Which Outcome Measure Will Actually Survive Renewal

A commissioner does not ask “did this help people.” They ask what changed, by how much, and whether it’s cheaper than the alternative. If your evaluation plan cannot answer that in one slide, it will not survive the renewal meeting. The measure you pick at the start of a social prescribing programme decides what you’re able to say at the end — and most programmes pick it badly, either because it was easy to administer or because it was the one used last time, not because it answers the question the funder is actually asking.

This note is about that choice: which outcome measures hold up under scrutiny, which ones look rigorous but don’t, and how to combine them so you have something to say to a commissioner who reads more evaluations than you do.

The three things commissioners actually want to know

Strip away the language of “impact” and “outcomes,” and a commissioner’s question usually decomposes into three parts:

1. Did the person’s situation change (loneliness, isolation, wellbeing)?
2. Did that change reduce demand elsewhere in the system (GP visits, A&E, social care contacts)?
3. Would it have happened anyway?

Almost every social prescribing evaluation answers only the first, and answers it with a single self-report score taken before and after. That is not nothing — but it is the weakest of the three answers, and commissioners who have seen enough of these know it.

The single-scale trap

The failure mode is predictable enough to name: the single-scale trap. A programme picks one measure — usually a loneliness or wellbeing scale — administers it at intake and at exit, reports the average change, and treats that number as the entire case for continued funding. It fails for three reasons.

First, self-report scores move for reasons that have nothing to do with the programme: regression to the mean, social desirability at exit (“I don’t want to seem ungrateful”), and the fact that people who drop out before exit — often the ones doing worst — aren’t in the average at all. Second, a single number can’t distinguish loneliness from isolation, and commissioners increasingly know the distinction matters. Julianne Holt-Lunstad’s 2015 meta-analysis found isolation, loneliness, and living alone all carry independent mortality risk, which means they are not interchangeable and a scale built for one doesn’t tell you about the other. Third, and most damaging for renewal conversations: a wellbeing score has no obvious currency conversion into money, and money is what commissioners are actually allocating.

What the evidence base can and can’t support

Before choosing measures, it helps to know what social prescribing research currently supports, because a commissioner who has read a systematic review will notice if you overclaim.

Claim	Evidence status
Social prescribing improves self-esteem and self-confidence	Reported across studies; a 2021 systematic review identified this as a consistent qualitative finding
Social prescribing reduces loneliness	Positive individual impacts reported in all nine studies in a 2021 systematic review, but heterogeneous designs, mostly uncontrolled
Social prescribing reduces service use (GP, A&E, social care)	Reported in 3 of 9 studies in the same review — real signal, but not consistent, and rarely measured with a comparison group
Structured, purposeful activity outperforms contact alone	Supported by qualitative synthesis (2022): participants describe benefit from restored participation and purpose, not proximity to other people as such
Social prescribing changes clinical outcomes (mortality, hospitalisation)	Not established. No study in this literature is designed to detect it

The pattern across the literature is consistent: qualitative and self-report signals are decent, service-use evidence is thin and inconsistent, and hard clinical outcomes are simply not measured. If you tell a commissioner social prescribing “cuts GP visits,” you are stating a possibility from three studies, not a finding. Say that directly rather than letting the claim harden into something the evidence doesn’t hold.

A measurement set that a commissioner will accept

Rather than one scale, build a small stack, each answering a different part of the commissioner’s question.

1. A validated baseline-and-follow-up scale, chosen for construct, not convenience. If the referral is for loneliness, use a loneliness measure, not a general wellbeing scale — the two move independently and conflating them invites the “which do you actually mean” question. The UK’s 2018 national loneliness strategy pushed for loneliness to be measured through Office for National Statistics-endorsed instruments precisely so local data could be compared against a national baseline; using a recognised instrument, rather than a bespoke questionnaire, is what lets your numbers be read against anything else.

2. A participation or engagement measure, not just a feeling measure. The 2022 qualitative meta-synthesis on social prescribing found that what participants describe as valuable is often restored meaningful participation and purpose — attending, contributing, having a role — rather than social contact as an abstract good. A dosage or engagement measure (sessions attended, roles taken on, activities sustained past a threshold) captures this directly and is far harder for a skeptical reader to dismiss than a self-rated happiness score, because it’s a count, not a feeling.

3. At least one system-contact measure, even an imperfect one. GP contacts, missed appointments, or self-reported use of unplanned care in the prior month. The evidence that social prescribing moves these numbers is weak — three of nine studies in the loneliness review found it — but a commissioner funding this from a health budget wants to see you are at least tracking the variable they care about, even where you can’t yet claim the effect.

4. A named comparison, even a weak one. Waiting-list controls, a delayed-start cohort, or simply reporting how outcomes differ for people who completed the programme versus those referred but not engaged. None of this is a randomised trial. But an evaluation with no comparison at all invites exactly the “would this have happened anyway” objection that ends funding conversations.

What this means in practice: stop choosing a single wellbeing scale because it is short and stop reporting service-use claims you can’t support with your own data. Build a four-part stack — validated scale matched to the actual referral reason, an engagement or dosage count, at least one system-contact metric even if the finding is null, and some form of comparison group. Report each honestly, including where the evidence is thin. A commissioner who sees calibrated confidence trusts the strong claims more, not less.

Why routine assessment changes what you can measure

One structural fix sits upstream of any individual programme’s measurement plan. The National Academies’ 2020 consensus report on isolation in older adults recommended that health systems routinely assess isolation and loneliness as part of standard care, rather than only at the point of a social prescribing referral. If your local health system has not adopted routine screening, you are working with a self-selected intake — people flagged as isolated only because someone happened to ask — and any before-and-after comparison you build on top of that is comparing a non-representative group against itself. Where routine assessment exists, push to be given the screening data as a matching baseline, not just your own referral list. Where it doesn’t exist, say so in your evaluation report as a named limitation rather than letting the sampling problem go unremarked.

What this does not solve

None of this produces evidence that social prescribing changes hospitalisation, mortality, or long-term health costs — the literature currently supplied does not contain a study designed to detect that, and a stronger measurement stack at the local level cannot manufacture population-level evidence that doesn’t exist yet. It also does not solve the reach problem underneath every social prescribing evaluation: the outcome data, however carefully chosen, only ever describes the people who were referred, showed up, and stayed long enough to be measured at exit. The people who declined the referral, or attended once and never returned, are the group commissioners should be asking about next, and almost no local evaluation is set up to say anything about them at all.