Purchase this article with an account.
Srinivas Rajaraman, Rupesh Singh, Siamak Yousefi, Christopher Bowd, Linda M Zangwill, Robert N Weinreb, Madhusudhanan Balasubramanian; Detecting Glaucomatous Progression from Localized Visual Function Changes with Corrections for Multiple Comparison. Invest. Ophthalmol. Vis. Sci. 2016;57(12):3919.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To detect glaucomatous progression from localized visual function changes with type I error controlled in a nonparametric framework (PixR) and compare to the Permutation Analysis of Pointwise Linear Regression (PoPLR) method.
The rate of visual threshold sensitivity at each of the Standard Automated Perimetry (SAP) retinal test locations were estimated using linear regression. For nonparametric analysis, regression errors were assumed to be independent and identically distributed (exchangeability criterion). The significance of rate of change (p-value) in each location was estimated using permutation tests with Monte Carlo sampling while accounting for multiple simultaneous comparison problem using Bonferroni correction. Using these p-values, glaucoma progression was detected once again nonparametrically based on the significance (at a level of 5%) of the observed number of progressing locations. Study eyes with at least 4 SAP exams from the UCSD Diagnostic Innovations in Glaucoma Study (DIGS) were included. 80 eyes of 74 participants were progressing based on stereophoto evaluation; and in 84 eyes of 45 participants all SAP measurements were within 3 months (stable group).
Sensitivity (95% CI) of PixR and PoPLR were 64% (53%, 75%) and 50% (38%, 62%) respectively. Specificity (95% CI) of PixR and PoPLR were 98% (94%, 100%) and 94% (88%, 100%).
With high specificity, correction for multiple comparison using Bonferroni correction provided a high diagnostic accuracy of detecting glaucomatous progression. While Bonferroni correction is generally conservative, it provided an optimal diagnostic accuracy due to relatively fewer number of locations simultaneously tested.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only