults built its clinical recommendations explicitly around treating isolation and loneliness as separate risk factors requiring separate assessment, rather than a single “connectedness” screen. The report’s estimate that roughly one quarter of adults 65 and older are socially isolated is a structural claim about network size and contact, not a claim about how lonely they feel. A health system that screens only for loneliness, using something like the UCLA Loneliness Scale that AARP’s 2018 national survey of adults 45 and older used, will miss isolated people who do not report feeling lonely, and vice versa.

The American Heart Association’s 2022 scientific statement went further, building the distinction into its own terminology: “objective and perceived social isolation.” Objective isolation is the AHA’s term for the structural condition, roughly equivalent to what the 2015 Holt-Lunstad paper called isolation; perceived isolation is their term for loneliness. The statement reports isolation and loneliness together carrying roughly a 30% increased risk of heart attack, stroke, or death from either, close to but not identical to the 2015 mortality odds ratios, because the outcome measured is cardiovascular events rather than all-cause mortality. The AHA statement is also unusually candid that the evidence gap is on the intervention side: there is very little trial evidence showing that reducing isolation or loneliness improves cardiovascular outcomes, as distinct from evidence that the two are correlated with worse ones.

The Surgeon General’s 2023 advisory inherited this same architecture and used it to build the smoking comparison that got widespread press coverage: that the mortality risk of social disconnection is comparable to smoking up to 15 cigarettes a day. That comparison draws on the disaggregated 2015 figures, not the 2010 composite, though public discussion of the advisory frequently does not distinguish which underlying estimate is being invoked.

What the split cost the field, and what it bought

The 2023 BMC Public Health review of the state of loneliness and isolation research names inconsistent measurement as a central barrier to comparing findings across studies. The 2015 disaggregation is part of the reason that barrier exists as clearly as it now does: once isolation, loneliness, and living alone were shown to behave differently, researchers could no longer treat any one instrument as a stand-in for the others. A study using a network-size questionnaire and a study using the UCLA Loneliness Scale are now understood to be measuring different constructs with different predictive profiles, not two versions of the same underlying variable. That is progress in precision, but it comes at the cost of comparability: a decade of studies that used composite or single-item measures of “social connection” cannot be cleanly mapped onto the three-construct framework retroactively.

It also means that population prevalence figures and mortality risk figures are answering different questions even when they come from the same research programme. The AARP survey’s finding that network size and diversity, along with physical isolation, are the strongest predictors of loneliness in its sample is a statement about what predicts the subjective state. The 2015 mortality odds ratios are statements about what the subjective state, and the structural condition, each independently predict downstream. Neither implies the other.

What would resolve the remaining ambiguity

The cleanest fix is not another meta-analysis pooling old measures differently. It would be a small number of large cohort studies that collect isolation, loneliness, and living-arrangement data on the same participants, using the same validated instruments, and follow those participants long enough to observe mortality and cardiovascular outcomes directly, rather than inferring them by combining studies that each measured only one construct. That would let researchers test whether the three factors are additive, whether one mediates the others, and whether the size of each effect has genuinely stayed stable across the fifteen years separating the two Holt-Lunstad papers, or whether some of the difference reflects changes in study populations rather than in the underlying phenomenon. Until that exists, the honest summary is that connection matters, isolation and loneliness are not the same thing, and any single number attached to either should come with the study that produced it.