The principal components analysis - method to reduce the collinearity in multiple linear regression model; application in medical studies


Dascalu C. G. , Cozma C. D.

2nd WSEAS International Conference on Multivariate Analysis and Its Application in Science and Engineering, İstanbul, Turkey, 30 May - 01 June 2009, pp.140-141 identifier

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.140-141

Abstract

During the statistical analysis of medical data, in many situations it is necessary to identify the multiple correlations established between the studied parameters. In this purpose, one of the most useful methods is to build a model of multiple regression, which allows the modeling of a dependant variable values having at least the ordinal type, based on its linear relation with more than one independent variables satisfying the same restriction, called predictors. The main problem which affects the efficiency of such a model is the collinearity between predictors - a clue that they are not selected well because they overlap each other. There are different methods to eliminate this difficulty, but a very efficient one is the principal components analysis, applied as a method to reduce the predictors number, in a step preliminary to the regression model building. We will present the results obtained using these procedures oil a set of 212 patients from three categories: healthy patients (65 cases), patients with liver cirrhosis (65 cases) and patients with liver hepatitis (82 cases), Included into a Study about the connections between the liver diseases and the heart's health - measured using detailed electrocardiograms. The study's purpose was to find if we can extract, using the heart's activity analysis, certain conclusions about the liver's state of health. The identified predictors are useful for further data processing.