Cluster analysis: A new technology for the evaluation of oximetry and airflow waveforms in obstructive sleep apnea

Abstract

A new diagnostic method is described which identifies the presence of waveform clusters indicative of OSA, and an illustrative OSA screening system is presented which applies the new cluster analysis method to the oximetry waveform. Rather than using the conventional identification of an excess average number of isolated events to infer the presence of upper airway instability, cluster analysis identifies a prolonged spatial pattern of clustered waveforms derived from the underlying mechanism of OSA. The illustrative screening system includes a microprocessor which divides the basic timed oxygen saturation waveform into slope defined dipoles and then further classifies sequential components of the waveform as slope and interval derived "objects" for microprocessor analysis. The microprocessor then performs an analysis of the objects to further derive more complex objects called coupled oscillations and slope clusters. Each new sequential object is added if it has suitable spatial relationships, and occurs in acceptable order and time interval. The building of complex objects from the repetitive derivation of basic objects along a timed waveform exploits the unique self perpetuating physiological events which underlie the propagation of OSA. The identification and comparison of these objects are then used to determine the presence of sleep apnea. The method can also be applied to a secondary plot of a flow signal parameter such as the maximum exhaled CO(2) or the mean attenuation of the inspiratory loop as a function of time. The utilization of cluster based recognition of events to improve the conventional AHI and DI averaging methods of diagnosis is also discussed. In addition, the differentiation of three basic oxygen saturation cluster types and their potential relevance to the induction of sustained ischemia and cardiovascular morbidity are discussed.