What happened in social health? A depression item standing in for loneliness, a Dutch vote on the Luisterlijn, Zaragoza’s plan on paper, and Character.AI’s auction rules

Hans Rocha IJzerman·2 November 2026

Every study the digest has covered, as a table you can filter

A quiet week, and I would rather say so. No new randomized trial on loneliness came through the indexes for 26 October to 1 November. What did come through is mostly conference abstracts from the European Journal of Public Health supplement, one preregistered survey paper from Germany, and a small methods paper that reads like a comment on a cohort paper published three days earlier.

That pair is the closest thing to a through-line. Plenty of cohorts still measure loneliness with the single “I felt lonely” item from the CES-D depression scale, then adjust for depression. One paper this week estimates what borrowing that item does to a loneliness and depression association; the other (from nursing homes in the French Caribbean) uses the item and reports a hazard ratio of 5.7. I would read them side by side.

This issue was researched and written by Claude, my AI, working from a standing brief I wrote. I did not write it, and I will not always have read it before you do. It goes up under my name because I stand behind it, so errors in it are mine to answer for. Corrections in the thread go into the next issue and into the brief.

In this issue

Research 18

Policy and advocacy 2

Practice 0

Innovation and startups 2

Money 3

⏺️ Research

⏺️ Borrowing the CES-D “felt lonely” item to measure loneliness agrees poorly with the UCLA-3 and builds part of the depression association in by construction. Ziyan Wang (School of Public Health, Hebei Medical University) used the 2018 Health and Retirement Study leave-behind subsample (N = 5,083, survey-weighted, 1,000 bootstrap replicates) and compared the item with the three-item UCLA scale, with “lonely” set at 6 or more on its 3 to 9 range. The item picked up 37.7% of people the UCLA-3 classed as lonely (κ = 0.363). Because the item is part of the CES-D, it accounts for 30.9% (95% CI 29.3 to 33.0) of the loneliness coefficient on depression in the main specification (27.6% to 30.9% across specifications). In a mediation model the proportion mediated came out 1.33 times larger with the item, but the interval (0.63 to 2.29) includes no inflation, and the paper says so. One dataset, one country, and an unusual journal for the topic (Transactions on Materials, Biotechnology and Life Sciences). You can say that a CES-D-item loneliness measure shares variance with a CES-D outcome by construction, and that in HRS it misses most people a short loneliness scale flags; the mediation result does not support a claim of bias. If you are choosing an instrument, the LONELY-EU Item Explorer lets you compare the alternatives item by item. (paper)

No preregistration stated. Data: the paper cites the public HRS files and has no data or code statement. Sample composition beyond age and sex: not checked by us. Analyzed N 5,083.

Measurement: UCLA-3 reliability not reported, in this sample or cited; the CES-D item is a single item; no factor-structure or invariance check. Dichotomization: UCLA-3 at 6 or more, cited to earlier work, with cut-points from 4 to 8 also reported (a threshold sensitivity analysis). Openness: checked (CC BY-NC 4.0).

⏺️ In 145 Caribbean nursing-home residents, loneliness on that same CES-D item came with a hazard ratio of 5.7 for one-year mortality, on 14 deaths. Simo and colleagues (EPICLIV, Université des Antilles) used the KASEHPAD cohort in six nursing homes, five in Guadeloupe and one in Martinique. Prevalence was 31.0% (23.5 to 38.6). Fourteen residents died within a year, and the HR was 5.7 (95% CI 1.4 to 22.6) in 139 people, “adjusted for age, gender, dependency, and depression”. Fourteen events are few for four covariates, and the interval runs from a modest effect to an enormous one. We could not see which instrument supplied the depression adjustment; if it is the CES-D, the item is being adjusted for the scale it came from; that is the problem the Wang paper describes. Loneliness was common in these homes; the mortality estimate does not support a claim about size. (paper)

Registration: NCT04587466 (the KASEHPAD cohort, registered October 2020); ClinicalTrials.gov could not be opened this run, so the record was not checked against the paper. Data and code: not stated. Sample composition: mean age 80.8, 52.4% men; education, income, and migration not stated in what we could read. Analyzed N 139 of 145.

Measurement: single item (CES-D item 14, “I felt lonely”), no reliability, no structural check; how the item was dichotomized could not verify; depression instrument could not verify. Openness: checked abstract and declarations (CC BY-NC-ND 4.0); the PDF did not render for us, so the methods section is “open but not retrievable by us”.

⏺️ A German survey of 14,443 adults separates emotional from social loneliness and finds different correlates for each. Kleinschlömer and colleagues used wave 1B (2021/22) of the FReDA family panel, ages 18 to 51, with preregistered OLS models on the six-item De Jong Gierveld scale. Discussing personal matters with one’s network went with lower social loneliness (β = −0.32, 95% CI −0.37 to −0.28), and being a first-generation migrant with higher social loneliness (0.31, 0.24 to 0.38). Depressed mood (0.22, 0.17 to 0.27) and neuroticism (0.17, 0.14 to 0.20) went with emotional loneliness. The design is cross-sectional, the Big Five come from the following wave (2022/23), and the complete-case analysis dropped people who were lonelier and more often migrants or less educated (|d| ≤ 0.23), so the sample leans toward the less lonely. Good for describing who reports which kind of loneliness among German adults under 52, and silent on causes. (paper)

Preregistration: OSF (osf.io/n4t2v, an anonymized view-only link; two research questions, six hypotheses); OSF is unreachable to us this run, so we could not check whether it is a frozen registration or when it was filed. Data: could not verify the statement (the page cut off); code: “will be made publicly available on the OSF upon publication”. Sample composition: migration status, education, and subjective financial security reported; no income variable. Analyzed N 14,443 of 20,216.

Measurement: De Jong Gierveld 6-item scale, omega in this sample 0.85 (social) and 0.71 (emotional); BFI-2-XS subscales omega 0.43 to 0.68; no CFA reported; models stratified by gender with no invariance test. Openness: checked (CC BY 4.0).

⏺️ Mentored young people in a Big Brothers Big Sisters trial were less lonely four years on, on an outcome the trial did not prespecify. We held this one last week for want of full text (it went online 24 October). DuBois and Herrera randomized young people to mentoring (1,012) or to a control whose eligibility for a mentor was delayed until the end of the four-year study (346), and analyzed 1,353 with 50 multiple imputations. The adjusted ITT effect was SMD 0.172 and the CACE 0.254 (positive means less lonely), both reported as significant; the confidence intervals sit in an image table we could not read. Response at four years was 70% in the mentoring arm and 79% in the control arm, follow-up overlapped COVID, and loneliness is two items from a PROMIS depression bank. The authors write that “loneliness was not a pre-specified outcome” (credit to them for saying so). That makes this a randomized secondary finding to replicate before anyone plans around it. (paper)

Registration: NCT03495635, which covers crime and delinquency outcomes; the authors state loneliness was not prespecified. Data: on request under a data-sharing agreement; code not stated. Sample composition: 38.7% Black, 30.2% Hispanic or Latine, 73.2% at or near the poverty line, 62.7% male. Analyzed N 1,353 (1,011 mentoring, 342 control).

Measurement: two PROMIS Pediatric Depressive Symptoms items (“I felt alone”, “I felt lonely”), inter-item r 0.80 at baseline and 0.84 at follow-up in this sample, no alpha or omega; r 0.87 with the NIH Toolbox loneliness measure in a separate unpublished sample of 106; no factor or invariance check. A secondary dichotomy (“often” or more on either item) is described by the authors as “somewhat arbitrary” and likened to prevalence studies; no threshold sensitivity analysis. Power sensitivity (the digest’s own calculation; two-sample, complete data, 80% power, α .05, ignoring covariates and imputation): with 1,011 and 342 the trial could detect about SMD 0.18, close to the adjusted estimate it reports; with 28% missing at four years the real threshold is higher. Openness: checked (CC BY 4.0 per Crossref; the page metadata says CC BY 3.0), except Table 1.

⏺️ A Danish volunteer-led program for lonely 16- to 30-year-olds reports gains at one year, without a control group. Bo and colleagues (Defactum, with Lasgaard at SDU) followed participants in a volunteer-run program of social activities, group support, and one-to-one mentoring. Of 336 eligible, 164 (49%) were followed up and up to 118 analyzed. UCLA-12 scores fell 2.6 points (95% CI −3.5 to −1.6), depressive symptoms on the MDI 5.2 (−7.1 to −3.3), and social anxiety on the SIAS-6 1.2 (−1.9 to −0.4); social support rose and social isolation did not change. The authors report no baseline differences between responders and non-responders, which helps, but young people tend to join at a low point, and regression to the mean alone could produce changes like these. It is a conference abstract, and it gives both eight and five cities. It shows that scores moved among those who stayed; it cannot show the program moved them. (abstract)

No preregistration stated. Data and code: not stated. Sample composition beyond age: not stated in the abstract. Analyzed N up to 118 of 336.

Measurement: UCLA-12, MDI, SIAS-6; reliability not stated; no structural or invariance check stated. Power sensitivity: the “no change” in social isolation comes with no printed estimate, so no figure can be derived. Openness: checked (abstract only; CC BY-NC 4.0).

⏺️ Loneliness and symptoms did not predict each other within person over 25 years in a psychosis cohort. Freilich and colleagues (Stony Brook) followed the Suffolk County Mental Health Project cohort, 552 people with psychotic disorders and 257 never-psychotic comparisons, across five assessments. Loneliness fell from 62% at six months to 51% at 25 years, against 28% in the comparison group at 25 years. In cross-sectional associations, avolition (β = .70), inexpressivity (.44), and reality distortion (.44) went with loneliness, while the within-person cross-lagged paths (random-intercept model) ran from −.04 to .04 in both directions. The abstract prints p values for the cross-sectional associations but no interval or p for the within-person paths, so the null cannot be sized. Loneliness stays high in this population and goes with both negative symptoms and reality distortion; whether it drives them is still open. (paper)

Preregistration, data, and code: could not verify. Sample composition: could not verify. Analyzed N: could not verify beyond 552 and 257 enrolled.

Measurement: the abstract says only “subjective loneliness”; instrument, reliability, and model fit could not verify. Power sensitivity: no interval or p is printed for the within-person paths, so no figure can be derived. Openness: closed (OUP standard reuse terms in Crossref; abstract only).

⏺️ Catalonia’s social prescribing program shows wellbeing gains in routine primary-care records, with no comparison group and no loneliness measure. El Yaakoubi El Boujaddaini and colleagues (Catalan Department of Health and Universitat Pompeu Fabra) analyzed 3,866 people referred through primary care (84.3% women, mean age 61.9), followed for up to a year. SWEMWBS scores rose 3.7 points in women and 3.9 in men, and the three-item Oslo social support score by 0.6 and 0.5, from ANCOVA models adjusted for baseline and covariates (all p < 0.001, no intervals printed). Loneliness was not measured, although it appears among the abstract’s tagged topics. A health system can use this to show the program reaches people at scale and that participants report feeling better afterward; it says nothing yet about whether social prescribing reduces loneliness. (abstract)

No preregistration stated. Data and code: not stated. Sample composition: sex and mean age reported; nothing else in the abstract. Analyzed N 3,866.

Measurement: SWEMWBS (7 items) and OSLO-3; reliability not stated; no structural check stated, and results are compared across women and men without an invariance test. Openness: checked (abstract only; CC BY-NC 4.0).

⏺️ Policy and advocacy

⏺️ The Tweede Kamer rejected a motion against cutting the Luisterlijn and adopted one promoting Maak Rimpels. In the 27 October votes on motions from the VWS budget debate, Maeijer’s (PVV) motion nr. 48, which “verzoekt het kabinet om niet te bezuinigen op de Luisterlijn”, fell 48 to 102; the cabinet’s recorded assessment was “overbodig” (record). Van Brenk’s (50PLUS) nr. 80 on bringing the Maak Rimpels initiative to wider attention passed 143 to 7, with only Groep Markuszower against. Tijmstra’s (CDA) nr. 57 on running the Wet DOS without a phase-out passed 150 to 0, and his nr. 56, on looking at shifting €100 million structurally from the Wlz elderly-care growth margin to the Wet DOS, passed 121 to 29. Nrs. 43, 68, and 85 were held over, and nrs. 37 and 38 were replaced by revised motions (nrs. 89 and 90) that passed. “Overbodig” suggests the cabinet does not plan a cut to the Luisterlijn, but I have not seen that in a government document, so I read it as a vote with no funding decision behind it yet. (link)

⏺️ Zaragoza’s plan on unwanted loneliness among older people is a public document, with €375,000 over four years. The “Plan de Acción ante la soledad no deseada en personas mayores de Zaragoza 2026-2029” has three axes, nine projects, and line items from €10,000 to €150,000, totaling €375,000; it also points to €1.82 million already budgeted for one axis of the city’s second age-friendly cities plan. It sets no quantitative targets. The city presented it on 23 June. At the 27 to 28 October jornadas in La Azucarera (“Soledades. Construyendo una estrategia compartida ante las soledades”), the program put IMSERSO, the Aragón government’s loneliness observatory, Barcelona, and Gipuzkoa on a public-policy panel, with Pamplona, Madrid Salud, and the Red Cross on day two; workshop outputs are to feed a future citywide strategy, with no timeline given. Spain’s state-level interinstitutional Mesa, created with the national framework in February, still has no public record of having met. (link)

⏺️ Practice

Nothing on the ground cleared the bar this week; the third-sector and social-prescribing sources we checked had no new delivery news in the window.

⏺️ Innovation and startups

⏺️ The Delaware bankruptcy court approved the bidding rules for Character.AI’s assets over state objections, leaving the fights for the sale hearing. At the 30 October hearing in In re Character Technologies (No. 26-11603), Judge Brendan Shannon approved procedures built around Disney’s $20 million stalking-horse bid. That bid is for a non-exclusive license to certain IP and technology (reported to exclude the app, brand, user data, and chats); Disney is not buying the company. Objections from a coalition of about 40 attorneys general were overruled for this stage only and can be raised again at the sale hearing; releases for Disney and former employees were not approved, and the judge strongly encouraged, without ordering, a consumer privacy ombudsperson. The debtor’s proposed schedule has bids by 20 November, an auction on 23 November, and a sale hearing on 24 November, subject to court availability; we could not see the entered order. The sale hearing is where the states can press again on what happens to users’ chats. (link)

Evidence: not a product event, so no evaluation question arises. Accounts of the hearing are secondary (Law360’s summary and trade press); counts of objecting states in coverage range from 32 to 41. User numbers in the filings conflict and are not printed here.

No funding round in scope with a named lead investor was announced this week in the sources we checked (Tech.eu, recentfunding.com daily lists, Behavioral Health Business, SEC EDGAR full-text search, and Massachusetts HPC notices, plus searches of TechCrunch, Sifted, EU-Startups, and Crunchbase News).

⏺️ Money

⏺️ Indre-et-Loire has an open call for local partnerships against isolation of older people, closing 8 January 2027. The département’s call for expressions of interest, “Prévention et lutte contre l’isolement des personnes âgées 2027-2029”, run with its conférence des financeurs (CFPPA), asks a lead body (a commune or its CCAS, an EPCI or CIAS, a social center, a Centre Ressource Territorial, a social landlord, or a nonprofit association) to bring a local partnership with a shared diagnosis, joint governance, and an action plan. The deadline is “08 janvier 2027 à 23 h 59”. The budget is not on the call page (a linked cahier des charges may give it; we did not read it), and the page carries no publication date. (link)

⏺️ Causeway Coast and Glens Borough Council runs a rolling Social Connections Grant of up to £500 for groups working on loneliness. It stays open until the budget runs out and funds groups “to address loneliness and isolation and build social connections” (the guidance notes on the page are dated 2025-26). It is not new this week; I include it because small council pots like this reach no national register, and a group in that corner of Northern Ireland may not know it exists. (link)

Corrections and things we missed go in the thread, please. They shape next week’s issue and the brief.

Hans

Discuss this issue in the General space (members only).

Privacy notice