Impact of Regression Methods on Detecting Increased Coronary Artery Calcification in Rheumatological Diseases

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Tebeb Gebretsadik*+ and Ayumi Shintani
• Companies: Vanderbilt University School of Medicine and Vanderbilt University
• Keywords: skewed response ; cluster zeros ; regression method ; Monte Carlo simulation ; coronary artery calcification

Abstract

The highly skewed distribution of an outcome variable with a large cluster of zeros has led to several analytic challenges. Our motivating example is the study of coronary artery calcification (CAC) among systemic lupus erythematous (SLE) and rheumatoid arthritis (RA) for markers of premature coronary disease. Many analytical methods have been used to analyze highly skewed Agatson score, a measure of CAC where the majority of patients have zero values along with few patients with high extreme values. We conducted Monte-Carlo simulation study to assess the performance of commonly used regression methods including linear regression with transformation (CAC+1), logistic regression (no CAC vs. any), the proportional odds logistic regression, the quantile regression, zero-inflated negative binomial regression and hurdle regression methods. Power and type I error rates were evaluated.