Assessment of angle closure spectrum disease as a continuum of change using gonioscopy and anterior segment optical coherence tomography.

Abstract

Studies examining the anterior chamber angle and angle closure disease often compare quantitative angle information obtained using anterior segment optical coherence tomography (ASOCT) with one of several ordinal scales derived using gonioscopy. We test the assumption that the ordinal gonioscopic angle grades have equal step sizes and can be analysed using metric statistics. The medical records of 214 consecutive patients who were referred for assessment of the anterior chamber angle were prospectively examined using gonioscopy and ASOCT (Spectralis Optical Coherence Tomography, OCT, www.heidelbergengineering.com). Anterior chamber angle parameters (angle opening distance, AOD, and trabecular-iris space area, TISA at 500 and 750 microns) were extracted from ASOCT images using a semi-automated segmentation algorithm written on MATLAB (www.mathworks.com). We first matched the quantitative values for each gonioscopic grade (0-4, from no structures visible to ciliary body visible) and described the frequency distributions to determine separability. We then applied a grade-agnostic clustering algorithm to determine the concordance between algorithm-clustered groups (using solely quantitative data) and those obtained using gonioscopy. The frequency distributions of the quantitative ASOCT parameters for each angle grade were mostly non-parametric and displayed unique distribution characteristics, with a floor effect seen for grade 0 and the lack of a ceiling effect seen for grades 3 and 4. Although we found significant differences in quantitative values across the five angle grades using the frequency distributions, some pairwise comparisons were indistinguishable (such as grades 0 and 1, and grades 3 and 4) due to the overlaps in distributions. On average, differences in quantitative values were consistent between gonioscopic grade steps, but there remained substantial variability that confounds prediction of change between ordinal steps. The clustering algorithm showed approximately 10% of cases with the same group assignment as that of the gonioscopic grade, improving slightly to 30% when the top 5% of quantitative data were excluded from analysis. Our results do not necessarily support the assumption that the ordinal scales used in gonioscopy can be interpreted using an interval scale. We highlight the need for better methods of describing the course and risk of angle closure spectrum disease to identify disease progression and conversion, where gonioscopy remains the gold standard.