Measuring the Health–Community Partnership, Not Just the Referral

A GP practice signs a memorandum with a community organisation. A link worker starts referring patients to a walking group, a men’s shed, a befriending scheme. Six months in, someone asks for evidence it worked. What you have is a referral count. What you need is something else entirely, and the gap between the two is where most social prescribing partnerships quietly fail to prove their worth.

The referral-count trap

Referral volume is the number every partnership reports first, because it is the number that is easiest to get. It is also the number that tells a funder least. A referral is a decision made by a clinician or link worker, not an outcome experienced by a patient. A pathway can generate 200 referrals a year and still be failing the people it refers, if half never attend, and the half who do attend never return after their first visit.

This is the failure mode worth naming: the referral-count trap — treating the act of referring as the unit of success, because it is the only step the health side fully controls. The community organisation controls what happens next, and that is precisely where most evaluations stop looking.

What to measure at each stage

A partnership evaluation needs to track the pathway in stages, not as a single funnel number at the end.

1. Referral appropriateness. Was the person referred to something plausibly matched to their situation, or was the referral a default when nothing else was available? This is usually invisible unless link workers record a reason.
2. Uptake. Did the person attend at all? The systematic review of social prescribing and loneliness found that among the nine studies it examined, all reported positive impacts for those who engaged — but engagement is the precondition, not the outcome, and dropout before a first visit is common and rarely reported separately.
3. Retention. Did they return after the first session? A single visit to a group activity is closer to a false start than a connection. Partnerships that report attendance but not the second, third, and fourth visit rates are hiding their real performance.
4. Individual-level change. Self-esteem and self-confidence are the outcomes most consistently reported across the social prescribing evidence base, according to the broader systematic review on wellbeing outcomes — a more modest claim than the loneliness reduction some programme literature implies, and worth stating accurately to funders rather than rounding up.
5. Service use. Three of the nine studies in the loneliness-specific review reported reductions in GP, emergency, social worker, or inpatient contact following social prescribing. This is the outcome commissioners care most about, and the one with the thinnest evidence base — small samples, no controls in most cases.

Evidence status table

Claim	Evidence status
Social prescribing improves self-esteem and self-confidence for people who engage	Reasonably consistent across studies, though few controlled trials
Social prescribing reduces loneliness	Positive in available studies, but the evidence base is small and heterogeneous
Social prescribing reduces GP or emergency service use	Reported in a minority of studies; promising but not established
Referral volume predicts programme value	Not supported — volume and outcome are frequently disconnected
Screening for isolation in clinical settings improves targeting	Recommended by the National Academies, but implementation guidance is still developing

What this means in practice: if your evaluation plan has one number — referrals made — replace it with four: referrals made, first-visit uptake, fourth-visit retention, and a before/after wellbeing or service-use measure for the subset who reach that fourth visit. That subset, not the full referral list, is where the partnership’s effect actually lives.

Whose data, and who owns the gap

The structural reason the referral-count trap persists is that health systems and community organisations rarely share a data system. The clinician logs a referral in patient records; the community organisation logs attendance in a spreadsheet, if at all. Nobody owns the join between the two. The UK’s national loneliness strategy pushed for loneliness measurement to be built into official statistics, which helps at population level but does nothing for a single partnership trying to track its own pathway. Building a shared, minimal dataset — referral date, first attendance date, attendance count at 4 and 12 weeks, and one pre/post wellbeing measure — is unglamorous infrastructure work, but it is the only way to move past self-reported success stories from the organisation delivering the activity, which is not a neutral source of that judgment.

Clinical commentary on the National Academies’ consensus report on isolation in older adults makes a related point for the health side: routine assessment only pays off if there is somewhere real to refer people to, and if what happens after referral is tracked with the same rigour as the screening itself. A screening tool without a measured pathway behind it just moves the referral-count trap one step earlier.

What this does not solve

None of this fixes the underlying evidence problem: most published social prescribing studies are small, uncontrolled, and drawn from programmes that were already motivated enough to publish their results. It also does not solve reach. Every metric above describes people who were referred and then showed up — it says nothing about who a link worker never reached, who declined referral, or who was never on a GP’s list in the first place. A partnership can perform well on every measure here and still be serving a narrow slice of the population it was meant to help.