Predicting Pathophysiological Events: Sleep Disordered Breathing, Vigilance Lapses, Driving Errors

Disrupted or insufficient sleep is a major cause of motor vehicle accidents, workplace injuries, and industrial disasters. Responding to this epidemic is a public health imperative. Two major causes of disrupted sleep and consequent impairment of daytime performance are obstructive sleep apnea (OSA) and sleep restriction or deprivation. OSA is a highly prevalent disorder in which patients experience repetitive total or partial collapses of the airway during sleep. These episodes of sleep disordered breathing (SDB) are commonly associated with arousal from sleep. Consequently, OSA patients experience disrupted sleep and are exposed to chronic intermittent hypoxemia, leading to impaired daytime performance and excessive daytime sleepiness. People with OSA are also at risk for heart disease, stroke, metabolic disorders, and cognitive dysfunction. Many of the adverse outcomes associated with OSA are also associated with sleep deprivation, even in the absence of a concomitant sleep disorder. This dissertation aimed to take a first step toward responding to the public health crises of OSA and drowsy driving by establishing the feasibility of individualized approaches to predict the onset of pathophysiological events in people with OSA and in acutely sleep deprived but otherwise healthy individuals. Our methods were also designed to expose precisely the signals and signal features that enable accurate and reliable predictions. The analysis of important predictors of pathophysiological events provided a novel approach to investigating the mechanisms that initiated the events. In particular, methods were developed to 1) predict episodes of SDB in the next 10 to 60 seconds in OSA patients, 2) predict lapses in vigilance, or sustained attention, within the next 30 seconds in OSA patients and in healthy individuals acutely deprived of sleep for 24 hours, and 3) predict mistakes during simulated driving within the next 30 seconds in OSA patients and acutely sleep deprived healthy subjects. Our methods were based on the analysis of multimodal physiological data using LArge Memory STorage and Retrieval (LAMSTAR) artificial neural networks, wavelet analysis, temporal networks constructed using measurements of phase coherence and empirical mode decomposition, and frequent subgraph discovery and prediction. To facilitate our work on predicting driving errors, we developed a driving simulator, called the Biobehavioral Effects of Disturbed Sleep (BEDS) Driving Simulator, and established its face validity and feasibility for use in sleep research.