Total retinal thickness is an important factor in evaluating diabetic retinal neurodegeneration

Abstract

Objective

Macular retinal nerve fibre layer (mRNFL) and ganglion cell-inner plexiform layer thickness (GC-IPL) measurements are important markers of diabetic retinal neurodegeneration (DRN). In this cross-sectional study, we aimed to quantify the contribution of total retinal thickness (TRT) and other factors in the variation of mRNFL and GC-IPL thickness among participants with diabetes.

Methods and analysis

We used macular-centred spectral domain-optical coherence tomography scans from participants with diabetes in the UK Biobank. Two multiple linear regression models (prior to and after adjusting for TRT) were used to determine factors associated with mRNFL and GC-IPL thicknesses. A p value of less than 0.05 was considered statistically significant.

Results

A total of 3832 eyes from 3832 participants with diabetes were analysed. Factors that explained the greatest variation in thickness were TRT (20.9% for mRNFL and 57.2% for GC-IPL), followed by spherical equivalent (8.0% for mRNFL only), gender (2.2% for mRNFL only) and age (1.4% for GC-IPL only). Other factors significantly associated with mRNFL and/or GC-IPL thickness explained less than 1% of the variation in their thicknesses. Self-reported ancestral background was not significantly associated with mRNFL thickness after accounting for TRT.

Conclusions

Although many factors were significantly associated with mRNFL and GC-IPL thickness in participants with diabetes, they accounted for a fraction of the variation in the thickness of both layers. TRT explained most of the variation in these measurements, hence accounting for TRT is needed when using these metrics to evaluate DRN.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Macular retinal nerve fibre layer (mRNFL) and ganglion cell-inner plexiform layer (GC-IPL) thicknesses are markers of diabetic retinal neurodegeneration. Demographic, systemic and ocular factors are significantly associated with mRNFL and GC-IPL thicknesses in participants with diabetes.

WHAT THIS STUDY ADDS

• While several factors are significantly associated with mRNFL and GC-IPL thickness, total retinal thickness accounts for the majority of the variation in layer thickness.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Adjustment for total retinal thickness may allow more accurate comparison of mRNFL and GC-IPL thickness across individuals. Other factors significantly associated with layer thickness contribute relatively little to the variation in thickness across individuals.

Introduction

Diabetic retinopathy (DR) is a leading cause of preventable vision loss in the working-age population.¹ An estimated 463 million people globally are living with diabetes mellitus (DM) and is expected to reach 700 million by 2045.² Between 1990 and 2020, the prevalence of blindness due to DR increased by 14.9%.³ Thus, there is an increasing need to develop measures to identify and potentially intervene in the early stages of DR.

While DR has traditionally been considered a microvascular complication, evidence from both human and animal studies support that retinal neurodegenerative changes may precede vascular changes in DR.⁴ These retinal neurodegenerative changes affect the thickness of the neural retina (ie, ganglion cell-inner plexiform layer (GC-IPL) and retinal nerve fibre layer (RNFL). Spectral-domain optical coherence tomography (SD-OCT) is a well-established method for evaluating quantitative structural changes in the retinal layers.⁵ Prior meta-analysis and cross-sectional studies have shown that RNFL and GC-IPL thickness measurements are thinner in participants with DM compared with those without DM,^{6 7} and that this thinning occurs before characteristic DR changes (ie, microaneurysms, dot-blot haemorrhages, cotton wool spots, and exudates) become visible on standard fundus photographs.8,11 Patients with diabetes also demonstrate progressive thinning of the neural retina at a faster rate than that expected from normal ageing and diabetes is a critical factor that accelerates age-related thinning of the GC-IPL.^{12 13}

However, the routine evaluation of diabetic retinal neurodegeneration (DRN) has not been widely adopted in clinical practice. One of the issues is that there is wide variation in the thickness of the inner retinal layers, making it difficult to use these thicknesses to determine the degree of DRN.^{14 15} Factors associated with RNFL and GC-IPL thickness include age, race, ethnicity, duration of diabetes, haemoglobin A1c, systolic blood pressure, body mass index, high-density lipoprotein (HDL) levels, glaucoma, spherical equivalent, and visual acuity.^{7 13 16 17} Our prior work has shown that many factors affect macular RNFL (mRNFL) and GC-IPL thickness, but mRNFL and GC-IPL thickness as a proportion of total retinal thickness (TRT) was consistent across age, gender, and ancestral backgrounds in healthy individuals.¹⁸ In this study, we aim to quantify the contribution of TRT and other factors in the variation of mRNFL and GC-IPL thickness among participants with diabetes in the UK Biobank.

Materials and methods

Description of the sample

The sample included adult participants with diabetes from the UK Biobank who underwent ophthalmic assessment in 2009. The UK Biobank is a population-based cohort of more than 500 000 participants. Detailed study protocols can be found online (http://www.ukbiobank.ac.uk/resources/ and http://biobank.ctsu.ox.ac.uk/crystal/docs.cgi). In brief, between 2006 and 2010, residents of the United Kingdom, ranging in age from 40 to 69 years, who were registered with the National Health Service participated in the study by answering a questionnaire related to demographics, medical history, medication use and lifestyle. They also participated in a verbal interview conducted by a nurse, a physical exam, a blood draw and additional measures of their health. In 2009, an ophthalmic assessment was conducted in selected assessment centres, which included measurement of visual acuity, intraocular pressure, refractive error, fundus photographs and SD-OCT imaging.

Assessment of participant characteristics

DM status was determined based on self-reported DM diagnosis, use of DM medications and the presence of DM complications, using a previously validated procedure.¹⁹ HbA1c ≥6.5% was also classified as DM as per the American Diabetes Association criteria for diagnosing diabetes.²⁰ We defined cognitive impairment as a failing score on 2 or more standardised cognitive tests as described in a prior publication.²¹

Assessment of ocular characteristics

Macular-centred SD-OCT volume scans acquired using the Topcon 3D OCT-1000 Mark II (Topcon GB) were used from the UK Biobank. Detailed protocols for measuring retinal layer thickness and the inclusion/exclusion criteria of the images have been described in previous publications.^{7 14 18} Briefly, the Iowa Reference Algorithm software V.3.8 was used. To minimise selection bias, automated measures of image quality were initially used¹: undefined region and² surface cost. Undefined region quantified the percentage area of a scan that had missing data or insufficient signal. Images with undefined region >0% were excluded from analysis. Surface cost was calculated using edge-based costs of the dark to bright and bright to dark transitions of the retinal sublayers. This is an inverted Gaussian-smoothed gradient magnitude of the OCT voxel intensities of the retinal sublayers. A lower value corresponds to more reliable segmentation.²² We randomly evaluated 100 OCT scans with a range of surface cost values and determined that scans with surface cost values >62 000 had unreliable segmentation. Subsequently, all scans with surface cost values >62 000 were excluded. For the remaining OCT images, any scans with thickness measurements of 2 or more SDs above or below the mean were manually reviewed. Scans with incorrect segmentation, poor signal or presence of pathology such as epiretinal membrane (ERM), macular oedema, vitreomacular traction (VMT), drusen, intraretinal or subretinal fluid, pigment epithelial detachment or decentration that could impact thickness measurements were excluded

If a participant had adequate SD-OCT measurements in both eyes, only the right eye measurement was used. Two independent graders determined the grade of DR on fundus photos based on the International Clinical Diabetic Retinopathy Scale as no/mild DR (no changes or microaneurysms only) or more than mild DR.²³ If the graders did not agree, the higher grade was assumed. We defined glaucoma cases based on having any of the following: (1) self-reported diagnosis of glaucoma, (2) self-reported glaucoma medication use or (3) self-reported history of glaucoma surgery.

Online supplemental figure 3 shows a representative scan from our study with segmentation of the retinal layers using the Iowa reference algorithm.²⁴ The retinal layer between the internal limiting membrane (ILM) surface and the boundary between the retinal nerve fibre (RNFL) and ganglion cell layer (GCL) is the macular retinal nerve fibre layer (mRNFL). The layer between the RNFL/GCL boundary and the IPL/inner nuclear layer boundary is the GC-IPL. The sum of all the layers from the ILM surface to the outer boundary of the retinal pigment epithelium was determined to be TRT.

Limitations of UK Biobank include low response rate (5.5%), and the participating population belonging to a healthier and more affluent sector compared with the general population. Additionally, eye measurements were performed in a few selected assessment centres and participants with eye measurements tended to have a higher education level compared with the general population. The data on self-reported eye diseases can be subjected to recall bias and misclassification error.²⁵

Exclusion criteria

We excluded participants with gestational diabetes, diabetes identified by ICD code only without any other indication of diabetes as defined above, pre-diabetes, other self-reported serious eye condition, optic nerve disease, chorioretinal disease, blindness or low vision, sickle cell disease, retinal surgery or retinal detachment. Also, eyes with poor-quality SD-OCT scans as described above, IOP <5 or >21, injury/trauma resulting in loss of vision, or macular degeneration were excluded. If on manual review OCT images were found to be incorrectly segmented, have intraretinal cysts present inside or outside the foveal centre, have subretinal fluid, VMT, ERM or any other pathology present, or have loss of foveal contour, they were excluded.¹⁸

Reasons and numbers of exclusions are mentioned in online supplemental figure 1.

Statistical analysis

Patient and eye characteristics are summarised by mean with SD, median with minimum and maximum value or frequency with percentage. Histograms were used to assess normality of variables. Unadjusted linear regressions assessed the association between these characteristics and retinal thickness. Significant factors were included in a multiple linear regression, and if the correlation between predictor variables was >|0.5|, only one was included in the model to avoid multicollinearity. Stepwise backward selection followed by forward re-entering of variables was used to determine which variables were consistently associated with retinal thickness. After the covariates were selected, we additionally adjusted for TRT.¹⁸ The per cent variation accounted for was reported as the increase in R² as a per cent. A p value of less than 0.05 was considered statistically significant. Analysis was done using Stata V.18 (Stata V.2023. Stata Statistical Software: Release V.18. College Station, Texas: StataCorp LLC).

Results

A total of 3832 participants with DM (3832 eyes) were included (table 1). The median age was 62.3 years (range 40.2–75.9 years), over one-third were women (37%, n=1406), and the majority (82%, n=3105/3788) were white. Most eyes (96%, n=2683/2796) had no or mild DR. The mean thickness of the mRNFL and GC-IPL was 29.2 (SD 3.4) μm and 68.5 (SD 6.4) μm, respectively (online supplemental figure 2). The mean central subfield thickness was 286.8 (SD 24.7) μm and TRT was 305.7 (SD 14.6) μm.

Table 1. Demographic and clinical characteristics of the study population (N=3832).

Characteristic	N	Median (min, max) or frequency (percentage)
Age (years)	3832	62.3 (40.2,75.9)
Women	3832	1406 (36.7)
Self-reported ancestral background white	3788	3105 (82.0)
Education score (England)	3748	11.9 (0.0,93.0)
Duration of diabetes (years)	3258	6 (0.0,68.0)
HbA1c (%)	3391	6.7 (4.1,15.7)
No or mild diabetic retinopathy	2796	2683 (96)
Presence of diabetic neuropathy	3832	70 (1.8)
BMI (kg/m2)	3804	30.1 (17.7,60.4)
Self-reported or chronic ischaemic heart disease	3832	745 (19.4)
HDL (mmol/L)	3296	1.2 (0.5,3.8)
Presence of hypertension	3832	2710 (70.7)
Systolic blood pressure (mm Hg)	3821	139.5 (84.5,202.5)
Cognitive impairment	3708	1081 (29.2)
Neurological disease	3832	432 (11.3)
Use of medication for hypertension	3832	2406 (62.8)
Use of medication for lowering cholesterol	3832	2608 (68.1)
Use of biguanides	3832	2035 (53.1)
Use of insulin	3832	571 (14.9)
Use of sulfonylureas	3832	736 (19.2)
Use of meglitinides	3832	16 (0.4)
Use of glitazones	3832	237 (6.2)
Use of SSRIs	3832	192 (5.0)
Cataract surgery	3832	137 (3.6)
Presence of glaucoma	3832	126 (3.3)
IOP, Goldmann-correlated (mm Hg)	3716	15.8 (6.1,21.0)
LogMAR visual acuity	3822	0 (-0.4,1.4)
Spherical equivalent (diopters)	3778	0.2 (-12.9,10.1)
Retinal thickness measurements (microns)	N	Mean (SD)
mRNFL	3832	29.2 (3.4)
GC-IPL	3832	68.5 (6.4)
Central subfield thickness	3832	286.8 (24.7)
Total retinal thickness	3832	305.7 (14.6)

BMI, body mass index; GC-IPL, ganglion cell-inner plexiform layerHbA1c, haemoglobin A1c; HDL, high-density lipoprotein; IOP, intraocular pressure; logMAR, logarithm of the minimum angle of resolution; mRNFL, macular retinal nerve fibre layer; SSRI, selective serotonin reuptake inhibitor

The independent linear regressions showed that most of the variation in mRNFL thickness was explained by TRT 18.1%, followed by spherical equivalent (3.7%). Age, gender, self-reported ancestral background, deprivation index, glycaemic control, cognitive impairment, hypertension, visual acuity, DR, cataract surgery, glaucoma, blood pressure medication, and use of biguanide, insulin and selective serotonin receptor inhibitors explained less than 1% of the variation in mRNFL thickness (complete list of potential factors in online supplemental table 1). Multivariable models comparing factors associated with mRNFL thickness in participants with diabetes with and without adjusting for TRT are shown in table 2. After adjustment for TRT, factors significantly associated with a thinner mRNFL were male gender, presence of glaucoma and an increasing spherical equivalent. Older age, having moderate to severe DR and increasing haemoglobin A1c were associated with a thicker mRNFL after adjustment for TRT. All the factors in this model explained less than 1% of the variation in mRNFL thickness except for TRT, which explained 20.9%, followed by spherical equivalent (8.0%) and gender (2.2%).

Table 2. Final multivariable models for mRNFL thickness.

Variable	mRNFL not adjusted for TRT			mRNFL adjusted for TRT
	Coefficient (95% CI)	P value	% variation explained	Coefficient (95% CI)	P value	% variation explained
Age (10-year increase)	−0.23 (−0.41 to−0.04)	0.017	0.23	0.24 (0.07 to0.41)	0.005	0.33
Gender (male)	−0.54 (−0.80 to −0.27)	<0.001	0.63	−0.91 (−1.15 to −0.67)	<0.001	2.24
Ancestral background other than white	−0.97 (−1.33 to−0609)	<0.001	1.10	−0.22 (−0.55 to0.11)	0.197	0.07
Glaucoma	−1.45 (−2.20 to −0.70)	<0.001	0.59	−0.86 (−1.52 to −0.19)	0.012	0.26
Cataract surgery	1.07 (0.23 to1.91)	0.013	0.25	0.66 (−0.09 to1.41)	0.084	0.12
More than mild DR	0.90 (0.26 to 1.55)	0.006	0.31	0.90 (0.33 to 1.48)	0.002	0.39
HbA1c (%)	0.14 (0.03 to 0.24)	0.009	0.28	0.14 (0.05 to 0.23)	0.002	0.39
Spherical equivalent	−0.25 (−0.30 to −0.19)	<0.001	3.06	−0.37 (−0.42 to −0.32)	<0.001	8.02
Total retinal thickness (µm)	–			0.11 (0.10 to 0.12)	<0.001	20.90

Variables for which the beta coefficient changed from negative to positive or statistical significance changed when TRT was added to the model are bold and italicizeditalicised.

DR, diabetic retinopathy; HbA1c, haemoglobin A1cmRNFLmacular retinal nerve fibre layerTRT, total retinal thickness

The independent linear regressions showed that, most of the variation in GC-IPL thickness was explained by TRT (58.2%), followed by age (3.9%), spherical equivalent (1.6%), glaucoma (1.2%) and blood pressure medication (1.2%). Self-reported ancestral background, education score, duration of diabetes, glycaemic control, high-density lipoprotein levels, systolic blood pressure, neurological disease, hypertension, diabetic neuropathy, ischaemic heart disease, intraocular pressure, logMAR visual acuity, cataract surgery, cholesterol lowering medications and use of selective serotonin receptor inhibitors explained less than 1% of the variation in GC-IPL thickness (complete list of potential factors in online supplemental table 2). Multivariable models comparing factors associated with GC-IPL thickness in participants with diabetes with and without adjusting for TRT are shown in table 3. After adjustment for TRT, ancestral background other than white and increasing spherical equivalent were associated with a thicker GC-IPL. Factors significantly associated with a thinner GC-IPL were older age, hypertension, glaucoma and increasing duration of diabetes. All factors in this multivariable model explained less than 1% of the variation except for TRT, which explained half (57.2%) and age (1.4%) of the variation in GC-IPL thickness.

Table 3. Final multivariable models for GC-IPL thickness.

Variable	GC-IPL not adjusted for TRT			GC-IPL adjusted for TRT
	Coefficient (95% CI)	P value	% variation explained	Coefficient (95% CI)	P value	% variation explained
Age (10-year increase)	−1.97 (−2.32 to −1.63)	<0.001	4.41	−0.73 (−0.96 to −0.50)	<0.001	1.40
Ancestral background other than white	−1.54 (−2.18 to−0.90)	<0.001	0.81	0.82 (0.40 to1.25)	<0.001	0.52
Hypertension	−0.92 (−1.45 to −0.39)	0.001	0.42	−0.37 (−0.72 to −0.02)	0.038	0.16
Glaucoma	−2.56 (−3.85 to −1.28)	<0.001	0.56	−1.13 (−1.98 to −0.29)	0.008	0.26
Diabetic neuropathy	−1.60 (−3.15 to−0.05)	0.044	0.15	0.15 (−1.87 to1.17)	0.777	0.00
Duration of diabetes	−0.03 (−0.06 to −0.01)	0.009	0.25	−0.03 (−0.04 to −0.01)	0.001	0.38
HDL	−0.87 (−1.55 to−0.01)	0.011	0.23	−0.19 (−0.63 to0.25)	0.402	0.03
Spherical equivalent	0.44 (0.34 to 0.53)	<0.001	2.75	0.07 (0.002 to 0.13)	0.044	0.15
Total retinal thickness (µm)	–			0.33 (0.32 to 0.34)	<0.001	57.17

Variables for which the beta coefficient changed from negative to positive or statistical significance changed are bold and italicizeditalicised.

GC-IPLganglion cell-inner plexiform layer thicknessHbA1c, haemoglobin A1c; HDLhigh-density lipoproteinIOP, intraocular pressurelogMAR, logarithm of the minimum angle of resolution; SSRI, selective serotonin reuptake inhibitor; TRT, total retinal thickness

Additionally, ocular and systemic factors associated with TRT per independent linear regressions have been shown in online supplemental table 3. Age, gender, ancesteral background and spherical equivalent explained more than 1% of the variation in TRT.

Figure 1 shows that in a model that did not adjust for TRT, predicted GC-IPL thickness was 1.5 microns lower and predicted mRNFL thickness was 1.0 micron lower among non-whites compared with whites. After adjusting for TRT, predicted GC-IPL thickness was 0.8 microns higher among non-whites compared with whites and predicted mRNFL was not significantly different between non-whites and whites.

Figure 1. Impact of adjusting for TRT on variation in RNFL and GC-IPL thickness among non-White compared with White participants. Comparisons significant at p<0.05 are marked with an asterisk.*GC-IPL models adjusted for age, HDL, glaucoma, duration of DM, diabetic neuropathy, hypertension and spherical equivalent. **mRNFL models adjusted for age, gender, cataract surgery, glaucoma, DR, A1c and spherical equivalent. DM, diabetes mellitus; DR, diabetic retinal; GC-IPL, ganglion cell-inner plexiform layer; HDL, high-density lipoprotein; mRNFL, macular retinal nerve fibre layer; TRT, total retinal thickness.

Discussion

We observed that TRT accounted for most of the variation in mRNFL and GC-IPL thickness among participants with diabetes in the UK Biobank. After accounting for TRT, spherical equivalent and gender explained more than 1% of the variation in mRNFL thickness, and age explained more than 1% of the variation in GC-IPL thickness. Although many factors have been reported to be significantly associated with these thicknesses, we found that they contribute to less than 1% in the variation in these measurements. As we move towards utilising these measurements to quantify DRN in clinical practice, it is important to account for those factors that contribute to most of the variation.

Our findings are consistent with our prior study in healthy participants (without diabetes), which showed that most factors reported to influence mRNFL and GC-IPL thicknesses may actually do so through their effect on TRT.¹⁸ Similar to our findings, most variables in the multivariable model adjusting for TRT explained less than 1% of the thickness variation in healthy individuals except for spherical equivalent (6.9% for mRNFL) and TRT (20.3% for mRNFL and 47.5% for GC-IPL). In healthy individuals, once TRT was accounted for, the association between ancestral background and mRNFL and GC-IPL thickness disappeared. In our cohort of patients with diabetes, GC-IPL thickness remained significantly associated with ancestral background after adjusting for TRT, however, it explained only 0.52% of the variation in thickness.

Among demographic factors, age affected GC-IPL and mRNFL in different ways. Increasing age was associated with a thinner GC-IPL, as reported previously.²⁶ However, unlike prior studies which have reported a decrease in absolute mRNFL thickness with ageing in healthy individuals,²⁷ increasing age was associated with a thicker mRNFL after accounting for TRT in our study. It is possible that among participants with diabetes, with ageing, the rate of decrease in TRT exceeds the rate of decrease in mRNFL thickness. Male gender was associated with a thinner mRNFL in participants with diabetes. Khawaja et al also showed that mRNFL was thinner in men regardless of diabetes status among UK Biobank participants.¹⁴ In our prior study in healthy participants, being man was associated with a thinner mRNFL in the multivariate model; however, thickness as a proportion of TRT was fairly consistent across age and gender groups.¹⁸ Ancestral background was not significantly associated with mRNFL once TRT was accounted for. It was associated with thicker GC-IPL, however it contributed to very small fraction of the variation in GC-IPL thickness (0.52%). This contrasts with prior studies that have emphasised racial and ethnic differences in retinal layer thickness.26,28 Among ocular factors, increasing spherical equivalent was significantly associated with a thinner mRNFL but a thicker GC-IPL. Khawaja et al reported a similar finding in subjects from the UK Biobank regardless of disease status (ref 14). The true effects of myopia on the thicknesses are hard to determine because we could not isolate the effects of magnification.²⁹

We found that more than mild DR was associated with a thicker mRNFL in participants with diabetes. DRN-related thinning has been reported to precede the development of DR in participants,^{12 13 30} and it is possible that once DR develops, mild macular oedema or increased capillary permeability may contribute to relative swelling of the mRNFL, even though scans with features of diabetic macular oedema (DMO) were excluded. Another factor is that mRNFL is a thin layer and may be affected by several factors such as VMT, ERM, or other vitreoretinal interface abnormalities. While we identified and excluded eyes with VMT and ERMs distorting foveal contour, it is possible that eyes with very mild abnormalities may have been included and may have affected the mRNFL measurements. Duration of diabetes was associated with a thinner GC-IPL in our study. This is in accordance with longitudinal studies showing a decline in GC-IPL thickness over time in participants with diabetes.¹³ Higher values of HbA1c were associated with a thicker mRNFL in our study. Studies evaluating the association between HbA1c levels and retinal layer thicknesses have shown mixed results.^{17 31 32} Sohn et al showed that HbA1c levels did not influence mRNFL and GC-IPL thickness in their cohort of participants with diabetes.¹² In our prior study from the UK Biobank, we reported that higher HbA1c levels were associated with a lower total retinal and GC-IPL thickness but a greater mRNFL thickness.⁷ Our findings may be explained by the possibility that participants with poor glycaemic control may have subclinical macular oedema contributing to increased mRNFL thickness. It is also possible that glycaemic control is independent of the pathology of DRN.¹² Further studies are needed to explore this association.

None of the medications, including the diabetes medications (insulin, biguanides, sulfonylureas, meglitinides and glitazones), showed any association with mRNFL and GC-IPL thicknesses. Recently newer classes of medications have been introduced that are extremely effective at glycaemic control. They include sodium–glucose-linked transporter (SGLT)−2 inhibitors, dipeptidyl peptidase−4 inhibitors, glucagon-like peptide (GLP)-1 inhibitors. Their effects on diabetic retinopathy and the neuroretina are less well established. A recent randomised control trial showed increased retinal thickness after 12 weeks of treatment with the SGLT-2 inhibitor dapagliflozin, versus a decreased thickness after 12 weeks of treatment with glibenclamide, a sulfonylurea drug.³³ The SUSTAIN trail showed that semaglutide, a GLP-1 analogue, was associated with an early worsening of DR.³⁴ As these medications become more commonly used to manage participants with diabetes, it will become important to understand their effects on DRN.

Strengths of our study include the large sample size, evaluation of a large number of demographic, systemic and ocular factors and strict criteria for defining DM, which included HbA1c lab values and ICD codes in addition to self-report. Limitations of our study include that participants were predominantly White, and additional studies are needed to assess the impact of TRT on mRNFL and GC-IPL thickness among people living with diabetes from other racial and ethnic groups. We did not have information on duration of use of medications or time elapsed since cataract surgery, and this could have potential effects on the association of these factors with retinal thickness measurements. In addition, we only had access to macular-centred OCT scans and thus do not know the effect of these factors on other regions of the retina, such as peripapillary area. However, macular scans are routinely used to assess DMO in patients with diabetes and may be more easily incorporated into routine clinical practice to evaluate DRN.

In summary, although many factors were significantly associated with mRNFL and GC-IPL thickness in participants with diabetes, they accounted for a fraction of the variation in the thickness of both layers. TRT explained most of the variation in these measurements. Our results support the need to account for TRT when utilising mRNFL and GC-IPL thicknesses and to consider using retinal layer as a proportion of TRT as opposed to the absolute value of layer thickness when comparing these measurements across subgroups. This is also important to keep in mind if these thickness measurements are used to quantify DRN. Further studies are needed to develop methods for analysing retinal layer thickness as a ratio or percentage of TRT when establishing normative and disease-specific databases.

Data availability statement

Data may be obtained from a third party and are not publicly available.