Hierarchical Modeling of Glaucomatous Visual Field Progression

• Activity Number: 424
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Luohua Jiang*+ and Gang Li and Joseph Caprioli
• Companies: University of California, Los Angeles and University of California, Los Angeles and Jules Stein Eye Institute
• Keywords: glaucoma progression ; visual field ; image analysis ; spatio-temporal data ; hierarchical model ; mixture models

Abstract

Reliable detection of glaucomatous deterioration remains one of the most difficult problems clinicians facing in glaucoma management. Visual field data, the data used to detect glaucoma progression, are hierarchical, with 15 or more longitudinal images available for most of the eyes, and within each image, 52 locations nested in 12 clusters. Most previous analyses for this type of data focused on simple linear regression at each location of an eye, which totally ignored the hierarchical structure and spatial-temporal correlation of the data. In addition, those methods did not incorporate the information of all the eyes when making diagnostic decisions. This work develops a Bayesian approach to model all the data simultaneously. The proposed method uses a multilevel random effects model with spatial mixture distributions to classify patients. In this framework one can estimate the mean profiles of stable eyes and progressive eyes and then calculate the posterior probability for an eye being progressive, which provides a more objective criterion than previous methods. The model also allows us to assess heterogeneity in profiles among clusters and among locations for a given cluster.