Nadler 2017.

Study characteristics
Methods	Design: RCT Date of study: 2014 (unclear ‐ not explicitly reported) Power calculation: yes (study underpowered, baseline rate of delirium lower than expected) Inclusion criteria: aged 50 years or older, at risk of obstructive sleep apnoea as defined by a STOP‐Bang score ≥ 3, scheduled for elective knee or hip arthroplasty, able to speak English, understand consent forms and give informed consent. Exclusion criteria: patients with severe tracheal or lung disease (e.g. bullous lung disease, pneumothorax, recent tracheal anastomosis) or contra‐indications to nasal‐mask CPAP (e.g. facial fractures/lacerations/burns, recent ENT surgery, basilar skull fracture, tracheostomy); Patients with previously diagnosed obstructive sleep apnoea
Participants	Sample size: 135 Country: USA Setting: perioperative hospitalised patients undergoing joint arthroplasty population could be shortened to elective orthopaedic Age: mean age 65.1 (SD = 8.4) in intervention group, mean age 66.3 (SD = 9.4) in control group Sex: male, 22 (32.4%) in intervention group, male 24 (35.8%) in control group Co‐morbidity: imbalance in rates of depression and visual or hearing impairment in intervention group compared to control. Dementia: dementia or significant cognitive impairment, 0 in intervention group, 2 (3%) in control group Frailty: not reported
Interventions	Intervention: CPAP before surgery and days 0 1 and 2 postoperative (variable amount of time between enrolment and surgery) Control: usual care
Outcomes	Outcomes reported: ‐Incident delirium using DRS‐R‐98 diagnostic assessment tool Outcomes from study not reported: ‐Incident delirium using CAM Frequency of outcomes assessment: once post op day 2
Notes	Funding source: Equipment loaned by Philips Respironics, Amsterdam. No other funding source Supported by the Department of Psychiatry Duke University Medical Centre, Durham, NC, USA Declarations of interest: none stated. Delirium not excluded at enrolment
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation performed by consulting a computer‐permuted sequence that guaranteed an equal number of patients in each arm within blocks of 20
Allocation concealment (selection bias)	Unclear risk	No information provided around how allocations were concealed.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants not blinded due to nature of intervention.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Investigators performing assessments were blinded to study group assignment and CPA devices were recovered from the group before assessment took place.
Incomplete outcome data (attrition bias) All outcomes	High risk	Assessments not performed in 10% of intervention group and 13% of control group and unclear why.
Selective reporting (reporting bias)	Low risk	Outcomes reported as per NCT protocol
Other bias	Low risk	No evidence of other bias