What Changed When Holt-Lunstad Split the Category

The single most quoted number in this literature is the claim that weak social relationships raise mortality risk by roughly 50%. It comes from Julianne Holt-Lunstad’s 2010 meta-analysis in PLoS Medicine, which pooled 148 studies covering 308,849 participants and found that people with stronger social relationships had a 50% greater likelihood of survival than those with weaker ones. The effect size was compared, at the time, to well-established risks like smoking and obesity, and that comparison has been repeated in nearly every subsequent report on the subject, up to and including the 2023 Surgeon General advisory.

What gets repeated less often is that the 2010 estimate rested on a single, broad category: “social relationships.” That category folded together measures as different as marital status, frequency of contact with friends, participation in social activities, network size, and subjective feelings of connectedness. A study measuring how many people lived in someone’s household counted toward the same pooled estimate as a study measuring how lonely someone said they felt. The 50% figure is real, and the underlying meta-analysis was rigorous by the standards of the field, but it answered a coarser question than most people citing it seem to think.

Five years later, Holt-Lunstad revised the category, and the revision is more analytically interesting than the headline number that preceded it.

Three risk factors, not one

The 2015 paper in Perspectives on Psychological Science, again led by Holt-Lunstad, did not simply update the earlier estimate. It split “social relationships” into three separate, measurable constructs and calculated a distinct odds ratio for each: social isolation (OR 1.29), loneliness (OR 1.26), and living alone (OR 1.32). All three remained significant predictors of early mortality after adjusting for baseline health status, but they were no longer treated as interchangeable indicators of the same underlying thing.

This matters because the three measure genuinely different phenomena. Social isolation is structural — an objectively small or infrequent network, regardless of how the person feels about it. Loneliness is subjective — a felt gap between the connection someone wants and the connection they have, which can exist inside a large network as easily as a small one. Living alone is a household-composition fact that correlates with both but is neither. The 2010 meta-analysis could not distinguish which of these was doing the work in any given study. The 2015 revision could, and it found that all three did independent work, which is a stronger and more useful claim than the 2010 estimate, even though the pooled 2010 number is larger and gets quoted more.

The 2015 paper added a second finding that the 2010 pooled estimate could not have surfaced: the mortality risk associated with these deficits was more pronounced in samples with an average age under 65 than in older samples. That is a genuinely counterintuitive result, given that isolation and loneliness are usually framed as problems of aging, and it depends entirely on the disaggregation. A single pooled category would have averaged that age effect away.

Why the split stuck

The distinction did not stay confined to psychology journals. The National Academies’ 2020 consensus report on isolation and loneliness among older adults adopted the isolation/loneliness split as its organizing frame, estimating that roughly one quarter of adults 65 and older are socially isolated as a structural matter — a number that says nothing directly about how many report feeling lonely, because those are different populations answering different questions. The American Heart Association’s 2022 scientific statement on cardiovascular and brain health went further, separately reporting a roughly 29% increased risk of heart attack or death from heart disease and a 32% increased risk of stroke associated with isolation and loneliness combined, while explicitly noting that the absence of intervention trials targeting either construct specifically was the field’s central gap. The 2023 Surgeon General advisory, while it leads with the memorable comparison to smoking 15 cigarettes a day, structures its own six-pillar strategy around actions that target network structure and actions that target subjective experience as distinct problems requiring different responses.

A 2024 study in Scientific Reports made the distinction its explicit subject, examining how the relationship between isolation and loneliness varies by age group and arguing that the two should be measured and targeted separately rather than treated as proxies for one another. That paper would not have had a research question to ask if the 2015 split had not happened first.

What the split changes in practice

The practical stakes are not abstract. A social prescribing program that increases someone’s contact frequency — enrolling them in a walking group, connecting them to a volunteer visitor — addresses isolation. It does not necessarily address loneliness, because loneliness is defined by the gap between wanted and actual connection, and a person can attend a weekly group and still feel that gap. A 2022 qualitative meta-synthesis on social prescribing found that participants described benefit extending beyond social contact itself, toward restored purpose and meaningful participation — which is a way of saying that contact alone was not the mechanism; something about the quality and structure of the activity mattered more than its mere occurrence. That finding only makes sense once isolation and loneliness are held apart as different targets with potentially different intervention logics.

The same distinction shows up in workplace research. A 2024 study of remote healthcare workers explicitly separated workplace isolation from loneliness as distinct constructs with different correlates, and found that perceived social support moderated the relationship between remote work and well-being differently depending on which construct was in play. An employer response built on the assumption that isolation and loneliness are the same thing — say, mandating more video calls — could improve contact frequency while doing nothing for the subjective experience, or vice versa.

The gap that remains

None of this means the field converged on clean, standardized measurement. A 2023 review in BMC Public Health mapping the current state of loneliness and isolation research still identifies inconsistent measurement as a structural barrier to comparing findings across studies, nine years after the 2015 split. Some surveys use the UCLA Loneliness Scale, others use single-item self-report questions, and household composition data get treated as a loneliness proxy in some analyses and excluded from it in others. The 2015 revision clarified what the three constructs are conceptually. It did not force every subsequent study to measure them the same way, and the result is a literature where isolation and loneliness are understood to be distinct in principle but are still, in a large share of published work, measured as if they were not.

What would settle this more convincingly is a longitudinal design that measures network structure and subjective loneliness on the same individuals with validated instruments at repeated intervals, and follows them long enough to separate which construct predicts which downstream outcome, and on what timescale. The 2015 meta-analysis was itself a synthesis of cross-sectional and prospective studies using inconsistent measures of each construct; the split it introduced was conceptual before it was methodological. Fifteen years after the original pooled estimate, the field has a better vocabulary for the problem than it has a common instrument for measuring it, and the vocabulary is the more durable achievement.