What the Trials Actually Show About Fixing Loneliness

Since 2018, when the UK published the first national loneliness strategy, “social prescribing” has become the default policy answer to loneliness across several health systems: a clinician identifies a patient who is isolated or lonely and refers them to a community activity — a walking group, a choir, a gardening scheme — instead of, or alongside, medical treatment. The policy has spread fast. The evidence that it works has not kept pace with the spread.

That gap matters because social prescribing is now embedded in the infrastructure the UK strategy created, echoed in the National Academies’ 2020 call for health systems to routinely assess isolation, and cited in general terms by the U.S. Surgeon General’s 2023 advisory as part of the case for treating connection as a health priority. It is the kind of intervention a policymaker points to when asked what, concretely, is being done. The question worth asking in 2026 is whether the randomised evidence supports that confidence, or whether the model has outrun its evidence base.

What social prescribing trials actually find

Two systematic reviews from 2021 looked at the existing literature and reached a consistent but modest verdict. A review in the International Journal of Environmental Research and Public Health found the strongest and most reliable outcomes were increases in self-esteem and self-confidence — not reductions in loneliness itself — and flagged limited trial evidence and substantial heterogeneity across programmes as a persistent problem. A second review, in Perspectives in Public Health, looked specifically at loneliness outcomes: all nine included studies reported positive individual impacts, and three reported reductions in use of GP, emergency, social worker, or inpatient services.

Read those two findings together and the honest summary is: something is happening, and it looks favourable, but it is not being measured with the tools that would let anyone compare one programme to another, or rule out the possibility that people who agree to be referred to a walking group were already inclined to feel better. A 2022 qualitative meta-synthesis in BMC Health Services Research adds a genuinely useful distinction here: participants described the benefit as extending beyond social contact itself to a restored sense of meaningful participation and purpose, and structured, purposeful group activity appeared to work better than unstructured social contact alone. That is a real finding, but it is a finding about mechanism, not about magnitude, and it comes from interviews, not outcome measurement.

By 2025, a systematic review protocol registered on medRxiv was still stating plainly that the effectiveness of social prescribing for older adults “remains unclear despite growing adoption,” and noted that only one peer-reviewed randomised controlled trial existed in this specific area at the time of registration. That is a striking sentence to find seven years into a national strategy built partly around the intervention. It is not that social prescribing has been shown not to work. It is that most of what has been published is uncontrolled programme evaluation — pre/post comparisons among people who opted in — which is exactly the design least able to separate the effect of the programme from the effect of being the sort of person who joins one.

What the actual randomised trials say

This is where the picture gets more interesting, because a small number of genuine randomised controlled trials of loneliness interventions now exist, and they do not point where the social prescribing literature’s optimistic tone would suggest.

The HEAL-HOA trial, published in The Lancet Healthy Longevity in 2024, tested prosocial engagement and volunteering against a control condition among lonely older adults in Hong Kong — one of the few instances of a randomised design in a field otherwise dominated by small, uncontrolled studies. A 2025 randomised trial in Clinical Gerontologist tested befriending in residential aged care and found it reduced scores on the UCLA Loneliness Scale — the standard instrument in this literature — by 2.39 points at eight weeks and 2.71 points at sixteen weeks compared with a control group. Befriending, in other words, does work by the only kind of evidence that can really support that claim.

But a follow-on HEAL-HOA trial published in early 2026 in a large sample — 1,151 older adults, all living in poverty, alone, and digitally excluded — pitted a structured, telephone-delivered programme of behavioural activation and mindfulness against befriending itself as the comparison condition, rather than against no treatment. Eight thirty-minute telephone sessions over a month, delivered by trained laypeople who were themselves older adults with lived experience of loneliness. The behavioural activation and mindfulness arm produced significantly greater reductions in loneliness at twelve months than befriending did. Befriending — the intervention most social prescribing schemes actually fund, and the one the 2025 aged-care trial had just validated against no treatment — lost the head-to-head comparison to a different, more structured psychological approach.

That is the crux of the problem for policy built around social prescribing as currently practised. The connective-tissue activities social prescribing typically funds — join a walking group, attend a choir, get matched with a befriender — sit closer to the befriending end of the intervention spectrum than to structured behavioural or psychological programmes. If befriending outperforms nothing but loses to behavioural activation, then a policy that funds befriending-style referral at scale is choosing a real but second-best intervention, based on evidence that was, until very recently, almost entirely uncontrolled.

Why this keeps getting missed

Part of the reason the gap between adoption and evidence has been easy to overlook is the conflation the field keeps making between isolation and loneliness. Holt-Lunstad’s 2015 meta-analysis established that social isolation, loneliness, and living alone are independently predictive of mortality, with broadly similar effect sizes — they are not interchangeable, and an intervention that changes one does not automatically change the other. Social prescribing referrals are typically triggered by isolation — a patient lives alone, has few social contacts — but the outcome policymakers actually care about, and the one national strategies cite in headline figures, is loneliness, the subjective experience. A programme can increase someone’s number of social contacts (isolation) without touching whether they feel lonely, and the 2021 qualitative synthesis’s finding about purpose and meaningful participation suggests that gap is exactly where many uncontrolled evaluations may be picking up something real but mislabelling it.

The National Academies’ 2020 consensus report and its 2020 clinician-facing commentary both called for routine assessment of isolation and loneliness in health care settings, which is a reasonable ask on its own terms — measurement is a precondition for evaluation. But routine assessment has often been implemented alongside referral pathways whose destination, social prescribing, has not itself been tested with the rigor the assessment step implies. The WHO’s 2025 Commission on Social Connection report, which estimated 871,000 loneliness-attributable deaths annually worldwide, calls on member states to make social connection a public health priority; it does not, and could not on current evidence, specify which interventions merit that funding priority over others.

What would actually settle this

The field does not need another uncontrolled evaluation of a social prescribing scheme reporting improved self-esteem among people who chose to attend. It needs more trials structured like the two HEAL-HOA studies and the 2025 aged-care befriending trial: randomised, with an active comparison condition rather than no treatment, using a consistent instrument like the UCLA Loneliness Scale so results are comparable across studies, and following participants long enough to know whether an effect at eight weeks survives to twelve months. That last point matters specifically because the one large trial that followed participants to a year found the more structured intervention pulling ahead of the simpler one over time — a pattern that a shorter trial would have missed entirely.

Until more of that evidence exists, the honest position is this: social prescribing has plausibility, some favourable uncontrolled evidence, and a genuine mechanistic account of why it might work. It does not yet have the kind of evidence that should make it the default answer over more structured alternatives, and the one recent trial built to test that comparison directly did not go social prescribing’s way.