Low Arousal Threshold Endotype and CPAP Adherence: Insights from a Large, Diverse Clinical Cohort

A low respiratory arousal threshold (ArTH) has been linked to reduced continuous positive airway pressure (CPAP) adherence in obstructive sleep apnea (OSA) via a multi-trait model developed in the RICCADSA trial. Our objective was to validate the prior model in a large, real-world cohort and explore alternative linear and non-linear approaches for predicting CPAP adherence.

Does a previously derived multi-trait model linking low ArTH to poor CPAP adherence remain valid in a diverse real-world population and do alternative linear or non-linear approaches offer improved predictive performance?

Adults with OSA from the SNOOzzzE-cohort who initiated CPAP within 1 year of in-lab polysomnography (2017-2019) were included. Pathophysiological traits (Vpassive, Vactive, loop gain, ArTH, ventilatory response to arousal) were estimated from polysomnography. Poor (vs good) adherence was defined as ≤2.48h/night at month 1 (1 ^st sample quartile). The prior model's performance was evaluated via the area under the curve (AUC) and adherence comparisons. Secondary analyses tested a de novo linear regression and a latent profile analysis (LPA)-derived trait-cluster model.

Among 744 participants (45% women, 47% non-White), median CPAP adherence was 4.8 h/night at 1 month. The prior model's AUC was 0.51 (95%-CI 0.46-0.55), with no usage differences between predicted poor vs good adherers. A new linear model overfit in training (AUC=0.85) but failed in testing (AUC=0.55). LPA identified a "Low ArTH Driven" cluster with persistently lower adherence at months 2-3 (P<.05) and a "Low ArTH & High loop gain" cluster whose usage stabilized after month 1.

The prior model did not generalize to this diverse clinical cohort. LPA identified a "Low ArTH Driven" endotype with persistently low CPAP adherence, suggesting potential for targeted interventions pending external validation.