Social Prescribing Has an Evidence Problem, Not Just a Funding Problem

Social prescribing asks a general practitioner or nurse to refer a patient not to a specialist or a drug but to a walking group, an art class, a befriending scheme, or a community garden. The UK folded it into national loneliness policy in 2018, and it has since spread into the vocabulary of health systems well beyond Britain. The question this article asks is narrow: does the trial evidence justify that spread, or has policy moved faster than proof.

What the policy actually claims

The UK’s 2018 loneliness strategy, published by the Department for Digital, Culture, Media & Sport, was the first national loneliness strategy from any government. It did two structural things: it embedded loneliness measurement into the Office for National Statistics, and it funded social prescribing as a delivery mechanism. The logic was straightforward. Loneliness and isolation are now established as independent predictors of poor health, so a health system should have a way of acting on that finding, and social prescribing offered a route already familiar to primary care.

The scientific case for treating loneliness as a legitimate target for the health system is not in dispute. Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science, covering studies with an average age under 65, found social isolation carried an odds ratio of 1.29 and loneliness 1.26 for early mortality, effects that held after adjusting for existing health status. The 2020 National Academies consensus report estimated roughly a quarter of adults 65 and older are socially isolated and called for routine assessment within health care. The 2023 U.S. Surgeon General advisory went further, comparing the mortality risk of social disconnection to smoking up to 15 cigarettes a day. None of this is controversial among researchers who work on the topic.

What is less settled is whether social prescribing, as a specific intervention, produces the benefit that the framing implies.

What the systematic reviews find

Two systematic reviews published in 2021 looked directly at this. A review in the International Journal of Environmental Research and Public Health reported that social prescribing programmes were associated with increases in self-esteem and self-confidence among participants. A separate systematic review in Perspectives in Public Health, focused specifically on loneliness outcomes, found that all nine included studies reported positive individual-level effects, and three of those nine reported reductions in use of GP appointments, emergency care, social worker contact, or inpatient services.

Read quickly, that sounds like a clean positive result. Read carefully, it is something narrower. Nine studies is a small evidence base for a policy now embedded across an entire national health system. Both reviews describe substantial heterogeneity in programme design, in what was measured, and in how long participants were followed. Neither review is built from randomised controlled trials in the conventional sense; the wellbeing review explicitly flags limited trial evidence as a constraint on what can be concluded. A programme evaluation that finds participants feel better after attending a scheme they were referred to, with no control group and no counterfactual, is evidence of something. It is not evidence that the scheme caused the improvement, as opposed to selection into a programme by people already inclined to feel better, or the effect of simply being asked how one is doing.

A 2022 qualitative meta-synthesis in BMC Health Services Research adds a useful refinement rather than a rebuttal. It found that participants describe the benefit of social prescribing as extending beyond contact itself, to a restoration of meaningful participation and purpose. Structured, purposeful group activity appears to do more than unstructured social contact. That is a plausible and important distinction, but it is drawn from qualitative interviews about perceived benefit, not from outcome measurement against a comparison group. It tells us what social prescribing feels like to people who complete it. It does not tell us what would have happened to the same people without it, or how many people referred into a scheme never attend at all, a dropout question none of these reviews resolves.

The National Academies’ more cautious framing

The 2020 National Academies report is worth reading against this backdrop because it is more circumspect than the policy enthusiasm around social prescribing might suggest. It calls for the health care system to routinely assess social isolation and loneliness in older adults, which is a measurement recommendation, not an endorsement of any particular treatment pathway. The accompanying 2020 commentary in the American Journal of Geriatric Psychiatry, written for a clinical audience, pushes on exactly this gap: it argues for routine assessment and discusses what that would require in practice, but it does not claim the intervention side of the equation is settled. The consensus report treats identification and intervention as two separate problems, and only the first has anything like a solid evidence base behind it.

This matters because policy documents tend to compress the two. A strategy that says “loneliness predicts mortality, therefore fund social prescribing” is smuggling an intervention claim into a risk-factor claim. The risk-factor claim is well supported. The intervention claim, that referring someone to a group activity measurably improves health outcomes rather than self-reported wellbeing, remains largely untested by the standard the rest of clinical medicine would require before wide adoption.

What a randomised trial actually shows, and its limits

The clearest counter-model in this list is the HEAL-HOA trial, published in The Lancet Healthy Longevity in November 2024. It is one of the few genuine randomised controlled trials in this entire literature, rather than an uncontrolled programme evaluation, and it tested prosocial engagement and volunteering against a control condition among lonely older adults in Hong Kong. Its existence is itself informative: the fact that a single dual-arm RCT stands out as unusual in a field this large tells its own story about how thin the intervention evidence is. HEAL-HOA is not a test of social prescribing as delivered through primary care referral, and it comes from a Hong Kong sample, so it cannot be read as direct validation of the UK model. But it demonstrates that randomised designs in this space are feasible, which raises the question of why so few exist for the schemes governments are actually funding.

Holt-Lunstad’s 2021 review in the American Journal of Lifestyle Medicine argues that social connection belongs alongside diet, exercise, and smoking cessation in preventive health frameworks. That argument is about where connection sits conceptually in prevention science, not a claim that any particular delivery mechanism works. It is frequently cited in support of programmes it does not actually test.

The honest summary

The evidence supports three separate claims at three different strengths. First, that loneliness and isolation predict poor health outcomes, including mortality: strongly supported, across large meta-analyses and a federal advisory. Second, that identifying isolation and loneliness in clinical settings is feasible and worth doing: reasonably supported, mainly by the National Academies’ consensus recommendation. Third, that referring identified patients into social prescribing schemes causes measurable improvement in health outcomes, as opposed to self-reported wellbeing: weakly supported, resting on nine heterogeneous studies without randomisation and a handful of qualitative accounts of perceived benefit.

A government building policy on the first claim is on firm ground. A government funding the third at scale, on the strength of evidence built for the first, is going further than the trial literature currently allows. What would close the gap is not more programme evaluation of the kind already published, but randomised trials of social prescribing itself, with health service utilisation and mortality as outcomes rather than self-esteem scales, run over years rather than months, and with attention to who never shows up at all. HEAL-HOA shows that design is possible. It has not yet been done for the intervention that several health systems have already committed to.