When the Evidence Base Doesn't Support What You're Being Asked to Fund

A commissioner asks for evidence that your walking group reduces cardiovascular events. A funder wants a cost-per-hospital-admission-avoided figure for your befriending scheme. A health board wants to know whether social prescribing “works” before it will renew a contract.

None of these questions can currently be answered honestly with a number. That is not a failure of your programme. It is a description of where the research actually is. The mistake is pretending otherwise, either by inventing precision you don’t have or by refusing to engage and losing the funding line entirely.

Where the evidence actually stops

The American Heart Association’s 2022 scientific statement on social isolation and cardiovascular health is unusually direct about this. It reports that isolation and loneliness carry roughly a 30% higher risk of heart attack, stroke, or death from either — but it explicitly names the absence of intervention evidence as the central research gap. The association evidence is real. The evidence that any specific intervention reduces that risk is not there yet.

The same gap runs through social prescribing. Two systematic reviews — one from 2021 covering wellbeing outcomes, one covering loneliness specifically — both report genuinely positive findings: increased self-esteem, self-confidence, and, in three of nine loneliness studies, reduced use of GP, emergency, or inpatient services. But both reviews flag the same limitation: heterogeneous programmes, small samples, and almost no randomised controlled trials. A 2022 qualitative meta-synthesis adds a useful nuance — participants describe benefit that goes beyond social contact itself, toward restored purpose and participation — but that is a qualitative finding about experience, not a quantitative one about health outcomes.

The National Academies’ 2020 consensus report on older adults is blunt in a different direction: it recommends the health care system routinely assess isolation and loneliness, while acknowledging that the intervention evidence to act on those assessments is thin. A 2020 clinical commentary on that report pushes further, asking what routine assessment would actually require of a clinic that has nowhere confident to refer a patient once isolation is flagged.

There is one genuine bright spot worth naming precisely because it’s rare: the 2024 HEAL-HOA trial in Hong Kong, a randomised controlled trial testing prosocial engagement and volunteering against a control group among lonely older adults. It matters less for its specific result than for its design — most of the intervention literature is uncontrolled programme evaluation, and this is one of the few studies that isn’t.

What this means for the ask in front of you

Claim you’re being asked to support	Evidence status
Social connection is associated with mortality and disease risk	Strong. Holt-Lunstad’s 2015 meta-analysis and the AHA statement both quantify this robustly.
Isolation should be routinely assessed in clinical settings	Recommended by the National Academies, but implementation guidance is thin.
Social prescribing improves self-reported wellbeing	Reasonable support, mostly small studies.
Social prescribing reduces health service use	Suggestive, not established. Three studies out of nine in one review.
A specific programme reduces cardiovascular events or mortality	Not supported by any current trial evidence.
A specific programme is cost-saving to the health system	Not supported. No robust intervention economics exist yet.

If someone hands you a target from the bottom two rows, you have three honest moves. Refusing the funding relationship outright is rarely one of them — it just hands the field to whoever is willing to overclaim.

Reframe the outcome to what the evidence can actually support. Instead of promising reduced cardiovascular events, commit to measuring change in validated loneliness or isolation scores, and cite the association with downstream risk as the reason that measure matters — without claiming your programme has proven it moves the risk itself.

Separate the population claim from the programme claim. You can say, accurately, that loneliness and isolation carry mortality risk comparable to established risk factors. That is not the same as saying your twelve-week programme reduces that risk, and a funder who understands the difference will respect you more for making it than for eliding it.

Name the gap in the proposal itself, and say what would close it. Point to the HEAL-HOA trial as the direction the field needs more of, and be explicit that your programme is contributing observational or pre/post data, not trial-grade evidence, unless you have actually built a control arm.

What this means in practice: When a funder asks for an outcome the evidence base cannot support — averted hospital admissions, reduced mortality, cost savings — do not manufacture a number to fit the ask. Offer the outcome that is actually measurable (validated loneliness or isolation scores, service contact patterns, retention), state plainly what that outcome is a reasonable proxy for and what it is not, and cite the population-level association evidence separately from any claim about your specific programme.

The failure mode: borrowing someone else’s number

Call this the borrowed-statistic problem. It shows up when a programme cites the mortality risk figures from Holt-Lunstad’s meta-analysis, or the AHA’s cardiovascular findings, as though they were evidence about the programme itself, rather than evidence about the general phenomenon the programme is trying to address. It’s an easy slide to make, because the two things sit one paragraph apart in most literature reviews and the strong number is far more persuasive than the weak one.

The problem surfaces later, not immediately. A funder renews once on the strength of a compelling headline statistic, then asks at renewal for the programme-specific data behind it — and there isn’t any, because the number was never about the programme. That conversation goes worse than if the honest limitation had been stated at the outset. It also erodes trust for the next organisation asking for the same kind of funding, because the commissioner has now learned to discount the whole category of claim.

The fix is procedural, not rhetorical: attribute every statistic to the population it describes, and never let a slide deck imply that a population-level hazard ratio is a programme-level outcome.

What this does not solve

None of this closes the actual evidence gap — it only stops you from misrepresenting where it sits. The field still lacks randomised trials for most social connection interventions, and that gap will persist until funders are willing to pay for the more expensive, slower study designs that could close it. Nor does honesty about the evidence guarantee funding; a commissioner under pressure for a specific metric may simply move to a provider willing to promise it. And this note only addresses the evidence relationship between you and a funder — it says nothing about whether your programme reaches the isolated people who most need it, as opposed to the socially connected people who find it easiest to sign up.