Health Econometrics: Respiration-Oxygenation Correlation through Spectral Models

J. Macbeth and M. Sarrafzadeh (USA)


Biomedical Modelling; Cardiovascular Modelling; Time Series Analysis; Respiratory Mechanics.


Medical embedded systems are capable of recording vast data sets for physiological and medical research. Linear modeling techniques are proposed as a means to explore relationships between two or more medical or physiologi cal signal measurements where a causal relationship is be lieved to be present. Multiple regression is explored for use in medical monitoring, telehealth, and clinical applications. Spectral regression methods for high-bandwidth med ical and physiological signals are demonstrated. The two stage method consists of performing an FFT over a time lagged window of the predictor signal, and constructing a model based on the FFT coefficients. The output of the re gression is used in a clustering to explore structure in the array of spectral predictors. It has been applied to medical and physiological time series data, specifically the link be tween respiration and blood oxygen saturation percentage in sleep apnea patients. Spectral predictors achieved a dramatically better goodness of fit than time-lagged predictors according to standard analysis of variance measures. In the dataset ex amined, the spectral model achieved a multiple R2 of 0.90, indicating that 90% of the variation in the dependent signal was captured by the model, while an ordinary distributed lag model had a R2 of only 0.016.

Important Links:

Go Back