Effect Sizes, Not P-Values: What the Isolation Evidence Actually Licenses in Older-Adult Policy

The most cited numbers in this field are odds ratios of 1.29, 1.26, and 1.32 — social isolation, loneliness, and living alone as predictors of early mortality, from Julianne Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science. They are robust, they survive adjustment for baseline health status, and they are also, in the ordinary language of epidemiology, modest. A 26% increase in the odds of dying within a study’s follow-up window is a real signal. It is not the signal that phrases like “as deadly as smoking” imply to a minister reading a briefing note.

Policy documents tend to cite this literature for the fact that it is significant rather than for how large it is. The UK’s 2018 strategy, A Connected Society, the first national loneliness strategy published by any government, and Japan’s creation of a dedicated ministerial post in February 2021 both rest on the claim that loneliness has health consequences. That claim is well supported. What follows from a given effect size for spending decisions, screening thresholds, and target populations is a separate question, and it is answered less often.

What 1.29 buys and what it does not

Holt-Lunstad’s earlier meta-analysis, published in PLoS Medicine in 2010, pooled 148 studies and 308,849 participants and found that stronger social relationships were associated with a 50% increased likelihood of survival. That is the figure most often quoted, and the framing direction matters: it is expressed as increased odds of survival for the well-connected, not as reduced mortality for the isolated, and readers routinely convert between the two as though they were symmetric. The 2015 paper, reporting the inverse direction, lands at 1.26 to 1.32. Same research programme, same lead author, a much less dramatic-sounding set of numbers.

The American Heart Association’s August 2022 scientific statement, led by Crystal W. Cene, gives the clearest cardiovascular magnitudes available: roughly a 30% increased risk of heart attack, stroke, or death from either, decomposing into 29% for heart attack and/or death from coronary disease and 32% for stroke. The statement also finds isolation and loneliness associated with worse prognosis among people who already have coronary heart disease or stroke, including recurrent stroke and mortality.

These are relative risks. They are reported without accompanying absolute risk differences, and for older-adult policy that omission does most of the work. A 32% relative increase in stroke risk applied to a 50-year-old’s low annual baseline hazard produces a small absolute difference. Applied to an 80-year-old’s much higher baseline, the same relative figure produces a considerably larger one. Neither the AHA statement nor either meta-analysis reports stratified absolute risk differences, so this remains an inference from the shape of the age–mortality curve rather than a finding. It is, however, the inference on which the case for concentrating resources in later life most plausibly rests — and it is not the case that policy documents actually make.

The relative effect is larger below 65, which is awkward

The 2015 meta-analysis contains a result that sits badly with how the field is organised. Social deficits were more predictive of death in samples averaging under 65 than in older samples. Whatever the mechanism — differential mortality selection, competing causes of death crowding out the social signal at advanced ages, or something about the meaning of isolation at different life stages — the relative effect size does not increase with age. It decreases.

Meanwhile the institutional apparatus is overwhelmingly geriatric. The National Academies’ 2020 consensus report, Social Isolation and Loneliness in Older Adults, is addressed to the health care system’s obligations to adults 65 and over. The clinician-facing commentary in the American Journal of Geriatric Psychiatry later that year extended it into what routine assessment would require in practice. The AHA newsroom summary of the 2022 statement singles out older adults and socially vulnerable groups as facing elevated risk of isolation.

That emphasis is defensible, but it is a prevalence argument, not an effect-size argument. The National Academies put roughly one quarter of adults 65 and older in the socially isolated category. If a quarter of a population is exposed and the exposure carries an odds ratio near 1.3, the aggregate burden is substantial even where the individual-level effect is unremarkable. The honest formulation is that older adults are a priority because exposure is common and baseline cardiovascular risk is high, not because isolation is more dangerous to them per person than to someone in midlife.

The prevalence figures for younger groups make the point sharply. Harvard’s Making Caring Common survey, fielded in October 2020 and published in February 2021, found 36% of Americans reporting serious loneliness and 61% of adults aged 18 to 25 — a higher rate than the older-adult isolation figure, in a group where the 2015 meta-analysis suggests the relative mortality effect is larger. Age-targeted policy built on effect sizes alone would not look like the policy that exists.

Two constructs, two instruments, one budget line

Comparing the AARP Foundation’s 2018 survey with the National Academies’ figure illustrates why prevalence-based targeting needs care. AARP surveyed 3,020 adults 45 and older using the 20-item UCLA Loneliness Scale and found one in three lonely. The National Academies figure — one quarter of those 65 and over — describes social isolation, a structural property of a person’s network measured by contact frequency, household composition, and organisational membership.

These are not the same quantity measured twice. Isolation is countable from the outside; loneliness is a subjective state that requires asking. They are independently predictive of mortality, which is precisely why the 2015 meta-analysis reports separate odds ratios for isolation, loneliness, and living alone rather than a single pooled figure. A programme that reduces isolation without touching loneliness, or the reverse, will register on one instrument and not the other. Any policy that specifies a target without specifying which construct and which instrument has left its own evaluation undefined.

The AARP survey does contain one of the more actionable magnitudes in the literature: 33% of respondents who had spoken to their neighbours were lonely, against 61% of those who never had. That is a 28-point gap on a validated instrument, cross-sectional and therefore uninterpretable as to direction, but larger than most differences reported anywhere in this field. It also points at exactly the kind of ordinary local contact that Eric Klinenberg’s work on social infrastructure treats as consequential, and that his account of the 1995 Chicago heat wave links to survival.

The intervention ledger has almost no numbers on it

Here is the asymmetry that should govern how the risk-factor literature is used. Exposure effects are quantified across hundreds of thousands of participants. Intervention effects are barely quantified at all.

The AHA statement is explicit that the absence of intervention evidence is the central research gap. The systematic review of social prescribing and loneliness published in Perspectives in Public Health in June 2021 found that all nine included studies reported positive individual impacts, with three reporting reductions in GP, emergency, social worker, or inpatient service use. A review in the International Journal of Environmental Research and Public Health the previous month reported gains in self-esteem and self-confidence, alongside limited trial evidence and substantial heterogeneity across programmes. The 2022 qualitative meta-synthesis in BMC Health Services Research found participants describing benefit that extended beyond contact into restored purpose and meaningful participation, and suggested structured, purposeful group activity outperforms contact alone.

“All nine studies reported positive impacts” is a count of directions, not a magnitude. None of these reviews yields a pooled standardised mean difference on a loneliness instrument, a number needed to treat, or a cost per case averted. A commissioner cannot tell from this body of work whether social prescribing moves a UCLA score by half a point or by five, or whether the reported service-use reductions would survive a control group.

Holt-Lunstad’s 2021 review in the American Journal of Lifestyle Medicine argues that social connection belongs alongside diet, exercise, and smoking in preventive frameworks. The analogy is apt in one respect and misleading in another. The exposure evidence is comparable in quality; the intervention evidence is not. Smoking cessation has trials with hard endpoints and known effect sizes. Connection interventions have promising directions and unmeasured magnitudes. John Cacioppo’s account of loneliness as an aversive biological signal with downstream effects on stress physiology, sleep, and immune function gives good reason to expect intervention to help. Expectation is not measurement.

What would make this a quantitative field on both sides

Three things, none exotic. Trials of connection interventions that pre-register a primary outcome on a named instrument and report a standardised effect size, so the results can be pooled rather than counted. Observational analyses that report absolute risk differences by age band alongside relative risks, which would either justify or undermine the geriatric concentration of current policy. And measurement that keeps isolation and loneliness separate throughout, since the 2015 odds ratios differ across them and the interventions plausibly do too.

The first bilateral ministerial meeting on loneliness, between the UK and Japan in June 2021, established the policy area as a formal object of government. It has not yet established what a unit of it costs to shift.