What to Fix and What to Flex When You Scale a Connection Programme

You have a group that works. One site, one facilitator, twelve people, a waiting list, and a set of before-and-after scores good enough that a commissioner has noticed. Now you are being asked to run it at thirty sites within eighteen months.

The decision in front of you is not whether to scale. That has effectively been made. The decision is which features of the current programme you are going to write down as non-negotiable, and which you are going to let each site decide for itself. If you do not make that list, the scale-up will make it for you — through contract terms, staff availability, and room bookings — and you will find out which features mattered by losing them.

You are being scaled on the strength of the problem, not the strength of your solution

Be clear-eyed about why the money has appeared. The risk-factor evidence is strong. Holt-Lunstad’s 2015 meta-analysis put the odds of early mortality at 1.29 for social isolation, 1.26 for loneliness and 1.32 for living alone, with effects persisting after adjustment for health status. The American Heart Association’s 2022 scientific statement associated isolation and loneliness with roughly a 30% increased risk of heart attack, stroke or death from either. The Surgeon General’s 2023 advisory framed the mortality risk as comparable to smoking up to 15 cigarettes a day, and the National Academies had already told the health care system in 2020 to assess isolation routinely.

That is enough to move a budget. It is not evidence that your programme works, and the AHA statement says so directly: the absence of intervention evidence is named as the central research gap.

So when you go looking for the literature to tell you what to hold fixed — optimum group size, number of sessions, facilitator qualification — it is not there. The 2021 systematic review in Perspectives in Public Health found all nine included social prescribing studies reported positive individual impacts, and three reported reductions in GP, emergency, social worker or inpatient use. Encouraging, and also nine studies. The parallel review in IJERPH the same year found gains in self-esteem and self-confidence alongside limited trial evidence and substantial heterogeneity between programmes. The 2023 BMC Public Health review of the field identified inconsistent measurement as a barrier to comparing findings at all.

The nearest thing to an active-ingredient finding comes from the 2022 qualitative meta-synthesis in BMC Health Services Research: participants described benefit that went beyond social contact into restored purpose and meaningful participation, and structured, purposeful group activity appeared to do more than contact alone. That is one plausible mechanism, from qualitative data. Treat it as the best available hypothesis, not a result.

Claim	Evidence status
Social disconnection is a population-scale risk factor for mortality and cardiovascular events	Strong. Multiple meta-analyses, an AHA scientific statement, a National Academies consensus report
Social prescribing improves self-reported loneliness, confidence and self-esteem	Consistent direction of effect, weak designs. Small study counts, high heterogeneity, little trial evidence
Purposeful, structured group activity beats contact alone	Qualitative synthesis only. Plausible mechanism, untested against a comparator
Programmes reduce use of primary, emergency or inpatient care	Suggestive. Three of nine studies in the 2021 review; no robust costing
Any specific group size, session count or dose threshold	No evidence. This is local judgment and should be labelled as such

The per-head problem

Here is the failure mode to watch for, and it is a funding mechanic rather than a delivery mistake.

Almost every commissioned connection programme is paid on volume: per referral accepted, per person starting, or per person completing. Two levers raise volume without new money. You can put more people in each group, and you can shorten the cycle so cohorts turn over faster. Both are invisible in the reporting, because the reporting counts people, not exposure.

They are also the two most likely carriers of whatever effect you had. A group of eight in which everyone speaks is a different intervention from a group of twenty-two in which four people speak. A twelve-week cycle in which relationships outlast the programme is a different intervention from a four-week cycle that ends before anyone exchanges phone numbers.

The per-head problem is not that anyone decides to dilute the programme. It is that dilution is the only available response to a volume target, and nobody in the delivery chain has the authority to refuse it. By month nine you are hitting throughput, your scores have flattened, and there is no way to reconstruct what changed because you never wrote down what the original dose was.

What this means in practice

Before you sign, put a dose floor in the specification and price it: maximum group size, minimum number of sessions, minimum facilitator hours per cohort. Then insist that volume targets are expressed as cohorts delivered rather than individuals seen. A commissioner can accept this — cohorts are a countable unit — but only if you raise it during contracting. Once the per-head price is set, the argument is over.

Write the fidelity specification before the contract

One page. Two columns. Every item in the left-hand column needs a stated reason, even a weak one, so that a site manager under pressure knows what they are trading away.

Fix	Flex
Maximum group size, at whatever your original was	The activity itself — growing, cooking, repair, choir, walking
Number of sessions and weeks per cycle	Venue type and day of the week
The session has a task with an output, not open-ended chat	Facilitator’s professional background
One named facilitator across a whole cycle	Referral mix across primary care, self-referral and community routes
A warm handover: someone the person already trusts introduces them	Group composition by age, condition or shared circumstance
Whether cohorts are closed or rolling — keep whichever you had	Language, cultural adaptation, and how the offer is described locally
The measurement instrument, unchanged	Whether refreshments, transport or childcare are provided

The warm handover is the item most often cut, because it consumes staff time that produces no countable output. The AARP Foundation’s 2018 national survey found 33% of those who had spoken with their neighbours were lonely, against 61% of those who never had. The people your programme most needs are the least likely to walk into a room of strangers on the strength of a leaflet. If the handover goes, the composition of your groups changes, and your outcome scores will move for reasons that have nothing to do with delivery quality.

Report by site, not by pool

Pooled means across thirty sites will hide everything you need to know. Report site-level results from the start, with the count of cohorts delivered, mean group size and mean sessions attended alongside each outcome.

Use a validated instrument and do not modify it. The AARP survey used the 20-item UCLA Loneliness Scale precisely so its findings sat alongside the academic literature; a shortened in-house version saves five minutes per participant and costs you the ability to compare anything to anything. The clinician-facing commentary on the National Academies report, published in the American Journal of Geriatric Psychiatry in 2020, is worth reading before you design the assessment step, because it is honest about what routine assessment actually demands of a service.

Expect two or three sites to underperform badly. That is the point of site-level reporting: those sites are your natural experiment in what fidelity was worth.

What this does not solve

Scaling to thirty sites gives you thirty copies of the same recruitment funnel, including its selection bias. If your original group filled through a link worker’s personal caseload and word of mouth, you have replicated a route that reaches people already in contact with services and already willing to be referred.

The Harvard Making Caring Common survey in 2021 found 36% of Americans reporting serious loneliness, rising to 61% of young adults aged 18 to 25 — and about half of lonely young adults said no one had spent more than a few minutes recently asking how they were in a way that felt genuine. Almost none of those people are on a social prescribing pathway. Volume growth in a referral-fed programme is not the same as reach, and no fidelity specification will fix that. It is a separate problem, and it needs a separate route in.