Risk factors for peripheral atherosclerosis. Retrospective evaluation by stepwise discriminant analysis

Abstract

To evaluate the optimal discriminators for peripheral atherosclerosis, we studied retrospectively 49 male patients and 39 male controls between 40 and 60 years of age. In addition to hypertension, cigarette smoking, diabetes mellitus, and hyperuricemia, we determined the most common lipids, lipoproteins, and apolipoproteins. Highly significant differences of median values between patients and controls in decreasing order of magnitude were recorded for apo A-II/apo B, apo A-I/apo B, apo B, total cholesterol, and LDL-cholesterol. A retrospective classification of patients and controls under optimal conditions with one variable (apo A-I/apo B) yielded an error rate of 25%. We found that apolipoproteins were better discriminators for peripheral atherosclerosis than than were lipids or lipoprotein lipids. The application of a linear regression discriminant analysis including 29 variables greatly decreased the rate of error and increased the sensitivity and specificity of the classification. From 229 possible models, we used an economic selection strategy to sort out those which either gave the best segregation or were considered the most practicable. The optimal model with 14 variables gave an error rate of less than 5% for the group studied. Suboptimal models yielded error rates between 13% and 18%. We conclude that a mathematical treatment of laboratory data which includes lipid parameters in addition to apolipoprotein values can improve the classification of peripheral vascular atherosclerosis.