Kennedy 2013.

Methods	Study design: cluster‐RCT Unit of randomisation: general practices Unit of analysis: patient Funding sources: National Institute for Health Research; National Primary Care Research and Development Centre Conflicting interests: none declared
Participants	Country: UK Setting: Primary care Conditions/numbers: 5599 patients with diabetes, COPD, or irritable bowel syndrome from 43 practices (2295 intervention, 3304 control) Multi‐morbidity: n/a Health literacy: Recruited from practices with high levels of socio‐economic deprivation (p. 2)
Interventions	Theoretical framework: Chronic Care Model, Normalisation Process Theory Focus: Both clinician and patient Type of intervention: Structured face‐to‐face coaching + staff training Clinicians involved: Nurse (usual), GP (usual) Tools: Whole System Informing Self‐Management Engagement (WISE). The intervention was intended to be feasible to implement widely in primary care, which put practical limitations on the intensity of the intervention. Aim was to take several components and deliver them as a comprehensive package under naturalistic conditions using routine care providers to maximise real‐world applicability. Two training sessions were organised for practice staff covering ways of embedding self‐management tools in practice systems (session 1) and using core self‐management skills in consultations and ensure participants received, or were directed to, appropriate resources (session 2). Fidelity checks and reinforcement sessions were scheduled after training. Two facilitators delivered the training and provided access to self‐management support activities and resources. These included a tool to assess patient support needs and priorities (PRISMS); self‐help guidebooks; access to community groups and programmes; and enhanced access to psychological therapists for IBS participants. Stages completed: Limited ‐ B, C, E, F Usual provider aware of patient's goals and action plans: Yes Standardisation of clinician input: Weak ‐ 2 training sessions + manual, but low levels of implementation Fidelity: Poor ‐ shared decision‐making at 6 months significantly less in intervention than control group (P = 0.05); only 2% of IBS participants referred to therapists; 42% of clinicians failed to use PRISMS tool (p. 4). Process evaluation (Kennedy 2014) examined reasons for failure to change practice and confirmed that very little personalised care planning took place. Attrition: 19% at 6 months and 27% at 12 months lost to follow‐up Comparison: Usual care, including information and support
Outcomes	Health status:subjective: Medical Outcomes Study short form (SF‐36), Euroqol (EQ5D)* Self‐management capabilities: self efficacy*, patient enablement Health behaviours: n/a Achievement of personal goals: n/a Service use: n/a Adverse events: none reported Length of follow‐up: Baseline, 6 months, 12 months
Notes	*Primary outcomes. Power calculation ‐ required sample of 40 practices and 48 participants per condition per practice (total participants = 5760), so slightly under‐powered.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Wait‐list comparator group. Used a minimisation procedure based on practice size, area deprivation and contractual status, practices were allocated 1:1 to intervention or control. Practices were paired as closely as possible according to their preferred training times, and using a minimisation procedure, 1 practice in each pair was allocated to training in the first year, with the other practice allocated to training at the same time the following year (p. 3).
Allocation concealment (selection bias)	Unclear risk	Research staff recruiting practices were unaware of the next allocation in the sequence at the time of recruitment (Bower 2012, p. 7). Baseline (and subsequent follow‐up) data collection then took place at both practices in a pair at the same time. Proved impossible to recruit participants prior to allocation. Practices required adequate advance notice of their training date, hence it became necessary to inform them of their group allocation prior to participant selection. Authors confident that any resulting bias is small. Recruitment was through electronic health records rather than by professional invitation, but practitioners could exclude patients after identification. These exclusions represented a relatively small proportion of patients (COPD 15% intervention, 11% control; diabetes 11% int., 10% cont; IBS 11% int., 18% cont.).
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Personnel were not blinded and outcomes were patients' self report.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Analyst blind to practice allocation (supplementary file).
Incomplete outcome data (attrition bias) All outcomes	Low risk	Intention‐to‐treat analysis. 81% completed 6 month follow‐up and 72.8% the 12‐month follow‐up. Few differences between intervention and control in completeness of outcome data. Missing values for outcome variables at follow‐up were not imputed, but addressed through covariate adjustment.
Selective reporting (reporting bias)	Low risk	Trial report matched published protocol apart from certain measures that were eventually omitted from the study to make the questionnaire shorter. No evidence of selective outcome reporting.
Other bias	High risk	Fidelity to the intervention was very poor ‐ shared decision making at 6 months significantly less in intervention than control group (P = 0.05); only 2% of IBS participants referred to therapists; 42% of clinicians failed to use PRISMS tool (p. 4). Kennedy 2014 confirms that very little personalised care planning actually took place, so we have excluded the study from the meta‐analysis. No evidence of selective recruitment by clusters. Two trial arms were reasonably well‐balanced on all variables at the participant level, but practices in the intervention group were on average slightly smaller (mean list size 4003 vs 4528 patients).