Following image acquisition at each study visit, all OCT images were exported from the instrument and analyzed using custom written software. The image analysis procedures employed in this study have been previously described in detail.
20 Briefly, an automated graph-based method was initially used to segment the outer boundary of the retinal pigment epithelium (RPE) in all OCT images. An experienced masked observer then manually segmented the chorioscleral interface (CSI), corrected any RPE segmentation errors, and marked the center of the fovea (defined as the position of the deepest portion of the foveal pit) in all OCT images. Following segmentation of the OCT images, the transverse scale of each subject's OCT data was adjusted to account for ocular magnification factors using their individual ocular biometry data from that study visit.
Choroidal thickness (defined as the distance from the RPE to the CSI) across each OCT image was then calculated to determine the subfoveal choroidal thickness, and the average choroidal thickness across a series of concentric annular zones around the fovea, including the central foveal zone (central 1-mm diameter), the inner macula zone (from an inner diameter of 1 mm to an outer diameter of 3 mm) and the outer macula zone (inner diameter of 3-mm to outer diameter of 6 mm). These data were further analyzed to determine the average thickness at eight locations (temporal, superior temporal, superior, superior nasal, nasal, inferior nasal, inferior, and inferior temporal) across each of the three zones (central fovea, inner macula, and outer macula).
All statistical analyses were carried out using IBM SPSS Statistics Version 21 (Armonk, NY, USA). The longitudinal changes in subfoveal choroidal thickness (and axial length) over the 18 months of the study (and the influence of various predictor variables upon the growth trajectory of the choroid) were examined using linear mixed model (LMM) analysis with restricted maximum likelihood estimation. The LMM examined the effect of study visit time (in years from baseline visit, as a time varying continuous variable) upon subfoveal choroidal thickness, using a first order autoregressive covariance structure (which assumes the correlation between measurements is higher for measurements taken closer together in time). Individual subject's slopes and intercepts were included as random effects in the model (allowing for any pattern of correlation between the random effects). In addition to classification according to refractive error group (i.e., myope or nonmyope), subjects were additionally classified based on their axial eye growth over the course of the study (linear regression analysis of each individual subject's change in axial length over time was used to derive an axial growth rate for each subject). This was based upon a tertile split of the axial growth rate data into groups exhibiting either slow eye growth (<25 μm/y, n = 33), or medium rate of eye growth (between 25 and 67 μm/y, n = 33) or fast eye growth (>67 μm/y, n = 33). Categorical predictor variables (refractive error group, axial eye growth rate, and sex) were included in the model as fixed factors, and continuous predictor variables (baseline axial length, change in axial length, and age at baseline visit) were included as covariates. Separate analyses were carried out for refractive error group, axial growth rate, and change in axial length because these factors are typically related. A similar approach was used for the analysis of the longitudinal changes in the parafoveal choroidal thickness, with the additional fixed factors of parafoveal zone and location included in the LMM.
Repeatability of the imaging and analysis procedures were assessed through analysis of the two repeated OCT measurements collected at each study visit. The mean difference and 95% limits of agreement between the differences were determined for these data at each visit using the methods of Bland and Altman.
30