MULTIVARIATE-ANALYSIS OF CHOLESTEROL DISTRIBUTION FOR MONITORING THE RISK OF CORONARY HEART-DISEASE

The screening of people for potential coronary heart disease and the m onitoring of subjects considered at risk have been performed for some time by measuring total serum cholesterol and its constituent lipoprot eins. However, these measurements vary substantially within subjects, making such assessments imprecise. It has been suggested that greater consistency can be achieved by analysing the joint distribution of the individual lipoproteins or of transformed variables derived from them . In this paper we present the results of a laboratory experiment to i nvestigate these ideas with a view to improving current methods of mon itoring patients at risk. Nested, random-effects, multivariate analysi s of variance and the log-ratio analysis of compositions are used to i nclude information on all three lipoproteins simultaneously, and ratio s of generalized variances are used to assess and compare the differen t response variables. The multivariate approach is seen to be far supe rior to the usual methods. Recommendations are made for routine monito ring and the practical implications are discussed.