Accelerated Peripapillary Retinal Nerve Fiber Layer Degeneration in Patients With Chronic Kidney Disease: A 2-Year Longitudinal Study

Abstract

Purpose: To evaluate the longitudinal changes in the peripapillary retinal nerve fiber layer (pRNFL) in patients with chronic kidney disease (CKD).

Methods: In this prospective cohort study, the CKD group consisted of patients with CKD stage ≥ 3. Age-matched healthy controls were enrolled at a 1:4 ratio. Spectral-domain optical coherence tomography was used to measure the pRNFL at baseline, 1 year, and 2 years. Within-group longitudinal changes and between-group comparisons were performed using linear mixed models.

Results: Overall, 152 patients with CKD and 40 controls were included (mean ages, 62.8 ± 9.1 years vs. 63.0 ± 9.3 years; P = 0.931). The CKD group showed faster loss of pRNFL than the control group (−0.87 µm/y vs. −0.26 µm/y; P = 0.004). Subgroup analysis found that the rate of pRNFL change was −0.41 µm/y in stage 3a CKD, −0.74 µm/y in stage 3b, −0.98 µm/y in stage 4/5, and −1.38 µm/y in end-stage renal disease. Multiple linear regression analysis revealed that CKD stage (coefficient = −0.549; 95% confidence interval [CI], −0.966 to −0.131; P = 0.010), hypertension (coefficient = −1.557; 95% CI −3.013 to −0.101; P = 0.036), and rim area (coefficient = −1.505; 95% CI, −2.940 to −0.070; P = 0.040) were factors associated with the pRNFL change over 2 years.

Conclusions: Patients with CKD experienced faster pRNFL loss than healthy controls did. Severity of CKD, hypertension, and rim area were independent factors associated with the loss of pRNFL.

Translational Relevance: This study contributes to our understanding of retinal neurodegeneration in normal aging and in patients with chronic kidney diseases.

Introduction

The global burden of chronic kidney disease (CKD) has substantially increased over the past three decades.¹ CKD affects approximately 10% of adults worldwide and is even more prevalent in older patients and in patients with diabetes mellitus (DM) or hypertension.² It can increase the risk of visual impairment and major eye diseases two to seven times.³ Patients with CKD are also more prone to have retinal neurodegeneration.^4–15 Optic atrophy with profound visual loss can occur in patients with end-stage renal disease (ESRD).^16,17 Understanding the natural course and risk factors associated with retinal neurodegeneration could be important for the preservation of visual function in these patients. 

The thickness of the peripapillary retinal nerve fiber layer (pRNFL) measured by spectral-domain optical coherence tomography (SD-OCT) parallels the axon loss in the optic nerve and can serve as a biomarker for retinal neurodegeneration.^18,19 Impaired renal function has been shown to be a factor associated with pRNFL thickness reduction.^6–9 Patients with diabetic or non-diabetic kidney diseases have been found to have a lower pRNFL thickness when compared to controls.^4–8 Patients with ESRD were also found to be more likely to have decreased pRNFL or present with multiple pRNFL defects.^10–15 

However, most of these prior studies were cross-sectional studies and could not determine whether the retinal neurodegeneration in CKD is stationary or progressive. The neural defects identified in those studies could be either sequelae of neural injuries from before the CKD diagnosis or caused by underlying pathological processes that led to accelerated neurodegeneration. The rates of pRNFL loss at different stages of CKD were also unclear in these studies. The purpose of this study was to evaluate the longitudinal changes in pRNFL at different stages of CKD and compare them to those that occur during healthy aging. A 2-year longitudinal design was used in this study to better illustrate the trends of change in each group. We also evaluated the risk factors associated with the longitudinal changes in pRNFL thickness in patients with CKD. 

Patients and Methods

Study Design

This prospective longitudinal study was conducted in a tertiary hospital (Keelung Chang Gung Memorial Hospital, Taiwan) between August 2017 and July 2021. The Chang Gung Memorial Hospital Institutional Review Board approved this study (IRB nos. 201602022B0 and 201702074A3). The study followed the tenets of the Declaration of Helsinki, and informed consent was obtained from each participant. Qualified subjects enrolled in this study received ocular examinations at baseline, 1 year, and 2 years (Fig. 1). 

Figure 1.

 

Flowchart of this study.

The estimated glomerular filtration rate (eGFR) was calculated from the serum creatinine concentration.²⁰ CKD was defined as exhibiting structural or functional abnormalities of the kidneys and decrease of eGFR to <60 mL/min/1.73 m² for >3 months.²¹ The stages of CKD were defined by eGFR values: 45 to 59 mL/min/1.73 m² (stage 3a), 30 to 44 mL/min/1.73 m² (stage 3b), 15 to 29 mL/min/1.73 m² (stage 4), and <15 mL/min/1.73 m² (stage 5). Subjects receiving hemodialysis or peritoneal dialysis were categorized as having ESRD. 

Patients who were ≥40 years of age with CKD stage ≥3 and no visual symptoms were enrolled in the CKD group. Age-matched healthy subjects were enrolled into the control group at a 1:4 ratio. The exclusion criteria were as follows: (1) lost to follow-up before 2 years; (2) presence of any significant ocular diseases that could affect the pRNFL thickness, such as high myopia < −6 D, long axial length > 26 mm, any optic nerve disease, glaucoma (or glaucoma suspected), retinal vessel occlusion, and diabetic retinopathy; (3) poor vision affecting fixation; (4) intraocular pressure (IOP) > 21 mmHg; (5) inadequate OCT image quality (signal strength index [SSI] < 40 or presence of significant artifacts) in any of the visits; (6) eyes that underwent cataract surgery within the study period; and (7) history of stroke, Alzheimer's disease, or dementia. Patients with cup-to-disc ratio ≥ 0.7, cup-to-disc ratio asymmetry ≥ 0.3, or glaucomatous visual field defect or who were using an anti-glaucoma medication were considered as glaucoma (or glaucoma-suspected) patients and were not eligible for this study. Only patients who completed all three visits (i.e., baseline, 1 year, and 2 years) with consistently good image quality in all visits were included for analysis. One eye of each individual was used for statistical analysis. For patients in whom both eyes were eligible, the eye with the more consistent OCT image quality was used as the study eye. 

Patients’ medical history and laboratory data were collected through a standardized questionnaire and medical records. An unattended automated office blood pressure (BP) measurement and comprehensive ophthalmic examinations were performed at baseline and during each visit. Best-corrected visual acuity (BCVA) was measured on a Snellen chart and then converted to the logarithm of the minimum angle of resolution (logMAR). IOP was determined using a noncontact tonometer (NT-3000; Nidek, Tokyo, Japan). An IOLMaster biometer (Carl Zeiss Meditec, Jena, Germany) was used for axial length measurement. 

SD-OCT Measurement

SD-OCT (RTVue XR Avanti OCT System; Optovue, Inc., Fremont, CA) was used to measure pRNFL thickness in this study. An optic nerve head (ONH) scan consists of 12 radial lines and six concentric rings. The disc area, rim area, and cup-to-disc ratio were automatically calculated from radial scans. The average pRNFL thickness was measured over 360° of a 3.45-mm-diameter circle centered on the ONH (Fig. 2). The pRNFL was then divided into eight sectors for sectoral analysis: superior nasal, nasal upper, nasal lower, inferior nasal, inferior temporal, temporal lower, temporal upper, and superior temporal. 

Figure 2.

 

A representative longitudinal pRNFL thickness measurement in a patient with end-stage renal disease. (A) The pRNFL thickness is measured over a 3.45-mm-diameter circle (black circle) centered on the ONH. (B) The pRNFL is autosegmented in a B-scan obtained from the location of the black circle in (A) and compared with the normative database. The ranges of normal, borderline, and outside normal values are represented in green, yellow, and red, respectively. (C) The sectoral measurements of pRNFL thickness at baseline and at 1-year and 2-year follow-up. The average pRNFL thicknesses are 77 µm, 75 µm, and 74 µm, respectively. Superior nasal, SN; nasal upper, NU; nasal lower, NL; inferior nasal, IN; inferior temporal, IT; temporal lower, TL; temporal upper, TU; superior temporal, ST.

Statistical Analysis

Baseline clinical characteristics and demographic data were compared between the CKD and the control groups using an independent sample t-test for continuous variables and Pearson's χ² test for categorical variables. To evaluate the rate of pRNFL change at different stages of kidney disease, all subjects were allocated into five subgroups: (1) controls, (2) CKD stage 3a, (3) CKD stage 3b, (4) CKD stage 4/5, and (5) ESRD. The time of measurement was coded as “time” (0, 1, 2) and was a continuous variable for this study. Within-group longitudinal changes in pRNFL thickness and sectoral pRNFL thickness were analyzed using a linear mixed model with time as a random effect (random slope) and with a random intercept at the subject level. Between-group (e.g., CKD vs. control) differences of longitudinal changes in pRNFL thickness were analyzed using a linear mixed model. The model included main effects of age, sex, axial length, SSI, group, time, and the interaction term (group × time), with time as a random effect (random slope) and with a random intercept at the subject level. Similar linear mixed models were applied to subgroup analyses. Simple linear regression models were used to determine the factors associated with the 2-year change in pRNFL thickness for all subjects and among patients with CKD. Age, gender, baseline logMAR BCVA, spherical equivalent, axial length, and any significant factors identified in simple linear regression were entered into a multiple linear regression model with backward elimination. The relationship between change in logMAR BCVA and change in pRNFL thickness over 2 years was evaluated by partial correlation after controlling for age, sex, and axial length in each group and subgroup. All analyses were performed using SPSS Statistics 26.0 (IBM Corp., Armonk, NY). A two-tailed P < 0.05 was considered statistically significant. 

Results

In total, 214 patients with CKD and 58 healthy controls were eligible at baseline (Fig. 1). Of these, 61 subjects were lost to follow-up and seven died within the 2-year study period. Also excluded were two patients who underwent cataract surgeries in both eyes during the study period and 17 patients who did not have adequate quality images for all three time points. In the end, 152 patients with CKD and 40 controls were included in this longitudinal data analysis. The mean serum creatinine concentration and the eGFR in the CKD group were 4.58 ± 4.68 mg/dL and 30.0 ± 20.9 mL/min/1.73 m², respectively. Forty-six patients had CKD stage 3a, 33 had stage 3b, 31 had stage 4/5, and 42 had ESRD. 

Table 1 summarizes the baseline demographic data and the clinical characteristics of the included subjects. The mean age ± SD was 63.0 ± 9.3 years in controls and 62.8 ± 9.1 years in patients with CKD (P = 0.928). There were no significant differences in sex, ever-smoker status, body mass index, systolic BP, diastolic BP, BCVA, IOP, axial length, disc area, rim area, cup-to-disc ratio, or SSI between the two groups at baseline. There was a trend that patients with CKD had lower pRNFL thickness than that of the controls at baseline (96.5 ± 10.0 vs. 99.2 ± 9.2); however, it was statistically insignificant (P = 0.125), probably due to the small case number in the control group. 

Table 1.

 

Baseline Demographic Data and Clinical Characteristics of the Included Subjects

 

Table 2 shows the mean pRNFL thicknesses at each time point and the longitudinal changes. The within-group longitudinal change was not significant for the control group (−0.26 µm/y; P = 0.142) but was significant for the CKD group (−0.87 µm/y; P < 0.001). Patients with CKD had faster pRNFL loss than subjects in the control group had (P = 0.004) after adjusting for age, sex, axial length, and SSI. The CKD subgroup analysis showed that the rate of change was −0.41 µm/y (P = 0.014) in patients with stage 3a CKD, −0.74 µm/y (P = 0.005) in those with stage 3b, −0.98 µm/y (P < 0.001) in those with stage 4/5, and −1.38 µm/y (P = 0.004) in those with ESRD. Compared with subjects in the control group, patients with CKD stage 3a (P = 0.518) or stage 3b (P = 0.109) did not show a significant difference in the rate of pRNFL loss, whereas patients with CKD stage 4/5 (P = 0.009) or ESRD (P = 0.009) had a significantly faster rate of pRNFL loss compared with the control group. Seven patients had baseline BCVA < 20/40 in this study. We conducted sensitivity analyses for the longitudinal changes of pRNFL by including patients with (1) baseline BCVA ≥ 20/63, and (2) baseline BCVA ≥ 20/40 only. The results were similar (Supplementary Tables S1 and S2, respectively). 

Table 2.

 

Longitudinal Change in pRNFL Thickness in the Control and CKD Groups

 

The results from the linear regression models analyzing the factors associated with the 2-year pRNFL change in the CKD group are shown in Table 3. In simple linear regression models, hypertension, CKD subgroup, and rim area were significant factors among all subjects and among patients with CKD. In the multiple regression model, CKD subgroup and rim area were significant factors among all subjects. Hypertension, CKD subgroup, and rim area were significant factors among patients with CKD. Other factors were not significant (all P > 0.05). 

Table 3.

 

Linear Regression Models Determining the Factors Associated With the 2-Year Change in pRNFL Thickness for All Subjects and for Patients With CKD

 

The rate of change in each sector is illustrated in Figure 3. The pRNFL loss was more prominent in the temporal sectors in the control group but was more uniform across sectors in the CKD group. Patients with CKD stage 4/5 or ESRD had more prominent pRNFL changes over the superior and inferior sectors, respectively. The details of longitudinal changes in individual sectors can be found in Supplementary Table S3. 

Figure 3.

 

Longitudinal changes in pRNFL. The average and sectoral longitudinal changes in the thickness of the pRNFL are expressed as micrometers per year (A) and percentage per year (B).

Table 4 illustrates the change in logMAR BCVA at 2 years and its partial correlation with the change in pRNFL thickness over 2 years after controlling for age, sex, and axial length. The partial correlation was significant in the CKD group (P = 0.004) but not in the control group (P = 0.877). In the subgroup analysis, the partial correlation was significant in the ESRD subgroup (P = 0.003) but not in the other subgroups. Supplementary Figure S1 shows the scatterplots for the correlation between the change in pRNFL thickness and the change in logMAR BCVA over 2 years in the CKD group and the ESRD subgroup. 

Table 4.

 

Partial Correlation Between the Change in logMAR BCVA and the Change in pRNFL Thickness Over 2 Years

 

Discussion

To the best of our knowledge, this is the first prospective longitudinal study evaluating the rate of change of pRNFL thickness in patients with different stages of CKD. We observed a significantly faster rate of pRNFL loss in the CKD group than in the control group. In subgroup analysis, there was a trend toward an increased rate of pRNFL loss in patients with more severe CKD. The presence of hypertension, more severe CKD, and a larger rim area were associated with faster pRNFL loss among patients with CKD. 

The thickness of the pRNFL may decrease gradually with normal aging.^22–28 The annual change was estimated to be −0.14 to −0.52 µm/y.^22–28 The rate of pRNFL change in our control group (−0.26 µm/y) is consistent with previous data for healthy aging. Previous studies have shown that the risk factors of non-glaucomatous pRNFL thinning may include older age, male, smoking, increased body mass index, DM, hypertension, increased IOP, myopic refractive error, longer axial length, and visual impairment.^27–34 

Signs of retinal neurodegeneration, such as decreased thicknesses of the pRNFL, ganglion cell complex, and central macula, have been demonstrated in patients with CKD.^6–11 The exact mechanism of neurodegeneration in patients with CKD is yet to be determined. In the present study, IOP was not associated with the change of pRNFL. Other ocular factors were balanced between the control and CKD groups; therefore, the accelerated pRNFL loss in the CKD group is likely attributable to systemic diseases. The possible mechanisms can be divided into two categories.³⁵ First, the neurodegeneration could result from the underlying systemic comorbidities, such as DM or hypertension.^27,35,36 DM can cause neurodegeneration in the eye.³⁶ Considering that DM is very common among patients with CKD and excluding diabetic patients might limit the clinical application of our results, we did not exclude diabetic patients in our study design. Instead, we evaluated the effect of DM in linear regression models. However, in our regression model, DM was not significantly associated with 2-year pRNFL loss. This could be because patients with diabetic retinopathy were excluded at baseline, which would reduce the influence of DM. 

The thinning of pRNFL may also relate to microvascular pathologies related to hypertension, such as atherosclerosis, increased resistance, rigidity, insufficient autoregulation, and ischemia.²⁷ In a longitudinal study, Lee et al.²⁷ reported that the rate of pRNFL reduction in patients with hypertension of ≥5 years’ duration was around 2.5 times higher than that of age-matched controls (−0.99 vs. −0.40 µm/y; P < 0.001). Our linear regression models also showed that hypertension is an independent factor associated with the change in pRNFL over 2 years in patients with CKD. 

The second category of mechanisms of neurodegeneration in CKD patients may involve the pathologic pathways related to CKD itself. Patients with CKD have higher visit-to-visit or 24-hour variability in BP,^37,38 which could increase arterial stiffness and maladaptive remodeling.³⁹ Intradialytic hypotension and hypertension are also common complications in ESRD patients,⁴⁰ which may also cause endothelial dysfunction and increase arterial stiffness.⁴¹ High BP variability in patients with CKD has been shown to be associated with decreased vessel density and increased flow void area over the superficial vascular plexus of the retina.⁴² These hemodynamic instabilities may compromise the perfusion of retinal neural tissue and accelerate neurodegeneration. Indeed, a U-shaped effect of BP on OCT metrics and retinal perfusion was demonstrated in a recent study that showed that low and high BP may both cause thinning of the ganglion cell–inner plexiform layer.⁴³ 

In addition, increased inflammation and oxidative stress, dysregulation of the renin–angiotensin system, and the presence of uremic toxins are common in CKD.^35,44 These may cause increased neuroinflammation, increased levels of free radicals, vascular dysfunction, or neuron and astrocyte death.^45–47 In our regression model, the severity of CKD was an independent factor associated with the 2-year pRNFL loss. The pRNFL loss in advanced CKD (3.6–5.1 times higher than that in healthy aging) seems much faster than the rate previously reported in hypertensive patients (i.e., 2.5 times higher than that seen during healthy aging),²⁷ despite the fact that data values are not directly comparable among studies. This may suggest that CKD-related mechanisms could contribute to the accelerated pRNFL loss. 

The regression models indicated that a larger neuroretinal rim area was also a significant factor associated with a faster rate of pRNFL loss. This is reasonable because a larger neuroretinal rim area may be associated with a shorter axial length²³ and a thicker pRNFL,²⁸ both of which were factors associated with faster pRNFL loss.^26–29 

Age, spherical equivalent, and axial length have been commonly found to be associated with pRNFL in large population studies.^30–34 However, they were not significantly associated with the 2-year pRNFL change in this study. This could be reasonable because patients with young age, high myopia, and long axial length had already been excluded at the baseline; therefore, the influence of these variables was weakened. 

Some eyes in the CKD group had unchanged pRNFL when we examined the pRNFL changes within each individual year. This may be because the value of the annual pRNFL change was small and could be masked by visit-to-visit variability, diurnal fluctuations, or transient papilledema related to acute blood pressure elevation.⁴⁸ Longer observation durations could provide more reliable trends in pRNFL change; therefore, we analyzed the trend of change with 2-year longitudinal data in mixed models. 

Patients with CKD may be at higher risks of visual decline.^3,6 The possible etiologies include cataract, glaucoma, age-related macular degeneration, diabetic retinopathy, and other retinopathies.³ Our results suggest that retinal neurodegeneration might also contribute to the visual decline in patients with CKD. The changes in BCVA were partially correlated with the changes in pRNFL thickness in the CKD group. Furthermore, there was a trend that ESRD patients, who had the fastest pRNFL loss, also had more prominent changes in BCVA than non-ESRD patients had. Further investigations would help confirm this structural–functional correlation. 

Our results showed that longitudinal changes in pRNFL were more uniform across sectors in patients with CKD than in healthy controls; however, patients with CKD stage 4/5 or ESRD had more prominent pRNFL changes over the superior and inferior sectors, respectively. The reason behind the faster pRNFL loss over the superior and inferior quadrants in advanced CKD has yet to be determined. The thicker pRNFL over the superior and inferior quadrants could play a role. Thicker pRNFL itself has been associated with faster longitudinal changes.²⁶ This may be supported by the fact that the pRNFL changes in advanced CKD become more evenly distributed when expressed as percent of change per year. Another possible reason could be that the occurrence of accelerated atherosclerosis in advanced CKD could lead to more prominent loss of non-neural components (i.e., decreased diameter of retinal major vessels) in these sectors.⁴⁴ 

Some population studies have reported that CKD or low eGFR levels are independent risk factors for glaucoma^49–53; however, other studies have suggested that the association may be the result of common risk factors.^51,52 Although the association between CKD and glaucoma is still controversial,⁵⁴ our study suggests that the CKD-related pRNFL loss should be taken into consideration when diagnosing glaucoma and monitoring glaucoma progression in patients with CKD. The overall rate of pRNFL loss in patients with CKD is similar to that in patients with glaucoma.⁵⁵ The sectoral changes in patients with advanced CKD were most prominent in the superior and inferior sectors, which might also mimic glaucomatous changes.⁵⁶ Therefore, ophthalmologists should interpret the changes in pRNFL cautiously in patients with glaucoma or suspected glaucoma who have CKD. 

Our study had several limitations. First, the sample size was relatively small with a short follow-up. The rate of lost to follow-up (22%) was relatively high because of the outbreak of the COVID-19 pandemic during the study period. Our results of pRNFL loss should be verified in further population-based studies that include subjects with different health statuses, such as healthy people, diabetic patients, and hypertension patients. Second, the neural parameters over macula, such as ganglion cell layer and inner plexiform layer, are also important for understanding retinal neurodegeneration. However, they were not analyzed simultaneously in this study because a different type of OCT scan would be required. It is challenging to obtain consistent and comparable change measures among the different image qualities generated by different types of OCT scans throughout the longitudinal follow-up. Third, visual field examination results were not available. A previous study found a higher risk of parafoveal scotoma in patients with pRNFL defects associated with systemic diseases.¹² The correlation between the structural and the functional effects of neurodegeneration in CKD should be investigated further. Fourth, we did not collect data on biomarkers of chronic inflammation or oxidative stress or the level of uremic toxins, which makes it difficult to identify the exact mechanism of accelerated pRNFL loss in CKD. However, our study is strengthened by its prospective longitudinal design. The longitudinal design offers more reliable data on chronological changes in pRNFL than cross-sectional studies offer. The quality control of the included images also helped determine real pRNFL changes. The linear mixed model showed no significant changes in the SSI of OCT images at baseline, 1-year follow-up, or 2-year follow-up in either the control group (63.6 ± 9.2, 64.1 ± 9.4, 62.7 ± 8.5, respectively; P = 0.512) or the CKD group (63.6 ± 8.7, 61.7 ± 9.5, 62.5 ± 9.2, respectively; P = 0.161). 

In conclusion, our study demonstrated that patients with CKD had faster pRNFL loss than what occurs during healthy aging. The average rate of pRNFL loss was −0.87 µm (−0.90%) per year in patients with CKD. There was a trend toward an increased rate of pRNFL loss with more severe CKD; the annual rate of loss of pRNFL was −0.98 µm (−1.02%) in patients with CKD stage 4/5 and −1.38 µm (−1.46%) in those with ESRD. Hypertension, severity of CKD, and rim area were independent factors associated with the degree of pRNFL loss over 2 years in patients with CKD. Ophthalmologists should consider the rate of pRNFL loss due to CKD when diagnosing glaucoma or monitoring glaucoma in these patients. 

Acknowledgments

The authors thank the Maintenance Project of the Center of Data Science and Biostatistics at Keelung Chang Gung Memorial Hospital (Grant Nos. CGRPG2F0011, CLRPG2C0021, CLRPG2C0022, CLRPG2C0023, CLRPG2C0024, CLRPG2G0081, CLRPG2G0082, CLRPG2G0083, CLRPG2L0021, and CLRPG2L0022) for the support in statistical analysis. 

Supported by a grant from the Ministry of Science and Technology of Taiwan (109-2314-B-182A-025). 

Disclosure: L. Yeung, None; I.-W. Wu, None; C.-F. Liu, None; Y.-T. Lin, None; C.-C. Lee, None; C.-C. Sun, None