Decomposition, Gradient, and Reduction Colinearity in High-Dimension Data from Case-Control Study

• Activity Number: 411
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yuanzhang Li*+ and David Niebuhr and Tianqing Liu
• Companies: WRAIR and Walter Reed Army Institute of Research and Walter Reed Army Institute of Research
• Keywords: Biomarker selection ; Reduction ; co-linearity ; case control study ; schizophrenia

Abstract

The effect of individual predictors in a multiple regression may be biased due to the multicollinearity. Multicollinearity often occurs in longitudinal studies, especially when the objective is to study the association between disease and biomarkers. In this study, we proposed space decomposition to group potential biomarkers according their association and use the gradient-nuisance vector approach to reduce the number of biomarkers included in the high-dimension regression. The co-linearity among the biomarkers was dramatically reduced by the method. We use this approach on a US military case control data set to evaluate the association of biomarkers on risk of schizophrenia. The linear correlation among biomarkers was as high as 0.8 and after decomposition, the correlation coefficient among the vectors was lower than 0.3. The predictive power of the model as a whole did not reduce. The