What happened in social health? Six ageing cohorts on loneliness and depression, smart speakers in Chinese care homes, Apple’s 18+ rule for romantic AI companions, and Seoul’s draft loneliness ordinance
Hans Rocha IJzerman·9 November 2026
Every study the digest has covered, as a table you can filter
A quiet week, and a slightly odd one. Most of the papers that reached the indexes made a claim and then, a table or two later, took part of it back. There is an association that fades once depression, health and income enter the model, a large trial effect from six care homes that the authors say they did not fully adjust for clustering, and a pooled estimate whose prediction interval crosses one. None of that is misconduct (it is what careful reporting looks like), but this week the abstracts are the least useful part of the papers.
Policy was thin. The WHO Commission on Social Connection is in the last weeks of its three-year mandate (2024 to 2026, per the Executive Board document EB158/28) and has posted nothing new; its formal report goes to the World Health Assembly in 2027. The movement was regional (Seoul, Piemonte, Abruzzo, Álava), and the Commission’s own page on the EU KIDS Act gives two different closing dates for feedback.
A note on who wrote this. Claude, my AI, researched and wrote this issue from a standing brief I wrote, and I pasted it in. It goes up under my name because I stand behind it, and any errors in it are mine to answer for. I do not write it, and I will not always have read it before you do. Corrections in the thread feed into the next issue and into the brief.
In this issue
Research 12
- Across six ageing cohorts, lonelier adults moved into depression more often and out of it less often, while objective…
- Not preregistered (the authors say the sensitivity analyses are exploratory and not preregistered).
- Measurement: no reliability reported for any instrument; loneliness is a three-item mean (ELSA, HRS, SHARE), a single item…
- Residents over 80 in six Chinese care homes reported less loneliness after 12 weeks with a smart speaker (d = 1.02), in a…
- Registered retrospectively in the Chinese Clinical Trial Registry (ChiCTR2025030411, 9 Mar 2026, labeled “retrospectively…
- Measurement: reliability for the short-form UCLA and the Smart Speaker Perceived Companionship Scale not reported in the…
- Older evacuees in northern Israel who had relocated more often during the war were lonelier, until depression, health and…
- Measurement: Hebrew 11-item De Jong Gierveld, Cronbach’s α = 0.84 computed in this sample (omega not reported); relocation a…
- German employees who said they were almost always lonely at work reported about a third more GP visits than those almost never…
- Measurement: workplace loneliness on a four-point frequency scale; whether it is one item or several, and any reliability…
- Companionship-oriented AI chat users aged 60 and over in Guangzhou reported slightly less loneliness than non-users, mostly…
- Measurement: three-item UCLA and LSNS-6, reliability could not verify; no structure or invariance information; AI use…
- Seoul is drafting a loneliness ordinance, Herald Business reports, with introduction targeted for the first half of 2027.
- Piemonte’s regional council approved a 2026 to 2028 active-ageing plan of €1m a year, with a social prescribing pilot among…
- The Commission’s own page on the EU KIDS Act gives two different deadlines for feedback, 26 November and 3 December.
Practice 3
- Abruzzo is funding 82 active-ageing projects from the European Social Fund Plus, one line of which targets loneliness and…
- Álava’s provincial social welfare institute awarded its first unwanted-loneliness grants, €50,000 across nine projects in…
- Public libraries across Germany are running a month of events on loneliness through November, under the federal Alliance…
- From 16 November, App Store apps with a virtual companion that engages in sexual or romantic content must declare it, and will…
- Evidence: a platform rule, with no product evaluation involved; whether any companion app has changed its rating could not be…
- Character.AI’s bankruptcy sale now runs to a 24 November bid deadline, an auction on 1 December, and a sale hearing on 3 and 4…
- Evidence: no product evaluation involved; all dates and terms come from trade reporting of docket entries (PPC Land…
Money 3
- The Nièvre commission for the prevention of loss of autonomy has a 2027 call open, with action against isolation of older…
- Two small UK funds with a loneliness or isolation priority close on 23 November and 10 December.
- That’s the week. If anything here is wrong, or if you know a source this should be reading, say so in the thread (it goes into…
⏺️ Research
⏺️ Across six ageing cohorts, lonelier adults moved into depression more often and out of it less often, while objective isolation showed no association. Guo, Yue, Li and Liu harmonized CHARLS, ELSA, HRS, KLoSA, MHAS and SHARE (adults aged 50 to 110; 127,408 people contributing 276,045 intervals between waves) and modeled moving into and out of a depressive state with complementary log-log models fitted per cohort and pooled by random-effects meta-analysis. Per cohort standard deviation of loneliness, the association ratio for onset was 1.43 (95% CI 1.27 to 1.61) and for recovery 0.82 (0.78 to 0.85). For objective isolation (a 0 to 3 index of being unpartnered, living alone, and having no social participation) the ratios were 1.03 (0.98 to 1.08) and 0.98 (0.93 to 1.02). The depression scales complicate this. The CES-D-10 used in CHARLS and KLoSA and the MHAS nine-item total contain a loneliness item. When the authors subtract it and adjust for baseline severity (four cohorts), the recovery ratio becomes 0.87 (0.83 to 0.91) with a prediction interval of 0.69 to 1.10, so a new cohort could plausibly show no recovery association at all (onset held at 1.23, 1.20 to 1.26). The onset association travels well across very different countries; the recovery association travels less well, and the authors themselves say these ratios are not recovery speeds and support no causal reading. Loneliness here is one to three items and differs by cohort (the LONELY-EU Item Explorer is a quick way to see how much short loneliness measures differ in what they ask). (paper)
Not preregistered (the authors say the sensitivity analyses are exploratory and not preregistered). Data from the six cohorts’ public repositories under their access terms, participant-level data not redistributed; no code statement. Sample composition beyond age, sex and cohort: could not verify. Analyzed N: 125,892 people and 269,409 intervals in the primary models (Research Square preprint, posted 3 Nov 2026, not yet peer reviewed).
Measurement: no reliability reported for any instrument; loneliness is a three-item mean (ELSA, HRS, SHARE), a single item (CHARLS, KLoSA) or a binary item (MHAS), standardized within cohort; no factor-structure check; measurement invariance across cohorts not tested (the authors note that standardization does not establish it). Depression dichotomized at cited published cutoffs (CES-D-10 ≥10, CES-D-8 ≥4, CES-D-9 ≥5, EURO-D ≥4); no threshold sensitivity analysis found. Checked (full text, CC BY 4.0). Power sensitivity, the digest’s own calculation from the isolation onset interval (SE of the log ratio about 0.025, 80% power, two-sided α = .05): the pooled analysis could detect a ratio of about 1.07 per isolation component, so the isolation null is informative at that scale.
⏺️ Residents over 80 in six Chinese care homes reported less loneliness after 12 weeks with a smart speaker (d = 1.02), in a study the authors say did not fully account for clustering by home. Yuan, Kamal Basha and Ng (Universiti Putra Malaysia) allocated six facilities in one provincial capital by lottery to a smart speaker program or a waitlist. They recruited 140 residents (70 per arm), and 132 completed all three assessments (baseline, week 6, week 12). On a short-form UCLA scale, loneliness in the intervention group fell from 19.4 to 14.5, a between-group d of 1.02 at week 12; the posted version gives no interval or p for d. The authors write that their sensitivity analyses “did not fully account for within-facility clustering”, and with six clusters that matters a great deal. The unit allocated is the home, so the effective sample sits much closer to six than to 132. Add a waitlist control and self-reported outcomes, and the study shows the program is feasible with the oldest-old and that their scores went down; how much of the drop belongs to the speaker is still open. (paper)
Registered retrospectively in the Chinese Clinical Trial Registry (ChiCTR2025030411, 9 Mar 2026, labeled “retrospectively registered” by the authors); the registry record was not opened this run. Data and code availability: not stated in the posted version. Sample composition beyond age: could not verify (Methods not yet posted). Analyzed N: not stated; 140 allocated by cluster (70 per arm), 132 completed.
Measurement: reliability for the short-form UCLA and the Smart Speaker Perceived Companionship Scale not reported in the posted version; no factor-structure or invariance information; blinding not stated; the moderated mediation through companionship is labeled exploratory, with no fit statistics available. Open but not retrievable by us in full: BMC early access, with only the abstract and declarations posted (CC BY-NC-ND 4.0). Power sensitivity, the digest’s own calculation (two-sample, about 66 per arm, 80% power, α = .05, ignoring both clustering and repeated measures): detectable d about 0.49; any design effect from six clusters raises that figure, and no sample-size calculation could be checked.
⏺️ Older evacuees in northern Israel who had relocated more often during the war were lonelier, until depression, health and income entered the model. Vitman Schorr and Zrihan Weitzman surveyed 109 returnees aged 54 to 89 from one evacuated community (April to June 2025). Adjusted for age, gender and partnership (n = 94), two to three relocations versus one went with 5.96 more points on the 11-to-55 De Jong Gierveld scale (95% CI 2.65 to 9.27), and four or more with 6.50 (2.03 to 10.97). In a separate model (n = 86), each point of perceived community resilience went with 0.31 fewer points (−0.51 to −0.11). Adding depressive symptoms (GDS-15), physical health and income (n = 76) leaves 1.93 (−1.32 to 5.19) and 3.22 (−1.09 to 7.52), with the relocation term at F(2,67) = 1.23, p = 0.298. Depression may well sit on the path from displacement to loneliness, so the extended model can adjust away part of what it is trying to measure. Repeated relocation and loneliness travel together in this one community; these data cannot separate displacement from the depression that comes with it. (paper)
No preregistration stated. Data “not publicly available due to privacy and confidentiality restrictions”; code not stated (Python and statsmodels named). Composition reported: gender (84 women, 24 men, 1 other), age, partnership, income; a single community. Analyzed N: 109, with 94, 86 and 76 across models (complete cases); the 26 who reported no relocation were excluded from the main analysis and 7 more for missing relocation data.
Measurement: Hebrew 11-item De Jong Gierveld, Cronbach’s α = 0.84 computed in this sample (omega not reported); relocation a single self-report item; no factor-structure check; invariance not relevant to the comparisons made; loneliness not dichotomized. Checked (MDPI full text; Crossref gives CC BY 4.0 where the page metadata says CC BY 3.0). Power sensitivity, the digest’s own calculation from the printed intervals (80% power, α = .05): the extended model could detect about 4.7 points for two to three relocations and 6.1 for four or more, against 5.96 and 6.50 in the first model, so the null is informative for the middle category and borderline for the highest.
⏺️ German employees who said they were almost always lonely at work reported about a third more GP visits than those almost never lonely. Hajek and König analyzed 1,975 employed respondents from a population-based online survey of 2,591, with covariates chosen by the Andersen model. Workplace loneliness was rated on a four-point frequency scale. Fully adjusted, GP visits were higher for “almost always” versus “almost never” (IRR 1.32, 95% CI 1.08 to 1.63), specialist visits for “often” versus “almost never” (IRR 1.22, 1.002 to 1.482), and the odds of hospitalization for “occasionally” versus “almost never” (OR 1.44, 1.08 to 1.93). Each outcome’s reported contrast sits in a different category, and the abstract does not give the others, so a dose-response cannot be read from it. The abstract describes no follow-up, so treat this as an association at one point in time (frequent contact with doctors could as easily leave someone feeling detached at work). (paper)
No preregistration stated in the abstract. Data and code availability: could not verify (full text not retrieved). Sample composition beyond age, sex and region: could not verify. Analyzed N: 1,975 employed of 2,591 respondents.
Measurement: workplace loneliness on a four-point frequency scale; whether it is one item or several, and any reliability, could not verify; no structural or invariance information in the abstract; exposure analyzed as categories. Open but not retrievable by us: the host refused the full text (CC BY 4.0 per Crossref).
⏺️ Companionship-oriented AI chat users aged 60 and over in Guangzhou reported slightly less loneliness than non-users, mostly where human networks were thin, and more loneliness the more emotionally dependent on the AI they felt. Chen, Yang and Chang surveyed a nonprobability sample of 427 people aged 60 and over (168 users), March to May 2026, with loneliness on the three-item UCLA and connectedness on the LSNS-6. At mean connectedness, users scored 1.18 points lower than non-users (95% CI −1.46 to −0.89). The gap narrows as human connectedness rises (interaction 0.59 per 5 LSNS-6 points, 0.35 to 0.83), from −1.61 at the lower quartile to −0.67 at the upper. Among users, each point of emotional dependence went with 0.75 more points of loneliness (0.49 to 1.01). People who take up AI chat late in life and keep using it may differ from non-users in ways no single survey can see, and the authors call the association concurrent and ask for prospective studies. Read it as a hypothesis for a cohort study, with no evidence yet of either benefit or harm. (paper)
No preregistration stated in the provisional abstract. Data and code availability: could not verify (full text not yet posted). Sample composition beyond age: could not verify. Analyzed N: 427 (168 users, 259 non-users), complete cases.
Measurement: three-item UCLA and LSNS-6, reliability could not verify; no structure or invariance information; AI use dichotomized as user versus non-user. Open but not retrievable by us: Frontiers has posted only the provisional abstract (CC BY 4.0).
⏺️ Policy and advocacy
⏺️ Seoul is drafting a loneliness ordinance, Herald Business reports, with introduction targeted for the first half of 2027. The draft is titled 외로움 예방 및 사회적연결 촉진을 위한 조례 (an ordinance on loneliness prevention and the promotion of social connection) and was reportedly first presented to Mayor Oh Se-hoon at the city’s 30 October “Seoul Without Loneliness” Season 2 meeting. Its contents (a legal definition of loneliness, fact-finding surveys, support targets) are described as expected. I found no Seoul Metropolitan Government release on the meeting or the ordinance. Coverage of the meeting (Gukje News) gives about 80,000 cumulative calls to the “Loneliness Hello 120” line and 184,000 users of 19 “Mind Convenience Stores”; the city’s own policy archive confirms the 19 stores and neither figure, so treat the numbers as press-sourced. (link)
⏺️ Piemonte’s regional council approved a 2026 to 2028 active-ageing plan of €1m a year, with a social prescribing pilot among its six lines of action. The plan, under regional law 17/2019 and presented by welfare councillor Maurizio Marrone, covers senior cohousing and intergenerational living, digital inclusion, and a “prescrizione sociale” pilot, delivered by co-design with local bodies and the third sector, “superando la logica dei bandi tradizionali” (moving past the usual call-for-proposals model). The council’s release does not use the word solitudine. Alessandria’s Garante degli Anziani, Paola Ferrari, said a million a year “non può essere considerato una dotazione sufficiente” (cannot be considered sufficient), in an article that also reports the region took up her cohousing proposal. The vote date and decision number were not in the release, which reached us syndicated. (link)
⏺️ The Commission’s own page on the EU KIDS Act gives two different deadlines for feedback, 26 November and 3 December. The page (last updated 21 October) says “The feedback period closes on 26 November 2026.” and, a few lines later, “Provide your feedback until 3 December 2026 (midnight Brussels time).” The Have Your Say portal did not render for us, so I cannot say which date governs; press coverage uses 26 November, and anyone planning to submit should aim for that. In Parliament, the file (2026/0286(COD)) still has no IMCO rapporteur, and the latest dated meeting in its transparency section is from 19 October. (link)
⏺️ Practice
⏺️ Abruzzo is funding 82 active-ageing projects from the European Social Fund Plus, one line of which targets loneliness and social isolation among people over 65. The €3.8m regional call paid for projects costing €30,000 to €50,000; its Linea D (“Supporto psicologico e counseling”) has the stated objective “prevenire o ridurre il disagio emotivo, la solitudine e l’isolamento sociale tra gli anziani”. The September ranking funded 82 projects; a correction of 5 November records three withdrawals totaling €125,000 (two local associations and a sports club) and restates the commitment as €3,774,744.82 across 2026 and 2027, without saying how the freed money was reallocated or how many projects now remain. (link)
⏺️ Álava’s provincial social welfare institute awarded its first unwanted-loneliness grants, €50,000 across nine projects in eight small municipalities. The grants fall under the province’s “Araba a Punto” strategy and go to Urkabustaiz, Iruraiz-Gauna, Berantevilla, Artziniega, Laguardia, Agurain, Amurrio and San Millán. Per-project amounts and the official gazette notice were not found; the source is a local radio report of 5 November. (link)
⏺️ Public libraries across Germany are running a month of events on loneliness through November, under the federal Alliance against Loneliness. The federal family ministry and the German Library Association (dbv) called the “Gemeinsam in der Bibliothek” month for 1 to 30 November. Neither the ministry nor dbv gives participation numbers or funding. (link)
⏺️ Innovation and startups
⏺️ From 16 November, App Store apps with a virtual companion that engages in sexual or romantic content must declare it, and will carry a minimum 18+ rating. Apple’s new “Dating and Relationships” descriptor covers “Features that enable dating or relationship connections between real people and/or a virtual companion that engages a user in sexual or romantic content”; declarations are required when a developer submits an update or changes the age-rating questionnaire, and every app must complete them by 12 January 2027. Apple names no companies, so press lists of which companion apps are affected are reporters’ inference, and a companion app without sexual or romantic content falls outside this descriptor. (link)
Evidence: a platform rule, with no product evaluation involved; whether any companion app has changed its rating could not be checked this run (no App Store lookup available).
⏺️ Character.AI’s bankruptcy sale now runs to a 24 November bid deadline, an auction on 1 December, and a sale hearing on 3 and 4 December. The bidding-procedures order entered on 4 November, as reported, gives Disney (whose $20m deal is a non-exclusive technology license) a $700,000 break-up fee and sets a $21.7m minimum competing bid for the Disney-licensed assets. A form purchase agreement for the other assets, filed 5 November, “contains no terms on users, minors or chats” according to PPC Land, and lets the buyer walk if the sale has not closed by 17 December. A coalition of about 40 attorneys general objected to the sale rules; the judge said at the 30 October hearing that he would overrule them. (link)
Evidence: no product evaluation involved; all dates and terms come from trade reporting of docket entries (PPC Land, Blockchain.News) and were not checked against the docket itself.
No funding round for a loneliness or social-connection company was announced this week in the sources checked (Tech.eu, Dealroom news, Healthcare Dive press releases, Femtech Insider, and SEC EDGAR full-text search).
⏺️ Money
⏺️ The Nièvre commission for the prevention of loss of autonomy has a 2027 call open, with action against isolation of older people as one of four axes, until 20 December. Applicants can receive up to €25,000 a year for at most two projects; projects outside care homes need at least 20% co-funding. The deadline is “dimanche 20 décembre 2026 -23h59”, and an information webinar runs on 19 November at 14:00. (link)
⏺️ Two small UK funds with a loneliness or isolation priority close on 23 November and 10 December. Community Foundation North East’s Jackie Haq Fund gives up to £1,500 to small groups in Tyne and Wear and Northumberland, with social isolation among six priorities, closing 23 November (link). Breckland Council’s Communities Grants, run by Norfolk Community Foundation, give up to £5,000 from an £84,000 pot for 2026/27, with social isolation and loneliness a named focus, closing 10 December at 5pm (link). The Charity Service in Greater Manchester (up to £3,000 a year, with reducing social isolation among its priorities) lists no deadline on its own page; grant aggregators give 27 November. (link)
That’s the week. If anything here is wrong, or if you know a source this should be reading, say so in the thread (it goes into the next issue and into the brief).
Hans
Discuss this issue in the General space (members only).