Gustafsson 2017.

Study characteristics
Methods	Randomised controlled trial
Participants	460 participants randomised (230 to medication reviews and 230 to control) Patients aged ≥ 65 years with dementia or cognitive impairment admitted to 3 wards at 2 hospitals located in Northern Sweden were included Mean age: 83.1 years 36% male Mean number of drugs: control group 8.3, intervention group 8.4
Interventions	Three clinical pharmacists with postgraduate degrees in clinical pharmacy and long experience in performing medication reviews in primary care and hospital wards conducted the interventions. The pharmacists were already part of the different ward teams at the time when the study started. The additional service provided by the clinical pharmacists consisted of medication reconciliation, medication review and participation in ward rounds. Participants in the control group received usual care.
Outcomes	Primary outcome: risk of drug‐related readmissions at 180 days Secondary outcomes: all‐cause readmission at 30 days and 180 days, mortality, cost analysis (not yet analysed), time to institutionalisation (not yet analysed) and adherence to quality indicators (not yet analysed)
Notes	Funding: grants from the Swedish Dementia Association, the County Council of Västerbotten, the Janne Elgqvists foundation, the Swedish Society of Medicine, and the foundation for Medical Research in Skellefteå
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	The randomisation sequence was prepared before study start using a throwing dice method by an independent person who was not engaged in the trial in any other way. The sequence was performed in blocks of 6 to 36 (each block contained between 3 and 18 intervention allocations and the same number of control allocations). Randomisation was stratified at ward level. To accomplish this, each ward used their own randomisation blocks, consecutively starting a new block after completion of the preceding, meaning that there were an equal number of control and intervention participants in each ward.
Allocation concealment (selection bias)	Low risk	When a patient formally entered the trial, an employee of the Department of Pharmacology and Clinical Neuroscience, who was not involved in the interventions, provided the treatment allocation according to the randomisation scheme.
Blinding of participants and personnel (performance bias) All outcomes	High risk	The participants and pharmacists were not blinded to treatment assignment.
Blinding of outcome assessment (detection bias) Mortality (all‐cause)	Low risk	An independent, blinded external expert group consisting of one specialist in geriatrics, one specialist in internal medicine and one clinical pharmacist working in another county assessed the outcomes.
Blinding of outcome assessment (detection bias) Hospital readmissions (all‐cause)	Low risk	An independent, blinded external expert group consisting of one specialist in geriatrics, one specialist in internal medicine and one clinical pharmacist working in another county assessed the outcomes.
Blinding of outcome assessment (detection bias) Hospital readmissions (due to adverse drug events)	Low risk	An independent, blinded external expert group consisting of one specialist in geriatrics, one specialist in internal medicine and one clinical pharmacist working in another county assessed the outcomes.
Blinding of outcome assessment (detection bias) Hospital emergency department contacts (all‐cause)	Low risk	Not described but will not likely influence ED contacts.
Incomplete outcome data (attrition bias) Mortality (all‐cause)	Low risk	One participant withdrew from the control group before discharge; none lost to follow‐up in both groups (why assessed as low risk of bias).
Incomplete outcome data (attrition bias) Hospital readmissions (all‐cause)	Low risk	One participant withdrew from the control group before discharge; none lost to follow‐up in both groups (why assessed as low risk of bias).
Incomplete outcome data (attrition bias) Hospital readmissions (due to adverse drug events)	Low risk	One participant withdrew from the control group before discharge; none lost to follow‐up in both groups (why assessed as low risk of bias).
Incomplete outcome data (attrition bias) Hospital emergency department contacts (all‐cause)	Low risk	One participant withdrew from the control group before discharge; none lost to follow‐up in both groups (why assessed as low risk of bias).
Selective reporting (reporting bias)	Low risk	Mortality was not prespecified as outcome on clinicaltrials.gov, but direction of effect does not suggest selective reporting. We therefore judged the outcome of mortality as having low risk of bias for selective reporting. The 30‐day readmission analysis was not pre‐specified in the trial registration at clinicaltrials.gov. The trial authors were contacted and replied “The 30‐days readmission analysis was not pre‐specified in the study protocol However, because of the increased use of 30‐ day readmission as an indicator of quality of care, the outcome was added as a post‐hoc analysis after the study was started”. Frequency of hospital visits (readmissions and emergency department) during the 6‐month follow‐up is a secondary outcome on clinicaltrials.gov, but emergency department contacts not reported in the publication. The trial authors sent data on readmissions and ED visits and the direction of effect does not suggest selective reporting. We therefore judged both outcomes as low risk of bias for selective reporting.
Contamination bias	High risk	No cluster‐randomisation.
Other bias	Low risk	No evidence of other types of bias.