Semiparametric Canonical Correlation Analysis for Prediction of Multiple Outcomes

• Activity Number: 574
• Type: Invited
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Denis Agniel* and Tianxi Cai
• Companies: Harvard Medical School and Harvard University
• Keywords: Multiple phenotypes ; Semiparametric transformation models ; Canonical correlation analysis ; Prediction ; Copula methods ; Resampling

Abstract

Considerable interest has recently been focused on studying multiple phenotypes simultaneously in both epidemiological and genomic studies, either to capture the multidimensionality of complex disorders or to understand shared etiology of related disorders. We seek to identify risk scores as linear combinations of a set of predictors that are highly predictive of a set of such phenotypes. Furthermore, in order to correctly characterize the complexity of a disorder, the phenotypes may be measured on completely different scales, i.e. some phenotypes may be binary, others continuous, and still others censored survival. We use marginal parametric and semiparametric transformation models to - in some sense - put all phenotypes on the same scale, and we use copula methods to estimate the correlation among phenotypes. We then perform a version of canonical correlation analysis which provides risk scores for predicting the multiple phenotypes. Resampling is used to quantify the variability in all estimates.