When the Threshold Moves: What HINTS-6 Changed About Counting Loneliness

The 2025 analysis of the 2022 Health Information National Trends Survey (HINTS-6) reports that roughly 1 in 7 U.S. adults experienced severe loneliness, 1 in 5 experienced moderate loneliness, and just over 37% experienced moderate-to-severe loneliness. Three numbers, one dataset. The choice of which one becomes the headline is not a matter of measurement precision. It is a matter of where the researchers decided to draw a line on a continuous scale, and that decision alone can move the prevalence estimate by a factor of two or more.

This matters because the loneliness literature already suffers from a well-documented instrument problem: different surveys ask different questions and get different numbers, and readers routinely treat those numbers as if they were measuring the same thing. HINTS-6 introduces a second, subtler version of the same problem. Even within a single survey, using a single instrument, the prevalence figure depends on how the underlying score is cut into categories.

What HINTS-6 actually did

Most national loneliness surveys report a single prevalence figure derived from a single question or a dichotomized scale: lonely or not lonely. The Cigna Loneliness Index in 2020 asked whether respondents sometimes or always feel lonely and got 61%. Cigna’s 2025 report, using fieldwork from mid-2024, put the figure at 57%. The 2023 Surgeon General’s advisory cited roughly half of U.S. adults. These are all binary framings, even where the underlying scale (often UCLA-derived) is continuous.

HINTS-6 instead reports loneliness as a graded state: none, moderate, or severe. This is a real methodological improvement in one sense. Loneliness is not binary in lived experience, and collapsing a continuous score into a single cut point discards information about intensity and, presumably, about who most needs intervention. But it also means the “headline” figure a journalist or policymaker reaches for is now a matter of choice among at least three defensible numbers: 14% (severe only), roughly 20% (moderate), or 37% (moderate-to-severe combined). None of these is wrong. All of them describe the same underlying distribution differently.

Compare this to the CDC’s 2024 MMWR surveillance report, published using 2022 data, which frames loneliness and lack of social and emotional support as a federal surveillance category tied directly to downstream outcomes – heart disease, stroke, dementia, depression. That report does not use HINTS-6’s tiered structure. So a reader trying to reconcile “1 in 7 severely lonely” from HINTS-6 with the CDC’s own 2022-data estimates is not comparing apples to apples, even though both describe the same year and the same country.

The AARP counterexample: what happens when the instrument doesn’t move

The clearest way to see the cost of shifting thresholds is to look at a case where nothing shifted. AARP’s 2018 survey of adults 45 and older used the 20-item UCLA Loneliness Scale and reported that one in three respondents was lonely. Its 2025 follow-up, “Disconnected,” used the same instrument on the same population segment. That consistency is unusual in this literature and it is exactly what makes the AARP time series trustworthy in a way that cross-survey comparisons generally are not. When AARP reports that loneliness rose among adults 45 and older between the two waves, that claim rests on a genuinely stable measuring stick. The HINTS-6 categories, by contrast, cannot yet be compared against an earlier HINTS wave using the same tiered structure, because the tiering is new. Whatever HINTS-7 reports in a few years, the comparison will only be clean if it keeps the same three-way split rather than revising it again.

This is the general lesson the field keeps relearning: a prevalence figure is only as useful as its comparability to the next one taken with the same yardstick. The 2023 BMC Public Health review of the state of loneliness and social isolation research names inconsistent measurement as a structural barrier to the field, and the HINTS-6 reclassification is a fresh instance of that same problem, arriving from an unexpected direction – not a new question, but a new way of scoring an old one.

Why the choice of cut point is not neutral

Severity tiers are attractive because they let a report and a headline point toward the group in most acute distress – the “1 in 7” severely lonely, who plausibly resemble the population the Surgeon General’s advisory had in mind when it compared the mortality risk of social disconnection to smoking up to 15 cigarettes a day. But reporting only the severe tier undercounts everyone in the moderate range, some of whom may carry comparable health risk given that Holt-Lunstad’s earlier meta-analytic work found loneliness effects on mortality risk that did not require clinical-level severity to register. Reporting only the combined moderate-to-severe figure, meanwhile, produces a number, 37%, that sits closer to the familiar “roughly half” language from the Surgeon General and Cigna reports, inviting exactly the false equivalence the tiered structure was designed to avoid.

There is no correct answer here, only a request that gets ignored too often: report the distribution, not just the category. A single headline percentage, whichever one gets chosen, will travel through press coverage and policy briefings stripped of the caveat about where the line was drawn. The number that survives is rarely the most informative one; it is the one that fit in a headline.

What would settle this

A study that reported the full continuous UCLA or comparable score distribution alongside any categorical breakdown, rather than the categories alone, would let readers apply their own threshold and compare directly against AARP’s dichotomized figures or Gallup’s single-item measures. Better still would be a national survey that adopted one fixed instrument and one fixed threshold and repeated it on a multi-year cadence, the way AARP has managed for its 45-plus population. Until then, every claim that loneliness prevalence has risen or fallen deserves a first question that has nothing to do with sample size: is this the same yardstick as last time, or a new one dressed up as the same measurement.