Retinal Pigment Epithelium Curvature Can Predict Late Age-Related Macular Degeneration

Abstract

Purpose: Outer retinal band integrity strongly predicts late age-related macular degeneration (AMD). However, it is often assessed subjectively as “continuity” using inconsistent definitions. Alternatively, “curvature” of the outer retinal bands is a quantitative metric that strongly correlates with AMD biomarkers and can screen for intermediate AMD. We evaluated the prognostic ability of retinal pigment epithelium (RPE) and ellipsoid zone (EZ) curvature for late AMD against outer retinal band continuity, pigmentary abnormalities, reticular pseudodrusen, and drusen volume.

Methods: Consecutive patients with intermediate AMD who progressed to late AMD (n = 17) or remained stable (n = 42) were recruited. RPE and EZ curvature were quantified as a ratio of their lengths over Bruch's membrane using the sinuosity method of assessing river curvature, where a ratio of ∼1 indicates no outer retinal pathology. RPE, EZ, and Bruch's membrane were manually segmented and their lengths automatically extracted. The primary outcomes were outer retinal sinuosity and the odds ratio of predicting late AMD.

Results: Mean follow-up time for progressors and nonprogressors was 4.4 and 3.6 years. RPE sinuosity was strongly associated with pigmentary abnormalities (P = 0.001) and drusen volume (P = 0.004) but not reticular pseudodrusen (P = 0.28). RPE sinuosity >1.03 was the strongest predictor of late AMD developing within 5 years (15 [2.9–75]) and across the study period (25 [2.3-282]). Drusen volume >0.03 mm³ was the strongest predictor of progression within 2 years (33 [2.5–426]), and RPD could not independently predict progression within any time frame.

Conclusions: RPE curvature is a promising, quantitative outer retinal biomarker that can prognosticate late AMD and potentially enhance prognostic models.

Recent work suggests that outer retinal band integrity on optical coherence tomography (OCT) is a strong prognostic biomarker of late age-related macular degeneration (AMD)—specifically, abnormalities of the external limiting membrane,^1,2 ellipsoid zone (EZ),^2–4 interdigitation zone (IZ),^2,3 and retinal pigment epithelium (RPE).^1,3,5 A recent systematic review found that these disruptions were more predictive of late AMD development than other AMD biomarkers more commonly used for clinical diagnosis and prognosis, including large drusen, reticular pseudodrusen (RPD), and pigmentary abnormalities.⁶ However, outer retinal integrity is frequently assessed subjectively^2,3,7 and defined inconsistently, including complete/incomplete,⁸ intact/disrupted,⁹ or preserved/irregular,³ to describe the degree of disruption. Although the interrater reliability of these approaches has been excellent (κ = 0.82–0.96),^10,11 this has not been established for raters in community settings with heterogeneous experience and training regarding OCT use, which may negatively influence interrater reliability.¹² Moreover, current methods of assessing the outer retinal bands are typically limited to the central 1 to 3 mm^5,8,9 or use complex image sampling methods.¹³ These could be difficult to implement in clinical settings, particularly as concerns about time pressure have increased in eyecare practices.^14–16 A computerized approach would be advantageous for overcoming these issues, particularly as the field is moving toward automated methods for image processing and artificial intelligence applications.^17,18 

Quantitative measurements of outer retinal integrity may provide opportunities to better characterize disease progression due to the continuous nature of measurements. Other continuous measures, such as drusen volume, are more sensitive indicators of longitudinal change than coarser categorical scales by enabling calculations of progression risk per unit increase.¹⁹ Existing quantitative methods have been developed to measure outer retinal band reflectivity^13,20 and thickness attenuation,^5,21 but they suffer from limitations in normative data and image quality control. For example, band reflectivity on OCT images can be affected by interdevice variability due to differences in image-processing methods, hardware, and acquisition parameters.²² Thickness-based measures are also impacted by physiologic factors such as age, ethnicity, and axial length/refraction²³ and require resource-intensive segmentation of numerous retinal boundaries across all B-scans in a macular cube—whether by human graders or via artificial intelligence methods. 

Previous work shows that RPE curvature can be quantified by adapting a method used in environmental science for measuring river curvature—sinuosity.²⁴ Briefly, RPE sinuosity was determined by calculating the ratio of RPE to Bruch's membrane length on foveal OCT B-scans. Sinuosity was found to significantly correlate with AMD lesions, including drusen, pigmentary abnormalities, and reticular pseudodrusen, and able to screen for intermediate AMD with excellent performance (area under the curve ≥0.8–0.9).²⁵ Importantly, RPE sinuosity was robust to physiologic variations in retinal thickness and other characteristics such as systemic disease and visual acuity²⁵ and, therefore, potentially more generalizable to different clinical populations than preexisting quantitative approaches for assessing outer retinal integrity. While sinuosity is a promising quantitative biomarker of outer retinal integrity, it has only been evaluated for the RPE band for diagnostic outcomes. Quantifying EZ integrity using sinuosity is also worth investigating, given previous findings that EZ continuity significantly outperforms other AMD biomarkers used for prognostication.⁶ 

Thus, this study explores sinuosity as an alternative metric for describing outer retinal integrity. Specifically, the prognostic capability of RPE and EZ sinuosity for late AMD was compared to qualitatively measured outer retinal band continuity and other current, clinical AMD biomarkers.²⁶ We hypothesized that (1) both quantitative (sinuosity) and qualitative (subjective) assessments of outer retinal band integrity could predict late AMD, and (2) their performance would be noninferior to current prognostic biomarkers. 

Methods

Study Population

Two groups of eyes with intermediate AMD (iAMD) were included: those showing no longitudinal progression to late AMD (stable) and those converting to geographic or neovascular late AMD within the study period (progressors). Inclusion criteria for all participants were age over 50 years, refractive error <6 D, iAMD at the baseline visit, at least 2 visits during the study period, and availability of a macular Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany) volume scan with B-scan quality score ≥15 at the first eligible visit. Patients with concomitant macular pathology or geographic atrophy were excluded. The study eye was chosen randomly if both eyes met eligibility criteria. 

The sample size (n = 60) was approximated based on the numbers of progressors and stable AMD eyes used in previous prognostic studies investigating ellipsoid zone and RPE integrity³ and drusen volume as risk factors.²⁷ The sample was obtained by screening consecutive, unique patients seen at the Centre for Eye Health (CFEH) between January 1, 2009, and December 31, 2022, for inclusion into the study (n = 41,156) until the target sample size was reached. CFEH is an optometry-led referral center providing advanced diagnostic imaging and/or disease management services in collaboration with ophthalmologists in the local health district.²⁸ All participants consented to use of their deidentified data for research, and the study was approved by a University of New South Wales Human Research Ethics Advisory Committee. 

For all included participants, date of birth, sex, ethnicity, systemic disease status, visual acuity, refractive error, study and fellow eye diagnoses, B-scan length, image compression factor (x-axis and z-axis scaling), the first consultation date with a diagnosis of iAMD and most recent consultation with iAMD (nonprogressors), or late AMD (progressors) during the study period were extracted from clinical records and patient reports using practice management software, Bp VIP.net (Version 2.1.530.017; Best Practice Software, Hamilton, New Zealand), and recorded in Excel (Microsoft, Redmond, WA, USA). Systemic conditions were defined as the presence or absence of cardiovascular disease or diabetes. 

Biomarker Grading

Eyes were assessed for three established AMD biomarkers (pigmentary abnormalities, RPD, drusen volume) and four measures of outer retinal band integrity (RPE continuity, EZ continuity, RPE sinuosity, EZ sinuosity). Pigmentary abnormalities were manually graded as present or absent based on color fundus imaging and the highest quality en face macular OCT scan available (Cirrus HD-OCT; Carl Zeiss Meditec, Dublin, CA, USA). RPD were manually graded as present if ≥5 hyperreflective mounds or triangular lesions above the RPE in ≥1 B-scan (Spectralis OCT) was observed within a macular OCT volume.²⁹ EZ and RPE continuity of the central 1 mm of the foveal B-scan were manually graded as disrupted if the bands were discontinuous, were absent, or showed heterogeneous reflectivity and intact if the bands were continuous with homogeneous reflectivity.^3,9 All manual gradings were performed at the baseline visit by two experienced independent clinicians (RC, MT) based on previous studies.^6,30 All disagreements were adjudicated by a third, senior clinician. 

Quantitative biomarkers, including drusen volume and EZ and RPE sinuosity, were extracted using automated methods. Drusen volume was extracted within the central 5 mm using advanced RPE analysis software (Cirrus HD-OCT; Carl Zeiss Meditec). 

Sinuosity Calculation

Foveal B-scans from the highest resolution Spectralis OCT (Heyex, Version 2.5.1; Heidelberg Engineering GmbH) macular volume available were automatically segmented for the EZ, RPE, and Bruch's membrane (BM) using proprietary software. Only B-scans of sufficient scan quality and free from artifacts were included for analysis. Foveal B-scans were selected for analysis as previous work suggests there was little benefit to using all B-scans within the central 1-mm vertical extent of volumes (spaced either 120 or 240 µm apart) to calculate mean sinuosity, which was not significantly different from foveal sinuosity or able to screen for intermediate AMD with superior performance.^25,31 

Automatic segmentations were reviewed independently by graders RC and MT, who manually corrected any segmentation errors by aligning the EZ, RPE, and BM segmentation lines to fit the center of hyperreflective bands. Although segmentation lines are often established at the inner and outer border of bands to measure thickness,^32,33 our method was deemed appropriate as curvature was the characteristic of interest (i.e., it reduces the risk of bias from choosing the inner or outer border, which may under- or overestimate curvature, respectively). This approach was performed in accordance with previous methods for assessing retinal layer thickness and established a consistent grading technique between investigators.^34–36 In areas of RPE-BM loss, the segmentation lines were interpolated between adjacent points to form a continuous line. Any disagreements that could not be resolved by discussion were adjudicated by a senior grader before further data extraction. 

Images of EZ, RPE, and BM segmentations were uploaded to MATLAB (MathWorks, Natwick, MA, USA) and their lengths extracted using a customized code to calculate sinuosity, as described in previous work (Fig.),²⁵ which can be provided on reasonable request. EZ and RPE sinuosity of the foveal B-scan was determined as the ratio of the length of segmented lines against BM length. The ratios were then expressed as an absolute percentage from 1.0 (i.e., an EZ sinuosity of 1.05 represented a 5% difference in curvature of the EZ to BM). 

Figure.

 

Sinuosity was calculated from the length of segmentation lines on foveal OCT B-scans. After extracting the raw scans (A, B), images were binarized to isolate (C, D) EZ, (E, F) RPE, and (G, H) Bruch's membrane segmentation lines. Baseline EZ and RPE segmentation lines for an eye with stable iAMD (left column) are contrasted against an eye with iAMD that progressed to late disease within the study period (right column).

Statistical Analysis

Independent and paired sample t-tests and Pearson's χ² test were used to assess differences in continuous and categorical primary outcome measures, respectively. Differences in the outer retinal band sinuosity and integrity and drusen volume between stable and progressor groups were examined, with P values <0.05 considered significant. Univariable and multivariable linear mixed-effects models were then used to assess associations between baseline characteristics and these measures. Outer retinal band sinuosity of the EZ and RPE, along with their prognostic ability, formed the primary outcome measures of the study, while qualitative measures (continuity) were secondary outcomes. RPD, pigmentary abnormalities, and drusen volume were recorded as covariables. 

The prognostic ability for conversion to late AMD within 2 and 5 years, as well as across the study period, was calculated as odds ratios (ORs) for outer retinal integrity and AMD biomarkers, which were adjusted for demographic factors, including age, sex, and systemic disease,^37,38 and any other significant demographic or eye factors found on multiple regression analysis. Best vision sphere (spherical equivalent), which is strongly associated with axial length,^39,40 was included in the analysis as axial length data were unavailable. Odds ratios of outer retinal integrity biomarkers were not adjusted for AMD biomarkers as sinuosity and continuity measures are highly correlated with these features and thus adjusting for them would confound the analysis in the same way that adjusting for drusen would confound associations between drusen volume and disease outcomes in AMD.⁴¹ 

RPE and EZ sinuosity measurements were transformed into categorical and continuous variables based on the mean difference in sinuosity found between progressor and nonprogressor groups across the study period to determine their prognostic ability (i.e., for a mean difference of 0.03 in RPE sinuosity between groups, values were binarized to categories using 1.03 as a threshold, and odds ratios were calculated for increments of 0.03). This approach reflects methods for assessing the prognostic ability of drusen volume (i.e., exceeding 0.03 mm³ in volume^27,42 and per 0.01-mm³ increase in volume^43,44) and allows appropriate comparisons to determine whether outer retinal sinuosity provides any prognostic advantage to existing metrics produced by OCT software. This analysis was post hoc due to lack of existing data on expected sinuosity differences between progressors and nonprogressors in AMD. All statistical analyses were performed using SPSS (version 25; IBM, Armonk, NY, USA). 

Results

Study Population

A total of 60 intermediate patients with AMD were consecutively recruited, including 17 progressors and 42 patients with stable diagnoses during the study period (January 1, 2009, and December 31, 2022). One patient with stable iAMD was excluded due to an image extraction error. The mean (SD) follow-up time was 4.4 (2.5) years and 3.6 (2.3) years for progressor and stable AMD groups, which was not significantly different (P = 0.27). Fellow eye late AMD (P = 0.002) and the presence of pigmentary abnormalities (P = 0.001) were significantly different between groups. No other demographic and imaging factors were significantly different (Table 1). 

Table 1.

 

Differences in Demographic, B-scan, and Disease Characteristics Between Groups

 

Differences in Sinuosity and Continuity Between Groups

The mean (SD) RPE and EZ sinuosity at baseline was 3.9% (4.2) and 1.4% (1.6), and 31% and 64% of all participants exhibited disrupted RPE and EZ, respectively. Differences in the primary outcome measures—baseline EZ and RPE sinuosity—and qualitative measures of outer retinal integrity (continuity) between stable and progressing patients with AMD were determined for progression within 2 years, 5 years, and across the study period (0.23–9.2 years). 

For progression within 2 years, neither sinuosity nor continuity measures were significantly different at baseline between stable patients versus those who progressed. For progression within 5 years, only RPE sinuosity (P = 0.01) was significantly different at baseline while RPE (P = 0.002) and EZ continuity (P = 0.01) were both significantly different at baseline. For progression across the study period, both RPE (P = 0.003) and EZ sinuosity (P = 0.04), as well as RPE (P < 0.001) and EZ continuity (P = 0.002), were significantly different between groups at baseline (Table 2). Overall, the mean RPE and EZ sinuosity of patients with stable iAMD were significantly lower at baseline, measuring as 3.6% and 0.93% less than those who progressed during the study period. Based on these results, RPE sinuosity was transformed to a categorical variable by using 1.03 as a threshold and assessed as a continuous variable using 0.03 as an increment for prognostic ability evaluation. Correspondingly, a threshold value of 1.01 and an incremental value of 0.01 were used to determine the prognostic ability of EZ sinuosity. 

Table 2.

 

Comparison of Sinuosity and Continuity Between Groups

 

Factors Associated With Sinuosity and Continuity

Regression analysis was then performed to identify potential confounders of RPE and EZ sinuosity and continuity. On univariable analysis, variables that were significantly associated with RPE sinuosity included fellow eye late AMD (P = 0.04), x-axis scaling (P = 0.005), pigmentary abnormalities (P < 0.0001), RPD (P = 0.03), and drusen volume (P < 0.001). The same variables were associated with EZ sinuosity (P < 0.0001 to 0.03), as well as best vision sphere (P = 0.04). 

For RPE continuity, fellow eye late AMD status (P = 0.003) and pigmentary abnormalities (P = 0.002) were significant variables, while visual acuity (P = 0.04), B-scan length (P = 0.02), pigmentary abnormalities (P = 0.01), and drusen volume (P = 0.01) were significantly associated with EZ continuity (Supplementary Table S1). 

Multivariable analysis was then performed, including any variables with significance levels ≤0.25 in univariate analysis. Demographic variables influencing RPE sinuosity were older age (P = 0.02) and female sex (P = 0.006), whereas no variables were significant for EZ sinuosity. Only the imaging variable, x-axis scaling, showed a significant, positive association with RPE (P = 0.03), while B-scan length did not affect any sinuosity measures. 

For AMD variables, fellow eye late AMD was significantly associated with increased EZ sinuosity (P = 0.05) and RPE continuity (P = 0.05). Pigmentary abnormalities were associated with RPE sinuosity (P = 0.001), RPE (P = 0.03), and EZ continuity (P = 0.04), while drusen volume demonstrated significant associations with RPE (P = 0.004) and EZ sinuosity (P = 0.01), as well as EZ continuity (P = 0.04). Meanwhile, reticular pseudodrusen was only significantly associated with EZ sinuosity (P = 0.03). None of the outer retinal biomarkers significantly correlated to all three AMD biomarkers (Table 3). 

Table 3.

 

Associations Between Variables and Primary Outcome Measures (Multivariable Analysis)

 

Predictive Performance

Finally, the prognostic ability of baseline RPE and EZ sinuosity and continuity was examined by calculating the adjusted odds ratio (OR) of late AMD development within 2 years, 5 years, and across the study period (up to 9 years) (Table 4). In addition to age, sex, and systemic disease,^37,38 ORs were also adjusted for best vision sphere, x-axis scaling, and fellow eye AMD status as significant variables identified on multivariable analysis. 

Table 4.

 

Prognostic Ability of Primary Outcome Measures

 

For progression within 2 years, no sinuosity measures could predict development of late AMD. Of the biomarkers currently used for prognosis, drusen volume exceeding 0.03 mm³ (33 [2.5–426]) and odds per 0.01-mm³ increase in drusen volume above zero (i.e., no drusen) at baseline (1.3 [1.0–1.6]) were also significantly associated with developing late AMD. The number of progressors was insufficient for calculating the OR for RPE sinuosity and EZ continuity. 

For progression within 5 years, both RPE and EZ sinuosity could predict conversion to late AMD. RPE sinuosity exceeding 1.03 (15 [2.9–75]) and EZ sinuosity exceeding 1.01 (5.0 [1.3–19]) were stronger predictors than continuous sinuosity values (per 3% increase in RPE sinuosity: 1.6 [1.1–2.5]; per 1% increase in EZ sinuosity: 1.2 [0.08–1.8]). RPE (8.7 [2.2–35]) and EZ continuity (10 [1.3–86]) were also significant predictors of late AMD, as were the AMD biomarkers, pigmentary abnormalities (11 [2.7–48]), drusen volume >0.03 mm³ (7.6 [1.9–30]), and per 0.01 mm³ in drusen volume (1.1 [1.0–1.3]). 

Across the study period, only sinuosity of the RPE could predict progression to late AMD, with RPE sinuosity >1.03 (25 [2.3–282]) demonstrating superior prognostic ability to 3% increases in sinuosity (1.7 [1.1–2.8]). Neither quantitative (continuous or categorical) measures of EZ sinuosity were successful in predicting progression. In contrast, qualitative assessments of RPE (6.5 [1.7–25]) and EZ continuity (10 [1.2–89]) were both significant predictors. For AMD biomarkers, only the presence of pigmentary abnormalities (5.7 [1.5–22]) and baseline drusen volume >0.03 mm³ (6.6 [1.5–29]) were significant predictors. 

RPE sinuosity greater than 1.03 demonstrated the strongest prognostic ability for late AMD development within 5 years or more and was superior to RPE and EZ continuity and traditional AMD biomarkers used for risk assessment. For short-term prediction of progression to late AMD within 2 years, drusen volume exceeding 0.03 mm³ showed the strongest prognostic ability. Notably, RPD was not a significant predictor for any time frame examined (Table 4). 

As some participants did not have sufficient follow-up time of 2 and 5 years, respectively, a subanalysis was conducted to evaluate the predictive potential of outer retinal biomarkers more rigorously. A further condition to only include data from OCT B-scans where the x-axis was ≥10 µm/pixel was also applied as this was substantially lower for 12 participants where the x-axis scaling was 5.52 to 5.96 µm/pixel compared to 10.51 to 12.36 µm for the rest of the cohort. A comparative analysis of sinuosity between converter and non-converters was firstly performed (Tables S2 and S5), followed by univariable and multivariable analyses for these cohorts (conversion within 2 years: Tables S3–S4, and within 5 years: Tables S6–S7) to identify factors needed for OR adjustment. The OR results from the subanalysis were similar to findings from the whole study cohort with RPE sinuosity >1.03 maintained as the strongest predictor of late AMD development within 5 years (44 [2.6–756]) and drusen volume >0.03 mm³ within 2 years (17 [1.3–225]) (Supplementary Table S8). 

Discussion

This study found outer retinal band integrity measured quantitatively as RPE sinuosity was the strongest predictor of progression within 5 years or more. In contrast, EZ sinuosity was not a consistent predictor of late AMD, with qualitative assessments of EZ continuity showing superior prognostic performance. Drusen volume exceeding 0.03 mm³ was found to be a strong predictor of late AMD development within 2 years and superior to all other outer retinal integrity biomarkers, while RPD presence could not predict progression within any time frame examined. This suggests that RPE sinuosity may be an effective long-term but not short-term predictor of late AMD development. 

Quantification of RPE Sinuosity Is Superior to Qualitative Measures of Outer Retinal Integrity for Predicting Late AMD

This was the first study to evaluate RPE and EZ sinuosity as prognostic biomarkers of late AMD. The mean RPE sinuosity and proportion of eyes showing outer retinal band disruptions in this study were comparable to previous estimates.^3,25 We found that 65% of progressors and 17% of nonprogressors exhibited disrupted RPE, which was similar to the 58% and 10% of eyes exhibiting disrupted RPE in a study by Ferrara et al.³ Although more eyes with disrupted EZ (52% vs. 27%) were identified, most progressors exhibited disrupted EZ in both studies (94% vs. 88%).³ 

In terms of prognostic ability, RPE sinuosity and continuity were both significant predictors of late disease within 5 years and across the study period. Specifically, RPE sinuosity exceeding 1.03 was associated with a 14 times increase in odds of developing late AMD within 5 years, while RPE continuity disruption was only associated with a 7.7 times increase. The odds of predicting progression across the study period were nearly four times greater for RPE sinuosity >1.03 (OR, 25) than for RPE continuity (OR, 6.5). Differences in prognostic performance between sinuosity and continuity measures could be attributed to the larger assessment area used to calculate RPE sinuosity compared to the 1-mm zone from which continuity was evaluated. Specifically, RPE sinuosity captures changes in both the “inner” and “outer” zones of the macula, which can help differentiate between eyes that will progress or remain stable. For example, eyes with geographic atrophy are more likely to have large drusen within the central 1-mm diameter compared to eyes without late AMD, where large drusen are more prevalent in the outer macula regions.⁴⁵ 

For EZ integrity, qualitative measures showed greater predictive ability for late AMD than quantitative measures. In fact, EZ sinuosity was only associated with significant odds for progression within 5 years. Poorer prognostic performance, despite assessment over a wider retinal area than continuity measures, could be attributed to the relatively anterior location of the EZ band compared to the RPE, leading to smaller curvature alterations from outer retinal pathology, which predominate in AMD.⁴⁶ Consequently, EZ sinuosity may also be more susceptible to drusen regression,⁴⁷ as curvature increases due to drusen are smaller to begin with (Fig.). In contrast, the superior prognostic performance of EZ continuity may be attributed to the fact that abnormalities in band appearance can persist after drusen regression and neovascular AMD treatment, even in the absence of curvature changes. For example, longitudinal assessments have found that the EZ band remains disrupted in a quarter of locations that have undergone drusen regression,⁹ and 40% of eyes treated with anti–vascular endothelial growth factor for neovascular AMD continue to exhibit EZ disruptions.¹¹ Nevertheless, the prognostic ability of RPE sinuosity was superior to both quantitative and qualitative measures of EZ integrity. 

RPE Sinuosity is Associated With Multiple AMD Biomarkers

RPE sinuosity and EZ continuity showed strong associations with pigmentary abnormalities and drusen volume—features currently used for estimating risk of progression to late AMD.²⁶ These findings suggest that they may be useful as summative measures for predicting late AMD and have the potential to enhance prognostic models, particularly if they can be automated. 

Interestingly, only EZ sinuosity was significantly associated with RPD, suggesting that there could be advantages to assessing the EZ band over a greater area as opposed to EZ continuity within the central 1 mm, which failed to show a similar association. The association between RPD and EZ sinuosity is logical, given that RPD appears as hyperreflective mounts above the RPE on OCT and primarily impact the EZ band.²⁹ Moreover, RPD is least likely to be found in the foveal region and presents most frequently within 1.5 to 2 mm from the fovea, superior and temporally,⁴⁸ supporting EZ sinuosity as a potential metric for capturing the noncentral distribution of RPD. Coincidentally, neither RPD presence nor EZ sinuosity were significant predictors of late AMD. Currently, there are no reliable and clinically accessible methods for quantifying RPD objectively, and subjective approaches are vulnerable to interreader variability.^49,50 Further exploration of associations between EZ sinuosity and RPD would therefore be extremely valuable in determining whether it could be useful for quantifying these lesions. 

Late AMD in the fellow eye is indicative of person-level AMD severity, and the significant associations with EZ sinuosity and RPE continuity reflect the tendency for study eyes to have increased drusen volume and a higher prevalence of RPD and hyperreflective dots (also known as hyperreflective foci) when the fellow eye exhibits late disease.^42,51 Examining these associations in larger sample sizes is needed to confirm these associations and whether RPE sinuosity and EZ integrity are also significantly impacted by fellow eye AMD status. 

Associations Between Outer Retinal Band Integrity With Demographic and Imaging Characteristics

This study also found significant associations between age, sex, and image compression to RPE sinuosity, contrasting with previous work, which showed that RPE sinuosity is robust to physiological and imaging characteristics.²⁵ This may be due to differences in study populations where this study included high-risk iAMD eyes (of which 29% converted to late AMD) versus the previous study, which used iAMD eyes from a consecutively recruited population. Indeed, the rate of iAMD eyes converting to late AMD in the consecutive list of eligible patients was only 4.7% (unpublished data).²⁵ It is known that age and sex are both independent risk factors of late AMD,^37,38 and therefore older and female patients are more likely to exhibit other AMD biomarkers that prognosticate progression, such as drusenoid pigment epithelial detachments⁶ and nascent geographic atrophy,⁵² which would increase RPE sinuosity. 

The significant association between image compression and RPE sinuosity may be attributed to the lower proportion of B-scan images extracted from macular cubes of the same dimensions (83%) compared to the previous study (93%), as image compression differs between scan protocols.⁵³ It is likely that continuity measures were unaffected by this factor as it was assessed within the central 1 mm rather than across the whole B-scan. Interestingly, no association between visual acuity and EZ continuity was found despite other groups reporting strong associations.^8,54,55 Lack of multivariable analysis⁷ to adjust for confounding factors (e.g., age) and investigating continuity in patients who have already developed late AMD^56,57 and therefore more likely to suffer significant VA loss in previous work may explain these differences. Overall, associations between patient and imaging characteristics need to be clarified in future investigations by repeating the evaluation in a larger sample size that is more representative of community patient populations, which will help determine whether sinuosity is robust to physiological and imaging factors in real-world scenarios. 

Limitations

The small sample size of progressors in some subgroups was a limitation as this prevented odds calculations of late AMD development within 2 years for categorical RPE sinuosity and EZ continuity measures. Although the study was sufficiently powered to evaluate the prognostic ability of RPE and EZ sinuosity across the study period, increasing the sample size would better estimate the short-term prognostic ability of these biomarkers. Extending recruitment to clinical settings beyond the CFEH would also improve the sample size as a high proportion of patients attending the clinic have glaucoma or are glaucoma suspects²⁸ with relatively few diagnosed with AMD compared to the general population.⁵⁸ Collection of lifestyle risk factors of progression, including smoking and diet, could also enhance the multivariable analysis and provide more robust evidence of the prognostic value of RPE sinuosity. Using a larger, more comprehensive retrospective data set that includes these characteristics in further investigations would be helpful for confirming our findings and evaluating the prognostic ability of outer retinal biomarkers comprehensively. 

Our calculation of outer retinal sinuosity over whole B-scans rather than the central 1 mm over which continuity was examined could limit comparisons between these two measures. However, this approach was chosen as drusen are highly concentrated in the central 1 to 3 mm of the retina, and previous work showed that the diagnostic accuracy of identifying iAMD from normal eyes did not significantly improve when images were cropped to the central 1 mm.²⁵ Furthermore, restricting integrity assessments to the central 1 mm of B-scans allowed for comparisons between our findings and previous work. It is worth noting that using the whole B-scan to generate sinuosity calculations rather than the central 1 or 5 mm likely underestimates its prognostic potential as the impact of drusen and other AMD lesions on outer retinal curvature would be diluted by including areas beyond this extent where there are typically less drusenoid changes.³¹ Thus, examining the prognostic ability of outer retinal sinuosity for B-scans cropped to different extents would be useful for refining the approach, as well as its ability to predict late AMD when derived from multiple B-scans from an OCT volume. 

Conclusions

RPE sinuosity is a strong predictor of late AMD, which outperformed qualitative measures of outer retinal integrity and existing AMD biomarkers used for prognosticating late disease over 5 or more years. These findings warrant further investigation, particularly as sinuosity is a quantitative measure that is amenable to automation and may improve clinical translation of outer retinal band evaluations to clinical settings. 

Acknowledgments

The authors thank Gonzalo Jacome for providing technical assistance. 

Supported by the Australian Government Research Training Program and, in part, by funding from the National Health and Medical Research Council (NHMRC #1174385, #1186915). Guide Dogs NSW/ACT provides support for the Centre for Eye Health, the location of recruitment. 

Disclosure: R. Cheung, None; M. Trinh, None; L. Nivison-Smith, None