The Glasgow deep end links worker programme evaluation: analysis implication of an unusual randomised study design

Background
Tackling health inequalities has been a Scottish Government priority since 2007. General Practitioners (GPs) at the Deep End is a collaboration of general medical practices serving the 100 most deprived populations in Scotland, based on the proportion of patients on the practice list with postcodes in the most deprived 15% of Scottish data zones. Eighty-six of these 100 practices are in the NHS Greater Glasgow and Clyde Health Board area. The Deep End Links Worker Programme was designed to prevent and reduce health inequalities in Scotland, and support people living in the most deprived areas of Scotland to ‘live well’ In their communities. It provides resources to General Practices serving these populations to develop a ‘links approach’. Practices can refer patients to a Links Worker, who works with the patient to identify and enable access to appropriate community-based resources. The argument is that increasing and enabling local activities can enhance community connection, trust and cohesion. Hence, the aim of Links Worker is to act as catalyst to hope and self-determination, by using the strong relationship with patients
that exists in General Practices as a natural community hub. Part of
the evaluation of the programme involved a comparison of patient
and staff outcomes measured in 7 Intervention practices, with those from 8 Comparator practices. These 15 practices were allocated to the Intervention and Comparison arms at random, prior to the involvement of the evaluation team.

Objective
The primary aim of the evaluation was to assess the effectiveness of the Links Worker Programme in achieving the intended outcomes at patient, practice and community levels.

Methods
A quasi-experimental evaluation design was decided by the Scottish government as part of their funding conditions. Among 15 General Practices that submitted a formal expression of interest in having a Links Worker attached to their team, patients’ outcomes in 7 Generals Practices with an attached Links Worker will be compared with those from 8 General Practices that do not have an attached Links Worker. The intervention group are patients who were referred to the Links Worker, but the comparator group are a random sample of the practice population. By design, the intervention group is very different from the comparator group. Therefore, simple between-group comparisons, which would be appropriate for most randomised studies, will not be appropriate. The main statistical analyses have been defined in terms of mixed effects regression models, to account for clustering of outcomes at the practice level, with adjustment for differences in baseline covariates. In addition, other methods will be explored,
aiming to control for the selection bias inherent in the study
design (e.g. With respect to age, gender, deprivation and comorbidities), such as nearest neighbour matching and propensity score analysis.

Results
The final analysis is underway. Some results of the evaluation will be presented, with particular focus on the implications of this unusual design, and the results obtained using alternative analysis approaches.