Yoav N. Nygate, MSc1 • Tom J. Vanasse, PhD1 • Sam Rusk, BSc1 • Chris R. Fernandez, MSc1 • Nathaniel F. Watson, MD, MSc2
Introduction
Machine learning (ML) techniques offer powerful tools for analyzing complex physiologic data. One such tool is the ML-based Brain Age Index (BAI) which quantifies the difference between a person’s ML-determined brain age (BA) using EEG signals collected during sleep versus their chronological age (CA). Prior research has associated BAI with various disease states. However, the effects of therapeutic interventions on BAI have yet to be explored.
Methods
A deep neural network model was trained on a dataset of 54,000 polysomnography (PSG) studies to predict patients’ BA. Additionally, a real-world dataset of 4,738 split-night PSG studies was collected, where patients underwent the first half of the study without positive airway pressure (PAP) therapy and transitioned to PAP therapy midway through the night. To evaluate the effect of PAP therapy on the BAI, ordinary least squares (OLS) regression was performed while controlling for sleep-related metrics, including the percentage of time spent in wake, N1, N2, N3, and REM.
Results
The OLS analysis found a significant association with a lower BAI after transitioning to PAP therapy (P-value = 0.001, OLS coefficient = -0.648). The mean BAI was -1.62 (-1.842, -1.398) and -1.584 (-1.806, -1.362) for the pre-PAP and PAP portions respectively. The dataset showed the following mean percentages of time spent in Wake, N1, N2, N3, and REM for the pre-PAP / PAP portions of the night: 38.1% (37.6%, 38.5%) / 25.7% (25.2%, 26.1%), 7.8% (7.6%, 8.1%) / 5.6% (5.5%, 5.8%), 73.2% (72.8%, 73.7%) / 60.7% (60.2%, 61.1%), 11.3% (10.9%, 11.6%) / 10.1% (9.9%, 10.4%), and 7.7% (7.4%, 7.9%) / 23.6% (23.2%, 24.0%).
Conclusion
The OLS analysis, controlling for sleep-related metrics, demonstrated improvements in BAI with initiation of PAP therapy. This suggests BAI is “malleable” and responsive to therapies that improve sleep and thus highlights its value as a potential biomarker of brain health. Overall, the study encourages further research into biomarkers for studying the neurocognitive benefits associated with Obstructive Sleep Apnea treatment.
1 EnsoData Research, Ensodata, Madison, WI, USA | 2 Department of Neurology, University of Washington School of Medicine, Seattle, WA, USA
