We will study a total of 400 subjects with OSA (200 at each site) from the Clinical Centers of Excellence (CCE) for responders (n = 2100) at NYU SOM and at EOHSI, Rutgers RWJMS (n = 1700). Recruitment for the study began in March of 2013 and will continue until August of 2016.

Inclusion criteria: 1) Member of the WTCHP at NYU SOM or Rutgers RWJMS, 2) OSA diagnosed on the basis of a 2-night home sleep study.

Exclusion criteria: (i) Gross skeletal alterations affecting the upper airway (eg. Micrognathia), (ii) Unstable chronic medical conditions known to affect OSA (congestive heart failure, stroke), (iii) Pregnancy or intent to become pregnant within the period of the protocol, (iv) Inability to sign informed consent form, (v) habitual snorer or diagnosis of OSA prior to 11 September 2001.

Anterior rhinoscopy is performed to identify erythema, edema, polyps or polypoid swelling, crusting, mucin, and or frank pus and documented.

In 2005, the Standardization Committee on Objective Assessment of the Nasal Airway recommended that 4-phase rhinomanometry be adopted as the universal standard for measurement of nasal resistance [22]. It consists of simultaneous measurement of airflow through the nose and differential pressure required for its generation. By dividing inspiration and expiration into increasing and decreasing phase, high-resolution rhinomanometry (HRR) introduces two new parameters: effective resistance and vertex resistance and has been validated in normal and pathological individuals. We assess nasal resistance using the 4-phase-rhinomanometer (RhinoLab GmbH, Rendsburg, Germany) following the detailed technical, practical methodology, and referenced normative data published in the literature [22]. Measurements are performed seated and in the supine position (before and 10 minutes after decongestion with 0.1 % xylometazoline solution) in a quiet procedure room with a constant temperature. Parameters obtained from the HRR include the unilateral effective and vertex resistances during inspiration, expiration and during the entire breath. “Effective Resistance (Reff)” describes the computerized measurement and calculation of 2000 effective flow and differential pressure measurements recorded for each averaged breath. “Vertex Resistance” is the resistance of the nasal airstream at the point of maximum flow during inspiration or expiration in a breath. Total nasal resistance is calculated using Ohm’s law modeling the nose as parallel resistors using the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathrm{Total}\ \mathrm{Nasal}\ \mathrm{Resistance} = \mathrm{Rright}\ *\ \mathrm{Rleft}/\left(\mathrm{Rright} + \mathrm{Rleft}\right). $$\end{document}TotalNasalResistance=Rright*Rleft/Rright+Rleft.

Effective and vertex resistance will be presented after the recommended logarithmic transformation. Log Reff ≤ 0.8 is defined as low resistance and > 0.8 as moderate-high resistance.

Subjects are given an ARES™ Unicorder (Watermark Medical Inc., West Palm Beach, FL, USA) to take home and wear for two nights, with a pre-addressed mailer to return the device to the sleep lab. The ARES Unicorder is worn on the forehead and does not require additional wires to external devices. It measures oxygen saturation and pulse rate from reflectance oximetry, airflow from a nasal cannula/pressure transducer, snoring via acoustic microphone and head movement actigraphy and head position from accelerometers. The device also provides audible alerts during the study if poor quality airflow or SpO₂ is detected so the subject can reposition the device.

Data from the monitor is autoscored and then manually reviewed by a single trained sleep technician at NYU SOM. There has been significant evolution of the definition of OSA in the epidemiological literature since the landmark study of Young et al. [5], where a reported incidence of 2–4 % was based on an Apnea + Hypopnea Index (AHI) > 5/hour and symptoms of excessive daytime somnolence. Apneas are scored when there is a reduction in airflow to less than 10 % of baseline. Hypopneas4 % are scored for > 30 % reduction in airflow associated with 4 % or more decrease in oxygen saturation. In most of the studies since, including the Sleep Heart Health Study [23], hypopnea has been defined as a reduction in airflow amplitude associated with desaturation and/or arousal, but the prevalence of OSA is five-fold greater when arousals are included without associated desaturation. The American Academy of Sleep Medicine has issued several definitions of OSA, most recently in 2007 [24] but all require an electroencephalogram (EEG) and there is no published consensus as to how to score home monitoring without EEG. The approach we take is to use AHI4 % (nearly identical to the recommended AHI of the AASM AHI) and to define the Respiratory Disturbance Index (RDI) as the sum of AHI4 % and the more subtle respiratory events (Hypopnea 1 %, which approximates the AASM “RERAs” or Respiratory Effort Related Arousals defined with an EEG). Hypopneas 1 % are scored for > 30 % reduction in airflow associated with 1 % decrease in oxygen saturation/or surrogate of arousal. AHI 4 % is calculated as apneas + hypopneas4 % divided by total valid recording time. The RDI is calculated as apneas + hypopneas4 % + hypopneas1 % divided by total valid recording time. Using these metrics, we define OSA as present when AHI 4 % > 5/hour or when RDI > 15/hour. We have extensive experience using the ARES device and have validated it against NPSG in over 300 subjects in multiple research studies [25, 26]. In our published validation study the diagnostic sensitivity for diagnosing OSA using a cutoff for RDI of 15/hour ranged from 85–95 % and specificity from 91–94 %. Of particular note the failure rate for acquiring scorable data is < 10 % [25].

We have previously compared ambulatory studies performed with the ARES to laboratory NPSG [25, 27], and shown that, with the ARES, AHI4% is well measured and AHI1% is functionally equivalent to the RDI.

After results of the ambulatory monitoring identify OSA, subjects return to the study center for discussion with study staff regarding therapy. If they agree to CPAP, they undergo mask-fitting and desensitization and CPAP education during this visit. CPAP education follows a written protocol common to both sites. During the first week, the research co-coordinator calls the patient twice (day 3 and day 7) to discuss any mask or CPAP-related issues. Period 1 of the intervention begins subsequent to this break-in week and lasts for 4 weeks. Treatment is switched remotely and period 2 of the intervention begins subsequent to that and last 4 weeks.

OSA patients are randomly allocated to CPAP or CPAP_Flexstratified by site and low (log[Reff] ≤ 0.8) and high (log[Reff] > 0.8) resistance, gender and AHI (> or < 30 events per hour of sleep) by the statistician. This is accomplished by a computer generated allocation table of randomly permuted blocks of assignments to study condition (CPAP or CPAP_Flex) for each site. Outcome of the randomized choice of treatment is provided to an unblinded individual who is not part of the analysis of the study data, when a subject arrives to the center, and who remotely sets the allocated CPAP device at the start of each treatment period. Devices are tracked by serial number. Trial participants are blinded to the therapy type but may be able to perceive a difference in the mask pressure during expiration. CPAP and CPAP_Flex machines are identical otherwise and no feedback is provided to subjects regarding the type of treatment. All investigators and research personnel who enroll and interact with the participants, outcomes assessors and data analysts are blinded to the intervention and allocation as it is remotely set and not evident on the device. The unblinded individual (not the PI) maintains a master password protected file containing the serial number, patient ID and designation code for each treatment period and subject. Except in a medical emergency situation this file will not be opened until the study has been completed and all data have been entered/cleaned. The unblinded individual will document any premature unblinding that may occur. The unblinded individual also reviews the data relevant to treatment efficacy during the first 5 days of autotitration and significant deviations from prescribed pressure due to leak during the rest of the intervention period.

All subjects diagnosed with OSA are given an autotitrating machine. We use the Respironics AutoCPAP device which has capability of autotitration of pressure with or without Cflex set at a level 3. The devices communicate with the sleep centers via modem or broadband (HIPPAA- compliant). The auto-titrating positive airway pressure (APAP) with a range of pressures from 5–15 cm H₂O with or without CPAPflex is set when the subject presses a modem button after turning on the machine at home. These machines have been used extensively [28, 29] and are used to determine a single optimal pressure after a period of autotitration. During the first 5 days of autotitration, summary raw airflow data collected are reviewed by an unblinded technician (see blinding above) to ensure efficacy of treatment. After 5 nights of at least 4 hours of use, the device automatically switches to a fixed pressure from the 90th percentile of pressures achieved and used for the remainder of the study. Patients who are non-compliant (use for less than 4 hours a night) during the first 5 nights remain on autotitrating pressures. The automated AHI is monitored to ensure efficacy of therapy (AHI of < 10 events per hour of sleep). If the AHI is > 10 events per hour of sleep the data is reviewed by a designated unblinded sleep physician and the pressure is modified or the patient is recommended for an in laboratory sleep study if indicated. All patients are provided heated humidification.

Efficacy will be evaluated by (i) reviewing residual AHI and inspiratory flow limitation at optimal pressure as recorded on the device and (ii) review of raw airflow signal. Nightly adherence at the optimal pressure is recorded on the device and also transmitted to the sleep center onto a HIPPAA-compliant website that is password protected.

Four weeks after the fixed pressure is achieved following the first intervention (ie CPAP or CPAP_Flex), the device will be switched to the alternate mode by remote modem connection and the autotrial is started again. This duration was chosen because long-term CPAP adherence has been consistently shown to be predicted primarily by early short-term CPAP adherence [30–33].

At the end of each intervention period subjects will fill out the Epworth Sleepiness Scale (ESS), Functional Outcome of Sleep Questionnaire (FOSQ) and a satisfaction questionnaire [29].

We will randomize 400 subjects with OSA, stratified by nasal resistance, into two sequences of treatment: CPAP_Flex followed by CPAP alone versus CPAP alone followed by CPAP_Flex. To determine if high nasal resistance is associated with decreased CPAP adherence, we based our power analysis on the method of Fisher’s Z test [34] for correlation coefficients. With 400 participants and setting alpha at 0.05 (2-sided), we will have 80 % power to test a small (negative) correlation of −0.14. To determine if use of CPAP_Flex will improve adherence with CPAP in subjects with high nasal resistance, but not in those with low nasal resistance, we assume that 50 % of our subjects (200) will have high nasal resistance, based on our pilot data in non-WTC subjects. In Aloia et al. [20], they found that CPAP_Flex significantly improved adherence of CPAP at 4 weeks (3.5 ± 2.8 hours for CPAP and 4.7 ± 2.2 hours for CPAP_Flex corresponding to Cohen’s d of 0.49). Assume that the effect size of CPAP_Flex improvement between high versus low nasal resistance is 80 % of the effect of Aloia et al., Cohen’s d = 0.40, representing an improvement of 20 minutes per night assuming the SD of the improvement to be 60 minutes (in paired/cross-over studies SD is usually smaller than that in unpaired studies). With set power = 80 % and alpha = 5 % (2-sided), to test an effect size d = 0.40 we will need 100 subjects each with high and low nasal resistance, based on the method of a 2-sample t test with equal variance. With a total of 400 subjects, we have enough power to test this hypothesis.

To determine if the benefit of CPAP_Flex on adherence will be greatest if it is offered at CPAP initiation rather than as a “rescue” therapy in subjects with high nasal resistance, we expect that 200 subjects will have high nasal resistance and half (100) of them will receive CPAP_Flex at the initiation of treatment per randomization. Using the method of a 2-sample t test, we will have 80 % power (alpha = 0.05, 2-sided) to test a difference of 23.9 minutes (SD = 60 minutes) in improvement between subjects with high nasal resistance and CPAP_Flex versus CPAP at initiation. After accounting for 40 % CPAP rejection, we will be able to test a difference of 27.6 minutes (SD = 60 minutes) in the CPAP_Flex improvement.

To determine if high nasal resistance is associated with decreased CPAP adherence we will first explore the distributions of nasal resistance and adherence of CPAP (number of hours/night), both in continuous scale and the discrete scale (nasal resistance (high versus low at the cutoff of log[Reff] = 0.8); CPAP rejection (rejection “yes” versus “no” at a cutoff of 2 hours/night). We will calculate the correlation between nasal resistance and CPAP when both are treated as continuous measures, and use the chi-square test when both are discrete variables. To control for age, gender, BMI, and AHI we will use the linear regression (in continuous case) and logistic regression (in discrete case) analyses as we evaluate these associations. To determine if the use of CPAP_Flex will improve adherence with CPAP in subjects with high nasal resistance, but not in those with low nasal resistance, we will use mixed model analysis for repeated measures resulting from the cross-over design. Specifically, we will use the average hours of CPAP/CPAP_Flex use over the last 2 weeks in each phase as the dependent variable. Treatment assignment (CPAP versus CPAP_Flex), nasal resistance (high versus low according to the cutoff of log[Reff] = 0.8) and their interaction will be modeled as fixed effects in the statistical model and the intra-subject correlation between repeated measures will be modeled using random effects. If the use of CPAP_Flex improves adherence to CPAP in subjects with high nasal resistance, but not in those with low nasal resistance, we expect to observe a significant interaction of treatment assignment and nasal resistance. Linear contrasts will also be constructed to evaluate improvement with CPAP_Flex in subjects with high and low nasal resistance separately. Variables such as age, gender, BMI, AHI, etc., will be further controlled for in these statistical analyses. Finally, to determine if the benefit of CPAP_Flex on adherence will be greatest if it is offered at CPAP initiation rather than as a “rescue” therapy in subjects with high nasal resistance, we will use linear regression analysis with the difference in the average use between CPAP and CPAP_Flex as the dependent variable, treatment sequence (CPAP followed by CPAP_Flex versus CPAP_Flex followed by CPAP), nasal resistance (high versus low) and their interaction as the independent variables. If the benefit of CPAP_Flex is greatest when it is offered at initiation in subjects with high nasal resistance, we expect the interaction of treatment sequence and nasal resistance to be significant. Linear contrasts will also be constructed to evaluate improvement with CPAP_Flex for each combination of treatment sequence and nasal resistance separately. If a substantial proportion of subjects reject CPAP and/or CPAP_Flex and results in excessive zeros (or low numbers close to zero) in the data, we will use the two-part model [35, 36] to analyze the data. In general, the term two-part model refers to having both a logistic model part to model the probability of non-zero versus zero outcomes (part 1) and a linear model part for the values of the non-zero outcomes (part 2). In our study, we will use the logistic model part to model the probability of rejection (defined by total CPAP/CPAP_Flextime < 2 hours, “yes”/”no”), and the linear model part to model the averaged daily use in hours after excluding those who reject. Treatment assignment (CPAP versus CPAP_Flex), nasal resistance (high versus low) and their interaction will be the independent variables in both model parts. Each hypothesis will be tested using either the score or the Wald test [36]. Subgroup analyses will be performed based on (i) sleep apnea severity (mild sleep apnea – AHI4% 5 to ≤ 15/hour and moderate/severe sleep apnea – AHI4% > 15/hour), (ii) baseline sleepiness (ESS < 10 and ≥ 10), (iii) excluding subjects who are using sedative drugs.