This prospectively registered trial (Australia and New Zealand clinical trial registry Number: ACTRN12616000203459) was reviewed by the relevant Ethical institutional review boards and written informed consent was obtained from participants. A detailed methodology paper has been published [16]. Briefly, 60 adults aged between 19–68 years were recruited with a BMI above the healthy range (>25–43 kg/m2 for Caucasians and in those of Asian or Indian descent BMI > 23 kg/m2); who were newly diagnosed with moderate-severe OSA (AHI > 20 events/hour [17]) and who were recommended CPAP therapy. Overnight polysomnography was completed at screening and after 12 months to determine AHI. Specific details of the polysomnography are included in the Supplementary file.

A stepped-wedge, randomised control design (Fig. 1) was implemented [15]. This study design was chosen because stepped-wedge randomised trials involve within-cluster comparisons between time periods when participants are and are not exposed to an intervention. Thus, when examining when the lifestyle intervention had an effect on change in weight, comparisons can be made between different time periods within groups. Within group contrasts offer substantial statistical power advantages compared to between group contrasts thus our trial could be constructed with a smaller overall sample size than what may be present in conventional trials. Randomisation occurred after enrolment (simple randomisation, implemented using Microsoft Excel by the study statistician (TH)). These were stored in opaque envelopes until after recruitment and the baseline assessment was completed, when they were revealed to the researcher and the patient at the same time who then knew how long they would wait for their weight loss intervention to commence. Participants were asked not to reveal their group allocation to their treating CPAP technicians or sleep physician who were blinded to group allocation. Visits for data collection at the university were staggered by appointment time, thus minimising risk of participants mingling. All participants undertook one month of CPAP alone and then subjects commenced their lifestyle intervention at the end of month one through to month six.

A cluster-randomised stepped-wedge design. The 6 groups allocated to a delayed lifestyle intervention start of between 1–6 months. Symbols represent data collection points. CPAP treatment, lifestyle intervention, weight measurement, overnight polysomnography, visit to Air Liquide centre for CPAP support.

Full Polysomnography was performed in an in-patient setting (level 1) or in the patient’s home (level 2). The recording included electroencephalogram (EEG), bilateral electrooculogram (EOG), mentalis/submentalis electromyogram (EMG), anterior tibialis (left and right) EMG and electrocardiogram (ECG). Respiration was assessed via nasal pressure cannula + /− oronasal thermistor, thoracic and abdominal respiratory inductance plethysmography (RIP) bands, and fingertip pulse oximetry.

Sleep stages, arousal, and respiratory events were scored according to American Academy of Sleep Medicine (AASM) 2012 recommended criteria using Profusion PSG3 software (Compumedics, Abbotsford, Victoria, Australia) [17]. Specifically, AHI was defined as the total number of apnoea’s and hypopneas per hour of sleep. Apnoea was defined as a ≥ 90% decrease in oronasal airflow for at least 10 s. Hypopneas were scored when nasal pressure signal dropped by ≥30% from baseline for ≥10 s and the event was associated with either a 3% or greater fall in oxygen and/or an arousal from sleep.

CPAP treatment was provided centrally for all participants by Air Liquide Healthcare (Melbourne, Australia). For the first week, participants were commenced on an auto titrating CPAP and afterwards were switched to a fixed pressure based on the 95th centile pressure. Routine care and decisions regarding treatment changes were provided by the participants’ treating physician, who were blinded to group allocation.

Participants were prescribed an intermittent energy restricted dietary pattern, which consisted of 5 days of restricted daily energy intake of 6300–7500 kilojoules (kJ) per day and were provided with meal plans to support their food selections and on 2 days per week, a very low energy intake (2200–2760 kJ per day) was prescribed by providing participants with a powered milk-shake that they made up at home along with multivitamins with fish oil, fibre and iron (women only). Participants were encouraged to eat low starch vegetables/salads and low joule fluids ad libitum. This procedure ensured overall good diet quality including micro-nutrient adequacy on the ‘fasting’ days [16]. One-to-one monthly appointments with the study dietician (KR) started when the participants were eligible to commence the active weight loss phase. For the duration of the six-month intervention, the dietician provided structured advice grounded in Michie’s behaviour change theory [18], with months one to three focusing on active weight loss, whilst months four to six focused on strategies to maintain lost weight. At the end of the 6-months of the intervention, participants were discharged from the study clinic having been prepared with strategies to maintain their weight on their own.

Participants were offered the use of a smartphone app (MyPace™) for communication with the dietician in between their monthly face-to-face appointments. On this App, participants could set and monitor their progress towards their own goals, track their weight loss and receive motivational texts. Participants were advised to increase their physical activity levels (i.e. 3 × 30 min/week moderate activity such as walking or swimming, a total of 90 min per week) by the dietician at their first appointment and were provided with a Fitbit© to help participants’ monitor their own physical activity levels. Participants provided separate consent for the research team to access their individual web based FitBit© account to monitor usage and activity. App usage and physical activity are provided in the Supplementary file.

Weight to the nearest 0.1 kg was measured monthly (SECA Clara 803) without shoes.

A detailed description of the statistical analysis protocol has been previously described [16] with further detail provided in the supplementary file. Multilevel, mixed effects generalized linear models were used to investigate the primary aims. These linear mixed models include all participants with data regardless of whether all follow-up data points were collected or not. Our primary analysis compared weight measurements taken during the intervention period of the stepped wedge, to those taken during the earlier control period. The planned approach was to undertake this contrast using one “change in weight (Tn minus Tn-1)” measurement for each participant for each month of the stepped wedge trial. A further analysis used data from both the stepped wedge portion of the design and the 12-month follow-up measurement. Participant number was treated as a random effect to account for dependency of observations within each participant, a complete case analysis was performed, and an intention-to-treat approach employed. A categorical fixed effect for month since trial commencement was included to account for temporal trends [15, 19]. Primary aim 2) was examined using analysis two, but with the addition of an intervention-by-month of commencement interaction effect. Primary aim 2) was also investigated using an ANCOVA-style linear regression approach (analysis three) using only the final assessment of weight treated as the dependent variable, the first weight measurement as a covariate, and month of commencement of the intervention as a categorical independent variable. Treating month of commencement as a linear variable + /− transformations was explored if analyses two & three revealed a linear, quadratic or other trend in the effect of month of commencement on the dependent variable (analysis four).

Secondary aim 1) was assessed using only pre-intervention period data from the stepped wedge portion of the trial using a multi-level, mixed effects, generalized linear model (analysis five). Trial participants were treated as a random effect.

Recruiting seven participants for each of the six starting time points of the lifestyle intervention provided 90% power to detect a standardised effect size of 0.40 for the rate of change in body weight outcome for the primary analysis. This assumes a conservative intraclass correlation coefficient of 0.10, two-tailed α = 0.05, and treats each individual participant as its own cluster (as this is the unit of randomisation). We collected 10 participants per group to maintain trial power allowing for a potential drop-out/missing data rate of over 30%.

Treating physicians were blinded to group allocation but participants were not due to the necessity for them to undertake the weight loss intervention.

We had intended to use an individual’s monthly average CPAP treatment adherence since the previous assessment as a covariate in analyses one to four, and for contamination-adjusted intention-to-treat analyses (planned analyses six to eight). However, data extracted from CPAP machines used in this trial were captured as summative across the trial period and not on a month-by-month basis, preventing us from being able to use this variable for these analyses. Planned analyses six to eight described in the protocol were not undertaken. One participant was recruited into the study but after further analysis of the overnight sleep study, was found to have only mild OSA and was therefore excluded from all analyses.

We undertook a post-hoc sensitivity analysis for comparisons of weight between intervention and control periods using raw weight as the dependent variable instead of change in weight from the previous assessment to assist with interpretation of the effect size estimates generated. We examined whether those who continued use of CPAP had different mean weight scores at 12-month follow-up compared to those who did not commence or discontinued CPAP, adjusted for weight at the baseline assessment using linear regression.