What happened in social health? A befriending trial that lands on its control, two measurement papers that disagree, Ireland's men's sheds, and a merger agreement blacked out in full
Hans Rocha IJzerman·28 September 2026
Four of this week’s items came apart when someone opened the document underneath them. A feasibility trial that reports meeting its progression criteria turns out, against its own registered thresholds, to have hit amber on the criterion that counts recruitment properly. A Dutch survey headline of 45% rests on 584 people, 92% of whom sit on the commissioning charity’s own supporter panel. The Commission’s plain-language page on its new children’s online safety proposal never mentions that Article 41 adds the regulation to the collective-redress directive. And the Swedish loneliness grant that two church-sector sites tell applicants closes on 11 October closes on the 15th.
These are ordinary gaps between a document and its summary, and closing them is most of what this digest is for. There is also a real through-line in the research. Two papers in three weeks put the structure of “social connectedness” to a confirmatory factor analysis, on 41,242 UK adults and on 779 Asian American adults, and reach different answers. Neither tests measurement invariance. Neither reports McDonald’s omega. WHO has meanwhile proposed a global index built on one of the two frameworks. Anyone planning to count social connectedness across Europe should first settle what exactly would be counted.
A note on how this is made. Claude, my AI, researches, writes and posts this digest each week from a standing brief I wrote. It goes up under my name because I stand behind it and any error in it is mine to answer for. I do not write it, and I will not always have read it before you do. Corrections and arguments in the thread feed into the next issue and into the brief itself, which is the main reason to post them here rather than email them to me.
⏺️ Research
⏺️ A UK feasibility trial of matched community participation for older adults with intellectual disability left both arms on identical loneliness scores at twelve months, and its abstract still calls the intervention promising. MATILDA randomized 57 of a 64 target (28 intervention, 29 active control) across Northern Ireland and London, matching participants to a community group with a mentor for six months. On the Modified Worker Loneliness Scale, Aloneness fell from 3.4 to 2.3 in the control arm and from 3.8 to 2.3 in MATILDA; Social Dissatisfaction landed on 1.1 in both. WHO-5 wellbeing ended slightly higher in the control arm (18.5 against 18.2). The authors performed no between-arm tests and say plainly that the study was not powered for them and that the findings “should not be interpreted as evidence of intervention efficacy”, so the null point estimates do not show that the intervention does nothing. The item is here for what the reporting leaves out. The published paper prints none of the thresholds, reports recruitment as 57/64 (89%) against target, and concludes the criteria were met; the registered criterion counts the proportion of those approached who consent, which is 57/95, or 60%, and falls in the protocol’s amber band of 40–69% rather than its green band of 70% and above. Retention at 12 months was green pooled (17/25, 68%) and red at the London site (5, 29%). A peer reviewer asked for the a priori thresholds and the published version still does not give them. If you run anything like this, the LONELY-EU Intervention Registry records designs and progression criteria alongside results, and has a contribution form for programs not yet listed.
Registered prospectively at ISRCTN 15294181, 20 December 2022; sponsor Ulster University, funder NIHR PHR 129125. The paper itself prints no registration number and cites only a protocol article (Pilot Feasibility Stud 2026;12(1):42), which carries it. Divergences from that protocol: the numeric progression thresholds are omitted; criterion 1 is reported against target instead of against those approached; the protocol was published 18 February 2026, roughly ten months after the registered trial completion date. A medRxiv preprint of the same paper gives 12-month attendance as 94% where the published version gives 68%. Data not shared, with access requests routed to Queen’s University Belfast; the published peer-review file records that the authors had indicated at submission they would release data on acceptance. Code not stated. Composition reported for sex (29 male, 28 female), age (mean 56.68, SD 8.6, range 45–81), ethnicity (White 49, Black 4, Asian 3, Other 1), living situation and site (33 Northern Ireland, 24 London); no education, income or deprivation. N = 57 analyzed, 28 and 29 per arm.
Measurement: no internal-consistency coefficient is computed in this sample for any instrument: “Cronbach”, “alpha” and “omega” appear zero times in the full text, covering the loneliness outcome, the Glasgow Anxiety Scale–ID and WHOQOL-DIS alike, all of which are supported by citation to prior literature. No factor analysis. Measurement invariance not tested, which matters here more than usual: the questionnaires were shortened mid-trial for the 6- and 12-month follow-ups, cutting WHOQOL-DIS to four items per subscale on the basis of loadings reported in a different study, so baseline and follow-up are not the same instrument and no sensitivity analysis accompanies the change. No latent-variable model, so no fit statistics. GAS-ID dichotomized at 13 or above, cited from the scale’s source and not justified here, with no threshold sensitivity analysis. Power sensitivity, the digest’s own calculation: with 28 and 29 per arm, a two-sample comparison at alpha .05 and 80% power would detect about d = 0.74, roughly 2.1 points on the Aloneness subscale at its observed SD of about 2.8; this ignores the power gained from repeated measures, and the paper states that no formal power calculation was conducted and that the target of 64 was set by the volunteer coordinator’s capacity. Openness: checked.
(paper · registration)
⏺️ Social connectedness has three dimensions on 41,242 UK adults and four on 779 Asian American adults, and neither paper tested whether its instruments behave the same way across the groups it compares. Smyth, Dulović and Waldeck fitted competing confirmatory factor models to Understanding Society waves L, M and N and retained a three-factor structure of relational connectedness, community connectedness and subjective disconnectedness, with χ²(24) = 2457, CFI .975, TLI .962, RMSEA .050 [.048, .051], SRMR .044, adequately summarized by a higher-order factor. That higher-order factor relates strongly to psychological distress (β = −.773, R² = .597), life satisfaction (β = .749, R² = .561), SWEMWBS (β = .770, R² = .594) and SF-12 mental health (β = .796, R² = .634), and weakly to SF-12 physical health (β = .175, R² = .031). The gap between .634 and .031 is the finding: the headline “half to two-thirds of variance explained” holds only for self-reported affect-laden scales measured in the same battery as the connectedness indicators, so the paper shows how these instruments hang together and offers no evidence that connectedness carries into physical health. The comparison paper, Villalonga-Olives and colleagues, is three weeks older and has had little coverage; it is the first empirical test of WHO’s structure–function–quality framework as a measurement model, on DREAMS wave 1. The social-capital model fits at CFI .991, RMSEA .058; the WHO model at CFI .990, RMSEA .071, with no RMSEA confidence interval for either and no convergence for the higher-order or bi-factor alternatives. Both models reach good fit only after indicators are dropped, and some of the drops were made on conceptual grounds. WHO quality items 6 and 7 were removed because they “aligned better” with a different dimension, so what survived is partly the authors’ reassignment of items. Anyone about to build or cite a cross-national connectedness index should read these two side by side, and the LONELY-EU Item Explorer holds the underlying instruments if you want to see what is actually being asked.
Smyth et al.: no preregistration statement. Data are Understanding Society via the UK Data Service under end-user license, which the authors may not redistribute; a supplementary file is posted (one .docx, 34.8 KB) and is claimed to contain complete analysis syntax, model specifications and variable-derivation information, which we could not open to check. Composition reported for sex (44.5% male, 55.5% female, 15 unreported) and mean birth year 1971.54 (SD 19.21) only; no ethnicity, education, income or UK region, in a paper whose central construct includes community connectedness. N = 41,242. Villalonga-Olives et al., posted 9 September 2026 and labeled older here: preregistration not mentioned; data available from the corresponding author on request; code not mentioned at all. Composition reported thoroughly for age, sex, education, income, employment, marital status, nativity (89.09% foreign-born) and English proficiency (31.58% poorly or not at all), though the Chinese, Korean and Vietnamese subgroups are named without numeric breakdown. N = 779, single metropolitan site. Neither is a trial.
Measurement: Smyth reports composite reliability computed in sample (relational CR = .65, subjective disconnection CR = .92, higher-order CR = .82) and no alpha or omega anywhere; the family-support indicator is available for 3,292 of 41,242 because of questionnaire routing, handled by FIML. Villalonga-Olives reports no reliability coefficient of any kind, neither alpha nor omega nor composite reliability, in a paper entirely about measurement quality. Dimensionality is the substance of both, and both compare competing structures empirically, which is the strongest part of each. Invariance not tested in either: Smyth pools three waves without testing across them; Villalonga-Olives tests nothing across the three ethnic subgroups, across nativity or across English-proficiency strata, and does not mention translation validation. Fit statistics as above; Smyth reports RMSEA 90% confidence intervals for all four models, Villalonga-Olives reports none for either and pre-empts the point by saying RMSEA was “interpreted cautiously”. No dichotomization in either. One reporting hazard in Smyth’s Table 4: standardized betas are paired with confidence intervals on the unstandardized scale, disclosed in the table note, so β = .749 sits beside [0.647, 0.715], an interval that does not contain its own point estimate; no standardized intervals are reported anywhere. Openness: checked for both.
(Smyth et al. · Villalonga-Olives et al.)
⏺️ Social isolation predicts mortality in 490,114 UK Biobank participants at a hazard ratio of 1.34, and how people get around barely moves it. Over a median 165 months and 42,913 deaths, isolation was associated with all-cause mortality at HR 1.34 (95% CI 1.31–1.38), cardiovascular at 1.49 (1.41–1.57), cancer at 1.22 (1.18–1.27) and chronic respiratory at 1.65 (1.49–1.83). The paper’s actual question is whether non-commute transport mode modifies that, and the answer is close to no: against car-only travel, mixed modes including a car gave an interaction HR of 0.93 (0.87–0.99) for all-cause mortality, public transport only gave 1.17 (1.00–1.37) for cardiovascular death with the interval touching one, and the authors report no evidence of modification for cancer or respiratory mortality. Both sides of the interaction are crude: isolation is a three-item index dichotomized at two or more criteria, and transport mode is a four-week recall with no frequency or duration, so an interaction hazard ratio of 0.93 contrasts two coarsely binned proxies and says nothing about dose. Read as a mortality association the estimate is consistent with a large existing literature; read as a claim that redesigning how isolated people travel would change their survival, the design carries no weight at all, because nothing here rules out that the people who take mixed modes were different before they started.
No preregistration or registration stated; sensitivity analyses are described as pre-specified with no registration named. STROBE checklist provided; ethics McGill Faculty of Medicine IRB A08-M55-24A. Data restricted, UK Biobank Application 45551; code “available upon request”, not deposited. Composition: ethnicity, Townsend deprivation, household income, education, population density, employment and urban–rural status are used as covariates or in sensitivity analyses, and decedent-versus-survivor breakdowns are reported for sex, education and income; the authors concede a highly educated, predominantly White cohort; no UK country or region breakdown. N = 490,114; not a trial. No sample-size or detectable-effect calculation is reported, and the null interaction terms for cancer and respiratory mortality are described without printed intervals, so no power sensitivity can be derived from what the paper prints.
Measurement: no psychometric instrument with a reliability coefficient is used. Social isolation is a three-item count adapted from the Berkman–Syme index (contact with family or friends at most monthly, no weekly social activity, living alone); loneliness in the sensitivity analysis is the single item “Do you often feel lonely?”, for which no internal-consistency estimate can exist. No alpha or omega for either. No factor structure or dimensionality check; the index is a conceptual count. Invariance not applicable and not tested. No latent-variable model; proportional hazards checked by visual inspection of scaled Schoenfeld residuals. Dichotomization twice: the index at two or more criteria, justified only by citation to prior method, and the loneliness item as yes or no, with no threshold sensitivity analysis and no analysis of the index as a 0–3 count. Openness: checked.
(preprint)
⏺️ In a Swedish twin cohort followed from 2004 to 2014, people already using a phone or computer daily in 2004 were less likely to land in the rising-loneliness trajectory, and the association nearly disappears once baseline loneliness is controlled. Growth mixture modeling across five waves of SATSA gave three classes: high and increasing loneliness (51 of 492, 10.4%), intermediate stable (162, 32.9%) and low stable (279, 56.7%). Daily digital use at baseline predicted lower odds of the rising class against the low stable class, OR 0.27 (95% CI 0.08–0.90, P = .03); the comparison left out of the abstract, intermediate against low stable, is OR 0.69 (0.40–1.17, P = .17). In a sensitivity analysis adding baseline loneliness, the digital-use association fell to P = .06. The exposure is measured once, in 2004, in a cohort of early adopters, and it is a binary for using either a phone or a computer daily, so the contrast is between people who were already that sort of person in 2004 and people who were not. The paper’s conclusion that digital device use “may represent a noninvasive approach to addressing loneliness” is an intervention claim, and there is no version of this design that supports it: a single baseline predictor in an observational twin cohort, with one interval running from 0.08 to 0.90 and evaporating under one covariate, describes who ended up where and cannot show what would happen if somebody were handed a tablet.
No preregistration stated. Data openly deposited in the National Archive of Computerized Data on Aging, ICPSR 3843. Code not stated. Composition reported for age (mean 69.64, SD 11.00 at Q5), sex (60.05% female), education (ISCED), living alone (33.29%), self-rated health, depressive symptoms, openness, extraversion and perceived social support; no ethnicity, income or region. Ethics Karolinska Institutet, Dnr 2007/151-31/4, 2010/657-31/3, 2015/1729-31/5. N = 771 with complete data; 492 in the trajectory subsample. Not a trial, but it makes an absence claim: power sensitivity, the digest’s own calculation from the printed standard error of 0.27 on the log-odds for the intermediate class, is that the analysis could have detected an odds ratio of about 0.59 or smaller, or 1.70 or larger, at conventional significance; the paper reports no sample-size calculation.
Measurement: loneliness is a two-item harmonized scale (“Are you ever troubled by feelings of loneliness?” and the CES-D item “How often in the last week have you felt lonely?”), each 1 to 4, summed to a 2–8 score. No reliability coefficient of any kind is reported; “Cronbach” and “omega” appear zero times. No factor analysis. Invariance not tested across the five waves, which a trajectory model rests on. Class enumeration reports log likelihood, LMR-LRT, SABIC and entropy but prints no AIC and no BIC despite naming both in the methods, and BLRT was not used; the three-class solution was chosen on minimized SABIC and on four-class entropy falling to 0.53. Three-class entropy is 0.77. No dichotomization of a continuous outcome; the exposure is a constructed binary, described above. Openness: checked.
(paper)
⏺️ A rapid review of goal-setting support for social connection in older adults found 13 studies, ran no quality appraisal, and says so in its own words. Chen, Fineberg and Ogrin searched MEDLINE and CINAHL only, screened with a single reviewer, extracted with one, restricted to English, and identified seven recurring implementation components interpreted through RE-AIM. There is no pooled estimate, no effect size and no heterogeneity statistic, because heterogeneity of designs, populations and outcomes “precluded quantitative synthesis”. Eleven of the 13 studies are from OECD countries (USA 3, UK 3, Australia 2, Canada 1, Portugal 1, Netherlands 1), with China and Malaysia making up the rest, and most interventions were delivered in urban or mixed urban–suburban settings. Taken for what it is, this maps what goal-setting interventions say they do and gives you a vocabulary for comparing them. It cannot tell you that any component works, because appearing in 13 unappraised studies says how often a component was used and nothing about its quality, and the authors state that the review “does not provide evidence of causal relationships or the comparative effectiveness of individual intervention components”. It is included because the authors report these limits so plainly.
Not registered and no separate protocol published, which the authors state explicitly; there is no PROSPERO record. Data “available on reasonable request”; search strategy and extraction framework in online supplementary material. Patient and public involvement: “None”, stated twice. No funding declared. 13 included studies.
Measurement, at review level: formal quality appraisal and risk-of-bias assessment were not undertaken, “due to the need for timely synthesis of evidence”; the reliability of the included studies’ instruments is not extracted or reported, the extraction framework covering study characteristics, context, goal-setting support, delivery and outcomes but not psychometrics; reporting bias due to missing results was not assessed, because no quantitative synthesis was attempted. Openness: checked.
(paper)
⏺️ A national analysis of England’s social prescribing link workers found gains on four patient outcomes and a precise null on isolation, and its equity companion, a conference abstract from the same team, reports “insignificant effects” in the most deprived and most urban practices while printing no confidence intervals at all. This is older, first published online 5 June 2025 and reposted as an early-access version on 3 November 2025, and it has not run here before, which is why it is in. Wilding and colleagues used 4,132,676 GP Patient Survey responses from 6,991 practices in 1,268 primary care networks between 2018 and 2023, with link worker provision measured as full-time equivalents per 50,000 patients. Good GP experience came out at OR 1.015 (95% CI 1.004–1.027), support from local services at 1.005 (1.001–1.008), and mental health needs understood at 1.012 (1.003–1.021). “Not feeling isolated from others” came out at OR 1.001 (0.996–1.005), p = 0.74, on a base where 92.0% already said they did not feel isolated. The authors write that their findings “were unable to support the NHS aim for patients to feel more connected, less lonely, or less isolated”, and attribute it partly to the survey’s question wording. Two things complicate the null. A supplementary analysis reports that networks where link workers make up a higher proportion of all additional-roles staff show a positive association with not feeling isolated. Supplementary Table S9 puts it at OR 1.041 (95% CI 1.001–1.083, p = 0.05) for all responders, 1.045 (1.002–1.089, p = 0.04) for people with a long-term condition, and 1.039 (0.990–1.090, p = 0.12) for people with a mental health condition. A lower bound of 1.001 at p = 0.05 is as marginal as a significant result gets, the association does not hold in the mental health subgroup, and the table does not say what unit of the proportion the odds ratio is per. And the companion conference abstract, from the same data with exposure scaled per 10,000 patients, reports positive significant effects across all five outcomes in rural areas (satisfaction 2.0pp, p<0.001) but effects in only one of five in the least deprived areas, concluding that “more deprived and urban practices [had] insignificant effects from link worker rollout”, with link workers themselves concentrated in the least deprived areas (concentration index 0.03, p<0.001) and no confidence intervals printed anywhere. Set that against Scotland, where a national GP survey published in July 2026 found 567 of 836 GPs with a link worker (67.8%) satisfied with their work and 587 (70.2%) believing link workers can reduce health inequalities. Practitioner belief and measured distributional effect are pointing in opposite directions, which is what to argue about before the next commissioning round.
Wilding et al.: no registration, no published protocol, no pre-specified analysis plan stated anywhere in the full text. No ethical approval required. Data cannot be shared directly; researchers may apply to use the GP Patient Survey linked with publicly available datasets. Code not stated; analyses in Stata 18. Composition beyond age and sex: ethnicity, employment status, deprivation and rurality are adjusted for, with descriptives in a supplementary table, and the authors report over-representation of older, White, female and rural responders. N = 4,132,676 responders, 6,991 practices, 1,268 networks; not a trial. The paper cites an estimated £130 million cost to the NHS in 2022/23 from NHS England sources, and says in the same sentence that this understates the total because it excludes the cost of onward referral destinations. Power sensitivity, the digest’s own calculation from the printed interval: the isolation analysis could have detected an odds ratio of about 1.006 per additional full-time-equivalent link worker per 50,000 patients, which on the paper’s own marginal-effect scale is roughly 0.06 percentage points; no sample-size calculation is reported, which is standard for a secondary analysis of this size. Donaghy et al.: no preregistration; data from the corresponding author on request; code not stated; composition reported for partner status, ethnicity, clinical sessions per week, practice list size, SIMD deprivation quintile and rurality; N = 836 for both outcome questions, from 1,380 respondents at a 31% response rate.
One internal inconsistency: the note to Table S9 says the adjustment factors are listed in Supplementary Table S7, while the published paper’s Table 1 note places them in Supplementary Table S3, and S7 holds the additional-roles summary statistics.
Measurement: the GP Patient Survey outcomes are single binary survey items, so no internal-consistency estimate exists for any of the five, and none is reported; the isolation outcome in particular is one item, which is the mechanism behind the authors’ own wording caveat. No factor structure, no invariance testing across the six survey years that are pooled, and no latent-variable model, so no fit statistics. The five outcomes are dichotomized from ordinal response scales at thresholds that are neither justified nor sensitivity-tested. In Donaghy, both headline measures are single items, satisfaction on a five-point Likert scale and the inequalities belief on four options, written for this survey, with the authors stating that “no validated questions on these items exist, as far as we are aware”; no reliability coefficient is reported, and none is computable. Openness: checked for both. Wilding’s supplementary file is missing from the PMC deposit but is on the article’s Figures & Data tab at bjgp.org.
(Wilding et al. · Agboraw et al. · Donaghy et al.)
⏺️ Policy and advocacy
⏺️ The EU KIDS Act would let qualified entities bring collective-redress actions over children’s online safety, which the Commission’s own summary of it does not mention. Article 41 of COM(2026) 681 final adds the regulation to Annex I of Directive (EU) 2020/1828, the Representative Actions Directive, and the European Parliament’s procedure file records the amendment in terms (“Amending Directive (EU) 2020/1828”). That is a materially different enforcement route from the Digital Services Act Chapter IV path in Article 34(1) that the proposal otherwise uses, and it is exactly the sort of provision that gets traded away in trilogue without anyone writing it up. Two further things the coverage has wrong or has not noticed. The lead directorate-general is GROW, under Commissioner Séjourné, not CNECT, which indicates how the Article 114 internal-market legal basis will be defended. And the widely reported “ban on social media for under-13s with a one-hour screen-time cap” came from a leaked draft: the adopted text sets a minimum age of 15 for autonomous account creation on services that meet defined risk conditions (Article 6(1)), permits guardian-created limited accounts for 13- to under-15s as a derogation (Article 6(2)), and the one-hour figure survives only as the ceiling on the daily limit a guardian must be able to set on such an account. Under-13s are dealt with separately in Article 7, which lets video-sharing services designed for that age group admit them through a guardian’s own account. Eleven days after publication the file still has no committee responsible, no committee for opinion, no rapporteur and no Council working-party document; all three committee rows read “Pending final decision on the referral”.
(procedure file · proposal text)
⏺️ Spain’s interministerial loneliness body was required to be constituted by early June, the Secretary of State said in July that it had been, and nothing about it has been published since. The Acuerdo of 24 February 2026, published as BOE-A-2026-5317, sets up the Mesa Interinstitucional de Soledades under the Secretaría de Estado de Derechos Sociales, with the Reto Demográfico secretariat and Imserso as vice-presidencies, ten civil-society seats on four-year terms drawn from four state councils, twelve central-government seats at subdirector-general rank or above, and a plenary at least annually. Paragraph Octavo required constitution within three months of the agreement taking effect. Paragraph Noveno says it runs on the ministry’s existing means, “sin que su actividad suponga gastos adicionales… ni incremento del gasto público”, which is worth holding against ministerial statements that the loneliness strategy “contará con presupuesto”. Those two are reconcilable only if the budget claim attaches to the Marco Estratégico and not to the Mesa. Rosa Martínez told Onda Cero in July that a mesa interinstitucional “se ha constituido”, describing municipal and third-sector participation that does not match the composition in the BOE. As of today there is no constitutive act in the BOE, no membership list, no meeting date, no minutes, no named secretary, and no announcement from the ministry, the Secretariat, Imserso, or any of the councils entitled to propose members. The body is constituted in law as a grupo de trabajo under article 22.3 of Ley 40/2015, the lightest form of collegiate body, for which no publication of constitution is required, so the absence of any record does not show that nothing has happened. What can be said is that the strategy carries no budget line and no quantified targets, and that its coordinating body is four months past its own constitution deadline with nothing published.
(BOE text)
⏺️ The US bill to fund social-isolation work through area agencies on aging has had one recorded action in 103 days. S.4816, the Addressing SILO Act of 2026, would amend title XX of the Social Security Act to authorize grants and training for area agencies on aging and other community organizations addressing social isolation among older people and adults with disabilities. It was introduced by Senator Murphy on 17 June 2026 with Senator Gillibrand as sole original cosponsor, read twice and referred to Finance the same day, and nothing has happened since: no hearing, no markup, no second cosponsor, no summary, and no CBO cost estimate received. The National Strategy for Social Connection Act was reintroduced the same day. Congressional inaction is not news on any given week, but a bill that routes money through the existing aging-services network instead of creating anything new, and that still cannot attract a third sponsor, is a reasonable read on where US federal appetite currently sits. Whether the referral is still the only entry when the session ends will be a clearer signal than anything said about it in the meantime.
⏺️ Practice
⏺️ A German Delphi study of 98 experts put meeting places first by a wide margin, and its most-named single measure was informal places where you do not have to buy anything. NEISS, run by the Institut für Gesundheitswissenschaften at PH Schwäbisch Gmünd under Marlen Niederberger, invited 761 experts from research, municipal practice and people with their own experience of loneliness; 150 registered and between 60 and 98 took part in individual rounds of a three-stage Delphi, rating action fields on political acceptability, practical feasibility, potential effectiveness and viability to 2031. It recommends 25 action fields in four areas. In the third round, 72% of 60 respondents put Begegnungsräume und -orte first, against 11% for knowledge and awareness, 10% for social and political participation and 5% for governance and quality assurance. The single most-named concrete measure, at 23% of nominations, was promoting informal and ideally consumption-free third places: free seating areas, chat tables with no obligation to order, upgraded libraries and public squares. The most useful finding is the gap the panel reports between effectiveness and deliverability, sharpest in the governance area: “Ohne verlässliche Finanzierung und Koordination bleibt die Wirkung vieler Maßnahmen begrenzt”, with financing across legislative periods the second most prioritized measure, and the flat recommendation that municipalities provide paid coordination because “Ehrenamtliches Engagement allein trägt Maßnahmen nicht dauerhaft”. Two caveats the authors are open about and one they are not. They say the study reflects mainly large-city perspectives and that separate funding lines for small and medium towns would make sense. They do not publish a consensus rule: “Konsens wurde erzielt, wenn Expert*innen sich in ihrer Einschätzung der potenziellen Wirksamkeit einig waren und diese hoch bewerteten”, with no agreement percentage, no interquartile-range or stability criterion and no scale points given, and no panel-composition table, so how many of the 98 were people with lived experience is unknown. Full results are promised in a journal publication.
(policy brief and Impulspapiere)
⏺️ Ireland’s department says reports that the men’s sheds grant has been dropped are “inaccurate”, and its own published payment table is headed “Grant 2024-2025”. The Irish Men’s Sheds Association wrote to members on 11 September saying Minister Dara Calleary had told a delegation at the Fianna Fáil think-in in Tullamore on 8 September that the 2025/2026 grant would not proceed and the next was expected in 2027. On 24 September the department told local radio that “no government decision has been made in relation to funding for 2025/26 and reports to the contrary are inaccurate”, described a constructive meeting with the IMSA on 22 September, said work would begin on “charting a way forward”, and added that all its funding programs are “subject to the Annual Budgetary Estimates process which is currently ongoing” so it cannot make commitments. Those two accounts are not quite contradictory: the association is reporting what a minister said in person, and the department is asserting the absence of a formal decision while declining to say the grant will proceed. What is documented is the gap. The May 2026 payment covered eligible expenditure from May 2024 to April 2025, and the department’s own shed-by-shed table is headed “Grant 2024-2025”, leaving sheds without dedicated overheads support from May 2025 to the end of 2026. Parsing that table gives numbers nobody has published: 369 sheds, total €755,163.45, which is below the €769,500 the department says it allocated for 2025 and well below the “up to €1 million” headline; 82 sheds (22%) received the €3,000 cap; the smallest award was €35.14, to Ballygar in Galway. Calleary has asked the IMSA to rework how it distributes the money after complaints that applications were “cumbersome and convoluted”, and the department is waiting on a submission. Budget 2027 is on 6 October, so this should resolve or harden within a week of publication.
(payment release and shed-by-shed table)
⏺️ The Dutch Week tegen Eenzaamheid produced two surveys, and the headline figure rests on 584 people, 92% of them recruited from the commissioning charity’s own panel. The Nationaal Ouderenfonds published its Nationaal Ouderenonderzoek 2026 on 24 September: 45% of over-65s report some loneliness and 18% strong loneliness, 26% want more contact, and of those, 43% would be helped by an invitation to visit and 41% by being asked to do something together. The factsheet states its own basis, and the basis is narrow. 628 people aged 65 and over completed it online, 578 of them from the Ouderenfonds’s own Ouderenpanel and 50 from an external panel, weighted afterwards to a Dutch distribution for age, gender and education; the loneliness scale is the De Jong Gierveld scale and is calculated on n = 584, not 628. Fielding dates are not stated and there is no methodological annex. Because it is a supporter panel and a convenience sample, it should not be quoted as a national prevalence estimate; as a repeated measure of the same panel it is more interesting, because 2025 gave 50% and 22%. A stronger line in the factsheet is political: 86% say politics pays insufficient attention to what matters to older people, and 83% want a minister for older people’s policy. A second survey, of 1,000 Dutch adults for Resto VanHarte, found 78% naming trust as a marker of real friendship and 71% being there for each other, against 26% naming how often people speak, a result that cuts against contact frequency as a program design target, though the organization has published no fieldwork agency, dates or question wording. Separately: Anja Machielse gives her afscheidscollege on 29 September at the Laurenskerk in Rotterdam, closing eleven years in the endowed chair in social resilience of older people that the Universiteit voor Humanistiek runs with the city, and presenting what she takes from 25 years of research on social relationships and loneliness. No text or recording has been announced.
(Ouderenfonds factsheet · Machielse lecture)
⏺️ Innovation and startups
No funding round in scope closed or was announced this week with a named lead investor and a verifiable amount; nothing was launched or deprecated with a social-connection mechanism actually claimed; no award or accelerator cohort in scope was announced; and there was no clearance, CE marking, UKCA marking, NICE early value assessment or DTAC assessment. On clearances specifically, openFDA’s 510(k) dataset runs roughly two weeks behind, so anything cleared inside this window would not yet be published there and its absence is not a finding. One item, and it comes from regulatory dockets rather than from anybody’s newsroom.
⏺️ Massachusetts has declined to review Sword Health’s acquisition of Headspace, Oregon’s review is open past its statutory window, and the merger agreement the companies filed is blacked out from cover to cover. (Hans co-founded Entrelacs, which sells conversational mental-health screening, care matching and continuous monitoring to employers, clinicians and health systems; the merged Sword Health/Headspace sells into the same buyers and is a direct competitor.) The Health Policy Commission’s transaction list now records the deal as received on 6 August 2026 with “No” against a cost and market impact review, consistent with its treatment of every recent behavioral-health consolidation, including Talkspace/UHS and Spring Health/Alma. Oregon is the only one of the three notified states that can approve, condition or block, and its docket still reads “currently conducting a preliminary review”. The 30-day window under OAR 409-070-0055 from its acceptance of a complete notice on 27 August would have expired around 26 September, and a response to a request for information now sits on the docket, which is the usual basis for tolling that clock. The date to use for the deal is 24 June 2026, the execution date on the merger agreement, three months before the 16 September announcement. The price is not recoverable: the agreement is withheld in its entirety as a trade secret under ORS 192.345, its 192 public pages carry nothing but Bates stamps, and the circulating “$300 million” traces to a single Axios Pro report of a $200–300M range that STAT and Healthcare Dive both declined to confirm. The new disclosure is a commitment Sword made to Oregon regulators on 21 August: it will extend outcome-based pricing to Headspace customers, defining a positive behavioral-health outcome as a reduction of at least 6 points on PHQ-9 or 4 points on GAD-7. It cites no published study in support, and as of today the deal has not closed.
Evidence: Sword filed three of its own published papers with Oregon as its clinical evidence, all of them musculoskeletal, pelvic or physical-health work, none concerning Sword Mind or mental health. Trial registries searched by sponsor and by product name: Sword Health, SA is lead sponsor on 14 ClinicalTrials.gov studies, thirteen of them musculoskeletal, rehabilitation or falls prevention, and the one mental-health entry, NCT07183085, is observational and enrolling by invitation. Headspace, Inc. and OrangeDot, Inc. are lead sponsor on zero studies; every Headspace trial in the registry is investigator-initiated, by UC Riverside, UC Merced, UCSF, Carnegie Mellon, Monash, UCL, Montefiore, Queen Mary and the University of Southern Denmark. No published evaluation of Headspace’s own product by the company was found. ISRCTN was not searched for these sponsors. No FDA clearance, CE marking under MDR, UKCA marking, NICE early value assessment or DTAC assessment was found for either product; Headspace’s Massachusetts filing describes it as a management services organization whose clinical services are delivered by affiliated medical practices, which describes a corporate structure and makes no device claim. Company-sourced and not independently verifiable: “more than 100 million people”, “20,000+ employer clients”, the 15,000-provider care network, the $200–300M price, and the Oregon activity counts supplied by the parties. No revenue figures for either party appear in any of the three state filings. The Minnesota Attorney General filing of 22 July 2026 has no public docket we could locate.
(Oregon docket · Massachusetts transaction list)
⏺️ Money
⏺️ Sweden’s national loneliness grant for civil society is open until 15 October and has 98 million kronor behind it, a figure that appears nowhere on the call page. Socialstyrelsen opened the 2027 round of Motverka ofrivillig ensamhet on 1 September under förordning (2019:474), for ideella föreningar, non-profit foundations that are neither state nor municipal, and religious communities and congregations. Money can go to activities that build community and stimulation, to creating meeting places and points of contact, and to measures that let more people volunteer or take part in voluntary activity. Decisions are planned for the first quarter of 2027, and the agency notes they may be affected by the volume of applications: 801 came in last year, against the instructions document’s stated range of roughly 750 to 800 a year. That document, Dnr 9.1–40748/2026, is where the total sits: “det totala beloppet som Socialstyrelsen fördelar för bidragsåret är 98 miljoner kronor”, conditional on the Riksdag appropriating the funds. Against roughly 800 applications that is an average of something like 122,000 kronor, which is our arithmetic; the agency publishes no average. Grants can only be spent to 31 December of the grant year, and anything at or above five prisbasbelopp needs a qualified auditor. One thing to check before you forward the link: two church-sector organizations that funnel applicants to this scheme, Equmeniakyrkan and Sociala Missionen, both currently display a deadline of 11 October, four days early. Neither page is dated, so these may be stale pages left over from an earlier round, but an applicant reading either of them loses four days.
(call page · Anvisningar 2027)
⏺️ Sweden’s municipal loneliness fund was oversubscribed 5.3 times, and not a single municipality got what it asked for. This is a closed 2025 round and it has not run here before, but the figure is agency-published and says more about unmet demand than any prevalence survey. Socialstyrelsen’s round for projects against involuntary loneliness among older people opened on 3 February 2025 and closed on 17 March; municipalities applied for 259,493,882 kronor against 48,500,000 available, and the agency states plainly that “eftersom det har funnits 48 500 000 kronor tillgängligt att fördela har det inneburit att ingen kommun har beviljats helt sitt sökta belopp”. Its rationing rule is stated too: priority went to applications focused on direct measures for older people in särskilt boende or receiving hemtjänst, and applications that looked like part of a municipality’s ordinary work, or that were not clearly enough about direct measures for the target group, were deprioritized. Few funders publish both the shortfall and the triage rule. Read the page carefully, though: the “available funds” section still says “för bidragsår 2024”, three references out of date on a page otherwise documenting the 2025 round, and the instructions PDF (Dnr 9.1–3171/2025) is the authority for the 48.5 million figure.
Nothing else opened. No new Horizon Europe topic in scope appeared this week; the CL2 2027 topics we are watching (HERITAGE-04, TRANSFO-01, TRANSFO-09) all open on 13 May 2027 and their budgets remain secondary-sourced, because the Funding and Tenders portal has now refused us five times and I will not print a figure I cannot see. Scotland’s Community Cohesion Fund application details for 2026-27 were due to be published “shortly” as of 17 September and eleven days later there is still no per-project amount, eligibility, opening or closing date; the Ministerial Taskforce on Community Cohesion still has no announced membership or start date.
Corrections, arguments and anything I have missed are welcome in the thread, and they feed both the next issue and the brief this is written from. If you know whether Spain’s Mesa has actually met, or what unit Wilding’s link-worker proportion is measured in, I would particularly like to hear it.
Hans
Discuss this issue in the General space (members only).