A partial least squares model of cephalometric data in obstructive sleep apnea patients. Pae, Eung-Kwon, Dds, Msc, Phd, Alan A. Lowe, Dmd, Phd, Frcd(c), John A. Fleetham, Md, Frcp(c) and Fred L. Bookstein, Phd. The University of British Columbia, 2199 Wesbrook Mall, Vancouver, B.C., Canada, V6T 1Z3
APStracts 2:0033L, 1995.
Since Obstructive Sleep Apnea (OSA) has a complex pathophysiology, studies of this disease demand an analysis of many variables which are usually interrelated. The analysis requires a large sample size to ensure an adequate statistical analysis and test of significance. Furthermore, interrelated variables can often be a source of multicollinearity. Landmark data for the face and tongue and interlandmark data for the pharynx obtained from upright and supine cephalometric measurements of 80 subjects with varying degrees of OSA were evaluated by means of a Partial Least Squares (PLS) analysis. PLS analysis is a statistical technique that is intrinsically insensitive to sample size and multicollinearity. The PLS model of cephalometric data from OSA subjects displays which part of the anatomical structure is most associated with OSA severity. The results show that pharyngeal length was proportionally associated with OSA indices. The pharynx variables obtained in the upright cephalograms may be useful in the prediction of OSA severety. Overall, the statistical tool adopted in this study appears to be practical for the analysis of complex diseases such as OSA.

Received 18 July 1994; accepted in final form 21 February 1995.
APS Manuscript Number L202-4.
Article publication pending Am. J. Physiol. (Lung Cell. Mol.
Physiology).
ISSN 1080-4757 Copyright 1995 The American Physiological Society.
Published in APStracts on 10 March 1995.