Blum 2021.

Study characteristics
Methods	European multicentre, cluster‐randomised, controlled trial
Participants	2008 participants randomised (963 to 54 intervention clusters and 1045 to 56 control clusters) Participants admitted to 4 study centres in Bern (Switzerland), Utrecht (the Netherlands), Brussels (Belgium) and Cork (Ireland); 85 medical clusters and 25 surgical clusters Median (IQR) age: 79 (74 to 84) years 56% male Median (IQR) number of drugs: medication review group 10 (7 to 13), control group 9 (7 to 12)
Interventions	Intervention group: Pharmacotherapy optimisation based on the Systematic Tool to Reduce Inappropriate Prescribing through (1) systematic medication review by a physician and a pharmacist, with support of the STRIP Assistant, a software‐based tool taking into account the predictable adverse medication effects, advising safe and appropriate therapy using established STOPP/START criteria, monitoring clinically relevant interactions and dosing appropriately in accordance with renal function, (2) drug discussion and adaptation with the prescribing physician, (3) shared decision‐making with the patient and (4) generation of a report with specific recommendations for the patient’s general practitioner. Control group: Usual practice and a sham intervention using a questionnaire (Medication Adherence Measure Questionnaire, ©MMAS30–32*) by a team member (the physician or the pharmacist) to mimic the intervention and improve blinding of the patient and other blinded team members.
Outcomes	Primary outcome: drug‐related hospital admission within 1 year after enrolment Secondary outcomes: number of any hospitalisations, mortality, number of falls, quality of life, degree of polypharmacy, activities of daily living, patient’s drug compliance, as well as the number of significant drug–drug interactions, drug overuse and underuse and potentially inappropriate medication
Notes	Funding: This work is part of the project OPERAM: OPtimising thERapy to prevent Avoidable hospital admissions in the Multimorbid elderly supported by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 634238, and by the Swiss State Secretariat for Education, Research, and Innovation (SERI) under contract number 15.0137. This project was also partially funded by the Swiss National Scientific Foundation (SNSF 320030_188549).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	In this study design, each prescribing hospital physician defines a cluster. Physicians were allocated 1:1 to either the intervention arm or the control arm, using a probabilistic minimisation method implemented by a web‐based clinical trial management system (WebSpirit hosted by the Clinical Trials Unit (CTU) Bern). Minimisation was done according to country to ensure a balanced distribution of hospitals. The minimisation algorithm was implemented using randomisation lists generated by an independent statistician in Stata (StataCorp., Stata Statistical Software Version 14).
Allocation concealment (selection bias)	Low risk	Only system administrators who were otherwise not involved in the conduct of the trial had access to the randomisation lists, to ensure concealment of allocation.
Blinding of participants and personnel (performance bias) All outcomes	High risk	The intervention team consisted of a doctor and a pharmacist; neither was blinded to enable direct interactions with both the attending hospital doctors and the participants. The participants, hospital doctors and general practitioners were partially blinded and received only general information on the trial without specific details about the intervention.
Blinding of outcome assessment (detection bias) Mortality (all‐cause)	Low risk	Outcome assessors were fully blinded.
Blinding of outcome assessment (detection bias) Hospital readmissions (all‐cause)	Low risk	Outcome assessors were fully blinded.
Blinding of outcome assessment (detection bias) Hospital readmissions (due to adverse drug events)	Low risk	An independent blinded adjudication committee at each trial site, consisting of doctors and pharmacists, consecutively adjudicated all hospital admissions (both medical and surgical) for drug relatedness according to a previously published standardised adjudication guideline. When a hospital admission (at the index hospital or any other hospital) was identified, a second unblinded team gathered data on hospital admission and concealed all information identifying the intervention allocation before sending it to the adjudication team outcome assessors who were fully blinded.
Blinding of outcome assessment (detection bias) Adverse drug events	Low risk	Outcome assessors were fully blinded.
Blinding of outcome assessment (detection bias) Health‐related quality of life	Low risk	The teams conducting follow‐up telephone calls were fully blinded.
Incomplete outcome data (attrition bias) Mortality (all‐cause)	Low risk	During follow‐up, 10 (0.5%) participants were lost to follow‐up (7 in the intervention group and 3 in the control group), and 118 (5.9%) withdrew from the trial (50 in the intervention group and 62 in the control group).
Incomplete outcome data (attrition bias) Hospital readmissions (all‐cause)	Low risk	During follow‐up, 10 (0.5%) participants were lost to follow‐up (7 in the intervention group and 3 in the control group), 118 (5.9%) withdrew from the trial (50 in the intervention group and 62 in the control group), and 385 (19.2%) died (179 in the intervention group and 206 in the control group).
Incomplete outcome data (attrition bias) Hospital readmissions (due to adverse drug events)	Low risk	During follow‐up, 10 (0.5%) participants were lost to follow‐up (7 in the intervention group and 3 in the control group), 118 (5.9%) withdrew from the trial (50 in the intervention group and 62 in the control group), and 385 (19.2%) died (179 in the intervention group and 206 in the control group).
Incomplete outcome data (attrition bias) Adverse drug events	Low risk	During follow‐up, 10 (0.5%) participants were lost to follow‐up (7 in the intervention group and 3 in the control group), 118 (5.9%) withdrew from the trial (50 in the intervention group and 62 in the control group), and 385 (19.2%) died (179 in the intervention group and 206 in the control group).
Incomplete outcome data (attrition bias) Health‐related quality of life	High risk	Loss to follow‐up was 41% for the medication review group and 38% for the control group.
Selective reporting (reporting bias)	Low risk	No unpublished outcomes when paper is compared to published protocol or trial registration at clinicaltrials.gov
Contamination bias	Low risk	Cluster‐randomised. Cluster‐randomisation was at the doctor and not hospital level. Physicians may have worked in the same department, but the wards were in distinct locations, and each physician was responsible for his/her own non‐overlapping ward.
Other bias	Low risk	No evidence of other types of bias. Recruitment bias: low risk of bias (support for judgement: to limit selection bias, the recruitment team were fully blinded). Baseline imbalance: low risk of bias (support for judgement: Table 1 suggests no major baseline imbalance). Loss of clusters: low risk of bias (support for judgement: no loss of clusters). Incorrect analysis: low risk of bias (support for judgement: adjusted for clustering). Comparability with individually randomised trials: low risk of bias (support for judgement: allocated by attending physician resulting in patients being admitted to different wards, so no risk of herd effect).