How to Report Attrition Without Burying It

You have twelve months of referral data. Ninety people were referred to your loneliness or social prescribing programme. Sixty-one attended a first session. Thirty-four completed the full course. Your end-of-year report needs to say something about outcomes, and the easiest thing to do is calculate improvement scores from the thirty-four and write the report around them.

Do not do that. Or rather: do it, but say what you did.

The completers-only report

This is the failure mode worth naming, because it is not fraud and it is not incompetence — it is the default output of ordinary reporting software, and almost every practitioner has produced a version of it without meaning to mislead anyone. You measure at intake and at completion. Anyone who did not complete has no exit score, so they fall out of the average by construction. The report then says “participants showed a 40% improvement in loneliness scores,” which is true of the people who finished and silent about the fifty-six who did not.

The systematic reviews of social prescribing are honest about how common this is at the evidence-base level, not just the individual-programme level. A 2021 systematic review in the International Journal of Environmental Research and Public Health found real gains in self-esteem and confidence across the studies it covered, but flagged limited trial evidence and heavy heterogeneity in how programmes measured and reported outcomes. A companion review focused specifically on loneliness found all nine included studies reported positive impacts — a suspiciously clean result that should make you more cautious about the field’s reporting norms, not less, since publication and reporting bias toward positive completer-based results is exactly what produces a run of nine out of nine.

What “honest” actually requires

Honest attrition reporting is not a confession paragraph buried in a methodology appendix. It is three specific things, all visible near the top of the report.

1. A cohort funnel, not a completer sample. State the number referred, the number who attended a first session, and the number who completed, as a simple sequence. Ninety referred, sixty-one attended once, thirty-four completed. Anyone reading your outcome data should see this before they see the outcome data.
2. A named reason category for each drop-off point, even if the categories are coarse. “Did not attend first session” is different from “attended once, did not return” is different from “completed fewer than half the sessions.” If you cannot distinguish these, say so rather than collapsing everyone into “attrition.”
3. A statement of what you don’t know about non-completers. Did they leave because the programme didn’t help, because life intervened, or because it worked well enough after one or two sessions that they didn’t need more? You usually don’t know. Say that you don’t know, rather than letting silence imply the first explanation didn’t happen.

What this means in practice: report the denominator you started with, not the denominator you ended with. A funder who sees “34 of 90 referred people completed the programme, and among completers, loneliness scores improved by X” trusts the X far more than one who sees only the X. The funnel is not a confession — it is the number that makes the outcome credible.

Why burying it backfires with the people you’re trying to convince

Commissioners and funders read a lot of these reports. They notice when a completion rate is missing, and the absence reads as evasion even when it wasn’t intended that way. It also makes benchmarking impossible: if every programme reports only completer outcomes, no commissioner can tell whether a 60% completion rate is normal or alarming, because nobody is reporting the denominator.

There is a second, more direct cost. The qualitative literature on how people experience social prescribing suggests that the benefit participants describe is not simple social contact but restored meaningful participation and purpose — engagement has to be structured and sustained to produce that, not just attended once. If that is the mechanism, then a low completion rate is not a footnote to your outcome data. It is evidence about whether your programme is actually delivering the thing that produces benefit. A programme with 90% first-session attendance and 30% completion has a different problem than a programme with 40% first-session attendance and 90% completion among those who show up, and a report that only shows the completion outcome erases that distinction entirely.

What a defensible report looks like

Element	Weak version	Defensible version
Denominator	“Participants reported improved wellbeing”	“34 of 90 referred (38%) completed the programme; outcomes below relate to this group”
Attrition detail	Not mentioned	Broken down by stage: referred, first attendance, mid-course, completion
Reasons	Absent or guessed	Categorised where known; explicitly marked unknown where not
Comparison	None	Compared to attrition in comparable published programmes, where such figures exist
Framing	Outcome presented as programme effect	Outcome presented as effect among completers, with a stated limit on generalisability

That last row matters because the underlying evidence base gives you very little to lean on for the counterfactual. The National Academies’ 2020 consensus report on isolation and loneliness in older adults called for routine assessment within health systems, but it was explicit that the intervention evidence supporting any particular programme design was thin. A 2023 review in BMC Public Health mapping the state of loneliness and isolation research reached a similar conclusion: inconsistent measurement across studies is a structural barrier, not a minor inconvenience, and it means you cannot borrow someone else’s effect size to fill the gap your own attrition created.

The rare exception worth citing is the HEAL-HOA trial, a randomised controlled trial of volunteering as an intervention for lonely older adults in Hong Kong, published in late 2024. It is useful precisely because it is one of the few loneliness interventions tested against a control group rather than reported as a single-arm before-and-after. If you have any access to a comparison group, even an informal one — a waiting list, a similar cohort that didn’t take up the referral — use it, because the completer-only average is otherwise doing the work a control group should be doing, badly.

Evidence status of the claims you’ll be tempted to make

Claim	Evidence status
Social prescribing programmes improve self-esteem and confidence among people who complete them	Reasonably supported across multiple reviews, with heterogeneous methods
Social prescribing reduces loneliness for anyone referred	Not established — most reviews measure completers only
Structured, sustained engagement produces more benefit than one-off contact	Supported by qualitative synthesis, not by controlled trials
Your programme’s completion rate tells you something about programme quality, not just participant circumstance	Plausible, but no established benchmark exists across programmes

What this does not solve

Honest attrition reporting will not make your completion rate higher, and it will not manufacture a comparison group where none exists. It also does nothing about the deeper reach problem in this entire field: referral pathways still depend on someone — a GP, a link worker, a self-referral form — reaching the person in the first place, and the sixty-one who attended a first session and the thirty-four who finished are already a subset of people who had enough stability, transport, and time to show up at all. A completion rate, however honestly reported, describes what happened to the people who could engage with the programme as designed. It says nothing about the people who never made it to session one, and no amount of careful attrition tables changes that.