Predicting Glaucoma Progression Using Random Forest Based on Correlated Binary Outcome and Longitudinal Covariates

• Activity Number: 544
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Juanjuan Fan*+ and Lucie Sharpsten and Xiaogang Su and Shaban Demirel and Richard A. Levine
• Companies: San Diego State University and San Diego State University and University of Alabama and Devers Eye Institute and San Diego State University
• Keywords: random forest ; joint modeling ; longitudinal data ; correlated binary outcome ; glaucoma progression ; sensitivity and specificity

Abstract

The prediction of glaucoma progression has been recognized as one of the crucially needed and most challenging facets of glaucoma research. This is largely due to the difficulties associated with dealing with correlated outcome data from two eyes of each patient and noisy, often longitudinally collected, visual field data. To this end, we develop a joint modeling approach by fitting longitudinal visual field data with random effects models, while incorporating the true rate of visual field deterioration as well as other baseline demographic and clinical data into a random forest framework. Variable importance measures are also derived from the random forest. The method is applied to the data from perimetry and psychophysics in glaucoma (PPIG) study. Using both eyes of each subject and longitudinal visual field data, our proposed method is able to predict glaucoma progression with near perfect sensitivity and specificity for the test sample.