What happened in social health? A social-prescribing pilot for people on opioids, living alone after a partner dies in Wales, a $10m bid for the rest of Character.AI, and Loiret’s isolation call
Hans Rocha IJzerman·16 November 2026
Every study the digest has covered, as a table you can filter
Hi all,
A quieter week than the last few. The papers that stood out share a pattern I keep running into: an association that looks solid when you compare people goes quiet once you ask the question at the level where it would matter for an intervention. Loneliness sits neatly between hearing loss and cognition in a cross-section of SHARE, and the preregistered test of whether changes travel together is null. Lonelier young adults trust people less, and a lonely moment doesn’t predict a less trusting next encounter. A three-arm pilot of group CBT and social prescribing comes back with intervals wide enough to hold almost anything. None of that refutes anything (the item notes say how much each study could have seen), but it is a good week to be careful about what we say out loud.
On policy, the WHO Commission on Social Connection was set up in November 2023 for three years, and its page has nothing newer than a June webinar, so there is no closing output to report yet. The EU KIDS Act still has no rapporteur in IMCO.
This issue was researched and written by Claude, my AI, working from a standing brief I wrote, and I pasted it in here. It goes up under my name because I stand behind it, and errors in it are mine to answer for. I don’t write it, and I won’t always have read it before you do. Corrections in this thread feed into the next issue and into the brief.
In this issue
Research 7
- A three-arm pilot of group CBT and enhanced social prescribing for lonely adults on long-term opioid therapy found neither arm…
- Loneliness sat between hearing loss and cognition in a cross-section of SHARE, while the preregistered test of whether changes…
- In 14 days of experience sampling, feeling lonely at one social interaction did not predict how 114 young adults appraised the…
- Linking Welsh administrative records to the National Survey for Wales, living alone within two years of losing a partner went…
- An Australian survey of 398 adults could not separate subjective social connectedness (the positive UCLA items) from affective…
- Across five harmonized ageing cohorts in 18 countries (N = 94,122), loneliness was tied to mental disorders far more strongly…
- A short goal-setting exercise raised middle-aged Japanese employees’ stated intentions to guard against their own loneliness…
- England put £40m into GP practices to keep at-risk patients out of hospital this winter, and the National Academy for Social…
- South Korea’s winter plan checks on 566,000 vulnerable older people during cold-wave advisories, with at least one call or…
Practice 3
- Bizkaia’s provincial council funded five community projects against unwanted loneliness, for €149,099.61 in total, and turned…
- Ourense is running TERnadal for a second year, subsidized Christmas spa stays for over-50s framed as a response to unwanted…
- Marmalade Trust in Bristol is taking referrals for Christmas Day lunches and doorstep visits until 12 Dec.
- Character.AI’s bankruptcy now has a second stalking-horse bidder, for the assets outside Disney’s license.
- Tuktu, a Canadian platform matching older adults with vetted community helpers, won AGE-WELL’s 2026 National Impact Challenge.
Money 3
- Loiret’s commission des financeurs has opened its 2027 call, with an axis on isolation of older people, closing 8 Jan 2027.
- Istituto Ganassini’s #MaiSoli prize (“never alone”), €5,000 for one non-profit supporting caregivers, closes this Friday, 20…
- If anything here is wrong, or you know the source behind a number I flagged as unverified, please say so in the thread.
⏺️ Research
⏺️ A three-arm pilot of group CBT and enhanced social prescribing for lonely adults on long-term opioid therapy found neither arm clearly better than usual care at two months. Keiser, Tong and colleagues randomized 102 adults in a Washington State health system (34 per arm) who screened lonely on the UCLA-3 (6 or more) to virtual group CBT, enhanced social prescribing, or usual care, and analyzed UCLA-20 scores by intention to treat in linear mixed models; 93 (91.2%) were retained at two months. Against usual care, CBT came in at −2.24 (−9.27 to 4.80), p = .53, and social prescribing at −6.47 (−13.61 to 0.67), p = .08. Uptake was decent: 85.3% and 79.4% started, and 64.7% and 61.8% got what the authors call an adequate dose. A reader can take feasibility from this (people in pain on opioids will start these programs and mostly stay), plus an interval for social prescribing that runs from a large benefit to a slight harm; as a conference abstract from a pilot, it says nothing yet about efficacy. If you run something similar, the Intervention Registry (https://ties-community.eu/intervention-registry/, 74 approaches from 24 countries at the moment) is where we collect them, and you can add yours at https://ties-community.eu/register-an-intervention/.
Registration not stated in the abstract. Data and code availability not stated. Sample composition: mean age 54, 68.6% women, 70.6% White, 58.8% living with others; education, income and migration status not stated. Randomized 34 per arm (102); 93 retained at two months, per-arm retention not stated.
Measurement: UCLA-20 outcome and UCLA-3 eligibility screen, with no reliability reported in the abstract; no structural check; invariance not tested. The UCLA-3 cut of 6 is an eligibility threshold, and the abstract does not say how it was chosen. Openness: conference abstract checked; no open license on the supplement, and the full paper is not yet published. Power sensitivity (the digest’s own calculation, assuming the printed intervals are 95%, which the abstract does not say, though the backed-out standard errors reproduce the printed p values): standard errors are about 3.6 UCLA-20 points per contrast, so the trial had 80% power (two-sided α = .05) for differences of about 10 points; it reported 6.5 and 2.2. The null is uninformative and refutes nothing. No sample-size calculation is reported in the abstract.
(paper)
⏺️ Loneliness sat between hearing loss and cognition in a cross-section of SHARE, while the preregistered test of whether changes in the three move together was null, and Mendelian randomization found no effect of loneliness on cognition. Shuai Yang, Grilli, Wootton, van de Weijer, Treur, Klimentidis and Sbarra used 10,375 SHARE participants (Alzheimer’s & Dementia, online 28 Oct; we held it two weeks for lack of numbers, and the full text is now readable). At wave 7, worse self-rated hearing went with more loneliness (B = 0.135, 0.101 to 0.168), loneliness with lower cognition (B = −0.468, −0.535 to −0.401), and the indirect path was −0.063 (−0.081 to −0.045). The change-to-change indirect effect was B = 0.080 (SE 0.051, −0.019 to 0.179), p = .114, opposite in sign to the cross-sectional one, and the authors note the slope result depended on model specification. In the MR, genetic liability to loneliness predicted nothing in cognitive performance (B = 0.018, SE 0.166, p = .914), and in a two-step MR the share of the sensorineural hearing loss association with cognition running through loneliness was 0.004 (p = .922). You can say lonely, hard-of-hearing older Europeans score lower on cognitive tests; this paper gives no support for the idea that treating hearing protects cognition by reducing loneliness, and the authors read the genetic results as favoring shared liability over a simple causal pathway.
Preregistration: the paper says the study was preregistered on OSF and lists deviations in Supplementary Methods 3, but gives no registration ID or link, so we could not check its date or contents. Data: SHARE and the GWAS data, access governed by the original providers. Code: “will be” posted at osf.io/xqdhc (promised future deposit). Sample composition: mean age 71.4, 58.2% women, education reported (mean 11.1 years, available for 9,848); country and ethnicity not reported. Analyzed N 10,375.
Measurement: UCLA-3 (range 3 to 9) for loneliness, with no in-sample reliability reported; hearing is a single self-rated item (excellent to poor); no structural check on either; invariance across waves not tested. Openness: checked (full text via PMC), CC BY-NC-ND 4.0. Power sensitivity (the digest’s own calculation from the printed SE): the change-to-change indirect effect could be detected at 80% power only at about 0.14 in absolute size; the cross-sectional indirect effect was 0.063, from a different model, so the null cannot rule out an effect of that order.
(paper)
⏺️ In 14 days of experience sampling, feeling lonely at one social interaction did not predict how 114 young adults appraised the next one or how much they trusted the other person. Merscher, Reinhard, Wolf and Weiß (LMU Munich and Würzburg) prompted participants twice a day (92% compliance, 2,169 social-interaction reports). With loneliness centered within person and lagged to the next interaction (a median 17.5 hours later, so usually across a night), the effect on appraisal was B = −.002 (−.04 to .04) and on trust B = .01 (−.05 to .07). Between people, those who were lonelier on average reported less positive appraisals (B = −.32) and less trust (B = −.24), all on 1 to 7 scales. The predicted moderation by trait loneliness ran opposite to the hypothesis (p = .057 and .070). So people who are lonelier on average report less trust, and this study gives no sign that a lonely moment erodes trust at the next encounter, in a sample of mostly female students (98 women of 114, mean age 22.6). Version 2 went up on F1000Research on 9 Nov; version 1 dates from 4 May 2026, so this is older work in revised form.
Preregistered on OSF (10.17605/OSF.IO/R46A2); the authors name three deviations (the primary question changed, the primary models moved from concurrent to lagged, and several variables were handled differently); we could not open the registration to check its date. Data and code: OSF (10.17605/OSF.IO/VP2EJ), not opened. Sample composition: gender, age and student status reported; education beyond that, income, migration status and country not stated. Analyzed N 114 of 127.
Measurement: state loneliness is a single 1 to 7 item; trait loneliness is the revised UCLA-20 with no in-sample reliability reported (ICCs .34 for loneliness, .19 trust, .27 appraisal); no structural check; invariance across days not tested. Openness: checked, CC BY 4.0. Power sensitivity (the digest’s own calculation from the printed intervals): the lagged appraisal effect has an SE of about .02, so lagged effects of about .06 were detectable at 80% power, against between-person effects of −.32 and −.24 on the same scales.
(paper)
⏺️ Linking Welsh administrative records to the National Survey for Wales, living alone within two years of losing a partner went with about five times the odds of emotional loneliness. Johnson, Bedston, Bailey, Fry and Griffiths (Swansea, SAIL Databank) show single-person households rising from 27.0% to 29.6% of households in Wales between 2008 and 2024, then link 49,167 survey records from 43,196 people. Against people not living alone, and adjusted for demographics, deprivation, health and the built environment, living alone went with emotional loneliness at OR 1.9 (1.8 to 2.0), social loneliness at 1.2 (1.1 to 1.3), and overall loneliness at 1.5 (1.4 to 1.6). By route into living alone, emotional loneliness ran from OR 5.1 (4.3 to 6.0) within two years of partner loss, through 3.2 (2.8 to 3.8) and 2.1 (1.8 to 2.4) as the loss receded, to 1.7 (1.6 to 1.8) after separation. The built environment added little. The design is observational and largely cross-sectional, and the routes into living alone are inferred from administrative records, so the paper supports targeting recently bereaved people who live alone; it cannot say living alone causes loneliness.
No preregistration stated. Data: SAIL Databank, behind its approval panel. Code: not stated. Sample composition: ethnicity from linked 2021 Census records (about 83% linked), plus education, tenure and deprivation in the models; about 9% had no linkage key and about 6% of survey records did not link. Analyzed N 44,907, 46,635 and 44,405 across the three outcomes.
Measurement: De Jong Gierveld 6-item scale, no in-sample reliability reported, no structural check; dichotomized at 2 or more on each subscale and 4 or more overall, cited to De Jong Gierveld and Van Tilburg (2010), with no threshold sensitivity analysis. The paper labels the partner-loss time bands inconsistently across sections. Openness: checked, CC BY 4.0.
(paper)
⏺️ An Australian survey of 398 adults could not separate subjective social connectedness (the positive UCLA items) from affective social connection (r = .85), and that same positive-item factor failed fit. Verhagen, Karg, Sivasubramaniam, Yeomans and Farmer (Swinburne) fitted CFAs to the UCLA-LS v3 (a two-factor 10-item version retained), a 16-item affective dimension scale, the 8-item Social Connectedness Scale (which held up as distinct), and the Personal Wellbeing Index. The affective scale’s composite reliability of .99 and AVE of .93 suggest redundant items. Each construct alone explained 27% to 54% of the variance in wellbeing, and together 63%, with loneliness at β = −.43 and cognitive connectedness flipping from +.52 alone to −.23, which the authors call suppression. If you score the positive UCLA items as “connectedness” next to an affective-connection scale, this paper suggests you are measuring one thing twice; with 40 responses removed after collection to match the population age distribution and a single cross-section, it cannot say which construct matters more for wellbeing. The LONELY-EU Item Explorer (https://ties-community.eu/lonely-eu-resources/#measures) lets you look at the items behind these scales side by side.
No preregistration stated. Data not public (ethics). Code not stated. Sample composition: gender (39% women) and metro, regional or rural residence reported; education, income and migration status not stated. Analyzed N 398.
Measurement: CFA reported for each instrument; the positive UCLA factor failed the TLI and RMSEA thresholds; discriminant validity failed between subjective connectedness and the affective dimension; configural, metric and scalar invariance by gender supported on change-in-fit criteria, with a significant chi-square difference at the scalar step. Openness: checked, CC BY 4.0, though the renderer cut off partway through the results.
(paper)
⏺️ Across five harmonized ageing cohorts in 18 countries (N = 94,122), loneliness was tied to mental disorders far more strongly than social isolation was. Miao and colleagues (Zhengzhou) pooled HRS, ELSA, SHARE, CHARLS and MHAS for 2010 to 2020. Social isolation ranged from 17.94% (England) to 38.01% (Mexico) and loneliness from 9.11% (Switzerland) to 35.23% (Greece). Pooled average marginal effects on mental disorders were 0.243 (0.214 to 0.271) for loneliness and 0.026 (0.014 to 0.037) for isolation (about nine to one, our arithmetic; for physical disease the gap is much smaller). The cross-country prevalence table is the part to handle carefully: it assumes a short-form UCLA and an isolation index mean the same thing in Greece and Switzerland, and the abstract reports no test of that.
Preregistration, data and code: could not verify (only the abstract was retrievable). Sample composition: country reported; other dimensions could not verify. Analyzed N 94,122.
Measurement: short-form R-UCLA for loneliness and a multidimensional isolation index; thresholds, reliability and structure could not verify; cross-national invariance not reported in the abstract. Openness: open (CC BY-NC-ND 4.0) but not retrievable by us; the publisher’s full-text page refused.
(paper)
⏺️ A short goal-setting exercise raised middle-aged Japanese employees’ stated intentions to guard against their own loneliness, and left ageism unchanged. Shimizu (Seijo University) ran a pre-post study with regular employees aged 40 to 64 (N = 200), then a quasi-randomized one assigning by odd or even day of birth (388 intervention, 371 control, all analyzed). Intentions rose (B = 0.19, 0.09 to 0.29); ageism did not move (B = −0.02, −0.08 to 0.04, p = .50). Attention-check exclusions would have differed by arm (32.7% intervention, 45.0% control), which matters for the sensitivity analysis only, since the primary analysis kept everyone. The outcome is intention, measured with two new items, so the exercise shifts what people say they plan to do; whether anyone ends up less lonely is untested.
No preregistration mentioned. Data and R code on OSF (view-only link, not opened). Sample composition: age band, employment status and recruitment panel reported; education, income and migration status not stated. Analyzed N 388 and 371 in Study 2.
Measurement: the intention outcome is two new, unvalidated items (r = .57 to .70); ageism uses the 14-item Japanese short Fraboni scale, α .90 to .93 in this sample; no structural check; invariance not tested. Openness: checked, CC BY 4.0. Power sensitivity (the digest’s own calculation from the printed interval): the ageism SE of about .03 means effects of about .09 were detectable at 80% power, against an intention effect of .19 that the study did detect (on a different scale).
(paper)
⏺️ Policy and advocacy
⏺️ England put £40m into GP practices to keep at-risk patients out of hospital this winter, and the National Academy for Social Prescribing argued that link workers should have a role in it. The government release (9 Nov) gives primary care networks £126 per eligible patient for up to 300,000 people with heart disease, frailty, emphysema or other COPD, and does not mention social prescribing, link workers or loneliness. The Academy’s response says link workers “should play a vital role in programmes like this one”, points to more than 3,300 link workers in England, and cites Ways to Wellness in Newcastle, where it reports a 27% fall in secondary-care costs (the Academy’s figure; I have not checked the evaluation behind it). The government calls the money ring-fenced and the Academy calls it reallocated from existing GP budgets; neither presents it as new money. The social angle here is advocacy so far, and nothing in the scheme earmarks money for it.
⏺️ South Korea’s winter plan checks on 566,000 vulnerable older people during cold-wave advisories, with at least one call or visit a day for those at highest risk. The Ministry of Health and Welfare’s plan (10 Nov) also covers 278,000 households of older people living alone and disabled people through the ICT emergency safety service, and pays 69,000 senior centres ₩400,000 a month toward heating for five months. The release never uses the words isolation or loneliness, and gives no total budget; it is a cold-weather check-in system that, by design, reaches many older people who live alone.
(link)
⏺️ Practice
⏺️ Bizkaia’s provincial council funded five community projects against unwanted loneliness, for €149,099.61 in total, and turned down ten others. The award was published in the provincial bulletin (BOB no. 218, 12 Nov, notice I-1303). I could only read it through a bulletin digest, since the official PDF blocks automated access, so project names and per-project sums are unverified. Separately, a foral order in October raised the province’s 2026 loneliness program to €1,381,553, with a €150,000 line left unchanged; this award may come out of that line, and I could not confirm it.
(link)
⏺️ Ourense is running TERnadal for a second year, subsidized Christmas spa stays for over-50s framed as a response to unwanted loneliness. The provincial council’s scheme, per Europa Press (9 Nov), covers seven spas, with aid of €75 per person per day for registered residents of the province and €25 for others, special meals on 24, 25 and 31 Dec and 1 Jan, and applications accepted (as reported) from 23 Dec until places run out. No total budget, number of places, or evaluation of last year’s edition was given, and I found no council record to check against.
(link)
⏺️ Marmalade Trust in Bristol is taking referrals for Christmas Day lunches and doorstep visits until 12 Dec. It covers Bristol, North Somerset and South Gloucestershire; last year it reached 213 people experiencing long-term loneliness with more than 165 trained volunteers. The page is undated, so I can’t say it opened this week, but the referral deadline is useful if you work in that area.
(link)
⏺️ Innovation and startups
⏺️ Character.AI’s bankruptcy now has a second stalking-horse bidder, for the assets outside Disney’s license. The docket shows an additional stalking-horse designation filed on 14 Nov (Dkt. 164). According to PPC Land (a secondary source; the purchase agreement was not yet filed), the bidder is Smirks and Pangs Acquisition LLC, at $10m plus assumed liabilities, with a $350,000 termination payment, and the assets include the apps, models and weights, user account data and chat content. Nobody has said who owns the LLC. Per the company’s 5 Nov notice, sale objections are due 18 Nov at 4 pm ET, bids 24 Nov, the auction is 1 Dec, and the sale hearing is 3 and 4 Dec. For anyone following the coalition of about 40 attorneys general, whether users’ chat content can be sold, and to whom, now has a price attached; competing proposed orders for a consumer privacy ombudsman were filed on 12 Nov (Dkt. 148, per PPC Land), and none has been appointed yet.
Evidence line: no product evaluation at issue. The buyer’s identity, price and asset list are single-sourced to PPC Land and unverified against the filing.
⏺️ Tuktu, a Canadian platform matching older adults with vetted community helpers, won AGE-WELL’s 2026 National Impact Challenge. The prize (announced 10 Nov) is a CAD 20,000 grant plus in-kind prizes and access to AGE-WELL’s Innovation Studio. The company says its services are aimed at “fostering connection and a sense of belonging”.
Evidence line: no published evaluation found (searched the web and the company’s release); ClinicalTrials.gov could not be opened this run and ISRCTN blocked automated search, so trial registration is unchecked; no clearance or marking, and the product makes no clinical claim. The belonging claim is company-sourced with no outcome data, and an award is not evidence.
(link)
No funding round in scope was announced in Tech.eu, Behavioral Health Business or SEC EDGAR filings this week, and searches of EU-Startups, Sifted, TechCrunch and recentfunding.com found none either.
⏺️ Money
⏺️ Loiret’s commission des financeurs has opened its 2027 call, with an axis on isolation of older people, closing 8 Jan 2027. The specification (published 9 Nov) funds individual or group actions against isolation (axis 6), including mobility support and volunteer training where the local offer is missing or insufficient, alongside caregiver and other prevention axes. Any public or private body can apply (residences receiving the forfait autonomie cannot lead actions), for one year or up to three. A second window, closing 2 May 2027, is for care-home residents, led by care homes and open to older people living at home. No total envelope or ceiling is given.
(link)
⏺️ Istituto Ganassini’s #MaiSoli prize (“never alone”), €5,000 for one non-profit supporting caregivers, closes this Friday, 20 Nov, at 16:00. It was published in May, so this is a reminder; I read it through CSVnet’s listing, since the funder’s rules have no text layer.
(link)
If anything here is wrong, or you know the source behind a number I flagged as unverified, please say so in the thread.
Hans
Discuss this issue in the General space (members only).