Freund 2016.

Study characteristics
Methods	Cluster‐RCT (115 clusters); follow‐up: 24 months; control: usual care
Participants	Eligible: 3065 Randomised: total 2076, I: 1093, C: 983 Randomised: COPD: 543, I: 321, C: 222 Completed: total 1718, I: 874, C: 844 (24‐month follow‐up) Completed: COPD: not reported Mean age: I: 71.6 years, C: 72.4 years Sex (% male): I: 48, C: 48 Inclusion criteria: 18 years or older and received medical treatment for ≥ 1 of the following index conditions at time of inclusion: type 2 diabetes mellitus, COPD, or chronic heart failure ; risk for future hospitalisation (i.e. predicted likelihood of hospitalisation within the upper quartile of the total population of health plan patients, as determined by analysis of data from the preceding 18 months Major exclusion criteria: active cancer (cancer diagnosis and current receipt of radiotherapy or chemotherapy), moderate to severe dementia, permanent residency in a nursing home, participation in a concurrent clinical trial (including telemonitoring studies), severe physical and mental disorders (such as dementia, psychotic disorder, or palliative care needs), other problems that hindered active participation in the intervention (such as language barriers), as assessed by the primary care physician
Interventions	Protocol‐based care management, including structured assessment, action planning, and monitoring delivered by medical assistants Intervention components were self‐management (education, action planning, exacerbation management), assessment of medical and non‐medical needs and resources, goal‐setting, follow‐up/communication tailored to patients' heath status (minimum every 6 weeks), case management, multi‐disciplinary teams. PCPs and HCAs were trained jointly in communication techniques and goal‐setting to enhance communication within the care management team, weekly review of patient progress between primary care physician and medical assistant practice teams received $135 per enrolled patient per year to cover staff costs as financial incentive Duration intervention: 12 months Involved disciplines: primary care physician, GP or general internist, medical assistant
Outcomes	Number of all‐cause hospitalisations at 12 months at the patient level (primary outcome); number of days in the hospital; hospitalisations related to index conditions; patient‐reported quality of life (SF‐12); general health (EQ‐5D); all‐cause mortality Intervention costs (estimation based on g standard wages for medical assistants' and physicians' working time). Only number of all‐cause hospitalisations (12 months, 24 months) reported for COPD separately
Notes	Unpublished data on COPD patients sought but not received Dominant component: self‐management
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "we used computer generated randomisation lists (SAS Version 9.2). Separate randomisation lists were prepared for urban and rural practices. A research assistant who was not otherwise involved in the project performed the central randomisation"
Allocation concealment (selection bias)	Low risk	Quote: "we concealed the allocation to intervention or control groups until each practice completed patient enrolment and baseline assessment"
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote: "because of the nature of the intervention, blinding primary care physicians, medical assistants, and patients was not possible" Comment: unlikely to affect primary outcome (number of hospitalisations derived from insurance data) but may affect some of the secondary outcomes (e.g. self‐reported quality of life)
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote: "we blinded the assessment of the primary and secondary end points as well as the responsible statistician to study group allocation"
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Quote: "for the quality‐of‐life measures, we performed analyses for the available cases (reported here) and used multiple imputation for incomplete data"; "results of the per protocol analysis and the multivariable models were similar to the results of the intention‐to‐treat analysis" Comment: furthermore, no missing data for 2 important outcome measures (all‐cause hospitalisation, number of days in hospital). Not all outcomes reported for participants with COPD specifically. Hence, impossible to conclude if missing outcome data are balanced in numbers across intervention and control groups
Selective reporting (reporting bias)	High risk	Comment: the PACIC, medication adherence, depression, self‐management capabilities, physical activity, activities of daily living, healthcare utilisation, total healthcare costs, blood pressure, MRC dyspnoea, forced expiratory volume, and number of exacerbations were mentioned in the protocol but were not reported in the results
Recruitment bias	Low risk	Quote: “we concealed the allocation to intervention or control groups until each practice completed patient enrolment and baseline assessment”; "we informed physicians about their allocation via an official letter and asked them to inform participating patients"
Baseline imbalance between groups	Unclear risk	Comment: practice and patient characteristics were similar between groups at baseline, with the exception of a slightly higher proportion of patients with COPD in the intervention group and a higher proportion from ethnic minorities in the usual care group. Investigators stratified randomisation according to population density of participating practice sites (urban vs rural) to minimise effects of population density on hospitalisation
Loss to follow‐up of clusters	Unclear risk	Comment: study authors describe a 10% attrition rate (see point 10 attrition), but loss to follow‐up of clusters is not mentioned, nor do study authors confirm that all clusters were present at follow‐up
Adequate analysis methods for CRT	Low risk	Quote: "we accounted for clustering within practices but were unable to account for clustering within physician/medical assistant teams within a practice (each of which had up to 2 teams)" Comment: intercluster correlation taken into account for sample size estimation