Chen 2017.

Study characteristics
Methods	Design: cluster‐RCT Date of study: August 1 2009 to October 31 Power calculation: not calculated before data collection Inclusion criteria: scheduled for elective abdominal surgery and expected LOS longer than 6 days (≥ 65 years of age) Exclusion criteria: not reported
Participants	Sample size: 557 Country: Taiwan Setting: two 36‐bed gastrointestinal wards of a 2000‐bed urban medical centre in Taipei Age: mean age 74.3 years (SD= 5.8) in intervention group, Mean age 74.8 (SD=6.0) in control group Sex: male, 111 (56.4%) in intervention group, male, 103 (57.2%) in control group Co‐morbidity: mean Charlson comorbidity index ‐ IG (1.6, 1.9), CG (1.5, 1.7) Dementia: cognitive, MMSE score range, 0‐30; 30 indicates no impairment, (mean, SD): IG (27.0, 3.8), CG (26.8, 3.1) P‐value 0.61 Frailty: not reported
Interventions	Intervention: the intervention was implemented by a mHELP nurse which is registered nurse who had 2 years of medical surgical experience and who was trained on site for 1 month before the intervention start. The intervention consisted of 3 protocols administered daily: orienting communication, oral and nutritional assistance, and early mobilisation. Intervention group participants received all 3 mHELP protocols with a median start time of postoperative day 1 (IQR 1‐3), in addition to usual care, as soon as they arrived in the inpatient ward and until hospital discharge. Control: usual care consisted of standard hospital care provided by surgeons, residents, nurses, and physical therapists (as needed) in the general surgery wards. All participants were encouraged to ambulate and did so as tolerated. The mHELP nurses did not provide services to participants assigned to the control group. However, the same attending physicians provided care to participants in the mHELP and control groups.
Outcomes	Outcomes reported: ‐ Incident delirium using CAM ‐ Inpatient mortality ‐ Length of hospital admission (days) ‐ Withdrawals (not explicitly included as an outcome but reported) Outcomes from study not reported: ‐ New diagnosis of dementia using change in MMSE at baseline, discharge, 4 and 6 weeks ‐ ADL using Barthel Index Frequency of outcomes assessment: daily from Monday to Saturday
Notes	Funding source: This study was supported in part by grants 98‐2314‐B‐002‐113‐MY3 from the Ministry of Science and Technology and NHRIEX‐9820PC from the National Health Research Institute in Taiwan (Dr C.C.‐H. Chen). Dr Inouye’s time was covered in part by grants R24AG054259, P01AG031720, K07AG041835, and R01AG044518 from the National Institute on Aging. Dr Inouye holds the Milton and Shirley F. Levy Family Chair. Declarations of interest: none reported Delirium not excluded at enrolment Intervention only delivered once participants discharged from ITU so varied from first postoperative day to after three days postoperatively
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Sequence generated based on computer‐generated list
Allocation concealment (selection bias)	Low risk	Cluster‐randomised to groups with an allocation ratio of 1:1. Cluster randomisation used to reduce risk of cross‐contamination due to shared occupancy rooms.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Due to nature of intervention, participants and personnel were unblinded. Intervention delivery was separate from outcome assessment.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Outcome assessors were masked to group assignment and room assignments were re‐randomised every 20 patients to minimise potential unmasking of the randomisation scheme
Incomplete outcome data (attrition bias) All outcomes	Low risk	All participants accounted‐for and low levels of attrition from analysis of outcomes, clearly described with reasons documented.
Selective reporting (reporting bias)	High risk	NCT protocol specifies other primary outcomes (frailty and bowel dysmotility) not reported here.
Other bias	Low risk	No evidence of recruitment bias or evidence of baseline imbalance associated with cluster methods. No clusters lost. Authors report that the intracluster correlation coefficient (ICC) was calculated for each outcome and were not significantly different from 0 and some were even less than 0, suggesting that the true ICCs are small and adjustment for cluster effect is not indicated. We thus analysed treatment effects using standard statistical methods not accounting for within‐cluster correlation.