Across 18 countries, loneliness predicted decline and death. Being alone did not.

Public health language treats loneliness and social isolation as a pair, usually joined by a slash. They are not the same measurement. Isolation is a count — how many people you see, how often, whether you live alone. Loneliness is a report — whether the connection you have feels like enough.

A registered report published in the Journal of Personality and Social Psychology separated the two across 11 longitudinal studies and 18 countries, and found that they behave differently [s1].

Why the design is unusual

Most of what is known about loneliness and health comes from individual cohorts using different loneliness scales, different covariate sets and different modelling choices. The authors note that this between-study variability in operational definitions, modelling approaches and covariate adjustments may itself be producing the mixed results in the existing literature [s1].

Their response was coordinated data analysis: fit independent but identical models across every dataset, then meta-analyse the outputs. Eleven longitudinal studies, representing participants from 18 countries, with a total sample of 175,070 [s1].

They used Cox regression, logistic regression and multistate survival models to examine loneliness, social isolation, cognitive ageing outcomes and mortality risk — including transitions between cognitive status categories, not just endpoints [s1]. Random effects meta-analyses synthesised the results [s1].

Registering the analysis plan in advance matters here. It removes the option of testing several specifications and reporting the one that worked.

The split

Loneliness was consistently associated with elevated risk of both cognitive impairment and mortality, across statistical approaches, and remained so after adjusting for social isolation [s1].

Social isolation was not consistently associated with cognitive impairment, and showed weaker associations with mortality risk [s1].

The authors' summary is that loneliness is a robust, proximal predictor of major ageing outcomes [s1] — and they draw an explicit policy implication about more efficient allocation of public health resources [s1].

What that does and does not mean

It does not mean isolation is harmless. It means that when both are measured in the same person and entered into the same model, the subjective report retains predictive power and the objective count largely does not.

It also does not establish causation. These are observational cohorts, and the direction of the arrow is genuinely uncertain for cognitive outcomes in particular. Early cognitive decline reduces the capacity to sustain relationships and can itself produce the feeling of disconnection. A multistate survival model that tracks transitions between cognitive states is better positioned to detect that sequencing than a simple endpoint model, but no observational design settles it.

What the split does have is a practical consequence. Interventions designed against isolation — transport schemes, group activities, contact frequency — are measured by whether people are in a room together. This analysis suggests the outcome those interventions should be judged on is whether the loneliness report changes, and that the two do not reliably move together.

The same pattern in a US sample

A cross-sectional study in JAMA Network Open examined where loneliness sits in the pathway from symptoms to suicidal ideation, using data collected between May 31, 2017 and October 1, 2023 from 62,685 US adults in the National Institutes of Health's All of Us Research Program [s2].

The sample had a mean age of 61.8 years (SD 16.1) and included 40,749 women, 65.0% [s2]. Anxiety symptoms were measured with the GAD-7, depressive symptoms with the first eight items of the PHQ-9, loneliness with the UCLA Loneliness Scale, and suicidal ideation with item 9 of the PHQ-9 [s2].

All three correlated with suicidal ideation: anxiety symptoms r = 0.33, depressive symptoms r = 0.39, loneliness r = 0.31, all P < .001 [s2].

The mediation results are the point. Loneliness partially mediated the association between anxiety symptoms and suicidal ideation, with a proportion mediated of 0.25 [s2], and partially mediated the association between depressive symptoms and suicidal ideation, with a proportion mediated of 0.10 [s2].

A quarter of the anxiety-to-ideation association running through loneliness is a substantial share for a variable that most clinical assessments do not ask about. The authors suggest that targeting and reducing loneliness may represent a transdiagnostic approach to interrupting the progression from anxiety and depressive symptoms toward suicidal ideation [s2].

The limits worth stating

The All of Us analysis is cross-sectional [s2]. Everything was measured at once, so the mediation model describes a statistical decomposition of correlations, not an observed sequence over time. The proportions mediated are estimates of how the variables sit relative to one another in a single snapshot.

Its sample also skews older than the US adult population, with a mean age near 62 [s2], and All of Us participants are volunteers rather than a probability sample.

What to watch

Both papers point in the same direction on measurement: the variable worth capturing is the subjective one, and it is not currently captured in most clinical or policy settings. Whether loneliness becomes a routine screening item — and whether anything reliably shifts it once identified — are the questions the intervention literature still has to answer.

This article is informational and is not medical advice.