Unraveling sleep apnea dynamics: quantifying loop gain using dynamical modeling of ventilatory control

Loop gain (LG) is a critical parameter for assessing ventilatory control stability in sleep apnea, with implications for personalized treatment. Existing LG estimation methods are hindered by complex processing and specialized equipment, limiting clinical applicability. This study aims to develop an automated method to quantify LG from respiratory inductance plethysmography (RIP) signals to enhance precision management of sleep apnea.

Polysomnography data from Massachusetts General Hospital, high-altitude studies at Beth Israel Deaconess Medical Centre, and patients with heart failure were analyzed. Cases included an apnea-hypopnea index greater than 15 and greater than 4 h of recorded sleep. RIP signals were filtered, normalized, and segmented into 8-min windows. LG estimation employed an augmented Mackey-Glass equation and an expectation-maximization algorithm. Simulation experiments on synthetic breathing data with known parameter values quantified the accuracy of our parameter estimates.

Data from 465 patients were analyzed, including 400 patients from the Massachusetts General Hospital dataset and 65 patients with heart failure. The method accurately estimated LG across diverse apnea phenotypes. Patients with a higher central apnea index, high self-similarity, or heart failure exhibited significantly higher median LG values (0.19, 0.27, and 0.41 respectively) compared to those with obstructive apnea (median LG = 0.11-0.14; p<.001). In addition, LG was significantly elevated during non-rapid eye movement sleep and at higher altitudes.

The automated LG estimation method developed in this study provides a scalable, non-invasive tool for endotyping in sleep apnea. By accurately modeling patient-specific ventilatory control, this approach supports personalized management strategies in apnea and broader clinical contexts. Statement of Significance This study presents an innovative method for estimating ventilatory control stability using respiratory inductance plethysmography signals, offering a practical, scalable solution for routine clinical use. By enabling detailed characterization of ventilatory control dynamics, the method can differentiate sleep apnea phenotypes and identify patients at elevated risk of ventilatory instability. This has direct clinical implications, such as guiding personalized treatment strategies, predicting continuous positive airway pressure tolerance, and flagging patients for possible adjunctive therapies like oxygen supplementation or carbonic anhydrase inhibitors. Furthermore, the fully automated nature of our approach enables repeated assessments over time, facilitating longitudinal monitoring of treatment efficacy and disease progression. By advancing diagnostic precision and treatment tailoring, this innovation has the potential to improve the management of sleep-disordered breathing and related conditions.