How the Surgeon General's Advisory Handled the Gaps in Its Own Evidence

The 2023 U.S. Surgeon General advisory on loneliness and isolation, Our Epidemic of Loneliness and Isolation, is the most widely cited document in the field, and for good reason: it compiles a genuinely large body of evidence into a single narrative and attaches a six-pillar national strategy to it. It is also, by design, a persuasive document, not a systematic review. That distinction matters for how it treats uncertainty, and it treats uncertainty unevenly.

Where the advisory is precise

On mortality risk, the advisory is careful. It states that the association between social disconnection and early death is comparable in magnitude to smoking up to 15 cigarettes a day, a figure drawn from Julianne Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science, which pooled effect sizes across 148 studies (in an earlier 2010 analysis) and reported odds ratios of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, adjusted for baseline health status. The advisory does not claim these are causal estimates from randomized evidence — because they cannot be, given that no one randomizes people into isolation — and it does not overstate the mortality comparison as a controlled experiment. It presents it as what it is: a comparison of relative risk magnitudes across two literatures that used different designs and different populations. That is the right level of caution, and it is stated once, clearly, rather than hedged in every sentence.

The advisory is similarly careful in distinguishing loneliness from isolation as constructs, a distinction the Center considers foundational and one that a surprising number of reports collapse. It treats subjective loneliness and structural isolation as separately measured and separately predictive, consistent with how the National Academies’ 2020 consensus report on older adults handled the same distinction. Roughly a quarter of adults 65 and older are isolated by objective network measures, a separate figure from the roughly half of U.S. adults the advisory says experience loneliness. These are not two ways of stating the same fact.

Where the caution thins out

The trouble starts in the sections on intervention. The advisory devotes considerable space to what should be done — its six pillars cover strengthening social infrastructure, enacting pro-connection public policies, reforming digital environments, deepening health-system knowledge, mobilizing culture, and building the evidence base. That last pillar is itself an admission: the advisory concedes, in effect, that the evidence for what actually reduces loneliness at scale is thin. But the document does not carry that concession forward consistently into the sections that recommend specific interventions.

Social prescribing is a useful test case, because it appears in the advisory as a promising practice, and it has a real evidence base to weigh against that framing. A 2021 systematic review in Perspectives in Public Health found that all nine included studies reported positive individual impacts, with three showing reduced use of GP, emergency, or social worker services. A separate 2021 review in the International Journal of Environmental Research and Public Health reported gains in self-esteem and confidence but explicitly flagged limited trial evidence and heterogeneity across programs — different countries, different referral pathways, different outcome measures, rarely a control arm. Nine studies, all showing some positive result, is suggestive. It is not the same evidentiary weight as the 308,849-person mortality meta-analysis cited two paragraphs earlier in the same document. The advisory moves between these evidence tiers without flagging the difference in confidence a reader should place in each.

The American Heart Association’s 2022 scientific statement on social isolation and cardiovascular risk, published the year before the advisory, is more explicit about this exact gap. It reports the isolation-cardiovascular associations in detail — roughly 30% increased risk of heart attack, stroke, or death from either — and then states directly that the absence of intervention evidence is the central research gap in the field. That is a stronger and more useful sentence than anything the advisory says about its own pillar six, because it names the missing study design rather than gesturing at “building the evidence base” as a future task.

What this looks like from outside

A 2023 review in BMC Public Health surveying the state of loneliness and isolation research more broadly makes the same point independently: inconsistent measurement across studies is a structural barrier to comparing findings, not an incidental one. Different surveys use different loneliness scales, different isolation indices, different reference periods. When a synthesis document draws on dozens of such studies and presents them as a single accumulating case, some of the apparent consensus is measurement convergence and some of it is measurement coincidence, and it is often hard to tell which from outside the underlying data.

None of this means the advisory’s core empirical claims are wrong. The mortality associations are well replicated across independent samples and different countries, and the advisory represents them accurately. The concern is narrower: the document is precise about the strength of the correlational evidence for harm and considerably less precise about the strength of the evidence for what fixes it, and a reader moving quickly through 80-plus pages of citations is unlikely to notice the shift in registers.

What would resolve it

A revised version, or a companion document, could do two specific things. First, tier the citations explicitly — meta-analytic mortality evidence in one category, single-arm or pre-post intervention studies in another, and controlled trials of interventions, where they exist, in a third. Second, state directly what an adequate trial of a connection intervention would need: randomization, a comparison condition, a validated loneliness or isolation instrument administered at more than one time point, and a follow-up window long enough to distinguish a durable effect from a novelty effect. Until that kind of trial exists at scale, “promising practice” is an honest label. It is a different label from “works,” and the distinction is worth preserving even in a document written to move people to action.