Censored Functional Data with Application to a Mortality Study in the ICU

• Activity Number: 323
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jonathan Gellar*+ and Elizabeth Colantuoni and Dale M. Needham and Ciprian Crainiceanu
• Companies: Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health and Johns Hopkins University and Johns Hopkins University
• Keywords: Functional Data analysis ; Survival analysis ; Biomarker ; Joint Modeling ; Intensive Care Unit ; Nonparametric regression

Abstract

We propose a method for the dynamic prediction of mortality based on a concurrently measured biomarker. Our model differs from existing methods in that it treats one's biomarker history as a functional predictor in a Cox proportional hazards model. At any given time, one's hazard function is allowed to depend on the subject's entire history of the covariate up to that point. The key feature of the model is the incorporation of a bivariate functional coefficient that allows the effect of the covariate history to be updated (smoothly) as time increases. Methods were motivated by a population of patients in the Intensive Care Unit (ICU). We relate daily measures of the Sequential Organ Failure Assessment (SOFA) score to one's risk of mortality, under the competing risk of hospital discharge. We show how our model can be used to predict outcomes based on partial trajectories of SOFA scores, and also compare our model to existing methods including distributed lag models and joint modeling approaches.