Multivariate Model-based Tool for Predicting Cardiovascular Outcomes Using Multimodal Physiological Sleep Data.

Area of Science:

• Sleep Medicine
• Cardiology
• Biomarker Discovery

Background:

• Obstructive sleep apnea (OSA) is linked to cardiovascular disease (CVD), but the Apnea-Hypopnea Index (AHI) may not fully reflect long-term CVD risk.
• Current risk assessment for OSA patients may benefit from incorporating more detailed physiological sleep data.

Purpose of the Study:

• To identify novel sleep-related physiological predictors of cardiovascular disease (CVD) outcomes in obstructive sleep apnea (OSA) patients.
• To develop a more accurate CVD risk stratification model beyond the standard Apnea-Hypopnea Index (AHI).

Main Methods:

• Utilized a multimodal database including polysomnography (PSG) metrics, clinical data, and risk factors with an average 10-year follow-up.
• Employed a two-step feature selection process: initial irrelevant feature removal followed by LASSO for informative feature selection.
• A wrapper-based method selected the final model, validated using repeated holdout, achieving 75.3% accuracy and 78.0% AUC with 10 features.

Main Results:

• Identified key predictors beyond traditional risk factors (sex, age, BMI, diabetes).
• Significant novel biomarkers include nocturnal average SpO2 (total and NREM), heart rate during apneic episodes with SpO2 drops, blood pressure, and central apnea frequency during REM/NREM sleep.
• The model demonstrated robust performance in predicting CVD outcomes.

Conclusions:

• Integrating multimodal sleep physiological metrics enhances cardiovascular risk stratification in OSA patients compared to AHI alone.
• These findings enable more precise clinical interventions for OSA patients at risk of CVD.
• Novel biomarkers identified offer potential for improved diagnostic and prognostic tools.