Spatial Cluster Patterns of Retinal Sensitivity Loss in Intermediate Age-Related Macular Degeneration Features

Abstract

Purpose: To examine spatial patterns of retinal sensitivity loss in the three key features of intermediate age-related macular degeneration (iAMD).

Methods: One-hundred individuals (53 iAMD, 47 normal) underwent 10-2 mesopic microperimetry testing in one eye. Pointwise sensitivities (dB) were corrected for age, sex, iAMD status, and co-presence of co-localized key iAMD features: drusen load, pigmentary abnormalities, and reticular pseudodrusen (RPD). Clusters (labeled by ranks of magnitude C₋₂, C₋₁, C₀) were derived from pointwise sensitivities and then assessed by quadrants and eccentricity/rings.

Results: Two clusters of decreased sensitivities were evident in iAMD versus normal: C₋₂, −1.67 dB (95% CI (confidence intervals), −2.36 to −0.98; P < 0.0001); C₋₁, −0.93 dB (95% CI, −1.5 to −0.36; P < 0.01). One cluster of decreased sensitivity was independently associated each with increased drusen load (13.57 µm increase per −1 dB; P < 0.0001), pigmentary abnormalities (C₋₁: −2.23 dB; 95% CI, −3.36 to −1.1; P < 0.01), and RPD (C₋₁: −1.07 dB; 95% CI, −2 to −0.14; P < 0.01). Sensitivity loss in iAMD was biased toward the superior and central macula (P = 0.16 to <0.0001), aligning with structural distributions of features. However, sensitivity loss associated with drusen load also extended to the peripheral macula (P < 0.0001) with paracentral sparing, which was discordant with the central distribution of drusen.

Conclusions: Drusen load, pigmentary abnormalities, and RPD are associated with patterns of retinal sensitivity loss commonly demonstrating superior and central bias. Results highlighted that a clinical focus on these three key iAMD features using structural measures alone does not capture the complex, spatial extent of vision-related functional impairment in iAMD.

Translational Relevance: Defining the spatial patterns of retinal sensitivity loss in iAMD can facilitate a targeted visual field protocol for iAMD assessment.

Introduction

Many studies have demonstrated overall retinal sensitivity loss in the early stages of age-related macular degeneration (AMD) using microperimetry.^1–6 However, the literature has been conflicted regarding which key intermediate age-related macular degeneration (iAMD) features^7,8 may confer functional deficit.^9–23 Increased drusen load has been associated with decreased retinal sensitivities,^9–15 although some studies have inconsistently reported that pigmentary abnormalities and/or reticular pseudodrusen (RPD) presence are associated with sensitivity loss,^9,10,17–22 whereas others have suggested no association.^{9,10,12,16,23} 

These issues may, in part, be attributed to the scarce and arbitrary application of spatial analyses to functional data. Assumed spatial groupings can lead to a statistical bias known as the modifiable areal unit problem (MAUP), leading to diluted results and erroneous interpretations.^24,25 This is particularly evident when retinal sensitivities are assessed according to abruptly delineated, symmetric, and concentric spatial templates such as the Early Treatment for Diabetic Retinopathy Study or Garway–Heath sectors,^26,27 despite well-established evidence that changes in visual sensitivity are relatively gradual, nonsymmetric, and nonconcentric.^28–33 Addressing the MAUP could clarify the relationship between functional impairment and key iAMD features, enhancing pathophysiological understanding and highlighting which retinal biomarkers and locations may be important to assess in iAMD. 

In a recent series of studies, we addressed the MAUP for structural retinal data by grouping basic (indivisible) unit data via statistical clustering.^34–39 Clustering has highlighted unique spatial patterns of structural differences emerging from the study of AMD and normal eyes, including paracentral photoreceptor degeneration^37–39 underscoring rod susceptibility in AMD⁴⁰ and inner retinal neuronal, synaptic, and vascular changes^36–38,41 supportive of retinal remodeling.^42–45 Both findings were exacerbated by specific iAMD features such as RPD.³⁹ 

Thus far, cluster analysis of functional outcomes has only been narrowly applied to photopic visual field sensitivities from a small sample of iAMD eyes.⁵⁴ This work indicated that cluster analysis enabled greater detection of sensitivity losses than traditional pointwise analyses, but it did not explore spatial relationships with any iAMD features⁵⁴ nor consider psychophysical testing under lower (mesopic) lighting, which may reveal greater functional deficits in AMD.^46–53 Henceforth, this study applied cluster analysis to mesopic visual field sensitivities from iAMD and normal eyes. We hypothesized that this would reveal spatially delineated associations between retinal sensitivities and key iAMD features, underscoring pathophysiological understanding of AMD and elucidating which iAMD features and retinal locations may be worth focusing on when assessing visual function. 

Methods

Study Population

The study population was a convenience sample prospectively recruited between January 4, 2022, to May 31, 2022, from pre-existing patients of the Centre for Eye Health (CFEH), Sydney, Australia. These patients had previously provided written informed consent for research recruitment. CFEH is a referral-only eyecare clinic with advanced diagnostic testing and management provided by optometrists and ophthalmologists.⁵⁵ Eligible participants provided written informed consent for prospective data collection and research use of their de-identified data, approved by the Biomedical Human Research Ethics Advisory Panel of the University of New South Wales and in accordance with the tenets of the Declaration of Helsinki. 

Eligibility required participants to be ≥50 years of age without any macular-involving disease or significant structural abnormality except iAMD with or without RPD. Participants under suspicion of macular-involving disease, such as from direct (ocular) or indirect (non-ocular) causes (e.g., from the presence of diabetes mellitus or hydroxychloroquine use), were also excluded. Diagnosis of iAMD was based on the Beckman Initiative classification⁸ requiring the presence of large drusen (≥125 µm diameter) or AMD-related pigmentary abnormalities with at least medium drusen (≥63 µm diameter) via color fundus photography.⁸ Participants 50 to 54 years of age (fulfilling the above criteria for iAMD and excluding diagnoses of AMD-mimickers⁵⁶) were included in concordance with other notable studies.^57–60 Optic neuropathies (suspected or confirmed) were excluded due to potential macula involvement. 

Diagnoses were formed by two non-blind CFEH clinicians (optometrists and/or ophthalmologists) and reconfirmed by investigator MT. Multimodal structural measures, including color fundus photography, near-infrared imaging, fundus autofluorescence, optical coherence tomography (OCT), and OCT angiography, were available for all patients and were used to exclude other macular-involving disease or significant abnormalities. A single eye was selected per participant using simple randomization if both eyes were eligible. 

Microperimetry Examination

All participants underwent microperimetry examination (macular integrity assessment [MAIA]; CenterVue, Padova, Italy) prior to other multimodal imaging to minimize fatigue and avoid bleaching of the photoreceptors. Neither pupil dilation nor dark adaptation was performed, and this was not expected to significantly affect microperimetry outcomes in healthy or diseased eyes,^61–63 further validated in a recent meta-analysis of intermediate AMD eyes.⁶⁴ Room lighting was turned off, and the non-study eye was occluded during examination. 

The MAIA uses light-emitting diodes for Goldmann III (GIII) size stimuli (π × [\(\frac{{0.43^\circ }}{2}\)]²) with a maximum luminous intensity of 318 cd/m², a mesopic background luminance of 1.27 cd/m², and a 36-dB dynamic range. A standard 10-2 grid with one extra foveal stimulus (69 stimuli total) was used for full threshold testing with a 4-2 staircase strategy. Automatic fundus tracking via scanning laser ophthalmoscopy was enabled to actively correct eye movements up to 25 Hz, ensuring that the stimuli were presented consistently on retinal locations. 

Microperimetry examinations with identical instructions for each participant were administered by a single operator. No participants had had prior experience with microperimetry. First, a 1-minute suprathreshold examination was performed followed by two consecutive full-threshold examinations on the study eye. At least 1 minute of rest was provided between examinations. 

Repeatability Analysis

To determine whether results from the two consecutive full-threshold examinations could be pooled, data were compared to establish any significant systematic difference,⁶⁵ and intrasession repeatability was then determined for global sensitivity (dB), pointwise sensitivities (dB), fixation loss (%), and 95% bivariate contour ellipse area (°²). Examinations with more than 25% of fixation loss/false-positive values were excluded.⁵ 

Cluster Analysis

Cluster analysis was performed to explore spatial patterns of retinal sensitivities for iAMD versus normal eyes and key iAMD features. First, for iAMD versus normal eyes, pointwise sensitivity differences (dB) were calculated with correction for the covariables of age, sex, and iAMD status (Fig. 1A), then clustered using unsupervised Two-Step clustering (Fig. 1B). This algorithm has proven robustness versus other cluster algorithms.⁶⁶ Point order was randomized,⁶⁷ and a log-likelihood method⁶⁸ was applied considering the minimum cluster number where each cluster was statistically significantly different (i.e., all cluster comparisons reached P < 0.05). Resultant functional patterns were presented graphically (Fig. 1C, left) and topographically (Fig. 1C, right). Clusters were labeled as C₋₂, C₋₁, C₀, C₁, and C₂ to reflect ranks of magnitude; that is, C₋₂ reflected greater reduced magnitude than C₋₁, C₋₁ reflected greater reduced magnitude than C₀, C₀ was not significantly different from zero, C₁ reflected greater increased magnitude than C₀, and C₂ reflected greater increased magnitude than C₁. Clusters with greater magnitude were assigned darker colors. Significant or non-zero clusters were further spatially assessed by retinal quadrants (superior, nasal, inferior, temporal) (Fig. 1C, right, white lines) and eccentricity/rings (central 2°, paracentral 4° and 6°, and peripheral macula 8° and 10°) (Fig. 1C, right, black lines). Note that all pointwise data were displayed from a retinal structural perspective (e.g., decreased sensitivities at the inferior retina would translate to a superior visual field defect). 

Figure 1.

 

(A) Microperimetry pointwise sensitivities were compared between iAMD and normal groups with correction for covariables. (B) Resultant pointwise sensitivity differences (dB) underwent cluster analysis, in which Two-Step clustering was applied considering the maximum cluster number where each cluster was statistically significantly different (n_max). (C) Functional spatial pattern of cluster means (95% CI) were then presented graphically (left) and topographically (right). Clusters with greater magnitude (mean [95% CI]) were assigned darker colors. All images in the right eye format and all pointwise sensitivities were displayed from a retinal structural perspective. Retinal quadrants are marked by white lines, centrality (eccentricity rings) is marked by black lines, and scale is on the bottom right. Stimuli diameters have been scaled to 2× diameter to improve visibility. (D) The iAMD features were then defined according to various multimodal structural measures, such as (top to bottom) for drusen load (color fundus photography and OCT B-scans); pigmentary abnormalities (OCT en face and B-scans); and RPD (near-infrared and OCT B-scans).^8,74,75 Cluster analysis was then repeated for each iAMD feature as in (B), presented as functional patterns and accompanied by corresponding structural patterns where possible.

Second, eyes were then grouped according to each key iAMD feature—drusen load, pigmentary abnormalities, and RPD (Fig. 1D).^7,8 Drusen load was defined according to retinal pigment epithelium to Bruch's membrane (RPE-BM) thicknesses^69–71; it was automatically segmented and then manually corrected by two trained graders⁷² and then colocalized to the same areas as each microperimetry GIII stimulus. Colocalized RPE-BM thicknesses were derived from SPECTRALIS OCT (Heidelberg Engineering, Heidelberg, Germany) macular cube scans (61 B-scans across 30 × 25° or ∼8640 × 7200 µm) via MATLAB 9.9 (MathWorks, Natick, MA) with code developed by DAC (co-investigator) which also adjusted for fovea-to-optic nerve head center tilt.⁷³ Pigmentary abnormalities and RPD were defined according to Laiginhas et al.⁷⁴ and Ueda-Arakawa et al.,⁷⁵ respectively, and colocalized to the same sectors as each microperimetry stimulus. Specifically, the presence of pigmentary abnormalities was determined using color fundus photography as the ground truth, with assistance from multimodal imaging, including fundus autofluorescence, near-infrared, and OCT en face images as needed. The presence of RPD was determined using OCT as the ground truth,⁷⁶ whereby five or more hyperreflective lesions had to be visible above the RPE. 

Cluster analysis was then repeated for each iAMD feature, with correction for all other covariables (age, sex, and co-present iAMD features). As the drusen load was a continuous variable, cluster analysis was applied to pointwise regression β-coefficients (across both AMD and normal eyes) for RPE-BM thickness (µm), to derive the associated functional pattern. The corresponding structural pattern of drusen load was also derived from cluster analysis of RPE-BM thickness. As pigmentary abnormalities and RPD presence were categorical variables, cluster analysis was applied to pointwise sensitivity differences (across AMD eyes with vs. without pigmentary abnormalities/RPD) to derive the associated functional pattern. The corresponding structural patterns of pigmentary abnormalities and RPD presence were derived from qualitative assessment of frequency (percentage of iAMD eyes) within each retinal quadrant and eccentricity/ring, due to qualitative definitions.^74,75 

Statistical Analysis

Statistical analyses were performed using Prism 9.4.0 (GraphPad, San Diego, CA), SPSS 25 (IBM, Chicago, IL), and Excel 2205 (Microsoft, Redmond, WA). Default statistical significance was considered as P < 0.05. Repeatability was described using two-way, mixed-effects model intraclass correlation coefficients (ICCs; average measures, absolute agreement),⁷⁷ paired t-tests, and 95% coefficients of repeatability (95% CoR).^78,79 Continuous values were expressed as means (95% CI) unless otherwise stated. Categorical variables including sex, iAMD status, pigmentary abnormalities presence, and RPD presence were dummy-coded for regression.⁸⁰ Normality was assessed using the D'Agostino–Pearson test and the appropriate statistical test (with or without assumption of normality) selected as described below. Summary values were derived using each participant as a single unit of observation to avoid combining interrelated data. Single comparisons of continuous values between groups were assessed using unpaired Student's t-test or Mann–Whitney U-test. Comparisons of continuous values within groups, as seen when comparing between clusters, were assessed using the paired t-test or Friedman test and Dunn's multiple comparisons test with adjustment. Comparisons of categorical values used Fisher's exact test. Further spatial delineations comparing quadrants and eccentricities/rings were assessed using McNemar's and Cochran's Q tests, respectively.⁸¹ Formal statistical comparison of qualitative structural distributions (for pigmentary abnormalities and RPD presence) were not performed to avoid inappropriate statistical interpretation (the MAUP)^24,25 due to lack of precise quantification methods. ICCs were interpreted as <0.5 = poor, ≥0.5 = moderate, ≥0.75 = good, and ≥0.9 = excellent.⁷⁷ Z-scores were interpreted as effect sizes, where ≥0.2 = small, ≥0.5 = medium, and ≥0.8 = large.⁸² 

Results

Participant Demographics

Single eyes from 53 iAMD and 47 normal control participants were included. Expectedly, age (P < 0.0001) and RPE-BM thickness (P < 0.001) were significantly greater in the iAMD group than the control group (Table 1). Sex and presence of cardiovascular disease were similar between groups (P > 0.99 and P = 0.65, respectively). Pigmentary abnormalities were present in 20/53 iAMD eyes (37.7%) and reticular pseudodrusen were present in 29/53 iAMD eyes (54.7%). Subsequent analysis of iAMD corrected for age, sex, and iAMD status (binary), and the analyses of each iAMD feature corrected for age, sex, and copresence of colocalized key iAMD features. 

Table 1.

 

Participant Demographics

 

Microperimetry Repeatability

From a total of 200 first and second examinations, 11 examinations from 11 eyes with fixation loss/false-positives > 25% were excluded. Pairwise comparisons between the remaining first and second examinations were performed for global mean sensitivity, pointwise sensitivities, fixation loss, and 95% bivariate contour ellipse area. None of these outcomes was significantly different (P = 0.07 to 0.99). Intrasession repeatability was then determined, whereby global mean sensitivity examinations showed good reliability (ICC = 0.85), and pointwise sensitivity showed up to good reliability (ICC up to 0.82; Supplementary Fig. S1). Subsequently, values between examinations were averaged for each participant for all further analyses. To confirm that the averaging of pointwise sensitivities did not alter cluster results, all cluster analyses were repeated using data from examinations one and two separately. There were no systematic differences in cluster outcomes when using pooled versus individual examination sensitivities (Supplementary Table S1). There were also no significant differences between iAMD and normal eyes regarding averaged fixation loss and 95% bivariate contour ellipse area (Table 1). 

Cluster Patterns of Retinal Sensitivity Loss in iAMD

Pointwise sensitivity differences between all iAMD versus normal eyes were calculated then clustered. Three resultant clusters were formed (C₋₂, C₋₁, and C₀) (Fig. 2). In iAMD eyes, two clusters demonstrated significantly decreased sensitivity: C₋₂, −1.67 dB (95% confidence interval [CI], −2.36 to −0.98; P < 0.0001) and C₋₁, −0.93 dB (95% CI, −1.5 to −0.36; P < 0.01). These clusters occupied 34.7% of the total stimuli (Table 2). The Z-scores for C₋₂ and C₋₁ revealed large effect sizes (−1.77 and −1.04 SD units from normal, respectively). Decreased sensitivity was more evident in the superior than inferior retinal quadrant (C₋₁ proportional area, 39% vs. 13% of respective quadrant stimuli; P < 0.001) and was most evident within the 2° rings (C₋₂: 80% of ring stimuli; P < 0.0001) and 4° rings (C₋₁: 63% of ring stimuli; P < 0.0001). 

Figure 2.

 

Spatial patterns of pointwise sensitivity differences between iAMD versus normal groups. Functional pattern of cluster differences (dB) are presented graphically (left) and topographically (right). Significance values (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001) were derived from unpaired Student's t-tests or Mann–Whitney U-tests. Z-scores are below the x-axis. Presentation is as in Figure 1C.

Table 2.

 

Cluster Analysis for iAMD Versus Normal and Spatial Delineations

 

Cluster Patterns of Retinal Sensitivity Loss Associated With Drusen Load

Cluster analysis was repeated for each key iAMD feature with adjustments for copresent localized features. First, for drusen load (RPE-BM thickness), cluster analysis was applied within group (across both AMD and normal eyes) to pointwise regression β-coefficients (continuous variable, using pointwise sensitivities as the outcome). 

Regarding drusen load, two clusters were formed (C₋₁ and C₀) (Fig. 3A), with one demonstrating a significant inverse relationship between colocalized RPE-BM thickness and pointwise sensitivity (C₋₁ β-coefficient: −7.37 × 10⁻² dB change per 1 µm; 95% CI, −8.21 to −6.53; P < 0.0001) covering 34.8% of total stimuli (Table 3). This was equivalent to a RPE-BM thickness increase of 13.57 µm per −1 dB. The relationship was more evident in the superior than inferior quadrant (C₋₁: 57% vs. 25% of respective quadrant stimuli; P < 0.001) and was most evident within the 2°, 8°, and 10° rings (C₋₁: 80%, 45%, and 63% of ring stimuli, respectively) and least evident within the 4° and 6° rings (C₋₁: 0% and 55% of ring stimuli, respectively; P < 0.0001). This was in contrast to primary cluster analysis between AMD and normal eyes, where sensitivity was relatively unaffected within the 8° and 10° rings. 

Figure 3.

 

Spatial patterns of pointwise sensitivity for drusen load. The functional pattern of cluster β-coefficients (A) and structural pattern of cluster drusen load (µm) (B) are presented as in Figure 2. Darker green indicates greater magnitude.

Table 3.

 

Cluster Analysis for iAMD Features and Spatial Delineations

 

To further explore the relationship between drusen load and pointwise sensitivity, the spatial distribution of drusen load was also assessed (i.e., structure only). Drusen load was comprised of two clusters: C₁, 14.43 µm (95% CI, 14.01–14.84) and C₂, 22.78 µm (95% CI, 21.03–24.54) (Fig. 3B); the drusen load was greater in the superior than inferior quadrant (C₂: 18.8% vs. 7.3% of respective quadrant areas; P < 0.05) and was most evident within the 2° and 4° rings (C₂: 100% and 75%, respectively; P < 0.0001). The structural pattern of drusen load was biased toward the superior and central macula, similar to the associated functional pattern. However, in the 8° and 10° rings, there was relatively lesser drusen load but decreased pointwise sensitivity, representing discordance between structure and (mesopic) function. 

Cluster Patterns of Retinal Sensitivity Loss Associated With Pigmentary Abnormalities

Cluster analysis was applied between groups (iAMD eyes with vs. without pigmentary abnormalities) for pigmentary abnormalities presence (categorical variable; using pointwise sensitivity differences as the outcome). In the iAMD group with pigmentary abnormalities (compared to without pigmentary abnormalities group), two clusters were formed: C₋₁ and C₀ (Fig. 4A). There was decreased sensitivity in one cluster (C₋₁: −2.23 dB; 95% CI, −3.36 to −1.1; P < 0.01) occupying 30.4% of total stimuli (Table 3) and with a large effect size of −2.23. Decreased sensitivity was more evident in the superior than inferior quadrant (C₋₁: 36% vs. 16% of respective quadrant stimuli; P < 0.001), and most evident within the 2° ring (C₋₁: 100% of ring stimuli; P < 0.0001). This was similar to the primary analysis, where decreased sensitivity was most evident within the 2° ring. The structural pattern of pigmentary abnormalities was also qualitatively assessed (Fig. 4B) and was greatest in frequency (percentage of iAMD eyes) within the 2° ring and superior quadrant, similar to the associated functional pattern. 

Figure 4.

 

Spatial patterns of pointwise sensitivity differences for pigmentary abnormalities. The functional pattern of cluster differences (dB) (A) and structural pattern of pigmentary abnormalities presence (percentage of iAMD eyes; qualitative assessment) (B) are presented as in Figure 2. Darker green indicates greater frequency.

Cluster Patterns of Retinal Sensitivity Loss Associated With RPD

Cluster analysis was applied between groups (iAMD eyes with vs. without RPD) as above for RPD presence. In the iAMD with RPD group (compared to without RPD group), two clusters were formed: C₋₁ and C₀ (Fig. 5A). Decreased sensitivity in one cluster (C₋₁: −1.07 dB; 95% CI, −2 to −0.14; P < 0.01) occupied 30.4% of total stimuli (Table 3) with a large effect size of −0.86. Decreased sensitivity was more evident in the superior than inferior quadrant (C₋₁: 55% vs. 0% of respective quadrant stimuli; P < 0.001) and most evident within the 4° ring (C₋₁: 38% of ring stimuli; P < 0.0001). This was in slight contrast to all other analyses whereby decreased sensitivity was most evident within the 2° ring. The structural pattern of RPD was qualitatively assessed (Fig. 5B), which was greatest in frequency (percentage of iAMD eyes) within the 4°, 6°, 8°, and 10° rings and superior quadrant, similar to the associated functional pattern. 

Figure 5.

 

Spatial patterns of pointwise sensitivity differences for RPD. The functional pattern of cluster differences (dB) (A) and structural pattern of RPD presence (percentage of iAMD eyes; qualitative assessment) (B) are presented as in Figure 2. Darker green indicates greater frequency.

Discussion

Drusen load, pigmentary abnormalities, and RPD in iAMD eyes were each associated with spatial patterns of decreased retinal sensitivities after adjustment for copresent features, with common bias toward the superior and central macula. Notably, the functional pattern of sensitivity loss associated with drusen load extended to the peripheral macula with paracentral sparing, which was discordant with its central structural distribution. Results suggested that a clinical focus on these three key iAMD features using structural measures alone does not capture the complex, spatial extent of functional impairment in iAMD. 

Drusen Load, Pigmentary Abnormalities, and RPD Are Each Associated With Functional Impairment

Retinal sensitivity has been commonly reported to be worse with increased drusen load^9–15 but, conflictingly, worse⁹ or nondifferent^10,12,16 with pigmentary abnormalities presence and worse^10,17–22 or nondifferent^9,23 with RPD presence. Our drusen results were similar, quantifying that a 14-µm drusen load increase (RPE-BM thickening) was associated with a 1-dB decrease in sensitivity. We acknowledge that this was an approximation based on cross-sectional data, and drusen regression may have confounded this measure.⁸³ Our cluster analyses highlighted that pigmentary abnormalities and RPD presence were also individually associated with retinal functional impairment, consolidating the three key features of iAMD^7,8 and suggesting that the copresence of these biomarkers may summate to complex patterns of functional impairment. 

Functional Impairment in iAMD Is Mostly Superiorly and Centrally Biased

Using cluster analysis, our results confirmed that functional impairment in iAMD (in totality and for each feature) was mostly biased toward the superior^3,84 and central macula.^3,17,84–86 Interestingly, the superior bias was in contradistinction to expected superior visual field decline (corresponding to decreased sensitivity at the inferior retina) seen in normal aging and glaucoma.³² As AMD is primarily a photoreceptoral rather than postreceptoral (optic nerve and ganglion cell) disease, superior retinal losses may instead be related to the greater superior distribution of conventional drusen,⁸⁷ RPD,^76,88 rods,^89,90 and/or cones.⁹¹ Meanwhile, the observed central bias aligned with reported peak central distributions of drusen load^87,92 and pigmentary abnormalities,⁹³ and the observed paracentral (4° ring) bias that was associated with RPD aligned with expected para-/pericentral RPD distribution.^76,88 These functional patterns corresponded to other studies that have demonstrated greater sensitivity loss within the central 2° to 8.5° (vs. outside) eccentricities in the early stages of AMD^3,17,84–86 and were further reaffirmed by our structural patterns and distributions of drusen load, pigmentary abnormalities, and RPD. These data show that, although the entire macula is important to monitor during AMD, greater attention may be warranted toward the superior and central areas. 

Structure–Function Discordance Associated With Drusen Load

Sensitivity loss associated with drusen load extended beyond the central macula (where structural drusen load was greatest) to the peripheral macula (8° and 10° rings), with paracentral (4° and 6° rings) sparing. Peripheral macula changes may partly reflect pathologic rod dysfunction in accordance with rod susceptibility in AMD,^89,90,94 particularly as cone populations steeply decrease and rod population steeply increases beyond ∼3.5° eccentricity.⁹⁰ Meanwhile the paracentral sparing may represent an area of lesser cone dysfunction^40,89,90 between greater drusen load (centrally) and greater rod density (peripherally). It was unlikely, however, that the peripheral macular sensitivity loss was an artifact of normal topographic variation in photoreceptor densities, as we compared corresponding pointwise sensitivities between iAMD and normal eyes prior to clustering. Relatedly, many studies have described the “winner takes all” mechanism,^95–98 whereby normal variations in photoreceptor populations are unlikely to significantly alter stimulus detection. 

This study did not explore methods of isolating cone versus rod function such as testing under scotopic lighting with varying chromatic stimuli and toward greater eccentricities,^99–103 so further suppositions require further study, particularly as mesopic testing is thought to be mostly cone driven.⁹⁹ Nevertheless, this finding intimates that current clinical reliance upon structural observation of the three key iAMD features may not adequately portray the full, spatial extent of functional impairment in iAMD. Thus, spatial, functional analyses may be able to provide supplementary contextual information regarding which retinal elements and locations may be changing in AMD. 

Limitations

The primary limitation of this study was that the spatial relationships between iAMD features and retinal sensitivities focused solely on the three key clinical biomarkers of iAMD. Visually, spatial patterns of sensitivity losses for each iAMD feature did not summate to the spatial pattern of sensitivity loss seen for iAMD (in totality), despite the use of multivariable linear regression, which describes additive/subtractive relationships between variables. Partly, this may have been due to our use of two-dimensional RPE-BM thickness as a surrogate measure of drusen load, discounting the potential effects of drusen load in three dimensions.¹⁰⁴ Additionally, in the future, larger study populations will enable further investigation into the spatial associations between retinal sensitivity and other (non-mutually exclusive) iAMD features such as atrophy subtypes reported in recent expert consensus meetings,^105,106 nascent geographic atrophy,¹⁰⁷ hyporeflective drusen cores, drusen calcification, and double-layer sign^108–111 and their potential interactive (multiplicative) effects.^112–114 Additionally, reliable methods of locally quantifying structural biomarkers such as pigmentary abnormalities, RPD, and even some atrophy subtypes may strengthen structure–function concordance.^74,115,116 

Second, our study population likely represented “milder” cases of iAMD due to the nature of the eyecare center used for recruitment,⁵⁵ thus resulting in a relatively small raw magnitude of sensitivity losses that were only slightly larger than the test–retest variability of the MAIA (∼1 dB).⁵ This was corroborated when observing the greater magnitude of sensitivity losses in other studies’ more “severe” cases of iAMD, such as presenting with nascent geographic atrophy (∼−7 dB using the MAIA)¹¹¹ and drusenoid pigment epithelial detachments (∼−5 dB using the MP-1,¹¹⁷ equivalent to ∼−9 dB using the MAIA),¹¹⁸ and reaffirmed when comparing average drusen height between our study population (∼14–22 µm) to other studies’ populations (∼14–134 µm).^59,70,71,119 The focus of this current study was to explore functional patterns of loss in iAMD features, and, regardless of magnitude or structure–function discordance, the magnitude of sensitivity loss is expected to increase over time with the chronic, progressive nature of AMD,^{2,4,120–122} consequently increasing in practical significance over time. 

Conclusions

Cluster analysis of iAMD versus normal eyes revealed that drusen load, pigmentary abnormalities, and RPD were each associated with spatial patterns of decreased retinal sensitivities after adjustment for copresent features and were commonly biased toward the superior and central macula. Clinical focus on the three key iAMD features could be expanded to incorporate mesopic functional testing to capture the complex spatial extent of visual functional impairment in iAMD. 

Acknowledgments

The authors thank research assistants Judy Nam for assistance with recruitment and Natalie Eshow for assistance with data collection. The authors also thank the volunteers for participating in this research. 

Supported, in part, by grants from the National Health and Medical Research Council of Australia (1186915 to MK and DAC, 1174385 to LNS). MT is supported by the Australian Research Training Program scholarship. Guide Dogs NSW/ACT provides support for the Centre for Eye Health (the clinic of recruitment). 

Disclosure: M. Trinh, None; M. Kalloniatis, None; D. Alonso-Caneiro, None; L. Nivison-Smith, None