O'Mahony 2020.

Study characteristics
Methods	Randomised controlled trial
Participants	1537 participants randomised (772 to medication review and 752 to control) Patients were admitted to 6 European medical centres (Ireland, Scotland, Spain, Italy, Belgium and Iceland), (13 medical and eight surgical clusters) Median (IQR) age: 78 (72 o 84) years 53% male Median (IQR) number of medications: 10 (8 to 13)
Interventions	Customised SENATOR software input information included patients’ diagnoses (ICD‐10 codes), prescription drugs (ATC codes) and doses, renal function (MDRD formula), liver function (liver transaminases, INR), cardiac rhythm (electrocardiogram) and complete blood count. By applying STOPP/START criteria (version 2) alongside potentially relevant drug‐drug and drug‐disease interaction information from local databases, SENATOR software produced an individualised medication advice report. Primary researchers subsequently notified senior attending physicians of intervention arm patients of the SENATOR reports and inserted copies into patients’ medical records, i.e. printed reports into paper‐based records or electronic reports into electronic records. Participants in the control group received standard pharmaceutical care as provided at each site at time of randomisation.
Outcomes	Primary outcome: the occurrence of probable or certain ADRs within 14 days of randomisation Secondary outcomes: primary endpoint derivatives Tertiary outcomes: all‐cause mortality, re‐hospitalisation, composite healthcare utilisation and health‐related quality of life
Notes	Funding: supported by the European Commission’s Seventh Framework Programme (FP7/2007–2013) (grant number 305930) as part of the SENATOR project
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	An independent statistician used random block sizes to generate stratum‐specific randomisation lists.
Allocation concealment (selection bias)	Low risk	The randomisation lists were integrated into the eCRF so that researchers could not access them, and any given allocations were only revealed once patients were unambiguously enrolled into the trial and their screening information was irreversibly entered onto the eCRF.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Primary researchers who recruited the patients were not blinded and attending hospital physicians were not blinded. However, the patients were blinded and those assessing the primary outcome (i.e. ADR) were also blinded.
Blinding of outcome assessment (detection bias) Mortality (all‐cause)	Low risk	Mortality data were assessed from patients’ medical records and national registries in all but one site, where death was established by telephone follow‐up. Since the outcome of mortality is assessed from records in most countries, lack of blinding will likely not influence mortality.
Blinding of outcome assessment (detection bias) Hospital readmissions (all‐cause)	Low risk	Readmissions were assessed by post‐hospital discharge follow‐up call by the local site research staff who were not blinded. However, the outcome of readmissions is objective and will likely not be influenced by the lack of blinding.
Incomplete outcome data (attrition bias) Mortality (all‐cause)	Low risk	Loss to follow‐up: 6.8% (52/766) in the medication review group, 7.4% (56/752) in the control group.
Incomplete outcome data (attrition bias) Hospital readmissions (all‐cause)	Low risk	Loss to follow‐up: 5.4% (41/766) in the medication review group, 5.3% (40/752) in the control group.
Selective reporting (reporting bias)	Low risk	Health‐related quality of life was prespecified in a trial registry, but according to the corresponding author it was decided not to measure the outcome due to lack of resources. Therefore we assessed as low risk of selective reporting.
Contamination bias	High risk	No cluster‐randomisation.
Other bias	Low risk	No evidence of other types of bias.