Roberts 2001.

Study characteristics
Methods	Cluster‐RT Study duration: 12 months
Participants	3230 participants Setting: nursing homes Diagnostic criteria: none provided Age: participant characteristics not provided in terms of mean age, just percent of sample in intervention and control groups that were in particular age ranges Sex: participant characteristics not provided Country: Australia Comorbidity: not provided Ethnicity: participant characteristics not provided Date of study: unknown although paper was initially received for publication in May 2000
Interventions	1 intervention group The 12‐month intervention involved 3 phases: introducing a new professional role to stakeholders with relationship building, nurse education, and medication review by pharmacists who had a postgraduate diploma in clinical pharmacy. The clinical pharmacy service model introduced to each nursing home was supported with activities such as focus groups facilitated by a research nurse, written and telephone communication, and face‐to‐face professional contact between nursing home staff and clinical pharmacists on issues such as drug policy and specific resident problems, together with education and medication review. This was a multifaceted intervention directly targeting nursing homes. Most of the contact with GPs was indirect using the existing relationships between nursing homes and visiting GPs. A number of focus groups and personal interviews about the project were conducted with GPs. Control nursing homes continued with usual care.
Outcomes	Mortality was collected at the end of the 12‐month study.
Notes	No significant changes were observed in annual mortality rates or frequency of hospitalisations between intervention and control nursing home groups. It is unclear from Table 5, which shows the mortality and hospitalisation data, how the study authors arrived at their figures or their conclusions. Therefore, we were unable to use the data to calculate hospitalisations.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Nursing homes were selected for the intervention treatment by random draws from a hat.
Allocation concealment (selection bias)	Low risk	Not clear if this was done although the homes were independently assigned to the control or intervention groups. However, according to the EPOC criteria, the risk of bias for this study is low because units, in this case nursing homes, were assigned rather than individuals.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Unclear, although with the objective outcomes that we are interested in this is less of a concern (according to EPOC risk of bias criteria).
Incomplete outcome data (attrition bias) All outcomes	Low risk	Control and intervention groups did not appear to differ in terms of attrition.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	This is less important for our purpose as we are looking at objective outcomes (hospitalisations, ED visits, and mortality).
Selective reporting (reporting bias)	Low risk	No evidence of this
Protection against contamination bias	Low risk	There is indication to suggest that the intervention was contaminated by the control group.
Other bias	Unclear risk	The limited duration of the study and size of the sample may have compromised the ability to detect an effect. We considered other bias due to cluster randomisation.