EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Machine learning prediction of the change in sleepiness following continuous positive airway pressure in obstructive sleep apnea: a multi-cohort analysis

Machine-learning-Prognose der Änderung der Schläfrigkeit unter kontinuierlichem positivem Atemwegsdruck bei obstruktiver Schlafapnoe: Multikohortenanalyse

Eric Staykov iD, Juha Töyräs iD, Dwayne L Mann iD, Timo Leppänen iD, Samu Kainulainen iD, Ali Azarbarzin iD, Scott A Sands, Philip I Terrill iD

SLEEP · 10.11.2026 · 49 (11)

DOI: 10.1093/sleep/zsag222 ↗

Auf einen Blick

Anhand von 5 Kohorten wurden Modelle zur Vorhersage der Schläfrigkeitsverbesserung unter CPAP geprüft. Baseline-Epworth-Score und ventilatorische Belastung sagten die Verbesserung am besten voraus.

Fragestellung

Lassen sich Baseline-Merkmale nutzen, um die Schläfrigkeitsverbesserung unter CPAP bei obstruktiver Schlafapnoe vorherzusagen?

Methode

Sekundäranalyse retrospektiver Daten aus 5 Polysomnographie-Kohorten; multivariable lineare Regression und LASSO; Modelltraining an 139 und Test in 74 CPAP-behandelten Teilnehmenden.

Ergebnisse

In der Cross-sectional-Analyse waren Baseline-Epworth-Score und ventilatorische Belastung, hypoxische Belastung, AHI, Zeit <90 % Sättigung und Flusslimitierung mit dem Epworth-Score assoziiert. Im Holdout-Test korrelierten vorhergesagte und beobachtete Verbesserungen (adj. R2 = 0,313).

Schlussfolgerung der Autor:innen

Modelle mit Baseline-Epworth-Score und ventilatorischer Belastung oder AHI sagen die Epworth-Verbesserung unter CPAP voraus und könnten helfen, geeignete Patient:innen zu identifizieren.

Kernaussagen

  • Baseline-Epworth-Score und ventilatorische Belastung waren die häufigsten Prädiktoren.
  • Vorhersage und tatsächliche Verbesserung korrelierten im Holdout-Test (adj. R2 = 0,313).
  • Das AHI-basierte Modell war etwas weniger genau, aber klinisch praktikabler.

Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.

Kurz gefasst

Fachartikel in der Zeitschrift SLEEP, online erschienen am 20.08.2026. Verfasst von Eric Staykov und 7 weiteren Autor:innen. Beteiligte Einrichtung: School of Electrical Engineering and Computer Science, The University of Queensland, Brisbane, Australia. Hauptthemen laut MEDLINE: Continuous Positive Airway Pressure, Machine Learning, Sleep Apnea, Obstructive, Sleepiness. Die Arbeit nennt 71 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: NHLBI NIH HHS, NCATS NIH HHS.

Auf einen BlickAbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study Objectives To predict the improvement in sleepiness following continuous positive airway pressure treatment for obstructive sleep apnea using baseline characteristics. Materials and Methods Data from five polysomnography cohorts (HomePAP, BestAIR, ABC, STAGES, and MESA; total n = 2332) were analyzed. Associations between polysomnographic metrics and Epworth Sleepiness Scale were assessed using multivariable linear regression. Cross-validated Least Absolute Shrinkage and Selection Operator regression was used to select predictor variables. Prediction of Epworth Sleepiness Scale improvement was evaluated in continuous positive airway pressure–treated participants (n = 213) using linear regression models trained on 139 participants and tested in a holdout group (n = 74).

Results

Cross-sectional analysis revealed significant associations between baseline Epworth Sleepiness Scale and ventilatory burden (0.56 points/standard deviation; 95% confidence interval 0.34 to 0.78), hypoxic burden (0.51; 0.28 to 0.74), the apnea–hypopnea index (0.45; 0.23 to 0.68), time below 90% oxygen saturation (0.36; 0.15 to 0.57), and flow limitation severity (0.31; 0.10 to 0.53). Baseline Epworth Sleepiness Scale and baseline ventilatory burden were the most frequently selected predictors of Epworth Sleepiness Scale improvement. Using these two variables in a regression model, predicted and actual Epworth Sleepiness Scale improvement were correlated in the holdout group (adjusted R2 = 0.313). The model classified continuous positive airway pressure responders (≥2 points Epworth Sleepiness Scale improvement) more accurately than current clinical guidelines (78.1% [3.5% standard deviation] versus 68.4% [3.7%]). Substituting the apnea–hypopnea index for ventilatory burden slightly reduced accuracy (77.3% [3.4%]). Charts were developed to assist in predicting Epworth Sleepiness Scale improvement from baseline metrics.

Conclusions

Models incorporating baseline Epworth Sleepiness Scale and either ventilatory burden or apnea–hypopnea index predict Epworth Sleepiness Scale improvement and may help identify patients with obstructive sleep apnea likely to benefit from continuous positive airway pressure. The ventilatory burden model demonstrated modestly better performance, while the apnea–hypopnea index–based model offers greater clinical applicability. Clinical Trials Information This was a secondary analysis of retrospective data. Some of the data utilized originated from clinical trials and is registered with www.clinicaltrials.gov:

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Continuous Positive Airway PressureMachine LearningSleep Apnea, ObstructiveSleepinessAgedFemaleHumansMaleMiddle AgedCohort StudiesCross-Sectional StudiesPolysomnographyPredictive Learning ModelsSecondary Data Analysis

Stichwörter der Autor:innen

OSA–PAP therapyexcessive daytime sleepinessflow limitationhypoxemiamachine learningobstructive sleep apnea endotypessleep disordered breathingventilatory burden

Autor:innen und Institutionen

  • School of Electrical Engineering and Computer Science, The University of Queensland, Brisbane, Australia.
  • Department of Technical Physics, University of Eastern Finland, Kuopio, Finland.
  • Science Service Center, Kuopio University Hospital, Kuopio, Finland.
  • Clinical Research Center, Kuopio University Hospital, Kuopio, Finland.
  • Division of Sleep and Circadian Disorders, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA, United States.

ORCID-Profile: Eric Staykov, Juha Töyräs, Dwayne L Mann, Timo Leppänen, Samu Kainulainen, Ali Azarbarzin, Philip I Terrill

Förderung

  • NHLBI NIH HHS
  • NCATS NIH HHS

Zitieren

APA

Staykov, E., Töyräs, J., Mann, D. L., Leppänen, T., Kainulainen, S., Azarbarzin, A., Sands, S. A., & Terrill, P. I. (2026). Machine learning prediction of the change in sleepiness following continuous positive airway pressure in obstructive sleep apnea: a multi-cohort analysis. SLEEP, 49(11). https://doi.org/10.1093/sleep/zsag222

Vancouver

Staykov E, Töyräs J, Mann DL, Leppänen T, Kainulainen S, Azarbarzin A, et al. Machine learning prediction of the change in sleepiness following continuous positive airway pressure in obstructive sleep apnea: a multi-cohort analysis. SLEEP. 2026;49(11). doi:10.1093/sleep/zsag222

BibTeX
@article{Staykov2026,
  author = {Staykov, Eric and Töyräs, Juha and Mann, Dwayne L and Leppänen, Timo and Kainulainen, Samu and Azarbarzin, Ali and Sands, Scott A and Terrill, Philip I},
  title = {Machine learning prediction of the change in sleepiness following continuous positive airway pressure in obstructive sleep apnea: a multi-cohort analysis},
  journal = {SLEEP},
  year = {2026},
  volume = {49},
  number = {11},
  doi = {10.1093/sleep/zsag222}
}
Kostenlos starten →