Assessment Predictive Power of Prognostic Models for Melanoma Survival Data

• Activity Number: 26
• Type: Contributed
• Date/Time: Sunday, August 3, 2008 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Shouluan Ding*+ and Seng-jaw Soong+
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham
• Keywords: Predictive Power of Model ; Concordance Correlation coefficient ; Melanoma survival Data ; Extended generalized Gamma Distribution ; Parametric Models ; Cox model

Abstract

We have developed the predictive Cox model and Extended Generalized Gamma (EGG) model in melanoma research. For model validation, which is the most overlooked, we used AJCC melanoma collaborative database divided into a training dataset with 14,760 patients from 10 institutions for Cox and EGG developing of models by thickness 7 subgroups and a test dataset with 10,974 patients from Sydney Melanoma Unit (SUM) for validating by the same groups. Concordance correlation coefficient (CCC) was used to assess the predictive power of Cox and EGG by comparing predicted survival rates with Kaplan-Meier observed survival rates at 5- and 10-year time point for individual patients with combination of significant clinical or pathologic prognostic factors within thickness groups for SUM dataset. Overall CCC of 5- and 10-year survival rates were excellent consistency, at 0.87 and 0.89, respectively.