Predicting All-Cause Mortality in Patients With Obstructive Sleep Apnea Using Sleep-Related Features: A...

Area of Science:

• Sleep Medicine
• Machine Learning in Healthcare
• Cardiology

Background:

• Obstructive sleep apnea (OSA) is linked to increased mortality risk.
• Machine learning (ML) shows promise for predicting clinical outcomes in OSA.
• Accurate mortality prediction is crucial for managing OSA patients.

Purpose of the Study:

• To develop and evaluate an ML algorithm for predicting 10- and 15-year all-cause mortality in OSA patients.
• To identify key sleep-related features for mortality risk stratification.
• To assess the model's performance using established statistical methods.

Main Methods:

• Stratified OSA patients into deceased and alive groups.
• Analyzed objective sleep measures and heart rate variability during sleep stages.
• Utilized the Light Gradient-Boosting Machine (LGBM) algorithm for risk prediction.
• Assessed model performance with Area Under the Curve (AUC) and survival analysis (Kaplan-Meier, Cox regression).

Main Results:

• Higher parasympathetic activity observed in OSA patients with worse outcomes.
• The LGBM model achieved a mean AUC of 0.806 for 10- and 15-year mortality prediction.
• Survival analysis confirmed LGBM's ability to significantly distinguish high-risk from low-risk groups.

Conclusions:

• Sleep-related feature analysis combined with the LGBM algorithm effectively evaluates mortality risk in OSA.
• The developed risk-stratification model provides an interpretable tool for clinicians.
• Patient-specific autonomic responses are significant predictors of long-term mortality in OSA.