Sleep measurement beyond AHI

The apnea hypopnea index (AHI) has long been the metric used to diagnose obstructive sleep apnea, judge its severity, predict its consequences, evaluate the probability of success for various non-CPAP therapies, and characterize the pathophysiology of the syndrome. Despite some degree of correlation to the above, the AHI shows only poor-modest ability to predict consequences in individual subjects. Approaches to improving on the AHI metric have varied the definition and cut points; added separate non-ventilatory metrics, combined with the AHI in statistical and other models; and sought to replace the AHI entirely by non-ventilatory measures. Proposed in this talk is the use of the all-night distribution of the amplitude of each breath, determined automatically from a flow recording. A single parameter, the ventilatory burden, can be derived from an amplitude histogram, or the entire distribution can be used as inputs to a machine learning approach. Finally, the combination of the ventilatory distribution with distributions of hypoxic and arousal metrics can be used in simple statistical combination, or can be used in a machine learning AI approach. Preliminary data suggests these approaches have good test-retest reliability, face validity, and show promise of better correlation to sleepiness and CV endpoints than the AHI and derived metics.