Table 1

Control groups

Intervention groupControl groupEstimateBias
Study-eligible patients in Ways to Wellness GP practices who were in receipt of the intervention at time t*Study-eligible patients in Ways to Wellness GP practices who were not in receipt of the intervention at time t and who go on to receive the intervention at time t+1* Embedded Image As eligible patients are assigned to the intervention at different time periods, we can estimate  τ using as a control group eligible patients who are yet to be treated but will go on to receive treatment. This approach attenuates some of the selection bias that may be present in treatment assignment and is a robust comparison. However, if eligible patients most likely to benefit are treated first, this may overestimate the effect size. Although if the selection mechanism is time invariant, much of this endogeneity will be ameliorated by using longitudinal data and this approach should provide the best estimate of the short-run effect of treatment.
If the intervention is randomly assigned across eligible patients, this should provide a consistent estimate of the short run effect of the intervention.
As t→t+1, this comparison estimates an intensity of intervention effect: eligible patients who have been on the programme for over a year compared with those who have just started the programme.
Study-eligible patients in Ways to Wellness GP practices who were in receipt of the intervention during the study period†Study-eligible patients in Ways to Wellness GP practices not receiving intervention during the study period† Embedded Image Treated eligible patients are compared with eligible patients in referring GP practices who are not referred into the intervention (this group may include individuals who later go on to be referred).
A group of eligible patients of particular interest includes those who are in GP practices that can refer to the intervention but do not (or that only refer a small number of patients). Living in the same area of the city and meeting the eligibility criteria means that comparing the outcomes of these control patients with the outcomes of the treated will provide an unbiased estimate of the treatment effect (assuming the assignment mechanism into participation is time invariant).
Individual and GP-level fixed effects will control for any selection effects. As with control group 1, it is possible that there are mechanisms for selection that are time-varying and lead to possible bias. The most likely source of bias is that those treated may be considered, by the referrer, to gain more from treatment than those eligible patients who remain untreated, or that assignment is affected by GP-level factors that are changing. This potential bias could lead to an overestimation of the treatment effect.
Study-eligible patients in Ways to Wellness GP practices receiving intervention over the study period†Study-eligible patients not in Ways to Wellness GP practices Embedded Image The first treatment group comprises all eligible patients in referring GP practices who receive treatment. This treatment group provides an estimate of the effectiveness of the intervention for eligible patients who actually receive treatment compared with similar patients who do not receive treatment because they are not in referring practices.
If, pretreatment, the treatment group and the control group have similar trends in their outcomes, and if there are no changes that may affect the control group differentially to the treatment group, this approach should provide the best estimate of the average effect of treatment on the treated.
If, however, patients in the control group are receiving alternative interventions that are beneficial, then we would underestimate the benefits of the social prescribing intervention.
Study-eligible patients in Ways to Wellness GP practicesStudy-eligible patients not in Ways to Wellness GP practices Embedded Image The second treatment group comprises patients in referring practices who are eligible for treatment, regardless of whether or not they receive treatment. If, pretreatment, the treatment group and the control group have similar trends in their outcomes, and if there are no changes that may affect the control group differentially to the treatment group, this approach should provide the best estimate of an intention-to-treat effect. This will be different, and we expect lower, than the average effect of treatment on the treated (Embedded Image ) (since our treatment group contains untreated individuals). However, this model has the benefit of overcoming any problems regarding intervention assignment in social prescribing practices.
  • *Time at which an individual engages with the intervention.

  • †1 April 2015 to 31 March 2018/2020.