Probability of Change in Optic Nerve Head Topography with High Dimensional Analysis of Variance

Abstract

Abstract: : Purpose: To develop new statistical methods to determine the overall probability of change in a series of confocal scanning laser ophthalmoscope (CSLO) measurements of the optic nerve head over time. Methods: 112 eyes from 112 patients from the glaucoma divisions at UCLA, Yale University, and the University of Antwerp, Belgium were evaluated. All patients had at least 3 imaging sessions (mean 4.2 ± 1.0, range 3-8) with the Heidelberg Retinal Tomograph (HRT) over at least 5 years (mean 6.4 ± 1.0 years, range 5.0-8.8). Serial HRT images from each patient were aligned. A Fourier transform was applied to the aligned areas to reduce the dimensionality of the data analysis. To decrease high frequency noise in a series, only the most important 61% of the transformed height data was further analyzed by applying Fan’s adaptive Neyman test and the p value was calculated. Results: For 92 eyes (82%) a statistically significant change (p ≤ 0.05) was observed in optic nerve head topography over time. Of those eyes demonstrating apparent change, 80% showed decreasing height measurements, while 20% showed increasing height measurements. Conclusion: Analysis of longitudinal series of CSLO measurements for significant change must take into account the temporal and spatial correlations in the data. Transforming optic nerve head height measurements into a frequency domain with the Fourier transform and applying high dimensional analysis of variance is a simple and powerful method to detect change. This initial approach overcalls change and needs further modifications to reduce sensitivity.

Keywords: optic disc • imaging methods (CT, FA, ICG, MRI, OCT, RTA, S • clinical research methodology