McGeoch 2004.

Methods	Study design: parallel‐group cluster‐randomised study in an intervention group of practices and a control group of practices Location, number of centres: participants attending 2 groups of general practices in Christchurch, New Zealand Duration of study: 12 months. Year study performed: July 2002‐December 2003
Participants	N screened: 257 N randomised: 159 N completed: 152. INT 84, 1 died, 1 withdrew consent; CONTROL 68, 2 died, 2 withdrew consent, 1 unable to be contacted M = INT 45 (52%), CONTROL 49 (67%) Age: INT 69.8 (11.6), CONTROL 72.1 (9.9) Baseline details: current smoker INT 27 (31%), CONTROL 17 (23%); ex‐smoker INT 59 (69%), CONTROL 56 (77%); pneumococcal vaccination (last 5 years) INT 34 (40%), CONTROL 30 (43%); FEV1 % predicted INT 54.6 (18.7), CONTROL 53.1 (18.1); BMI INT 25.9 (4.6), CONTROL 25.4 (4.1); HADS anxiety INT 6.2 (4.2), CONTROL 5.3 (3.6); HADS depression INT 4.6 (3.7), CONTROL 4.1 (2.9); SGRQ total INT 43.3 (18.8), CONTROL 36.8 (17.6); P = 0.03 Inclusion criteria: GP database searched for diagnosis or use of bronchodilator and inhaled corticosteroid prescriptions. COPD according to ATS criteria (history of cough, sputum, SOB, > 10 pack‐year smoking); plus FEV1/FVC < 70%, weekly symptoms, history or 1+ exacerbations in previous 12 months requiring increased therapy Exclusion criteria: unable/unwilling to sign consent, primary diagnosis asthma, other primary functionally limiting disease, other medical condition likely to affect patient mortality, hospital level residential care, already using self‐management plan, on domiciliary O2, attending GP who already uses self‐management plans more than occasionally, exacerbation of COPD requiring increased treatment within 6 weeks or admission to general hospital within 3 months, cognitive impairment as per 3 MS < 75%, alpha1‐antitrypsin deficiency
Interventions	Intervention: AP intervention: usual care and individual standardised educational session from practice nurse or respiratory educator on the use of a self‐management plan, which includes methods of early recognition of exacerbations and appropriate self‐initiated interventions including antibiotics and short course oral corticosteroids; instruction to make early contact with GP. Control: usual care, specifically denied access to written self‐management plan. Non‐standard education on smoking cessation, exercise, controlling breathlessness, nutrition, use of inhaled therapy and immunisation was given according to practice standards. Treatment period: 12 months Follow‐up time points: assessments at baseline, 12 months; telephone interviews at 3, 6 and 9 months
Outcomes	Medications: % people used courses of antibiotics and oral steroids at 6 and 12 months HRQoL: SGRQ measured at 6 and 12 months Healthcare utilisation: % participants who attended GP visits, ED visits and hospital admissions at 6 and 12 months; % participants who took courses of antibiotics/prednisone at 12 months Hospital Anxiety and Depression Scale (HADS): recorded at baseline and at 12 months COPD Self‐Management Interview (COPD‐SMI): 30‐minute structured interview at baseline and at 12 months, comprising 3 written descriptions of situations (read to participants) based on stages of an exacerbation. Feeling of wellness Early exacerbation Severe exacerbation In each scenario, investigators assessed 3 self‐management domains of medication use, healthcare‐seeking decisions and self‐care. They scored each of 13 items per situation on a 3‐point scale (0–2), separately scoring responses for knowledge (knowing what to do) and actions (whether participants would actually do the task and when they would do it), yielding a maximum possible score of 26 for each in all 3 situations. Study visits at baseline and at 12 months, with telephone interviews at 3, 6 and 9 months
Notes	Funding: Study was funded by Pegasus Health, an independent practitioner association, The Canterbury Respiratory Research Trust and The Asthma and Respiratory Foundation of New Zealand. No funding was received from any pharmaceutical company.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomisation sequence generation was not described. Practices were randomised via 1 investigator. Individual participants were also randomised by a random numbers table if too many were included in a single practice. Participants were screened after randomisation by standardised history and spirometry.
Allocation concealment (selection bias)	Unclear risk	Participants were allocated by practice attendance, but information on allocation of practices was not available. If too many patients were identified in each practice, a random numbers table was used to allocate individual participants. An aspect of concern regarding this method was that if the same GP was implementing both intervention and usual care, confounding between treatment methods may occur, possibly diluting effects of active intervention.
Blinding (performance bias and detection bias) Participants	High risk	Researchers were unable to blind participants to educational intervention; patient questionnaire outcomes may be influenced by perception of receiving extra intervention.
Blinding (performance bias and detection bias) Study personnel	Unclear risk	Nursing staff administering assessments were not blinded to whether participants were included in intervention or control groups.
Blinding of outcome assessment (detection bias) Objective outcomes, e.g. healthcare utilisation	Low risk	Although it was not clear how healthcare utilisation data were collected, this was unlikely to be affected by bias.
Blinding of outcome assessment (detection bias) Subjective outcomes eg quality of life, anxiety	Unclear risk	Nursing staff administering assessments were not blinded to whether participants were included in intervention or control groups; this may potentially affect collection of questionnaire data.
Incomplete outcome data (attrition bias) Health care utilisation (objective)	Low risk	Analysis: INT 84/86 (1 death, 1 WD consent), CONTROL 70/73 (2 WD consent, 1 no contact). Small losses to follow‐up, balanced across groups
Incomplete outcome data (attrition bias) Subjective e.g. Quality of life	Low risk	Analysis: INT 84/86 (1 death, 1 WD consent), CONTROL 70/73 (2 WD consent, 1 no contact). Small losses to follow‐up, balanced across groups
Selective reporting (reporting bias)	Low risk	Study protocol was not available, but all expected outcomes were reported.
Other bias	Unclear risk	Sample size calculation was based on the assumption that about 10 patients would be recruited for each surgery, and that no additional between‐participant variation would be due to clustered‐randomisation of surgeries. Analysis of the 12‐month change in outcome variables was based on a mixed‐model repeated measures ANOVA. This analysis enabled estimation of any additional variation in outcome measures as a consequence of clustered‐randomisation of surgeries rather than individuals. Analyses of outcome variables showed no additional variation from this source beyond that anticipated by between‐participant variation. Analysis of the 12‐month change in outcome variables was based on a mixed‐model repeated measures ANOVA. This analysis enabled estimation of any additional variation in outcome measures as a consequence of clustered‐randomisation of surgeries rather than individuals. Analyses of outcome variables showed no additional variation from this source beyond that anticipated by between‐participant variation. For this reason, all analyses were based on use of participants as replicates. When baseline differences in outcome measures were evident, ANCOVA for repeated measures was used to test the relative effects of treatments.