Lung Cancer Risk Prediction with Stochastic Covariates

• Activity Number: 31
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Denise M Danos*+ and Neal Simonsen and Elizabeth Fontham and Evrim Oral
• Companies: Louisiana State University Health Sciences Center and Louisiana State University Health Sciences Center and Louisiana State University Health Sciences Center and Louisiana State University Health Sciences Center
• Keywords: Modified Maximum Likelihood ; Order Statistics ; Stochastic Covariates ; Logistic Regression ; Robustness ; Outliers

Abstract

Lung cancer is the leading cause of cancer death in the US. Previous studies on the nutritional etiology of lung cancer may be inconsistent partially due to inadequate control for smoking through the methods used to do so. Ignoring the stochastic nature of risk factors may contribute to this inconsistency as well. We propose an enhancement of logistic regression analysis that could be used to assess the association of nutritional risk factors with lung cancer. We consider stochastic non-normal covariates and utilize modified maximum likelihood methodology. We show that the proposed estimators are highly efficient, and treating the risk factor as non-stochastic results in loss of efficiency. We illustrate the method using data collected from a population-based case-control study, namely the Lower Mississippi River Interagency Cancer Study (LMRICS), wherein 892 subjects with complete information on diet and smoking habits were interviewed from 1998 to 2001.