Abstract

Rationale: Increase in respiratory effort (RE) is a key feature of obstructive sleep apnea (OSA) and is participating in sympathetic overactivity and remodeling of vascular walls. We aimed to evaluate the impact of percentage of sleep time in RE (RE_dt) as a predictor of prevalent hypertension (HT) in adults with OSA.

Methods: A machine learning model was built to predict HT from clinical profiles including age, sex, BMI, conventional PSG indices and RE_dt derived from mandibular movement signals (Sunrise, Namur, Belgium).

Results: In 1126 patients referred for suspected OSA (M/F ratio = 1.2; BMI = 31.4 ± 7.6; apnea hypopnea index (AHI) = 24.6 ± 20.9), HT prevalence was 30.8%. Percentage of TST spent in RE was significantly associated with a higher likelihood of prevalent HT (OR = 17.6; 95%CI: 9.7-31.9).

The classification rule allows for predicting HT with high accuracy (85.9%). Shapley additive explanation process was conducted to assign a score to each feature values depending on their contribution to the prediction. This analysis (Figure) revealed that RE_dt was the best predictor among sleep test driven metrics, its contribution was even better than that of ODI, AHI or RDI.

Conclusion: The percentage of sleep time spent with an increase in RE derived from mandibular movement signals appears as a new relevant metric to assess OSA impact in cardiovascular risk.