Developing and Evaluating Exercise Services

A defensible service starts with need rather than with an idea. Needs assessment establishes who is inactive in this population, which groups are least served, what the barriers are locally and what already exists, drawing on routine data, local surveys and the people the service intends to reach. Outcomes come next and must be specified before anything is built, because a service designed to increase activity looks different from one designed to reduce isolation or to improve a clinical measure. A programme theory or logic model then sits between need and design, setting out how the proposed components are expected to produce the intended outcomes and what assumptions must hold for that to happen. The intervention is designed around recognised behaviour change techniques rather than assembled from available facilities, and it is piloted to test specific feasibility uncertainties such as recruitment, acceptability, delivery fidelity, data collection and progression criteria, rather than simply run as a smaller version of the full programme.

Several design decisions determine whether a service will reach the people it is meant to. Referral routes shape who arrives, so a scheme reachable only through general practice will miss those who do not attend. Cost, transport, timing and setting all filter participants, and a service delivered in a leisure centre at midday reaches a very different population from one delivered in a community hall in the evening. Staff mix matters, both for competence and for whether participants feel the service is for people like them. Recruitment, retention and the transition after the programme are all common failure points, and building in follow-up contact, gradual progression and a planned handover into ongoing community activity addresses the most common reason schemes fail to produce lasting change. Leading such a service means governance as well as design: clear clinical accountability, agreed inclusion and exclusion criteria, staff training and supervision, incident reporting, data protection, and a plan for sustainability beyond the initial funding period.

Design decisions at service level do much of the work that no individual consultation can. Five useful local audit domains are follow-up, consistent messaging, relevant signposting, accessible patient-facing resources and reinforcement. These form a pragmatic service design checklist rather than a validated minimum set, and the distinction matters: a systematic review of physical activity referral schemes identified 19 distinct components across programmes and concluded that the evidence is not yet sufficient to determine which individual components drive uptake, adherence or behaviour change.

Taking each in turn, follow-up matters because a single contact rarely sustains change, so the service should define when the person will next be seen or contacted. Consistent messaging means the general practitioner, the practice nurse, the referral scheme and any written material say the same thing, since contradictory advice undermines all of it. Signposting should carry current information on cost, accessibility, eligibility, location and suitability, without which it is unlikely to result in uptake. Patient-facing resources should be available in a form the person can take away, understand and return to, with attention to reading age, language, digital exclusion, disability accessibility, alternative formats and a contact point for clarification. Reinforcement means building in recognition of progress through review, recorded goals or feedback on outcome measures, noting that feedback and review are recognised components of effective goal regulation rather than a guaranteed effect, since their value depends on goal quality, commitment, context and the nature of the feedback.

The checklist is not comprehensive, and other components recur in the evidence: person-centred assessment, individualised content, explicit behaviour change techniques, screening and a baseline consultation, a written prescription or action plan, an actual connection to activity opportunities, a defined action after non-attendance, exit and maintenance planning, and feedback to the referrer. Auditing a service against these domains is a quick way to generate hypotheses about why a scheme with good uptake is nonetheless failing to produce sustained change, which evaluation can then test.

Evaluation has three components and confusing them is the commonest error. Process evaluation asks whether the service was delivered as intended, and covers reach, fidelity, the dose delivered and the dose received, adaptations made during delivery, participant experience, context and the mechanisms through which the service is expected to work, alongside referrals, uptake, completion and dropout. Outcome evaluation asks whether the intended outcomes changed, which requires a baseline measure and a defined follow-up point; a comparison group is not strictly needed to document change, but it is needed to claim credibly that the service caused it. Economic evaluation asks whether the benefit justified the cost, and while it is not essential for every small service evaluation, it is usually decisive for commissioning and scaling. Attendance is one process measure, but attendance alone does not provide a complete process evaluation and says nothing about effectiveness or value for money. The methodological framework most often cited for this work is the guidance on developing and evaluating complex interventions, which emphasises context, programme theory, stakeholder involvement, identifying key uncertainties, refinement, economic considerations and implementation, and treats these services as complex because they have multiple interacting components and depend heavily on context.

The practical difficulties are worth anticipating. Selection bias is pervasive, because people who accept a referral and attend differ systematically from those who do not, so before-and-after comparisons in completers flatter the service. Regression to the mean affects any service recruiting people at their worst. Loss to follow-up is typically substantial and non-random, since those who dropped out are also least likely to return a questionnaire, and analysing only completers converts a modest effect into an impressive one. Where individual randomisation is impractical, consider cluster randomisation or a stepped-wedge trial, in which the order in which clusters receive the service is commonly randomised, where research approvals are appropriate. Non-randomised alternatives include a controlled interrupted time series, difference-in-differences, and carefully matched comparison areas. Note that introducing random allocation or additional research procedures may convert a service evaluation into research, which then requires the relevant sponsorship and approvals. Reporting should be honest about all of this, because a service evaluation that overstates its effect makes the next commissioning decision worse rather than better. A minimum evaluation dataset should also capture adverse events, unintended consequences, equity of reach and outcomes, and data completeness. Finally, evaluation findings should feed back into the design cycle rather than being filed, which is the difference between a service that improves and one that simply continues.

• •Start from needs assessment, define outcomes, then set out a programme theory explaining how components produce outcomes and what assumptions must hold, before designing around behaviour change techniques.
• •Referral routes, cost, transport, timing, setting and staff mix all determine who actually reaches the service.
• •Recruitment, retention and the transition after the programme are all common failure points, so plan follow-up contact and a handover into ongoing activity.
• •Evaluation has three parts: process, covering reach, fidelity, dose, adaptations, experience and context; outcome, requiring baseline and follow-up, with a comparison group needed to claim causation; and economic, usually decisive for commissioning.
• •Attendance is a process measure only: it does not establish outcomes, effectiveness or value for money, and completer-only analysis inflates apparent effect through selection, regression to the mean and non-random loss to follow-up.
• •Where individual randomisation is impractical, consider cluster randomisation or a stepped-wedge trial, which commonly randomises the order clusters receive the service, or non-randomised options such as controlled interrupted time series, difference-in-differences or matched comparison areas.