Research ethics approval.

This study is conducted in accordance with the current version of the Declaration of Helsinki, the International Council for Harmonisation Good Clinical Practice (ICH-GCP) guidelines, the Human Research Act (HRA), and all other locally applicable regulatory requirements. The trial protocol and related study materials were reviewed and approved by the Kantonale Ethikkommission Zürich (Zurich Cantonal Ethics Committee), with the reference number 2024-01948. No participant recruitment or data collection activities commenced prior to obtaining this ethics approval.

Protocol amendments.

Any substantial amendments to the protocol, including changes to study objectives, design, methodology, study site(s), sponsor, or key study personnel, will be submitted for review and approval by the Ethics Committee prior to implementation. Exceptions may only occur under emergency circumstances to ensure participant safety and well-being; such deviations shall be documented and reported to the Ethics Committee as soon as possible.

Substantial amendments are defined in accordance with ClinO (Ordinance on Clinical Trials in Human Research), Art. 29, and include changes affecting participant safety, rights, or obligations. A list of substantial changes is maintained on www.swissethics.ch. Non-substantial amendments will be compiled and submitted annually to the Ethics Committee together with the Annual Safety Report (ASR).

Confidentiality.

Participant confidentiality will be maintained throughout the trial in compliance with Swiss data protection regulations. Identifiable information is stored separately in a subject identification log, linking names to PIDs assigned in sequence of enrolment. This log also records reasons for exclusion, early withdrawal, or non-adherence.

Access to identifiable data is restricted to authorized personnel involved in the study. Any disclosure to third parties requires prior written permission from the data owner. Study data may be accessed by monitors or the Ethics Committee solely for verification purposes.

Harms.

This study is classified as category A, involving only minimal risk to participants. The procedures conducted across the seven study visits consist predominantly of standardized, non-invasive assessments, performed under continuous supervision by trained research personnel.

The use of a driving simulator may occasionally cause dizziness, nausea, or discomfort due to simulator sickness. These symptoms typically subside promptly upon test completion or leaving the simulator environment.

EEG monitoring and blood collection via a peripheral venous catheter are performed by qualified personnel. Minor adverse events, such as hematoma at the venipuncture site or skin irritation from EEG electrodes, may occur but are generally transient and resolve without complications. The total blood volume collected (9 mL) poses no significant health risk. The collection of saliva, urine, DBSs, and exhaled breath is entirely non-invasive and carries no known risks.

Participants assigned to the intervention arm will discontinue their nightly CPAP therapy at home for a period of ≥7 days, during which a temporary reactivation of OSA symptoms is expected. This CPAP-withdrawal model is clinically validated as a safe and effective method for studying the pathophysiology of OSA [24,27]. Only patients deemed low risk for adverse outcomes by physicians at the Sleep Disorders Center, based on clinical judgment and prior screening, will be included. Participants will resume their established long-term CPAP therapy immediately after the study.

All procedures, with the exception of at-home PSG recordings, will be conducted in controlled environments by trained staff, with medical oversight provided as needed. Participants are closely monitored for any adverse effects, and appropriate support will be made available. Individuals retain the right to withdraw from the study at any point without consequences for their ongoing medical care.

Participation in the study offers no direct therapeutic benefit to the individuals involved. However, the findings may contribute to the advancement of diagnostic and safety tools in traffic medicine and sleep disorder management, potentially benefiting future patients and public health initiatives.

Intervention group (OSA patients).

Participants in the intervention group will be recruited from a restricted pool of patients treated at the Sleep Disorders Center and Pulmonary Division of the University Hospital Zurich. This population is characterized by high adherence to CPAP therapy and regular clinical follow-up. Eligible individuals are preselected by the partner institution based on clinical criteria, including the absence of substantial comorbidities (e.g., major affective, somatic, or neurological disorders), cognitive impairments, or increased risk associated with CPAP withdrawal.

Although gender balance is desired, it cannot be guaranteed in the intervention group due to the significantly higher prevalence of OSA in males. Given the limited sample size and inter-individual variability of outcome measures, retrospective stratification by gender is not planned at this stage.

Healthy comparison group.

Healthy participants will be recruited from the general population. Recruitment will be facilitated through institutional communication channels of the University of Zurich, including digital platforms and email distribution. Additionally, advertisements will be placed in local schools and public areas throughout Zurich to inform about the open recruitment status.

Recruitment process.

First contact will be offered by a flyer containing contact information to the investigators. Interested individuals can reach out to the study team via phone, email, video call (optional), or in-person at the study site. Investigators will provide detailed information about the study objectives, design, procedures, duration, risks, benefits, and any potential discomforts. These explanations can be delivered individually or in groups, using the participant’s preferred communication method.

All participants will receive a comprehensive information sheet and informed consent form, ensuring that they have sufficient detail to make an informed decision regarding their participation. Enough time, as much as the participant desires, will be given for consideration, and all participants will be informed that participation is entirely voluntary, with the option to withdraw at any time without consequences for their ongoing medical care.

As part of the screening process, all prospective participants will complete a brief, five-minute simulated drive to assess for motion sickness (kinetosis or dizziness), which constitutes an exclusion criterion. This session also ensures that participants fully understand the simulation task requirements before enrollment.

Consent and assent.

Verbal and written informed consent will be obtained by the sponsor-investigator or his designee (certified by GCP) prior to any study-specific procedures, including the collection of biological samples. The informed consent process will be documented in the participant’s file. Any deviations from the consent process as outlined in the protocol will be recorded and justified.

The participant will receive a copy of the signed consent form; the original will be retained in the study files. Consent includes permission for data collection, storage, analysis, and verification by study monitors or the ethics committee. For any ancillary studies not explicitly covered in the primary study’s informed consent form, additional written informed consent will be obtained from each participant prior to their involvement.

Data collection.

All data is recorded semi-automatically (driving data, EEG, ET) and/or without interference from the investigator. To maximize both temporal accuracy and data reliability, the study employs a dual-recording strategy. First, all multimodal data ET, EEG, and driving simulation output (SILAB)—are recorded and synchronized in real time using the Lab Streaming Layer (LSL) framework [36]. This ensures precise temporal alignment across systems during active data collection. Concurrently, each subsystem independently records high-resolution raw data in its native format, providing a redundant backup to safeguard against potential streaming failures. This local storage approach guarantees data completeness.

To promote the quality and integrity of the collected data, the study team is trained in standardized procedures for assessment and documentation. The course of data collection is documented using standardized paper-based Case Report Forms (CRFs), completed by study investigators for each participant. These forms record all trial-relevant tests and assessments. Participants are identified by a unique participant ID (PID), which is assigned sequentially. A separate subject identification log links PIDs to participant identities and tracks screening outcomes, withdrawals, and dropouts.

All entries into the CRFs are verified against the source documents, which include medical records, laboratory results, psychological assessments, and other original documentation. In cases where data are directly extracted from instruments and not transcribed into the CRF—such as EEG or eye-tracking recordings—these are clearly identified as source data. The specific location of such source data is documented in the CRF in the designated data storage log.

Data management.

All aspects of data management are coordinated by the Division of Traffic Medicine, which also serves as the central coordinating body for the study. Data are handled in compliance with Swiss data protection regulations. CRFs and other study documents are stored in locked areas with access limited to authorized personnel. Biological samples are identified solely by PID, stored under secure conditions, and transported personally by authorized staff to the designated analysis laboratory with authorized restricted access. No genetic data are collected, and no biobank will be established.

Data quality assurance procedures include double data entry, automated error-check algorithms, manual data audits after each experimental session, and automated post-acquisition audits after completion of all study visits to identify missing or inconsistent data.

In case of participant’s discontinuation, data and samples collected up to the point of withdrawal may be retained and analyzed. If a participant leaves early, no additional final visit is planned, but support is provided for a safe departure if needed.

After study completion, all study documents and data are archived for 20 years at the Institute for Forensic Medicine, University of Zurich. This includes CRFs, consent forms, medical records, and laboratory results. Data analysis will be performed only after trial completion and independently of funding sources. Data will be coded, processed, and analyzed in accordance with the predefined statistical analysis plan.

Trial instruments and measures.

For the primary outcome, EEG recordings are required to determine MWT-latencies, based on precise identification of sleep onset (SO). According to AASM classification, SO is defined as the start of the first epoch scored as any stage other than stage W (wake). In most individuals, this will usually be the first epoch of stage N1, characterized by SEM (slow eye movements) and LAMF (low-amplitude, mixed-frequency) in EEG activity [29]. For the calculation of DS-MWT-derived latencies, the classical EEG-based SO is complemented by the capture of SA in a driving simulator [5].

As for secondary outcome measures:

1. In addition to determining primary sleep latency outcomes, electrophysiological data collected during both MWT and DS-MWT sessions will provide insights into the dynamics of the sleep-wake transition. Of particular interest are MSEs —brief intrusions of sleep-like EEG activity during otherwise wakeful states —which may precede or occur independently of formal SO. Automatically detected MSEs will be analyzed in this study to characterize their distribution, duration, and relationship to subjective and behavioral sleepiness [33].

All EEG recordings, accompanied with electrooculogram (EOG), electromyogram (EMG), and electrocardiogram (ECG) —including those conducted during in-laboratory test sessions and overnight at-home PSG —will be performed using the Nox A1 PSG System (Nox Medical, Reykjavík, Iceland). This device has been validated for clinical sleep research and supports high-fidelity multi-channel EEG acquisition in both ambulatory and controlled lab environments [37]. Recordings will be analyzed using the dedicated Noxturnal software (Noxturnal, Nox Medical, Reykjavík, Iceland) and custom analysis scripts implemented in MATLAB (Version R2023a, The MathWorks, Inc., Natick, MA, USA).

1. 2. Driving performance will be assessed using a custom-modified full-chassis BMW i3 simulator, engineered to deliver a high degree of ecological validity. The simulator is equipped with actuator systems replacing standard shock absorbers to simulate realistic motion cues, including acceleration, deceleration, road texture, and cornering forces. The vehicle is embedded within a 270-degree projection system offering 360-degree field-of-view coverage, thereby immersing participants in a virtual driving environment that closely mimics real-world conditions.

The simulated environment is powered by SILAB Driving Simulation Software (Scenario Package “Driver Fitness and Ability,” Version 7.0; Würzburg Institute for Traffic Sciences GmbH, Veitshöchheim, Germany), which delivers a suite of validated scenario modules for evaluating FTD [38]. SILAB records a broad set of vehicle dynamics and behavioral parameters, including SDLP, SDS, number of lane departures, etc.

Participants will perform a customized driving task, which consists of an approximately 40-minute, highly monotonous car-following task in a simulated nighttime setting, chosen to increase the likelihood of drowsiness-related events. The scenario involves continuous driving without stops, traffic lights, or turns on a two-lane motorway separated by a double solid centerline, with low oncoming traffic on the left. This low-stimulation environment requires sustained attention and minimal steering correction, closely mimicking real-world conditions where sleep-related driving incidents are most likely to occur.

1. 3. Subjective sleepiness will be evaluated using computer-based questionnaires (KSS, ESS), administered and scored according to validated standard procedures.
2. 4. Eye-tracking metrics will be continuously recorded during DS-MWT and MWT sessions using high-resolution eye-tracking systems Tobii Pro Glasses 3 (Tobii AB, Danderyd/Stockholm, Sweden). A range of ocular motor features, including blink rate, saccade latency, saccade peak velocity and smooth pursuit velocity gain, will be analyzed. These parameters will be examined to determine whether they reflect levels of sleepiness and whether they hold predictive value for SO or MSE during the driving simulation [39,40].
3. 5. Unstimulated oral fluid will be collected using neutral Salivette devices (Sarstedt, Sevelen, Germany), following standardized procedures for non-invasive saliva sampling [41]. Samples will be immediately frozen and stored at until the conclusion of the trial. After completion of data collection, samples will undergo targeted metabolomic analysis using ultra-performance liquid chromatography coupled to high-resolution mass spectrometry (UPLC-HRMS) [19].
4. 6. Dried blood spot (DBS) samples will be collected using MICROLET lancets (Bayer HealthCare LLC, Whippany, NJ, USA) applied onto DBS collection cards (CAMAG AG, Muttenz, Switzerland), which allow for precise, standardized blood volume collection. All DBS samples will be stored under controlled conditions and subsequently processed and analyzed analogously to oral fluid samples.
5. 7. Exhaled breath samples will be collected using MIR Spirobank Smart spirometer (MIR, Rome, Italy) and the BreathExplor device (Munkplast AB, Uppsala, Sweden). These devices are designed for safe, efficient collection of aerosol particles from the exhaled air, suitable for subsequent biochemical and metabolomic analyses.

Sample size calculation.

The total sample size is N_total = 54, consisting of the intervention group N_I = 36 and a healthy comparison group N_H = 18.

Sample size of N_I = 36 for the intervention group was estimated according Formula (1), taking into account the adapted randomized crossover RCT design [42]. The study involves a cohort of OSA patients randomized to both the CPAP starting condition and the measurement sequence, resulting in four groups (W-DM, W-MD, C-DM, C-MD) across two phases. A mixed-model ANOVA will be used for data analysis, as it appropriately accounts for the correlation of repeated measures within individuals, handles missing data more effectively, and incorporates both fixed and random effects.

(1)

Where,

• n = Required number of participants per group.
• = Z-value corresponding to the chosen significance level (e.g., 1.645 for two-sided) [43].
• = Z-value corresponding to the desired power (e.g., 1.28 for 90% power) [43].
• = Standard deviation of the measurements = 13.5 minutes (from Bonnet & Arand [16]).
• = Intra-class correlation (ICC) = 0.5, i.e., the correlation between repeated measures within the same participant.
• = Equivalence margin or effect size = 1 pooled SD = 13.5 minutes (from Bonnet & Arand [16]).

For the sample size calculations, a two-sided = 0.1 was used, assuming that a potential difference between M and D would have a clear direction and a maximal value (bounded test range) of 40 min for the test latency. When conducting an equivalence test, using a 0.1 two-sided alpha-level effectively matches the 0.05 significance level for each side of the distribution (one for each tail), aligning the rigor of the test with the traditional hypothesis test for no difference. This assumption holds, as the reference standard deviation of 13.5 min was obtained with an equivalently bounded test range.

A graphically extracted and pooled standard deviation from Bonnet & Arand [16] (Fig 2, baseline and wake conditions, p. 388) was used, representing the only available data approximating repeated MWT test–retest measurements. One standard deviation was additionally defined as the equivalence margin. Sample sizes were calculated for both 80% and 90% power assuming a mixed-model ANOVA analysis, which allows pooling of variances in within-subject designs and thereby reduces required sample sizes. An interim analysis will be performed at the sample size corresponding to 80% power to evaluate study feasibility and reassess assumptions regarding the intra-class correlation (). A conservative within-subject correlation of was prespecified; however, a higher observed intra-class correlation would indicate lower within-subject variability across conditions and may reduce the required final sample size.

Similarly, a sample size of N_H = 18 for the comparison group (healthy) was estimated for the comparison group of healthy participants, based on a crossover design without retest, as applied in the test-retest framework. This corresponds to 9 participants per group to achieve 90% power, assuming a two-way ANOVA (see Formula (2), adapted from Chow & Liu [44]), resulting in a total of 18 participants across both groups. Although a smaller standard deviation ( = 5.9 minutes) was used — derived from normative data on healthy individuals reported by Banks et al. [12] — the required sample size remains comparable to that of the OSA cohort. This is primarily due to the absence of additional variability-reducing conditions that were present in the OSA sample.

(2)

Where,

• n = Required number of participants per group.
• = Z-value corresponding to the chosen significance level (e.g., 1.645 for two-sided) [43].
• = Z-value corresponding to the desired power (e.g., 1.28 for 90% power) [43].
• = Standard deviation of the measurements = 13.5 minutes (from Bonnet & Arand [16]).
• = Intra-class correlation (ICC), i.e., the correlation between repeated measures within the same participant.
• = Equivalence margin or effect size = 1 pooled SD = 13.5 minutes (from Bonnet & Arand [16]).

Statistical analysis plan.

Based on the design of this prospective, randomized, crossover trial, statistical analysis will be performed to detect both intra- and inter-individual (i.e., individual and group-level) differences. All analyses will be conducted in relation to two key contextual variables:

• Sleep quality of the night preceding each test session, as assessed by PSG, and
• Time of day at which each test run was performed (9 a.m., 11 a.m., 1 p.m., 3 p.m.).

The primary hypothesis will be tested solely based on the determined mean sleep latencies obtained from the MWT and DS-MWT, without incorporating additional secondary or exploratory parameters. For the primary outcome analysis, linear mixed-effects models (LMMs) will be applied, with test latency as the dependent variable and test condition (MWT vs. DS-MWT) as the main fixed effect. Interindividual variability will be accounted for by including participant as a random effect. Additional covariates, including previous-night PSG-derived sleep quality parameter, test sequence, and time of day, will also be included in the models. To account for potential non-linear variation across the day, time of day will be treated as a categorical factor. Possible interactions between condition and time of day will be explored, based on the assumption that time-of-day–related fluctuations in behavioral performance and sleepiness may differentially affect test latency depending on the testing modality. Results will be considered significant if p < 0.05.

All secondary variables will be analyzed post hoc, with the goal of generating new hypotheses rather than performing formal confirmatory tests. These analyses will include individual evaluations of each collected secondary parameter in reference to:

• The experimental condition (MWT vs. DS-MWT),
• The treatment state (CPAP continuation vs. withdrawal) or healthy comparison,
• The test sequence (DM vs. MD).

Where appropriate, group-level comparisons will be followed by multiple comparison tests to examine differences in greater detail. For example, time-course effects in driving behavior parameters—such as SDLP —will be examined using repeated measures ANOVA across ten 4-minute segments within each test session, or through paired t-tests comparing the first and final modules of each session.

For these secondary exploratory analyses, linear mixed models will be used as the initial framework, with stepwise model testing based on biologically plausible relationships rather than relying solely on statistical significance. Where applicable, Akaike Information Criterion (AIC) will be used to identify the most parsimonious model, and AICc will be applied in cases involving small sample sizes (N = 18). Cross-validation techniques, such as k-fold cross-validation, will be implemented to evaluate predictive performance and reduce overfitting.

Beyond aggregated latency measures, predictive models for SO and/or SA will be explored at finer temporal resolutions, including:

• Per 40-minute run (N = 4 runs × 54 participants = 216 observations),
• Per 4-minute driving module (N = 10 modules × 216 runs = 2,160 observations),
• Or via dynamic time-series analysis using shorter time units still to be defined.

These exploratory analyses will aim to maintain an events-per-variable (EPV) ratio between 10 and 15, depending on the number of predictors included. These models will evaluate whether secondary variables collected during each run can predict the occurrence of SO or MSE in real time.

While p-values will not be the primary decision metric for exploratory models, False Discovery Rate (FDR) corrections may be applied to account for multiple testing while preserving statistical power. Bonferroni corrections, due to their conservativeness, will be avoided to reduce the risk of Type II errors (false negatives). Significant findings from exploratory analyses will be interpreted cautiously and treated as hypothesis-generating, requiring independent replication in future studies.

Although our study is not powered to detect gender differences, sex/gender will be included as a covariate in all relevant models for control purposes.

All statistical analyses will be performed using R, Python, MatLab and GraphPad Prism. Any deviations from this predefined statistical analysis plan will be documented and justified in the final trial report.

The study is intended to be completed with N = 54 participants, unless recruitment proves infeasible within the projected time frame. The last participant’s last visit (LPLV) is planned for 30 April 2027, though an extension of the recruitment period may be considered if necessary.

Missing datasets due to participants’ withdrawal from the trial will not be multiply imputed due to the low sample size and inherently low statistical power. All available data will be used, if possible, for statistical analysis. Drop-outs during the study phase will be replaced by recruitment of new participants from the accessible participant pool, if possible.

Monitoring.

Internal monitoring duties will be carried out by Dr. med. Kristina Keller, who is not actively taking part in recruitment, data acquisition or data analysis. There is no additional external monitoring foreseen or required. Internal monitoring will be conducted to ensure adherence to the study protocol, regulatory requirements, and data quality standards. All source data and relevant documents will be accessible to the internal monitor. Monitoring visits may take place without prior notice and will include a review of available documentation and clarification of any issues directly with the site team.

Given the exploratory nature of this study, the establishment of a steering committee, an endpoint adjudication committee, or an independent data monitoring committee is not required and has therefore not been implemented.