Brown 2013.

Methods	Parallel randomised trial Method of randomisation: randomised stated, method unclear Method of concealment: used sham control with device of identical appearance of that for active CPAP Blinding: double‐blind Analysis: only reported data of participants who had completed the 3‐month follow‐up (available‐case analysis)
Participants	Location: USA Setting: enrolled from a single centre of inpatient neurology service, the treatment was conducted in the community Number of participants: 32 at randomisation (59% male, median unknown), 19 completed study, case‐available analysis for 19 participants Treatment group: 15 (33% male, median age 61 years, IQR 46 to 76) Control group: 17 (76% male, median age 74 years, IQR 55 to 81) Stroke criteria: ischaemic stroke based on accepted clinical criteria Time since stroke onset at randomisation: median time from stroke onset to CPAP titration was 4 days Fatigue criteria: participants did not have to have fatigue to be recruited Other entry criteria: had sleep apnoea; modified Rankin Scale ≥ 2 Comparability of groups: unclear
Interventions	Treatment intervention: active CPAP Control intervention: sham CPAP Treatment duration: 3 months Delivered by: voluntary use by participants
Outcomes	Time for fatigue assessment: by the end of 3‐month use of CPAP Primary outcome: objective usage data of CPAP Secondary outcome: FSS score
Notes	Only people with sleep apnoea were recruited This study was primarily aimed at testing the feasibility of CPAP to treat sleep apnoea in people with stroke. Fatigue was measured as 1 of the symptoms of sleep apnoea for the efficacy of CPAP The investigators reported median scores and IQR of FSS and we requested mean scores and SD from the investigators Funding: National Institutes of Health Grant K23 NS051202 and National Center for Research Resources Grant M01‐RR000042
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment (selection bias)	Low risk	Quote: "The active and sham CPAP units were identical in appearance, with the exception of the barely discernible augmented leak port on the latter, and made similar sounds while on"; "blinding to subject condition was successfully maintained"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote: "Outcome assessors were masked to intervention allocation"
Intention‐to‐treat	High risk	Available‐case analysis: only reported data of participants who had completed the 3‐month follow‐up