The relationship between ON–OFF function and OCT structural and angiographic parameters in early diabetic retinal disease

Abstract

Purpose

This study measured associations between ON and OFF functional indicators and structural optical coherence tomography (OCT) and OCT angiography (OCTA) markers in diabetic retinal disease.

Methods

Fifty‐four participants with type 1 or type 2 diabetes (mean age = 34.1 years; range 18–60) and 48 age‐matched controls (mean age = 35.4 years, range 18–59) underwent visual psychophysical testing, OCT and OCTA retinal imaging. Psychophysical tasks measuring (A) contrast increment and decrement sensitivity and (B) response times to increment and decrement targets were assessed as surrogate measures of ON and OFF retinal ganglion cell function.

Results

The group with diabetes had worse foveal contrast increment and decrement thresholds (p = 0.04) and were slower to search for increment and decrement targets relative to controls (p = 0.009). Individuals with diabetes had a less circular foveal avascular zone (FAZ) (p < 0.001) but did not differ from controls in foveal vessel density and FAZ area. Functional and structural outcome measures related to the peripheral retina were also comparable between those with and without diabetes. Functional responses to increments and decrements were not significantly correlated with FAZ circularity or vessel density in individuals with diabetes.

Conclusions

Diabetic retinal disease results in impaired performance on measures of inferred ON and OFF pathway function in addition to vascular deficits measurable with OCTA. Future longitudinal studies may determine the temporal relationship between these deficits, and whether they predict future diabetic retinopathy.

Key points.

• In this cross‐sectional study, participants with diabetes exhibited varying losses in functional ON–OFF performance and vascular OCTA deficits.

• Inferred ON pathway function is more likely to be associated with damage to retinal blood flow compared to inferred OFF pathway function.

• The findings from this study contribute to the accumulating evidence revealing visual function deficits in individuals with diabetes without manifest vascular retinopathic changes.

INTRODUCTION

Diabetic retinopathy (DR) is recognised as a neurovascular complication in individuals with diabetes mellitus (DM). ¹ Structural thinning of retinal neural layers and aberrant changes in retinal blood flow can be quantified prior to the onset of DR using optical coherence tomography (OCT) ² , ³ , ⁴ and OCT angiography (OCTA), ⁵ , ⁶ , ⁷ respectively. In parallel, diabetic‐related retinal neuron dysfunction is confirmed by reports of contrast sensitivity deficits, ⁸ perimetric defects ⁴ and impaired electrophysiological (ERG) responses. ⁹ Assessing visual function alongside retinal imaging is an approach that provides a more complete clinical picture of the diabetic retinal disease process. The challenge of associating function with structure, however, is that the strength of the relationship is reliant on the measurement capabilities and dynamic range of the existing measures which have their respective limitations.

Arrangements of retinal ganglion cell (RGC) receptive fields into ON‐ and OFF‐centre with antagonistic surrounds form the foundations of contrast perception, ¹⁰ , ¹¹ which is impacted before changes to visual acuity in individuals with DM. ⁸ , ¹² , ¹³ ON and OFF RGC are key retinal constituents of parallel ON and OFF visual channels responsible for perceptions of contrast increments and decrements, respectively. ¹⁴ , ¹⁵ Healthy individuals demonstrate an increment–decrement response asymmetry showing higher contrast sensitivity ¹⁶ as well as faster temporal responses to decrement (OFF) visual stimuli. ¹⁷ Following reports of morphological and physiological damage to ON RGC (and to a lesser extent in OFF RGC) in animal models of DM, ¹⁸ , ¹⁹ a study was undertaken to determine whether DM had a greater impact on the ON versus the OFF pathways. It was found that functional performance to both contrast polarities were affected in earlier stages of DR. ²⁰

The pathogenesis of diabetic retinal disease is complex, and although the mechanistic links between neuron damage and microvascular alterations remain controversial, ²¹ , ²² the fact that both pathophysiological features are involved in the disease's natural history suggest they are interrelated. Due to the anatomical proximity of OFF RGC dendrites to the intermediate capillary plexus (ICP), vascular alterations to the ICP have been proposed as a potential mechanism for functional and morphological changes to OFF RGC in glaucoma work. ²³ However, there is limited evidence to explain ON RGC vulnerability in diabetic animal models ¹⁸ , ¹⁹ , ²⁴ and whether ON RGC damage in diabetes is associated with alterations to retinal blood flow. The spatial organisation of ON and OFF RGC and their corresponding bipolar cell partners to the inner and outer halves of the inner plexiform layer raise the possibility for ON or OFF cells to be differentially affected by retinal blood flow, as ON synaptic connections are adjacent to the deeper vessels of the superficial vascular plexus (SVP). Hence for this study, the association between both ON and OFF functional indices and quantitative measures of retinal structural and angiographic markers was evaluated. In line with findings from diabetic animal models, ¹⁸ , ¹⁹ , ²⁴ it was hypothesised that the relationship between ON functional measures and OCTA blood flow metrics would be stronger than that of OFF functional measures.

METHODS

Participants

To evaluate whether ON–OFF function correlated with structural OCT and OCTA measures in the group with DM, a sample size estimation was obtained for testing a correlation coefficient using Fisher's z‐transformation. ²⁵ Assuming an alpha of 0.05, a power of 0.8 and an underlying correlation of 0.4, a sample size of 47 is required within the DM group.

Participants were the same individuals recruited in our previously published study. ²⁰ Fifty‐four individuals with DM and 48 age‐matched individuals without DM between 18 and 60 years of age participated. Inclusion criteria were Snellen VA of 6/9 or better and refractive errors between ±6.00 dioptres (D) sphere and <2.50 D cylinder. Individuals with type 1 and type 2 diabetes participated and diabetes status was confirmed by an endocrinologist (SF) or confirmed with a current medical history of DM. All participants, including controls, who did not report a glycated haemoglobin (HbA1c) value obtained within the previous 3‐months underwent a finger‐prick test using the Quo‐Lab point‐of‐care HbA1c analyser (EKF Diagnostics, ekfdiagnostics.com/quo‐lab.html) to estimate their HbA1c.

Individuals having co‐existing ocular diseases that could affect the retina or vision, diabetic macular oedema (DMO), a history of DR or DMO treatment, were pregnant or breastfeeding or experienced migraines were excluded. Individuals who were using any medication known to affect cognitive processing or vision were also excluded, along with those with a recent history of a cardiovascular event or on active treatment for cancer. The International Clinical Diabetic Retinopathy and Diabetic Macular Edema Disease Severity Scales was used to classify DR severity. ²⁶

This study was approved by The University of Melbourne Human Research Ethics Committee (identifier: 1955267). All participants gave written informed consent and all procedures complied with the Declaration of Helsinki.

Overview of the testing session

Participants attended for a single test session involving ophthalmological examination, visual psychophysics testing and a suite of retinal imaging, including OCT and OCTA with the Spectralis HRA + OCT (Heidelberg Engineering GmbH, business‐lounge.heidelbergengineering.com/us/en/products/spectralis/spectralis/). All functional and structural measurements were performed monocularly with the eye having the better VA chosen for testing. Where VA was comparable between both eyes, the eye with the lower refractive error was selected.

Structure–function data were collected by spatially mapping 3 × 3 mm OCTA scans at the fovea and three other retinal locations (15° superior, inferior and temporal to the fovea), as illustrated in Figure 1, to the visual field locations in the functional tasks. The choice to include both central and mid‐peripheral locations was owing to a predominance of microvascular abnormalities (areas of nonperfusion and retinal haemorrhages) reported in the retinal midperiphery; these changes are postulated to occur due to the capillary's lower capacity to transport oxygen. ²⁷ , ²⁸ The retinal midperiphery also has a sparser RGC distribution (and thinner ganglion cell layer–inner plexiform layer [GCL‐IPL] thickness). Thus, any early damage to RGC in this region could potentially reveal more pronounced structural or functional deficits.

FIGURE 1.

Four 3 × 3 mm optical coherence tomography angiography (OCTA) scans were separated by a visual angle of 15° from the centre of the fovea. The Gabor visual stimuli in Task A (contrast increment and decrement sensitivity task) were scaled and positioned to correspond with the peripheral OCTA scans presented in the figure. Suprathreshold visual stimuli in Task B were scaled to the central OCTA scan.

The analysis first focused on functional and structural correlations in the macula region before analysing functional and structural data for the peripheral retina. The analysis and results are organised in this manner because one of the functional tasks only measured contrast processing features foveally (Task B described in Figure 2, panel b) and additional OCTA metrics related to the foveal avascular zone (FAZ) were also analysed.

FIGURE 2.

Schematic illustrating two psychophysical assessments measuring contrast increment and decrement processing. (a) Task A measured contrast sensitivity to increment and decrement Gabors. Gabor presentations were interleaved at four visual field locations (separated by a visual angle of 15°); Gabors were presented with an onset and contrast ramped across 200 milliseconds (ms). A single time course for Task A is illustrated for the right eye. (b) Task B measured response times to increment and decrement suprathreshold targets embedded in a binary noise background; an example of stimuli sequence and correct responses is presented. A single participant's data was fitted to an exponential‐Gaussian function (equation 1) shown on the line and histogram plots. Visual stimuli in task A and the noise background in task B subtended a visual angle of 10° × 10° to spatially map to high resolution 3 × 3 mm en face optical coherence tomography angiography (OCTA) scans (see Figure 1 for the spatial locations of OCTA scans). Note that Task B was a foveal task.

Psychophysical assessments

Detailed methods of each of the psychophysical tasks have been published. ²⁰ Tasks were presented using PsychoPy ²⁹ with custom software written in Python interpreted language. Experiments were performed on a gamma‐corrected Display++ LCD monitor (Cambridge Research Systems, crsltd.com) (screen resolution: 1920 × 1080 pixels; dimensions: 39.5 cm × 71 cm, refresh rate: 120 Hz, mean luminance: 100 cd/m²). Experiments measured observers' (A) contrast thresholds at four visual field locations (Figure 2, panel a) and response times to increment and decrement visual stimuli in central vision (Figure 2, panel b). Task A measured contrast increment and decrement sensitivity in a detection task while task B was designed in a method of constant stimuli and measured response times to suprathreshold increment and decrement targets.

For task A, vertically oriented Gabor stimuli (visual angle: 10° × 10°) with a sinewave component modulated around positive or negative phases of the sinusoid were presented in a similar manner to perimetry. Stimuli presentations were interleaved between four locations: at the centre, 15° nasal, 15° superior and 15° inferior to the central point of the screen (Figure 2, panel a). Participants were tasked to fixate at a fixation point and respond to the Gabor stimulus. Contrast increment and decrement Gabors were defined as Weberian contrast and had a spatial frequency of 3 cycles per degree based on prior work on contrast sensitivity in central ³⁰ and peripheral vision. ³¹ Contrast thresholds were returned from a staircase procedure with a ‘one‐up, three‐down’ decision strategy; the staircase had four reversals and logarithmic unit step sizes of 0.2, 0.2, 0.1 and 0.1. Stimulus contrast at the last two reversal intensities of each staircase were geometrically averaged and the final contrast threshold estimates were the average of two runs per increment or decrement condition. Contrast thresholds were converted to contrast sensitivity in decibel (dB) units using the formula 10 × log10(1/threshold).

Task B was a foveal task where participants visually searched for one to three dark or light suprathreshold targets that were embedded in a noise background composed of black and white pixels of the same luminance. Observers were instructed to respond as rapidly and as accurately as possible using a numerical keypad (Figure 2, panel b). Successive presentations were determined based on the participant's response times. Response times were fitted to an exponential‐Gaussian (ex‐Gauss) function in R (r‐project.org, version 3.6.0) (Figure 2, panel b) to obtain outcome response times estimates for increment and decrement stimuli for each participant. ³² The ex‐Gauss function, described in Equation 1, is a convolution between an exponential (with a mean of τ) and Gaussian distributions (with a mean of μ and standard deviation of σ). ³²

$$f\left(t\right)=\frac{1}{\tau }\exp \left\{-\frac{\left(t-\mu \right)}{\tau }+\frac{{\sigma }^2}{{\tau }^2}\right\}a*\oint \left\{\frac{t-\mu }{\sigma }-\frac{\sigma }{\tau }\right\}$$ (1)

The ex‐Gauss function well‐represents behavioural tasks that employ response time data by modelling response times as a combination of both decision (exponential) and human performance (Gaussian) components. ³² , ³³ The three parameters (μ, τ and σ) defining the ex‐Gauss function were estimated using the optim function in R, which performs a minimisation of the residual sum of squares. The sum of μ and τ was calculated as the primary outcome of the response time task as per previous authors, ¹⁷ , ³⁴ , ³⁵ but μ and τ were also compared separately. Previous studies employing experiments similar to task B have also used the ex‐Gaussian function to fit distributions of response times for increments and decrements. ¹⁷ , ³⁴ , ³⁵ In practice, there is negligible difference between analysing the raw average response and the sum of μ and τ components, but these values were not mathematically identical (see Figures S1–S4 for further clarity).

Structural assessments

OCT and OCTA imaging were performed after pharmacologic pupil dilation with tropicamide 1%. Participants were examined using the Spectralis OCT's Glaucoma Module Premium Edition which comprises a high‐resolution circular scan centred on the optic nerve head (768 A‐scans, 3 circle B‐scans and 24 radial B‐scans) and a macula cube scan centred on the fovea (768 A‐scans, 61 B‐scan sections 120 μm apart) with a scan area of 30° (height) × 25° (width). For the OCTA scan pattern, 10° × 10° (2.9 × 2.9 mm) volumetric scan images centred at the fovea and at three peripheral retina locations described in Figure 1 were taken (isotropic lateral solution of 5.7 μm/pixel, 512 × 512 A‐scans). ³⁶

Segmentation of individual retinal neural layers and vasculature was automatically performed by the Spectralis' inbuilt image processing software and manually cross‐checked by author VTST. OCTA macula images were segmented according to its trilaminar vascular distribution ³⁷ : the SVP slab was defined as the inner 80% of the ganglion cell complex (GCC) (consisting of the RNFL + GCL + IPL), the ICP slab was defined between the outer 20% of the GCC to the inner 50% of the inner nuclear layer and the deep capillary plexus (DCP) described the outer 50% of the inner nuclear layer and the outer plexiform layer. ³⁷ For the peripheral OCTA en face scans, the superficial vascular complex (SVC) (consisting of the radial peripapillary capillary plexus and SVP) and the deep vascular complex (DVC) (consisting of the ICP and DCP) were segmented according to anatomical data. Prior studies have shown that the ICP and DCP merge at approximately 7 mm temporal to the fovea. ³⁷ The Spectralis' built‐in projection artefact removal software was used on the ICP, DCP and DVC.

All OCT and OCTA image quality were above 20‐dB signal‐to‐noise ratio (median image quality OCT: 34 dB, range, 27–40 dB; OCTA: 38 dB, range 30–44 dB). For further quality control, the 5‐point grading scale described by Hogg et al. ⁶ was employed where two independent, masked graders judged image quality based on the proportion of microvasculature that was not obscured by artefact. Where the graders disagreed on the image grading, a third grader (author RCAS) made the final decision for inclusion. A total of 13 OCTA macula scans from six eyes were excluded with 599 scans remaining. For the peripheral retina, following OCTA grading and image quality control, 536 slabs (SVC and DVC) remained (superior: 184, inferior: 169 and temporal: 183 images).

For image analysis at the fovea, the GCL‐IPL thickness was considered as well as vessel densities of the SVP, ICP and DCP. This differed for the three peripheral retinal locations where the retinal nerve fibre layer (RNFL) to the inner plexiform layer (IPL) was analysed for the OCT scans. To obtain vessel density estimates, OCTA images were imported into the Erlangen‐Angio‐Tool (EA‐Tool). ¹⁷ Vessel densities were then calculated as the percentage of ‘white area’ in the ‘total area’ within a region of interest (ROI). For the macula OCTA scan, this was an annular ROI with inner and outer ring diameters of 0.80 and 2.9 mm, respectively (Figure 3, panel a). ³⁸ The macula annulus was referenced to the anatomical fovea for each macula OCTA scan, which required manual selection; participants' en face and cross‐sectional OCT and OCTA scans were compared side‐by‐side to select the location of their anatomical fovea. Peripheral OCTA scans had a circular area of 2.9 mm in diameter (no inner annulus). Vessel density was returned by the EA‐Tool in four quadrants, which were summed to obtain a total vessel density estimate for the ROI.

FIGURE 3.

Image analysis performed by the Erlangen‐Angio‐Tool. (a) Vessel densities were taken within the annular region of interest (inner ring diameter 0.80 mm, outer ring diameter 2.8 mm) for the macula optical coherence tomography angiography (OCTA) en face scan based on Hosari et al. ³⁸ (b) To calculate the foveal avascular zone (FAZ) circularity and area, the perimeter of the FAZ was marked using a composite image of the three different retinal plexi present at the macula (depicted by different colours).

In order to spatially align the structural data to the area for OCTA vessel densities approximately, central GCL‐IPL thickness within the fovea were estimated using the standard Early Treatment for Diabetic Retinopathy Study (ETDRS) grid, ³⁹ and approximated the area for the peripheral RNFL‐IPL thickness based on the 9 mm² OCTA scans described in Figure 1. For the FAZ circularity and area, composite images of all maculae retinal plexi were generated per participant using the EA‐Tool (Figure 3, panel b). Composite images were then imported back into the EA‐Tool for manual delineation of the FAZ circularity and calculation of the FAZ area by two independent markers. FAZ circularity is a ratio defined as:

$$\frac{\mathrm{FAZ}\text{perimeter}}{\text{Perimeter of the circle with equal area}}$$

(a perfect circular FAZ equals 1). ⁴⁰ Combining SVP, ICP and DCP into a composite image best represents FAZ borders based on primate histology studies. ³⁷ , ⁴⁰ Intraclass coefficient between the two masked graders were 0.822 and 0.959 for FAZ circularity and area, respectively. Images from four participants showed capillaries traversing the foveal pit which rendered the task of determining FAZ area and circularity impossible; thus, these images were excluded.

Statistical analysis

Linear mixed model analysis (LMM) was fitted in R using the lmer function in the lme4 package. ⁴¹ LMM was estimated to predict fixed variables including contrast increment/decrement sensitivities (1) foveally and (2) at three peripheral locations as well as (3) response times to increments/decrements. Separate models were also fitted to predict foveal vessel densities and vessel densities for each peripheral retinal location (superior, inferior and temporal) as well as structural RNFL‐IPL thickness. All models had Group as a fixed effect and included Participant as a random effect. An example for one of the model specifications were as follows: (dB ~ group + ON_OFF + group: ON_OFF + (1|Participant)). T‐tests were performed to describe group differences and correlation analyses were performed with either Pearson or Spearman correlations, depending on whether the data met normality assumptions. p < 0.05 was considered statistically significant.

RESULTS

Results are presented for functional measurements first in central vision and then peripheral vision, as well as structure/microvasculature measurements centrally then peripherally because some outcome measures were only measured foveally (i.e., psychophysical task B and FAZ integrity).

Table 1 presents the clinical characteristics of groups with and without DM. Controls and participants with DM were of a similar age range with normal visual acuity. Participants with DM included type 1 and type 2 individuals; most DM participants had no to mild non‐proliferative DR.

TABLE 1.

Clinical characteristics.

	Individuals without DM n = 48	Individuals with DM n = 54	p‐Value
Age, y (range)	35.4 (18–59)	34.08 (18–60)	0.79
Visual acuity, logMAR (range)	0.00 (−0.20– +0.18)	0.00 (−0.20– +0.20)	0.45
HbA1c (range)	5.0 (4.4–5.8) a	7.5 (5.9–11.5)	<0.001
Individuals with DM
Diabetes type	Type 1	44 (81%)
	Type 2	10 (19%)
Duration of diabetes, y (range)		14.9 (0.5–43)
Diabetic retinopathy severity		No DR	n = 33 (61.1%)
		Mild NPDR	n = 13 (24.1%)
		Moderate NPDR	n = 8 (14.8%)

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; HbA1c, glycated haemoglobin; NPDR, non‐proliferative diabetic retinopathy.

^a n = 47 for this measure as HbA1c data was unavailable for 1 person.

Psychophysical tasks

It was first determined whether the groups differed in functional performance for the two tasks. At the central location for task A, the DM group had reduced contrast sensitivity (mean increments: 11.17 dB; mean decrements: 10.18 dB) relative to the controls (increments: 12.13 dB; decrements: 10.84 dB), supported by a significant main effect of group (beta = −0.77, 95% CI [−1.48, −0.05], t(194) = −2.11, p = 0.04) by the model. Although there was a significant main effect for contrast polarity (beta = 1.30, 95% CI [0.87, 1.72], t(194) = 6.03, p < 0.001), the group by contrast increment and decrement interaction was not statistically significant (beta = −0.36, 95% CI [−0.94, 0.23], t(194) = −1.19, p = 0.23), suggesting that there was no preferential decline in functional response to either increments or decrements in the DM group.

For task B, the response time outcome measure was log‐transformed so that the data met assumptions underpinning LMM analysis. For task B, there was a significant main effect of group (beta = 0.06, 95% CI [0.01, 0.1], t(184) = 2.62, p = 0.009) and contrast polarity (beta = 0.19, 95% CI [0.16, 0.21], t(184) = 15.36, p < 0.001) suggesting that participants with DM were slower to respond to both increment (median response times: 1.64 s vs. 1.42 s for controls) and decrement stimuli (1.06 s vs. 0.94 s for controls). The group by contrast polarity interaction was not significant (p = 0.70), demonstrating that increment and decrement response times were similarly impacted in the DM group.

To explore whether the functional decline in mean response times for the DM group was due to the Gaussian or the exponential component, the ex‐Gauss parameters, μ and τ, were analysed separately. It was found that the DM group was slower for both perceptual and decision components; there was a statistically significant effect of group for the μ (beta = 0.08, 95% CI [0.04, 0.12], t(198) = 3.17, p < 0.001) and for τ (beta = 0.08, 95% CI [0.0008, 0.15], t(198) = 2.20, p = 0.03).

Peripheral contrast sensitivities were measured in task A (task B was a foveal task) and these results are described in Table 2. Group, contrast polarity and location were used as fixed effects, and the interaction between contrast polarity and location was investigated, as well as group and location, including participants as a random effect. For peripheral locations, the effect of group was not significant (beta = −0.28, 95% CI [−0.93, 0.37], t(588) = −0.84, p = 0.40). However, there was an effect of location for the superior visual field location (beta = −2.03, 95% CI [−2.44, −1.61], t(588) = −9.64, p < 0.001), driven by reduced contrast sensitivities in the superior visual field relative to the inferior and nasal visual fields.

TABLE 2.

Mean and 95% confidence intervals for functional and structural outcome measures in the foveal location and in three other peripheral locations.

					Individuals without DM			Individuals with DM
					Mean	95% confidence intervals		Mean	95% confidence intervals
Function	Foveal location		Contrast sensitivity task (dB)	Increment	12.13	11.59	12.68	11.17	10.54	11.81
				Decrement	10.84	10.43	11.24	10.18	9.64	0.71
	Peripheral locations	Superior location	Contrast sensitivity task (dB)	Increment	5.57	5.03	6.12	5.24	4.59	5.89
				Decrement	5.41	4.87	5.94	5.49	4.95	6.03
		Inferior location		Increment	7.58	7.01	8.08	7.39	6.78	8.00
				Decrement	7.47	7.04	7.91	7.21	6.70	7.72
		Nasal location		Increment	7.30	6.83	7.78	6.66	6.05	7.26
				Decrement	7.12	6.68	7.57	6.52	6.05	6.99
Structure	Fovea		Vessel density (%)	Superficial vascular plexus	30.57	29.87	31.28	30.15	29.24	31.06
				Intermediate capillary plexus	21.84	21.21	22.48	21.29	20.5	22.09
				Deep capillary plexus	24.32	23.62	25.02	23.74	22.77	24.70
			GCL‐IPL thickness (μm)		93.63	91.85	95.41	94.38	92.22	96.53
			FAZ circularity (ratio)		0.71	0.66	0.77	0.58	0.54	0.63
			FAZ area (mm2)		0.29	0.25	0.33	0.27	0.23	0.30
	Peripheral locations	Inferior retina	Vessel density (%)	Superior vascular complex	26.67	25.63	27.71	26.48	25.3	27.65
				Deep vascular complex	19.71	18.83	20.6	19.09	18.12	20.06
			RNFL‐IPL thickness		92.85	90.18	95.52	94.15	91.56	96.75
		Superior retina	Vessel density (%)	Superior vascular complex	26.27	24.10	28.45	26.41	24.57	28.24
				Deep vascular complex	21.78	20.78	22.79	21.37	20.30	22.44
			RNFL‐IPL thickness		92.70	90.16	95.24	91.85	89.08	94.61
		Temporal retina	Vessel density (%)	Superior vascular complex	15.22	14.27	16.17	15.81	15.01	16.6
				Deep vascular complex	22.44	21.48	23.40	22.38	21.46	23.3
			RNFL‐IPL thickness		66.4	64.59	68.22	69.14	67.67	70.60

Abbreviations: dB, decibels; DM, diabetes mellitus; FAZ, foveal avascular zone; GCL‐IPL, ganglion cell layer–inner plexiform layer; RNFL‐IPL, retinal nerve fibre layer–inner plexiform layer.

Note: Bold font represents the mean.

Structural OCT and OCTA

Group differences in foveal structural OCT and OCTA measures were explored to compare with prior literature. Central GCL‐IPL thickness were comparable between individuals with DM (94.38 μm, 95% CI: 92.22–96.53) and without DM (93.63 μm, 95% CI: 91.85–95.41), t(86) = −0.49, p = 0.63.

For the model investigating group differences in mean macula vessel density, vessel density, group and retinal plexus were fixed effects while participant was included as a random effect (i.e., vessel density ~ Group + Retinal_Plexus + Group:Retinal_Plexus + (1|Participant)). Mean macula vessel densities were lower in the group with DM for SVP, ICP and DCP but not significantly different (beta = −0.80, 95% CI: [−1.87, 0.27], p = −0.14) (refer to Table 2 for vessel density data).

Differences in FAZ geometry were evaluated between controls and participants with DM. Significant group differences were observed in FAZ circularity, t(100) = 3.72, p < 0.001, but not for FAZ area, t(100) = 1.41, p = 0.23.

In the peripheral retina, there was no significant difference in the RNFL‐IPL thickness between DM and those without DM (beta = 1.23, 95% CI [−2.04, 4.49], t(166.97) = 0.74, p = 0.46). The only statistically significant effect was location, primarily driven by a thinner RNFL‐IPL thickness at the temporal retina (15° from the fovea) relative to the superior and inferior locations (main effect of location (temporal): beta = −26.44, 95% CI [−28.52, −24.37], t(196.45) = −25.19, p < 0.001). A thinner RNFL‐IPL layer temporally was expected considering the superior and inferior vascular arcades contribute to the overall retinal thickness at the superior and inferior retinal regions.

For vessel densities in the peripheral retina, fixed effects were vessel density, group, location and retinal plexi (SVC and DVC). In this model, group and location were also included as interaction terms and participant as a random effect. No statistically significant effect was observed for group with the model (beta = −0.82, 95% CI [−2.24, 0.61], t(527) = −1.13, p = 0.26). There was an effect of retinal plexus (beta = 1.77, 95% CI [1.07, 2.47], t(527) = 4.95, p < 0.001), suggesting vessel density differences between the SVC and DVC at the peripheral locations.

Function–structure correlations in those with diabetes

Correlations were performed for the group with DM and only on functional and structural data in the central location because some of these measures revealed a between group difference. Given that the peripheral structure–function data did not distinguish between those with versus without diabetes, no correlation analyses were performed on the peripheral data.

There were no statistically significant correlations for either contrast increment or decrement sensitivity measures with macular vessel density in any of the retinal plexi (Figure S1). There were also no significant correlations between contrast sensitivity with FAZ circularity or between response times and FAZ circularity (Figure S2). No significant correlations were found between functional outcomes and FAZ area (Figure S3). A trend towards a significant correlation was found between DM participants' slower response times to increments and reduced vessel densities at the SVP and DCP (Figure 4, panel a).

FIGURE 4.

(a) Increment and (b) decrement response times versus vessel density in each macular capillary plexus. Only data from participants with diabetes mellitus (DM) are shown. Filled triangles: Individuals with DM without diabetic retinopathy (DR). Unfilled triangles: Individuals with DM with DR. DCP, deep capillary plexus; ICP, intermediate capillary plexus; SVP, superficial vascular plexus.

DISCUSSION

Despite finding differences between the DM and control groups for some outcome measures, there were no significant correlations between visual function measures and either macular vessel density or FAZ circularity in this DM group. While the DM cohort had a more irregular FAZ relative to the control group, vessel densities were comparable with controls; these findings were mixed considering the prior literature. ⁵ , ⁶ , ⁷ , ⁴² , ⁴³ , ⁴⁴ The DM group also demonstrated decreased contrast increment and decrement sensitivity as well as slower response times to high contrast increment and decrement stimuli. Functional performance to either ON or OFF pathways did not predict OCTA metrics; instead, individuals with DM (with and without clinical signs of DR) had a range of ON–OFF functional deficits and microvasculature damage.

Given that the ability to expose structure–function relationships is contingent in part on the sensitivity of measurement techniques, this study focused on correlations foveally because some of the structural ² , ³ , ⁷ and functional ⁴⁵ , ⁴⁶ indices were impacted in individuals with DM without DR, whereas none of the peripheral outcome measures were different between groups. These results indicate that functional responses to contrast increments and decrements can differ between those with and without DM before microvascular change or structural GCL‐IPL thinning. Visual processing of contrast increments and decrements relates to standard contrast sensitivity which has been used as a functional index in a handful of similar studies. Srinivasan et al. ⁴⁷ found a weak positive correlation (rho = 0.27) between Pelli–Robson measured contrast sensitivity and FAZ circularity in their DM group without DR. Meshi et al. ⁴⁸ did not find a significant correlation between contrast sensitivity and their chosen OCTA measures, but reported worse contrast sensitivity in their DM cohort without DR. Other functional indices used to find microvasculature‐function associations include multifocal electroretinography (mfERG), ⁴⁹ , ⁵⁰ pattern electroretinography (ERG), ⁵¹ microperimetric retinal sensitivity ⁵² , ⁵³ and full‐field ERG. ⁵⁴ The consensus from these studies suggests a stronger positive association between visual dysfunction and OCTA microvascular parameters when individuals with more advanced DR are included. However, this investigation deliberately recruited individuals who exhibited earlier signs of DR in order to explore structure–function relationships that might show potential for prediction of future DR.

These data do not reveal any obvious structure–function relationships. However, there was a trend for a weak relationship between increased increment response times and decreased vessel densities that approached statistical significance, especially at the SVP and the DCP (Figure 4, panel a). The sample size was determined to provide adequate power for a clinically meaningful correlation strength of 0.4 in the absence of prior data to inform. The results suggest that these parameters may be related, but that the association is weaker in strength. Given the early stage of diabetic retinal disease in most of the tested participants, we cannot determine whether a stronger neuron‐microvasculature relationship would manifest for these measures in more advanced disease.

The results also revealed that in the cohort with DM, dysfunction occurred in some individuals who did not manifest losses in vessel density, FAZ circularity and FAZ area (refer to Figures S2–S4). Conversely, some individuals with DM demonstrated irregular FAZ circularity without any associated dysfunction. Structural and functional signs of retinal damage do not always manifest in all individuals with DM, and curiously, a few DM individuals had no measurable retinal damage 30 years after diagnosis. ⁹ , ⁵⁵ These observations have led some researchers to suggest a phenotypic heterogeneity in DR (see review by Cabrera et al. ⁵⁶ ). Another consideration is that the present study was cross‐sectional in nature, making it difficult to disambiguate whether some individuals with DM had already experienced changes in their functional responses and/or to their retinal microvasculature, despite their functional and microvasculature indices remaining within ‘normal’ ranges. Moreover, the impact of demographic factors such as diabetes type ⁷ , ⁵⁷ and age, ⁵⁸ on the strength of the relationship between loss of microvascular integrity and neuronal dysfunction is not entirely clear. Future longitudinal studies are needed to address the relationship between the time courses of functional and structural damage in diabetes.

In this work, stringent image quality control was applied to avoid misinterpretation of the OCTA results and the chosen OCTA parameters are more commonly used in the literature to enable comparisons with other studies. However, OCTA can only capture ‘erythrocyte blood flow’ above a minimum threshold as it relies on contrast between consecutive B‐scans. For example, it has been reported that some vascular features such as microaneurysms with a turbulent blood flow tend to be missed. ⁵⁹ It is also not yet established which OCTA parameters are useful as clinical biomarkers or have the highest diagnostic value for diabetic retinal disease; as such, there is a possibility that commonly used OCTA metrics might not adequately quantify blood flow deficits. ⁶⁰ As OCTA technology matures, more refined approaches for extracting blood flow information OCTA images are expected. ⁶¹ , ⁶²

This study used large 10° × 10° stimuli to test contrast sensitivity and compared the functional outcome to the underlying structural or vascular data from the same region. The functional data suggests non‐specific (both ON and OFF pathway) neuronal dysfunction or loss. While the primary aim was to relate functional performance to vascular estimates, OCT structural data was also investigated. Because of the large size of the psychophysical stimulus, compensation was not made for potential displacement of RGC bodies when making the mapping comparison between structure and function, which would be necessary if looking for focal loss using small stimuli such as those used in perimetry in the macula region. In the periphery, the visual stimuli were large (3 × 3 mm) and positioned 4.5 mm from the fovea, so while the RGC spatial offset described by Drasdo et al. ⁶³ was detectable at this eccentricity (peak displacement was 1.15 mm temporal from the fovea), the practical implication of a slight misalignment between structure and functional data at this eccentricity is likely to be negligible.

An alternate analytical approach that has been used to explore structure–function relationships (rather than vascular‐function relationships) both in early glaucoma ⁶⁴ and diabetic eye disease ² has been to interpret the relationship using a neural model whereby local GCL thickness is mathematically converted to an estimate of the number of RGC. This approach was used by Montesano et al. to show structure–function relationships in early diabetic eye disease between estimated RGC counts with microperimetric and frequency double perimetry sensitivities. ² This approach was not used here because the stimuli were large and four times the area of Frequency Doubling Technology grating stimuli (5°). Visual stimuli were also tested in the retinal midperiphery, where an approximation of RGC counts cannot be reasonably made given the key simplifying assumption underpinning GCL thickness conversion to RGC counts is the tightly packed, dense number of RGC cell bodies around the central ±10° eccentricity. Interestingly, Montesano et al. ² showed the largest functional differences between those with and without diabetes using Frequency Doubling Technology perimetry, which is a low spatial frequency contrast sensitivity test. This result aligns with the present finding of non‐selective contrast sensitivity deficits.

In planning future work, a greater focus should be on evaluating the progression of ON or OFF dysfunction and microvascular changes longitudinally to provide a clearer picture of the mechanisms underlying diabetic retinal disease, and to determine the predictive value of increment and decrement contrast processing as potential biomarkers for both DR progression and visual loss. New functional biomarkers could be beneficial in clinical trials of therapeutic interventions in early DR and could be useful in risk stratification in DR screening programmes.

AUTHOR CONTRIBUTIONS

Vanessa T. S. Tang: Conceptualization (supporting); data curation (lead); formal analysis (lead); investigation (lead); methodology (equal); project administration (lead); validation (lead); visualization (lead); writing – original draft (lead); writing – review and editing (lead). Robert C. A. Symons: Conceptualization (equal); methodology (equal); resources (supporting); supervision (equal); writing – original draft (supporting); writing – review and editing (supporting). Spiros Fourlanos: Investigation (supporting); methodology (supporting); resources (supporting); writing – review and editing (supporting). Daryl Guest: Conceptualization (supporting); methodology (supporting); supervision (supporting); writing – review and editing (supporting). Allison M. McKendrick: Conceptualization (lead); formal analysis (equal); investigation (equal); methodology (equal); software (equal); supervision (equal); validation (equal); visualization (equal); writing – original draft (equal); writing – review and editing (equal).

FUNDING INFORMATION

Australian Government Research Training Program Scholarship, The University of Melbourne, Parkville, Australia. The sponsor or funding organisation had no role in the design or conduct of this research.

CONFLICT OF INTEREST STATEMENT

Vanessa T. S. Tang: none. Robert C. A. Symons: Bayer. Spiros Fourlanos: none. Daryl Guest: none. Allison M. McKendrick: Heidelberg Engineering GmbH.

Supporting information

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Tang VTS, Symons RCA, Fourlanos S, Guest D, McKendrick AM. The relationship between ON–OFF function and OCT structural and angiographic parameters in early diabetic retinal disease. Ophthalmic Physiol Opt. 2025;45:77–88. 10.1111/opo.13394