Modeling Risk Reduction in Limited Distribution--Coronary Artery Calcification in the DCCT/EDIC Study

• Activity Number: 302
• Type: Contributed
• Date/Time: Wednesday, August 11, 2004 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Wanjie Sun*+ and Patricia Cleary
• Companies: George Washington University and George Washington University
• Keywords: Tobit regression ; limited distribution ; left censoring

Abstract

Limited distributions are observed in medical studies, where a large percentage of the dependent variable is censored below or above a threshold value, thus generating a mixture distribution. Ordinary least square estimates or logistic regression are inconsistent or inefficient for censored data. The Tobit model was designed for censored normal data. Tobit models the probability of being censored and if uncensored, the variability of the continuous variable simultaneously. In the DCCT/EDIC study, we used the Tobit model to perform a statistical analysis of a highly skewed Coronary Artery Calcification (CAC) dataset with 70% left censoring of zero CAC scores. Natural log transformation was applied in order to preserve the non-negativeness of CAC scores and to conform to the normal assumption of the Tobit model. The adjusted geometric mean ratio of CAC scores between the two treatment groups was used to assess the group effect for each combination of the effect modifiers. Overall treatment effect was assessed using a likelihood ratio test.