0090 Sleep Architecture Associations with Brain Age: A Multi-Site Model Validation

Abstract Introduction Machine Learning (ML) can draw upon complex patient data to predict current and future health status, but the best performing ML models are often difficult interpret. One such ML-defined marker is brain age index (BAI), difference between ML-predicted chronological (CA) using electroencephalogram (EEG) during sleep. BAI associated with multiple disease states including increased risk for all-cause mortality. Still, validation of published has been limited, underlying associations polysomnography sleep architecture metrics estimation not well understood. Methods A deep neural network (DNN) model was trained (BA) patients raw EEG signals recorded clinical (PSG). The DNN on N=54,000 PSGs. then validated two independent datasets: a Kaiser Permanente (KP) Historical PSG Dataset (N=10,694), multi-site retrospective dataset (N=15,158). To test association BAI, ordinary least squares (OLS) regression performed these datasets basic features derived from staging arousal data. Results In datasets, we observed mean absolute error (MAE) 6.67 (6.69/6.49 F/M, KP dataset) 5.65 (5.56/5.44 dataset). OLS analysis, consistent, negative were detected both N3 REM percentage vs. BAI. Furthermore, consistently positively N1 percentage. Each effects significant at p< 0.001. Conclusion This study builds expands prior research by evaluating large assessing relationship N3/REM duration predicted age. Using model, significantly concomitant decreased N3/REM. Further needed determine if BA malleable potentially reduced intervention positive changes in lifestyle wellness-related behavior. Support (if any)