Diagnosing Glaucoma Progression With Visual Field Data Using a Spatiotemporal Boundary Detection Method

Abstract

Diagnosing glaucoma progression is critical for limiting irreversible vision loss. A common method for assessing glaucoma progression uses a longitudinal series of visual fields (VF) acquired at regular intervals. VF data are characterized by a complex spatiotemporal structure due to the data generating process and ocular anatomy. Thus, advanced statistical methods are needed to make clinical determinations regarding progression status. We introduce a spatiotemporal boundary detection model that allows the underlying anatomy of the optic disc to dictate the spatial structure of the VF data across time. We show that our new method provides novel insight into vision loss that improves diagnosis of glaucoma progression using data from the Vein Pulsation Study Trial in Glaucoma and the Lions Eye Institute trial registry. Simulations are presented, showing the proposed methodology is preferred over existing spatial methods for VF data. Supplementary materials for this article are available online and the method is implemented in the R package womblR.

Multivariate Longitudinal Modeling of Macular Ganglion Cell Complex

“…Such correlations have been investigated within the central 24° and 10° visual fields. 44 Asaoka reported that sectorization of 30-2 and 10-2 visual field locations were distinct and that the identified 10-2 sectors were stable on bootstrapping. 45 Such sectors tended to follow patterns of central RNFL bundles.…”

Bayesian Non-Parametric Factor Analysis for Longitudinal Spatial Surfaces

“…More details about the glaucoma population are given in Section 1 of the online Supplementary Materials. See Berchuck et al (2019b) for a more in depth introduction to visual fields for statisticians.…”

A Discrete Kernel Stick-Breaking Model for Detecting Spatial Boundaries in Hydraulic Fracturing Wastewater Disposal Well Placement Across Ohio

“…Lee & Mitchell, 2012; Lu et al., 2007; Ma et al., 2010), with extensions to the spatiotemporal setting (e.g. Berchuck et al., 2019; Rushworth et al., 2017). Other methods have attempted to avoid some of the computing and identifiability issues suffered by these high‐dimensional models by viewing the entire neighbourhood structure as a single random quantity (Lee et al., 2014; Li et al., 2011).…”