Generalized estimating equation modeling of choriocapillaris flow deficits with an SS-OCTA normal database

Abstract

Purpose : To establish a generalized estimating equation (GEE) model to estimate choriocapillaris (CC) flow deficits (FDs) in all age groups with swept-source optical coherence tomography angiography (SS-OCTA).

Methods : A database of 164 normal adult subjects from all age groups was used for the modeling of the relationship between FD size (FDS), intercapillary distance (ICD), age and distance to fovea (DF) center (i.e. radial distance). All subjects were imaged using an SS-OCTA (PLEX^® Elite 9000 (ZEISS, Dublin, CA)), with both 3x3 mm and 6x6 mm scans. CC was defined as a 20μm thickness slab located beneath Bruch’s membrane. CC structure image was used for signal attenuation compensation; retinal projection artifacts were removed and the regions corresponding to the large retinal vessels were excluded from further analysis. A previously published fuzzy C-means algorithm was implemented to segment FDs. The size and minor axis of each FD were calculated to represent FDS and ICD, respectively. Distances from the centroid of each FD to fovea was calculated.

Results : The GEE model is described as FDS/ICD = Intercept + age*coefficient 1(age) + DF (i)*coefficient2i (DF) + error, in 3x3 mm scans, i=1 to 4, in 6x6 mm scans, i=1 to 8. The modeling showed significant correlation of FDS and ICD to age/DF in both 3x3 mm and 6x6 mm scans (all p<0.001). In 3x3 mm, coefficient for age is 11.67 (robust standard error (R.S.E.): 1.33) for FDS and 0.12 (R.S.E.: 0.012) for ICD. These results mean that for the same DF, we estimated 11.67 µm² (95% CI: [9.07 - 14.28]) increase in FDS and 0.12µm (95% CI: [0.09 – 0.14]) increase in ICD every year in aging. For 6x6 mm, coefficient for age is 22.55 (R. S.E.: 4.07) in FDS and 0.15 (R. S.E.: 0.02) for ICD. Similarly in 6x6 mm scans, we estimated 22.55 µm² (95% CI: [14.57 - 30.53]) increase in FDS and 0.15µm (95% CI: [0.10 – 0.19]) increase in ICD every year in aging. 95% confidence interval (CI) of FD size and ICD for each age was calculated. See Fig.1 and Table 1&2 for details.

Conclusions : GEE modeling can predict the normal ranges of FDS and ICD with all age groups. Such model could be used to identify abnormally large FDS in SS-OCTA scans and be potentially useful in disease diagnosis and treatment monitoring.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.