Multiply Imputing Ordinal Variables Using Latent Continuous Variables in Tissue Microarray Data

• Activity Number: 194
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jun Xing*+ and Thomas R. Belin and Steve Horvath
• Companies: University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles
• Keywords: multiple imputation ; latent variable ; joint modeling ; tissue microarray

Abstract

Tissue microarrays (TMAs) are a new high-throughput tool for study of protein expression patterns that are increasingly being used to evaluate the diagnostic and prognostic importance of tumor biomarkers. Protein expression variables are typically measured on an ordinal scale and typically have non-normal distributions, frequently with substantial skewness. The fact that protein expression values may be missing creates challenges when looking for associations between tumor biomarkers and clinical outcomes. The motivating example for the present work comes from TMA data with six candidate biomarkers obtained from bladder cancer patients. We propose to impute missing values by simultaneously modeling of protein expression covariates, survival time, and other covariates, with interest focusing on the connection between protein expression covariates and disease risk. This presentation will contrast results from simple approaches such as median imputation and multivariate normal imputation and will outline a joint modeling approach that we anticipate will offer advantages over simpler techniques.