Purchase this article with an account.
S. Demirel, B. Fortune, J. Fan, R. A. Levine, R. Torres, H. Nguyen, S. L. Mansberger, S. K. Gardiner, G. A. Cioffi, C. A. Johnson; Predicting Progressive Glaucomatous Optic Neuropathy Using Standard Automated Perimetry. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1153.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To test the hypothesis that specific locations and patterns of threshold findings within the standard automated perimetry (SAP) visual field (VF) have predictive value for progressive glaucomatous optic neuropathy (pGON).
We used the initial SAP threshold data of 168 individuals with high-risk ocular hypertension or early glaucoma, along with intraocular pressure and central corneal thickness in a classification tree (CART) model to predict pGON. Two fellowship trained glaucoma specialists independently determined pGON based on masked evaluation of longitudinal stereo optic nerve head photos. The mean interval between photos used to determine pGON was 5.5 years (range 2.0 - 7.9). Data from the worst eye of each individual was included. If only 1 eye of an individual displayed pGON then it was considered the worst eye. If neither or both eyes of an individual displayed pGON then the worst eye was randomly selected. The random selection process was repeated 10 times, with 10 CART models consequently constructed.
Patterns of findings within the VF (e.g. being in the lower quartile of normal values at one location while being in the upper tertile of normal values at another location) carried information predictive of pGON. Initial thresholds conveyed information that altered the risk of pGON even when these thresholds were not abnormal at the p<5% level. A subset of VF locations was identified that had most influence on pGON prognosis (figure). These locations were along the nasal horizontal meridian, clustered in the inferior temporal quadrant and scattered within the superior field. In the figure these locations are shown in the 24-2 pattern (OD configuration) with darker shading signifying greater importance.
CART models can be used to determine which VF locations are most predictive of pGON. Patterns of threshold findings within the VF convey useful information even when thresholds are within the classically defined normal range.
This PDF is available to Subscribers Only