Match the Intervention to the Problem, Not the Referral Category

A GP practice refers a housebound 78-year-old to a befriending service. A social prescribing link worker refers a 40-year-old who lives with three other adults but says she has no one to talk to. Both referrals land in the same “loneliness” category on the commissioner’s spreadsheet. They should not.

The first person has an isolation problem: too little social contact, structurally. The second has a loneliness problem: contact is present, but it does not feel like connection. These are different conditions with different mechanisms, and the intervention evidence increasingly shows they do not respond to the same fix.

Two constructs, not one

Julianne Holt-Lunstad’s 2015 meta-analysis separated these out formally: social isolation carried an odds ratio of 1.29 for early mortality, loneliness 1.26, and living alone 1.32 — comparable magnitudes, but three distinct measures, not interchangeable proxies for each other. Isolation is objective and countable: how many people you see, how often. Loneliness is subjective: the gap between the connection you have and the connection you want. John Cacioppo’s framing of loneliness as an aversive internal signal, similar to hunger, is useful here precisely because it locates the problem inside the person’s appraisal of their situation, not in a headcount of their contacts.

A 2024 study in Scientific Reports examined how isolation and loneliness interact differently by age, and a 2026 study of risk factors in socially isolated older adults reached a similar conclusion from the other direction: isolation does not reliably predict who feels lonely. Some isolated people are fine. Some densely connected people are miserable. The National Academies’ 2020 consensus report on older adults made the operational point explicit: health systems need to assess isolation and loneliness separately, because routing everyone through one instrument misses half the picture.

What this means in practice: before you assign someone to a befriending scheme, a volunteering programme, or a structured group activity, screen for which problem you’re actually looking at. A contact-frequency question (how many days in the past week did you have a meaningful conversation) tells you about isolation. A UCLA Loneliness Scale item or equivalent tells you about the subjective gap. Do not let one stand in for the other on an intake form.

What the evidence says works for which

The intervention literature is thin and mostly uncontrolled, but a small number of randomised trials now let you say something more precise than “social contact helps.”

Claim	Evidence status
Isolation and loneliness are distinct constructs requiring separate measurement	Well supported — consistent across Holt-Lunstad’s meta-analysis and multiple 2024–2026 studies
Befriending reduces loneliness scores in aged care settings	Supported by one RCT (2025), effect size modest (2.4–2.7 points on UCLA scale)
Structured psychological approaches (behavioural activation, mindfulness) outperform plain befriending for loneliness	Supported by one large RCT (HEAL-HOA, 2026, n=1,151) — befriending was the losing comparison arm
Volunteering/prosocial engagement reduces loneliness in older adults	Supported by one RCT (HEAL-HOA dual trial, 2024), conducted in Hong Kong
Structured, purposeful group activity outperforms unstructured contact for social prescribing outcomes	Suggestive — qualitative meta-synthesis (2022), not a trial
Social prescribing reliably reduces loneliness in older adults generally	Unclear — a 2025 systematic review protocol notes only one peer-reviewed RCT exists in this area despite widespread adoption

The pattern worth sitting with: befriending is not a null intervention. It works, modestly, on its own terms. But in the one trial that pitted it directly against a more structured alternative — eight telephone sessions of behavioural activation and mindfulness delivered to 1,151 older adults living in poverty and digitally excluded — the structured approach won on loneliness reduction at twelve months. Befriending was the control condition it beat. If your only tool is befriending, you are giving people a real but comparatively weak intervention, and you should know that going in rather than discover it from a disappointing evaluation.

The default-to-contact failure mode

Call this the more-contact fallacy: treating every referral as solvable by adding social contact time, regardless of whether the person’s problem is a shortage of contact or a poor quality of the contact they already have. Programmes built entirely around volume — more visits, more calls, more group sessions — will show real gains for isolated people and disappointing or null results for lonely-but-connected people, because the second group’s problem was never about quantity.

The qualitative meta-synthesis on social prescribing found something consistent with this: participants described benefit coming from restored purpose and meaningful participation, not contact as such. That is a loneliness-shaped finding, not an isolation-shaped one, and it points toward structured, purposeful activity rather than generic social time for that subgroup.

A triage sequence you can actually run

1. Screen isolation and loneliness separately at intake. Use a contact-frequency measure for the former and a validated loneliness scale (or a short proxy from it) for the latter. Two questions, not one blended one.
2. Isolated but not lonely: prioritise access — transport, physical infrastructure, proximity to a place people already go. AARP’s 2018 survey of adults 45 and older found network size, diversity, and physical isolation were the strongest predictors of loneliness in that population, which argues for interventions that widen access before anything more elaborate.
3. Lonely but not isolated: this is where befriending alone is least likely to be sufficient. Consider structured psychological or purpose-based approaches before defaulting to more social time.
4. Both: this is the group most likely to need a combination, and the group where a single-intervention programme is most likely to under-deliver against expectations.
5. Re-measure on the same two axes at follow-up, not just on a single composite wellbeing score. If isolation has dropped but loneliness hasn’t, that tells you something the composite would hide.

What this does not solve

None of the trials underlying this triage are large or numerous enough to generalise confidently across settings — the strongest evidence comes from older adults, several in aged care or low-income contexts, and it is not established that the same hierarchy of interventions holds for younger adults or working-age populations. The social prescribing evidence base as a whole remains, by the commissioning bodies’ own admission, built on one RCT and a stack of uncontrolled evaluations. And triage at intake only works for people who make it to intake: it does nothing for the isolated or lonely person who never gets referred at all, which is most of them.