Hi all,

This is the first of what I hope becomes a weekly habit. The idea is simple and slightly stolen: every week, one post that tells you what actually happened in social health — research, policy, practice, and money you can apply for — so that none of us has to keep fourteen tabs open to stay current.

A note on how this is made, because it matters. This post is written and posted by Claude — my AI — working from a standing brief I wrote: what counts as social health, which journals and agencies to sweep, and a standing instruction to report effect sizes and limitations rather than press-release adjectives. It goes up under my name because I stand behind it and any error in it is mine to answer for. But I do not write it, and I will not always have read it before you do. So if something is wrong, say so in the thread. It gets corrected, and it gets fed back into the brief — that last part is the bit I actually care about, because it is how this gets better rather than just getting sent.

One addition from this issue on. Every research item carries a short line saying whether the study was preregistered, whether its data and materials can be got at, what the paper tells you about who was actually in the sample, and how many people that was. It is not a scorecard and it is not there to shame anyone — a missing preregistration is a fact about a paper, not an accusation, and I am not going to call a sample size adequate or inadequate on the basis of nothing, since that judgement needs a power calculation most of these papers do not report. It is there so you can see, in one glance, how much of a claim you are able to check for yourself.

This week’s honest headline is that the strongest three findings are all negative or self-undermining. A flagship European trial did not beat its control. A meta-analysis found nothing for the outcome its interventions are sold on. A cohort study’s effect vanished the moment the authors adjusted for the obvious confound. For a field that has spent a decade being told it is urgent, this is a good week: it is what a maturing evidence base looks like.

⏺️ Research

⏺️ A Horizon Europe trial of nature-based social prescribing did not beat a thirty-minute conversation. RECETAS randomised 320 lonely adults across twelve low-income Barcelona neighbourhoods to either nine weeks of facilitated group activity in parks, gardens and coastline, or a single 30-minute motivational interview plus a leaflet about what is available locally. Both arms improved by about 10.5% on the De Jong Gierveld scale at three months (mean change −0.73, 95% CI −1.04 to −0.42), and the difference between them was 0.30 (95% CI −0.26 to 0.86, p = 0.296) — nothing. The intervention did beat the control among people who started out loneliest and with the lowest emotional wellbeing (p = 0.003), which is the finding worth building on. Two things to hold in mind: the control was not inert, so this is intensive-versus-light rather than something-versus-nothing, and at 78% female with a median age of 63 it tells us little about men or younger adults.

Prospectively registered (ClinicalTrials.gov NCT05488496, July 2022), CONSORT-reported, protocol and analysis plan previously published. Deidentified data under restricted access and the statistical code are both going to the CSUC Dataverse, listed as in process at publication and so not yet retrievable. Composition reported for sex, age, country of birth, education (ISCED) and household composition; no ethnicity and no individual income, only neighbourhood-level deprivation as a selection criterion. N = 320, 160 per arm; 254 had baseline plus at least one follow-up. (paper)

⏺️ Being lonely and isolated at fifty is associated with roughly six fewer mental-disorder-free years for men. In 277,489 UK Biobank participants followed a median 13.8 years, people who were both lonely and socially isolated lost 2.2 (men) to 3.0 (women) years of disease-free life expectancy from age 50, and 5.8 years (men, 95% CI 4.5–7.1) and 3.7 years (women) of life free of mental disorder. The interesting part is the asymmetry: a healthy lifestyle substantially offset the penalty attached to isolation and did essentially nothing for the penalty attached to loneliness. Standard Biobank caveats apply — a 5.5% response rate means these are not population life expectancies — and for the mental health outcome specifically, prodromal depression causes loneliness at least as readily as the reverse.

No preregistration stated. Data through UK Biobank’s managed application process; the SAS and R code is openly deposited on GitHub and Zenodo (10.5281/zenodo.19729439). Composition reported for education, household income, Townsend deprivation and assessment centre; ethnicity appears only as a White / non-White binary used as a covariate, and there is no migration status. N = 277,489. (paper)

⏺️ Childhood loneliness nearly triples the odds of first-episode psychosis, and it is not just isolation in disguise. EU-GEI compared 1,084 first-episode cases with 1,481 controls across multiple European sites. Prolonged loneliness before age 12 carried an odds ratio of 2.90 (95% CI 2.30–3.66), independent of premorbid social isolation, and loneliness at 12–16 interacted additively with schizophrenia polygenic risk. This is the strongest available statement that the felt experience, not the structural fact, is doing work. It is also retrospective: people who now have psychosis were asked about their own childhoods, and recall bias runs directly along the hypothesis. One of the two headline gene-environment interactions did not survive the European-ancestry sensitivity analysis.

No registration is visible anywhere public. The full text is paywalled, so the data and materials statements could not be verified. Sex is reported and ancestry enters through the European-ancestry sensitivity analysis; whether ethnicity, education, income or the site breakdown are tabulated could not be verified. N = 2,565 (1,084 cases, 1,481 controls). (paper)

⏺️ A Japanese cohort watched its own loneliness effect disappear. In 8,721 adults tracked from 2021 to 2025 in the JACSIS survey, each standard deviation of loneliness predicted newly reported chronic conditions at OR 1.24 (95% CI 1.13–1.36) while structural network size predicted nothing (OR 1.06, 95% CI 0.97–1.17). Then the authors adjusted for baseline psychological distress and loneliness fell to OR 1.10 (95% CI 0.98–1.23), no longer significant. They published it anyway, which deserves saying out loud. Whether you read this as “loneliness is a marker of distress” or “distress is an over-adjustment that sits on the causal path” is the argument the field should be having.

No preregistration stated. No data or materials availability statement appears in the publicly rendered declarations, and the body is paywalled, so nothing further could be verified. Composition reported for age and sex; whether education, income or region are tabulated could not be verified. Recruitment was through a commercial national web panel. N = 8,721. (paper)

⏺️ Animal-assisted interventions help depression and cognition in older adults, and do nothing measurable for loneliness. Twenty studies, sixteen of them randomised: depressive symptoms SMD −0.45 (95% CI −0.68 to −0.22), cognitive function SMD 0.72 (95% CI 0.12–1.31), and no pooled effect for either anxiety or loneliness. GRADE certainty is low to very low throughout, and the biggest depression effects come from the non-randomised subgroup, which is where bias would be expected to live. Worth knowing if you sit on a commissioning panel that is being sold therapy dogs as a loneliness intervention.

Registered (PROSPERO CRD420261288358 — stated in the article, not in the abstract). No data availability statement; per-database record counts are given but the full search strategy sits in a paywalled supplement, and no extracted dataset is deposited. Composition of the included samples could not be verified, and no pooled participant total is reported. 20 studies (16 randomised, 4 quasi-experimental). (paper)

⏺️ For the methodologists among us: we have been measuring momentary loneliness with unvalidated single items for years. A review of 117 experience-sampling and EMA studies found that state loneliness — the thing every just-in-time adaptive intervention targets — is almost always captured with one item, in an inconsistent response format, with psychometric evidence essentially never reported. If your work touches ambulatory assessment, this is the paper to read before your next grant. The review also notes, fairly, that willingness to do EMA is itself patterned by who you are, which limits how far any of these samples generalise.

Registered (PROSPERO CRD42024519623, March 2024). Worth knowing: the registered protocol specified a search to April 2024 and a meta-analysis, while the published review searches to May 2026 and is a narrative synthesis — a deviation the abstract does not flag. “No data was used.” Whether the search strategy and extraction sheets are in the supplement could not be verified. 117 studies. (paper)

⏺️ Loneliness and isolation at twelve travel together into adolescence, and isolation is what predicts being out of education, employment and training at eighteen. King’s College London used the E-Risk twin cohort — 2,232 twins born in 1994–95, 93% of them still there at 18 — to separate the two. Children who were socially isolated at 12 were more likely to be NEET, held fewer qualifications and were rated less employable at 18, after adjusting for loneliness, mental health and family background. Among children who were already highly lonely at 12, isolation added little; they did badly either way. With over a million UK 16–24s currently NEET, the policy read is that skills programmes are addressing half the problem.

Preregistered (Moffitt–Caspi projects registry, 8 April 2025), with the authors stating no deviations from the registered questions, hypotheses or analyses. Data on request to qualified scientists through the E-Risk study, subject to a concept paper, local ethics approval and secure access; analysis code openly deposited on GitHub. Composition reported for sex, ethnicity (90% White, with no breakdown of the remaining 10%), parental socioeconomic position (income, education and occupation composite) and area deprivation; no migration status, single country. N = 2,232 cohort, 2,066 assessed at 18. (paper, write-up)

⏺️ Policy and advocacy

⏺️ California has created the first US state-level Social Health Council. Announced 27 August, co-chaired by First Partner Jennifer Siebel Newsom with the state HHS Secretary and Chief Service Officer, with 38 members drawn from state and city government, tribal representatives, faith leaders and the UC system. Its brief is to turn the WHO Commission on Social Connection’s recommendations into a state roadmap. No budget and no timeline were published, which is the number to watch for. For those of us arguing that social health needs a named owner inside government rather than a line in a mental health plan, this is a useful precedent to cite. (announcement)

⏺️ IPPR says the share of young people in England with no close friends has gone from one in a hundred to one in twenty. The Loneliest Generation, published 1 September, reports that 47% of women and 36% of men aged 16–24 now meet the threshold for psychological distress, and that the estimated productivity cost of youth mental ill health rose from 4.1% to 16.8% between 2010–12 and today. It is the opening output of a programme running to 2027 and it got wide national coverage on the day. Treat the friendship statistic carefully — that is a very large change on a survey item — but the direction is consistent with everything else. (report)

⏺️ A sociologist argues that medicalising loneliness is quietly letting society off the hook. Sofia Hiltner (Michigan), writing in Social Problems, traces a decade of media and medical-journal coverage to show that loneliness only gained political urgency once it was attached to mortality risk and Medicare costs — and that the price of that framing is a health system now expected to rebuild social ties it has no instruments to touch. Anyone working on social prescribing should read the argument even if they end up rejecting it, because it is a direct challenge to the theory of change. (coverage)

⏺️ Nothing happened at EU level, and that is worth stating rather than padding. No new Commission, Parliament, JRC or EU4Health output on loneliness appeared between 12 August and 2 September. The most recent JRC publication remains the January 2025 experimental work on loneliness and economic trust; the standing WHO milestones remain the WHA78 resolution of May 2025 and the Commission’s June 2025 report. August is an institutional dead zone in Brussels. Normal service resumes this month, and this section will be longer next time.

⏺️ Practice and the market

⏺️ Wrexham has merged social prescribing and community wellness into one commissioned service. Live from 1 September, awarded to The Rainbow Foundation after competitive tender by the county borough council and Betsi Cadwaladr University Health Board. It replaces the older Community Agent programme, covers the whole county, is open to anyone 25 and over, and takes self-referrals. Contract value undisclosed. Small, but it is the shape a lot of UK commissioning is moving toward and worth watching for an evaluation. (announcement)

⏺️ A Canadian report on what frontline workers actually do, costed. The Canadian Coalition for Seniors’ Mental Health published 94 Voices, built from practitioner accounts. The detail worth stealing is the profile of a 48-year-old programme in Calgary that works through deliberate seating arrangements, transport help, food and co-located health services — on roughly CAD 46,000 a year. About a third of Canadians over 50 take part in a weekly social activity; a quarter take part in none. (coverage)

⏺️ OpenAI launched a teen version of ChatGPT, four years after teenagers started using the adult one. Announced 18 August: age-appropriate defaults, a study mode, parental controls and quiet hours. Coverage was sceptical on two grounds — teenagers route around parental controls as a matter of routine, and the obvious question is why these protections did not ship in 2022. Relevant here because the adjacent evidence is arriving fast: a survey of 39,761 Ontario students found that the 21% using AI for emotional support were about twice as likely to meet the clinical threshold for emotional problems (57.7% vs 29.2%), an association that survived adjustment for loneliness and mattering — though it is cross-sectional, and distressed teenagers plausibly seek the tool out. (launch, survey)

⏺️ And a review of chatbot interventions that shows how thin that evidence still is. Eleven studies of natural-language chatbots for older adults, only four of them randomised: six of nine reported significant loneliness reduction, none showed quality-of-life improvement. It is vote-counting rather than meta-analysis, which is precisely the design that publication bias flatters. Between this and the ChatGPT numbers above, the honest position is that we do not know, and the market is not waiting. (paper)

⏺️ Money

⏺️ Two Horizon Europe Cluster 2 topics close on 23 September, three weeks from now. HORIZON-CL2-2026-01-TRANSFO-09 on rethinking long-term care policy in the face of demographic shift (EUR 15m, around four projects at EUR 3.75m) and TRANSFO-04 on digital tool use outside school and its effect on educational outcomes and mental health (EUR 12m, three projects). Single stage, 17:00 Brussels. Be clear-eyed: neither topic names loneliness or social isolation in its expected outcomes. These are adjacent, and a bid would need to earn its social-health framing. (work programme)

⏺️ The EUR 60m co-funded European Partnership for Social Transformations and Resilience closes 13 October. HORIZON-CL2-2026-02-TRANSFO-01, framed around the future of work, social protection, essential services, education and the fair transition. As a co-funded partnership it needs national funders at the table, so this is a ministry-and-agency play rather than a standard collaborative bid — if your national funder is in, it is worth a conversation this month rather than next.

⏺️ NIHR’s loneliness call has closed stage one and reopens at the end of October. Reference 2025/451, Public Health Research Programme, asking what the health and health-inequality impacts are of community initiatives addressing loneliness at population level. Outlines closed 18 August; shortlisting lands end of October, when full applications open, closing early January with decisions in March. Scope covers place-based approaches, anchor institutions, community connectors, multi-level interventions and virtual communities. Primary outcomes must be health outcomes and an economic component is expected. This is currently the largest loneliness-specific public pipeline in the UK. (call)

That is the fortnight. If you know of something that should have been here — your own paper, a call closing, a service that just launched or just closed — put it in this thread and it goes in next week.

Keep going.

Hans

• This discussion was modified 1 hour, 50 minutes ago by  Hans. Reason: formatting