One Questionnaire, Two Different Risks

The American Heart Association’s 2022 scientific statement on social isolation and cardiovascular health uses a phrase that most coverage of it drops: “objective and perceived social isolation.” The statement treats these as separate exposures with separate mechanisms, worth separate risk estimates. Most of the reporting that followed collapsed them into a single line about loneliness and heart disease. That collapse is not a rounding error. It is the most common measurement problem in this literature, and it recurs across surveys, reviews, and federal reports that otherwise handle their data carefully.

Two constructs, one questionnaire

Isolation is structural: how many people someone sees, how often, how varied that network is. It can be counted from the outside, by asking about contact frequency or network size, without ever asking how a person feels about it. Loneliness is the subjective gap between the connection someone wants and the connection they have. A person with a small network may not feel lonely at all. A person embedded in a large one may feel intensely isolated. Cacioppo and Patrick’s 2008 account of loneliness makes this the starting point of the whole argument: loneliness is a signal, akin to hunger, that can fire independently of a person’s actual circumstances.

Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science is unusually disciplined about keeping the two apart. It reports separate pooled odds ratios: 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, each predicting early mortality. The three measures move together but are not interchangeable, and the paper treats them as three distinct risk factors rather than three estimates of the same thing. That distinction matters more than it sounds. If isolation and loneliness carried identical risk and identical causes, a single combined index would be a defensible shortcut. Because they do not, combining them can obscure which lever an intervention is supposed to pull.

Where the conflation happens in practice

The National Academies’ 2020 consensus report on older adults is careful in its own text, defining isolation and loneliness separately and estimating that roughly one quarter of adults 65 and older are socially isolated by objective network measures. But the figure that circulated in coverage and policy citation afterward was frequently rendered as “a quarter of older adults are lonely,” which is not what the report measured. The instrument behind that quarter-figure counts contact and network size. It says nothing about how those adults felt about their circumstances.

The AARP Foundation’s 2018 survey of adults 45 and older avoids this by using the 20-item UCLA Loneliness Scale, a subjective instrument, and being explicit that the resulting one-in-three figure is a loneliness estimate, not an isolation count. That survey also reports a structural correlate — 33% of respondents who had spoken with a neighbor in recent weeks were lonely, against 61% of those who had never done so — which is itself an interesting demonstration of how isolation and loneliness relate without being identical: contact reduces the odds of loneliness considerably but does not eliminate it, and a third of the well-connected group was lonely anyway.

The CDC’s June 2024 surveillance report is a useful recent case of the conflation happening at the level of report design rather than analysis. Its title bundles “loneliness” and “lack of social and emotional support” into one surveillance product drawn from 2022 data, and the resulting statistics are presented as a single connected narrative about mental health risk. Lack of social and emotional support is closer to a structural measure — whether a person has someone to call on — while loneliness is the subjective item. The report’s own tables keep the two variables distinct, which is good practice, but a reader skimming the topline is likely to walk away with one number where the underlying data supports two.

Why this is not pedantry

The practical stakes are about what an intervention is built to fix. A program that increases contact frequency — a phone-based check-in service, a transportation subsidy that gets someone to a senior center — is aimed squarely at isolation. It may do nothing for loneliness if the person’s dissatisfaction was never about frequency of contact but about the quality or reciprocity of it. Conversely, a therapeutic or arts-based program aimed at loneliness may leave someone’s actual network size untouched while still reducing distress. Evaluating either kind of program against the wrong construct will produce a false null. If a contact-frequency program is scored against a loneliness scale, and loneliness has other unaddressed drivers, the trial will look like it failed when the isolation metric it was actually designed to move might have shifted substantially.

The 2023 review in BMC Public Health surveying the state of loneliness and social isolation research names inconsistent measurement as a structural barrier to comparing findings across studies, without resolving which inconsistency matters most. The isolation/loneliness split is one candidate answer. A study using a network-size instrument and a study using the UCLA scale are not measuring the same underlying phenomenon even when both papers use the word “loneliness” in their titles or discussion sections.

What would settle this

None of the sources reviewed here report a single study that measures both constructs on the same sample using validated instruments for each, then separately tracks which one predicts which outcome, and which one an intervention actually moves. The AHA statement comes closest by naming isolation and loneliness as distinct exposures in its risk estimates, but it explicitly flags the absence of intervention evidence as its central gap — meaning even that statement cannot say whether a program built to reduce objective isolation also reduces perceived loneliness, or whether the two require entirely different designs.

A study that would close this gap would need three things current work rarely combines: a validated structural measure of network size and contact frequency, a validated subjective loneliness scale such as the UCLA instrument, and repeated measurement before and after an intervention so that the two trajectories could be compared rather than assumed to move together. Until that exists, reports that quote a single “loneliness” figure without specifying which instrument produced it, and what that instrument was built to detect, are asking one number to answer two different questions.