What the KFF Survey Means for Deciding When to End a Group

KFF’s survey, published on 1 June, breaks loneliness and social support down by race and ethnicity and connects experience of discrimination to the size of a person’s support network. Most national loneliness surveys don’t do this. It matters for social prescribing not because it tells you who is lonely — you likely already suspected that — but because it exposes a measurement gap that makes it hard to tell whether a referral pathway is working for the people it was built for, or just working, full stop, for whoever stayed.

That gap has a practical consequence most programmes never confront directly: nobody has a clear rule for when to close a group down.

The default is to keep things running

Social prescribing programmes are built to refer people in. Almost nothing in the infrastructure is built to tell you when a group has stopped doing its job. Funding cycles reward continuity — a group that has run for three years looks more credible on a renewal application than one that ran for six months and stopped. Link workers are measured on referrals made, not on referrals that turned out to be the wrong fit. The result is a system with strong incentives to keep everything open and almost no incentive to ask whether a specific group, for a specific population, has become a place people attend out of habit rather than benefit.

The KFF data gives this problem a sharper edge. If loneliness and support-network size vary by race and by experience of discrimination, then a group’s average attendance and average wellbeing scores can look fine while masking the fact that it is failing a subset of the people referred into it — the people for whom entering an unfamiliar room already carries a cost the average member doesn’t pay. Aggregate retention numbers are exactly the kind of measure that hides this. A group at 70% retention with a fairly even demographic drop-off looks identical, on paper, to a group at 70% retention where every departure comes from the same population.

The stopping question, stated properly

The question is not “is this group working.” Almost any structured group activity produces some positive self-report — the qualitative literature on social prescribing consistently finds participants describing benefit that extends past mere contact, into restored purpose and participation. That finding makes people confident that “it’s helping somebody” is enough to justify continuing. It isn’t the right test.

The right test is comparative and time-bound: is this group, for this population, still producing more benefit than the next-best use of the same referral slot, the same room, and the same link worker’s caseload? That reframes closing a group not as an admission of failure but as an ordinary resource decision — the same kind of decision you’d make about any other intervention that has plateaued.

Three signals suggest a group has reached that point rather than merely hit a rough patch:

1. Selective attrition along a demographic line, not random attrition. If departures cluster by ethnicity, age, or another characteristic tracked at referral, that’s a structural problem with the group, not a motivation problem with the individuals who left.
2. Flat outcome measures across three or more consecutive cohorts, where the group has stopped producing new gains and is instead just maintaining people who were already stable — worth doing, but worth doing consciously, not by default.
3. A widening gap between who gets referred in and who is still there at week twelve. This is the group-level version of the group-therapy “week four problem”: early dropout concentrated among people from a specific background, while a self-selected core keeps the average numbers looking healthy.

What this means in practice: don’t ask “should we keep this group open” as a single yes/no. Track attrition by the same demographic categories you use at referral, not just overall. If attrition clusters along a line that also predicts loneliness risk in the wider evidence — as KFF’s breakdown by race and discrimination suggests it might — that is a signal to redesign or close the group, not a signal to run another retention campaign for the people already staying.

Evidence status

Claim	Evidence status
Loneliness and support-network size vary by race and discrimination experience	Direct finding, KFF 2024
Social prescribing groups produce self-reported benefit beyond contact alone	Supported by qualitative synthesis, but heterogeneous programme evidence
Structured/purposeful activity outperforms unstructured contact	Suggested by qualitative synthesis, not tested experimentally
Selective attrition predicts poor demographic fit, specifically	Plausible inference from the above, not directly tested in social prescribing literature
Social isolation and loneliness carry comparable mortality risk to smoking	Well-established meta-analytic finding, unrelated to the attrition question but relevant to why any of this matters

That last row matters for tone: the stakes of getting referral and retention wrong are not trivial. Holt-Lunstad’s 2015 review found isolation and loneliness carrying mortality risk on the order of established risk factors. A group that quietly fails one demographic while looking successful in aggregate isn’t a minor programme design flaw — it’s a resource that isn’t reaching the people whose risk is highest, while consuming budget that could.

What this does not solve

None of this tells you what to build instead of the group you close, and KFF’s survey doesn’t measure programme outcomes at all — it measures population-level loneliness and support, not what happens inside a specific referral pathway. The demographic breakdown by race is a national one; it will not map cleanly onto a single borough’s caseload or a single employer’s workforce, and applying it without local data risks assuming a pattern that isn’t there. Closing a group also does nothing for the people already inside it who were benefiting — that transition needs its own plan, not just a decision memo. And the systematic reviews of social prescribing are explicit that the underlying trial evidence is thin and heterogeneous across programmes, so treat any of these attrition signals as a prompt to look closer, not as proof on their own.