The main outcome measure was the macular curvature defined by the second derivative of a quadratic polynomial fit of the automatically generated BM segmentation. The polynomial fit was calculated using R (R Core Team [2021]; R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, URL
https://www.R-project.org/, R version 4.1.2 [2021-11-01]). Descriptive statistics were computed and stratified by the clinical group. Absolute and relative frequencies were calculated for dichotomous parameters, and the mean and standard deviation were calculated for approximately normally distributed variables (otherwise median and interquartile range). Linear regression models with general estimating equations (GEEs) were used to assess associations and account for correlations between corresponding eyes. As previous authors have mentioned in their research with a Topcon OCT, raster scans did not undergo any sort of spatial processing, however, we do not know if this also accounts for imaging post-production in the HEYEX algorithm as well. Furthermore, as also pointed out by Müller et al., varying optical path distortions by OCT system optics could not be excluded.
31 We have therefore decided to also adjust for mean corneal power as well as axial length to account for these possible influences in the perinatal model. For the ocular morphology model, we broke down the highly associated axial length into central corneal thickness, lens thickness, anterior chamber depth, and posterior segment length. First, univariable analyses of the macular curvature and sex (female), age (years), mean corneal power (diopter), axial length (mm), GA (weeks), birth weight (kg), birth weight percentile, ROP (yes), ROP treatment (yes), perinatal adverse events (yes), placental insufficiency (yes), pre-eclampsia (yes), breastfeeding, (yes) and maternal smoking during pregnancy (yes) were computed. Additionally, we calculated univariable models with adjustment for age, sex, axial length, and mean corneal power. Then, only perinatal parameters associated in the univariable analyses were included in the first multivariable model with additional adjustments for sex, age, corneal power, and axial length, but excluding ROP treatment due to high collinearity of ROP and ROP treatment. In the second multivariable model, multivariable associated perinatal parameters and the potential effect of ROP treatment (yes) were analyzed with additional adjustments for age, sex, axial length, and mean corneal power. Birth weight was excluded from the multivariable models to avoid collinearity, which was strong between GA at birth and birth weight. In the analysis of ocular geometry, the associations of ocular geometry with macular curvature were analyzed including only subjects without ROP treatment. First, univariable associations were calculated between macular curvature and sex (female), age (years), mean corneal power (diopter), central corneal thickness (µm), anterior chamber depth (mm), lens thickness (mm), posterior segment length (mm), foveal retinal thickness (µm), and subfoveal choroidal thickness (µm). We then calculated univariable models with additional adjustments for age, sex, and mean corneal power. In the multivariable model, all ocular geometric parameters were included that were associated in univariable analyses with additional adjustments for age, sex, and mean corneal power. A sensitivity analysis was performed with the inclusion of foveal hypoplasia in the model testing the association with ocular geometry. As this is an explorative study, a significance level was not defined and no adjustment for multiple testing was conducted. Thus,
P values are reported only for descriptive purposes and should be interpreted with caution.
32 Calculations were performed using commercial software (IBM SPSS 20.0; SPSS, Inc., Chicago, IL, USA).