Long-term risk assessment of living kidney donors: a retinal study

Abstract

Purpose : Living kidney donors are typically considered to have normal kidney function. Recently, concerns have been raised on the long-term risk of both cardiovascular and chronic kidney disease (CKD) following kidney donation. However, quantitative risk assessment tools are lacking. We propose a retinal imaging approach to gain insights into the timing of vascular changes in kidney donors.

Methods : We prospectively recruited three groups: 30 healthy subjects, 30 patients with stable CKD, and 30 kidney donors. Optical coherence tomography angiography images of one eye per participant were acquired. Image processing was performed using a U-Net neural network to segment the vasculature. The binarised image was converted into a vascular network object (graph) from which vessel characteristics were extracted. We obtained retinal measurements (e.g., nodes in the graph, size and shape of intercapillary space) from each region of interest in the retina, up to a total of 593 phenotypes. Multivariate models, corrected for age and gender, assessed statistically significant retinal characteristics that discriminated CKD from health.
Correlated vascular measures were removed and the remaining measures used to perform principal component analysis (PCA) to visualise patient distributions in a 2-dimensional (2D) space. Finally, clustering analysis was performed to investigate the separability between these two groups. Using the same PCA transformation, we projected kidney donors in our 2D space to observe their distribution in the healthy-CKD plane. Also, we investigated if donor PCA projection was associated with estimated glomerular filtration rate (eGFR), a routine clinical measure of kidney function.

Results : Our results show that PCA separates healthy subjects from those with CKD (accuracy=0.82). When we projected kidney donors into the same 2D space, they distributed across the two previous clusters, suggesting that some donors have similar retinal vascular characteristics to CKD whilst others resemble health. These differences were not associated with eGFR.

Conclusions : Our data suggest that retinal imaging may be used to discriminate those with normal kidney function from those with CKD. Whether these differences might link to patient outcomes remains unclear. Retinal imaging may identify living kidney donors at risk of future CKD and cardiovascular disease.

This is a 2021 ARVO Annual Meeting abstract.

 

(A) PCA; (B) histogram of clusters.