Shapiro 2017.

Methods	Randomsied parallel‐group trial
Participants	N = 46 newly‐diagnosed with OSA and prescribed CPAP for the first time. Inclusion criteria: ≥ 18 years; newly‐diagnosed by PSG; commencing CPAP for first time; able to read/speak/understand/write English; CPAP with smart card technology Exclusion criteria: requires BiPAP, significant craniofacial abnormalities, Downs syndrome, cognitive delay, hypotonia, neuromuscular degenerative disorder, taking anti‐anxiety medication, pregnant. Baseline characteristics: 45.5% female, Mean age 51.8 (13.1). Mean AHI 26.2. Mean ESS NR. Mean BMI 35.7. Country: USA
Interventions	Participants were randomised to standard care (SC, n = 33) or CPAP‐SAVER Intervention (CI, n = 33). SC: basic OSA and CPAP teaching and follow‐up provided by respiratory therapist/CPAP education employed by home medical supplier. CI: Standard care plus airway model, video education sheet, report card components, support calls.
Outcomes	CPAP usage (hours/night) at 1 month Anxiety symptoms (BAI) at 1 month Withdrawals
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation procedures adequately described and were sufficient.
Allocation concealment (selection bias)	Unclear risk	There is insufficient information regarding allocation concealment to make a determination as to whether it was possible for enrolling investigators/staff to have knowledge of forthcoming allocation.
Blinding (performance bias and detection bias) All outcomes	Unclear risk	Authors report participants were "masked as to group assignment," but do not describe the methods by which this masking was carried out for a behavioural intervention. Study had subjective outcomes, so knowing group assignment could influence these outcomes.
Incomplete outcome data (attrition bias) All outcomes	Low risk	1 of 33 (intervention) and 0 of 33 (control) were lost to attrition.
Selective reporting (reporting bias)	Low risk	This is a dissertation and the author presents a comprehensive protocol with design and analytic plan. Thus, it is likely that analytic plan was finalized before unblinded outcome data were available for analysis.
Other bias	Low risk	No baseline imbalances reported for variables of interest. Although there was a deviation from protocol (reduction of number of study sites) this seems to not have resulted in a deviation from the intended intervention.
Bias arising from the randomisation process (ROB2, primary outcome)	Unclear risk	After consent, participants were randomly assigned to either the intervention or control group, and were masked as to group assignment. Randomisation procedures: the investigator prepared 33 manila clasp envelopes; the contents included a sheet with the word Intervention printed on it...The envelopes were numbered sequentially from one through 33; this number served as the participant's identification number.... The investigator prepared another 33 manila envelopes; the content included a sheet with the words Standard care printed on it...The envelopes were numbered sequentially from 34 through 66; this number served as the participant's identification number. The investigator mixed all the envelopes into one batch and shuffled them 10 times. The investigator divided the shuffled envelopes into four piles, one pile for each home medical supply facility site (Sites A, B, C, and D). According to this description, the investigator who prepared the envelopes would be aware of the allocation contained within each envelope based on viewing the outside of the envelope. It is not clear whether any other investigator/staff (e.g. those responsible for selecting the next envelope from the stack) was also aware of the sequence correlation (i.e. that lower numbers were assigned to intervention and higher numbers to control). Additionally, author reports that, due to withdrawal of sites from the study, "unused envelopes from Sites B, C, and D were taken to Site A." It is unclear who delivered those unused envelopes. There is insufficient information regarding allocation concealment to make a determination as to whether it was possible for enrolling investigators/staff to have knowledge of forthcoming allocation. Mean and SD for all key baseline characteristics (age, gender, BMI, AHI) were reported for all randomised participants. Authors reported that differences were non‐significant: "After comparing the frequencies (chi‐square tests) and descriptives (independent samples t‐tests) analyses by group, the groups were determined to be homogeneous. There were no statistically significant differences between the intervention and standard care groups as to general and sleep demographics."
Bias due to deviations from intended interventions (ROB2, primary outcome)	Low risk	Authors report participants were "masked as to group assignment," but do not describe the methods by which this masking was carried out for a behavioral intervention. Authors report, "The initial number of sites was four (Sites A, B, C, and D). Two sites were lost to attrition (Sites C and D) due to the unavailability of a respiratory therapist to provide CPAP teaching. As these two sites withdrew from the study, recruitment slowed at Site B, and recruitment progressed at Site A, unused envelopes from Sites B, C, and D were taken to Site A. Additional envelopes were provided to sites as needed; all 66 envelopes were used." Though this is a deviation from planned randomisation protocol, it did not appear to result in a deviation from the intended intervention. Moreover, "[I]ntervention fidelity was maintained with the use of a protocol training manual and initial and booster training sessions to prepare the research assistants; monthly fidelity checks were conducted with no problems noted." (p. 102) From Methods, "If participants were noted to have values that appeared missing at random, their data was included in the analyses. Other missing data were evaluated on a case‐by‐case basis and excluded from analyses" From Results, "One intervention group participant was lost to attrition midway through the study. Missing data were determined to be missing at random and were included in the statistical analyses; where applicable, cases were excluded pairwise." These descriptions do not describe how missing data were 'included,' e.g. last observation carried forward, imputation, set to 0, excluded. However, there is no evidence that trialists used either 'as‐treated' or naive per‐protocol analysis. Rather, report suggests that all participants were analysed in the group to which they were randomised.
Bias due to missing outcome data (ROB2, primary outcome) All outcomes	Low risk	1 of 33 (intervention) and 0 of 33 (control) were lost to attrition.
Bias in measurement of the outcome (ROB2, primary outcome) All outcomes	Low risk	Outcome measured using objective CPAP usage data. Each intervention group outcome data ascertained via automated CPAP device monitoring; devices identical or sufficiently similar (i.e. similar distributions of CPAP device make) across groups. The distribution of Philips and ResMed devices were equal in each arm. Outcome "assessor" is CPAP device: no knowledge of allocation possible.
Bias in selection of the reported result (ROB2, primary outcome) All outcomes	Low risk	Authors report, "Before data was collected, IRB approval and informed consent were obtained." Additionally, this is a dissertation and the author presents a comprehensive protocol with design and analytic plan. Thus, it is likely that analytic plan was finalised before unblinded outcome data were available for analysis. Methods section indicates that two time points were planned; each planned outcome reported. Methods section indicates that one, commonly‐employed threshold adherence definition was planned; this outcome was reported in Results.
Overall risk of bias (ROB2, primary outcome) Machine usage	Unclear risk	‐