An Updated Simplified Severity Scale for Age-Related Macular Degeneration, Incorporating Reticular Pseudodrusen: Age-Related Eye Disease Study Report No. 42

Abstract

Purpose

To update the Age-Related Eye Disease Study (AREDS) Simplified Severity Scale for risk of late age-related macular degeneration (AMD), including incorporation of reticular pseudodrusen (RPD), and to perform external validation on the AREDS2.

Design

Post hoc analysis of two clinical trial cohorts: AREDS and AREDS2.

Participants

Participants with no late AMD in either eye at baseline in AREDS (n=2719) and AREDS2 (n=1472).

Methods

Five-year rates of progression to late AMD were calculated according to levels 0–4 on the Simplified Severity Scale, following two updates: (i) non-central GA considered part of the outcome rather than a risk feature, and (ii) scale separation according to RPD status (determined by validated deep learning grading of color fundus photographs).

Main outcome measures

Five-year rate of progression to late AMD (defined as neovascular AMD or any GA).

Results

In the AREDS, following the first scale update, the five-year rates of progression to late AMD for levels 0–4 were 0.3%, 4.5%, 12.9%, 32.2%, and 55.6%, respectively. Following both updates, the proportion progressing to late AMD by five years was 8.4% in participants without RPD and 40.6% in those with RPD. As the final Simplified Severity Scale, the five-year progression rates for levels 0–4, respectively, were 0.3%, 4.3%, 11.6%, 26.7%, and 50.0%, for participants without RPD at baseline, and 2.8%, 8.0%, 29.0%, 58.7%, and 72.2%, for participants with RPD at baseline. In external validation on the AREDS2, for levels 2–4, the progression rates were similar, at 15.0%, 27.7%, and 45.7% (RPD absent) and 26.2%, 46.0%, and 73.0% (RPD present), respectively.

Conclusions

The AREDS AMD Simplified Severity Scale has been modernized with two important updates. The new scale for individuals without RPD has five-year progression rates of ~0.5%, 4%, 12%, ~25%, and 50%, such that the rates on the original scale remain accurate. The new scale for individuals with RPD has five-year progression rates of 3%, 8%, ~30%, ~60%, and ~70%, i.e., approximately double for most levels. This scale fits updated definitions of late AMD, has increased prognostic accuracy, appears generalizable to similar populations, but remains simple for broad risk categorization.

Précis

The AREDS Age-Related Macular Degeneration Severity Scale and associated five-year rates of progression to late disease have been modernized to include reticular pseudodrusen and to fit updated definitions of late age-related macular degeneration.

Introduction

In clinical practice, having a readily available estimate of an individual’s risk of progression to late age-related macular degeneration (AMD) is highly useful. These prognostic predictions are helpful in multiple ways: to inform and counsel patients, to make tailored plans for follow-up and reimaging at appropriate intervals and/or for home monitoring, and to justify evidence-based interventions.^1–6

However, for such algorithms, an important trade-off exists between simplicity and predictive accuracy. For this reason, different styles of risk calculator are appropriate in different scenarios, including those suitable for use in research, reading centers, or clinical practice settings. For example, longitudinal data from the Age-Related Eye Disease Study (AREDS) have previously been used to develop the AREDS 9-Step Severity Scale⁷, the Advanced AMD Risk Calculator⁸, and the AREDS Simplified Severity Scale⁹. The 9-Step Scale requires detailed quantitative grading of color fundus photographs (CFP) and is considered appropriate only for reading center or research use.⁷ The Advanced AMD Risk Calculator generates predictions over a continuous percentage scale, but is intermediate in complexity; it necessitates access to the calculator and requires information on age, smoking history, and family history.⁸ By contrast, the AREDS Simplified Severity Scale was designed for ease of use in routine clinical practice, such that a clinician could perform broad risk categorization using risk estimates that are easy to memorize, based on simple grading possible even without CFP.⁹ The latter two approaches use information from both eyes. However, all three approaches are based on just two macular features: soft drusen and pigmentary abnormalities.

Two important limitations apply to the AREDS Simplified Severity Scale. First, the risk estimates are based on the outcome of progression to advanced AMD, i.e., neovascular AMD or central geographic atrophy (GA).⁹ However, for GA, the outcome of interest in modern clinical practice is any GA, i.e., central or non-central.^8,10 This is because the majority of GA cases are non-central at incidence; non-central GA progresses almost inevitably to central involvement (over a median of 3 years); visual function is often poor even with non-central GA; and potential therapeutic approaches to slow GA enlargement may be appropriate for non-central GA as much as for central GA.^10,11 However, in the current AREDS Simplified Severity Scale, non-central GA counts as a risk feature (within the category of AMD pigmentary abnormalities), rather than as part of the outcome.⁹

Second, reticular pseudodrusen (RPD), also known as subretinal drusenoid deposits (SDD), are now recognized as an additional feature strongly associated with increased risk of progression to late AMD.^12–14 Their presence confers excess risk of late AMD above that already conferred by soft drusen and/or pigmentary abnormalities, i.e., even when these features coexist.¹² Hence, severity scales and risk calculators not taking RPD into account are expected to have suboptimal accuracy; they are likely to overestimate risk in eyes without RPD and underestimate risk in those with RPD. However, adding this third risk feature into severity scales should not substantially affect their ease of use. First, assessing RPD as a binary feature (i.e., presence or absence in the eye) appears sufficient for risk categorization, such that complex quantitative grading for RPD area or number of SDD lesions is not required.¹² Second, they can be readily detected in routine clinical practice, since they are observed with highest sensitivity and specificity on optical coherence tomography (OCT), which represents the standard of care imaging modality in retinal clinics.¹⁵ Detection of RPD by OCT may also be less susceptible to media opacity or differences in retinal pigmentation than detection by other imaging modalities. In addition, artificial intelligence algorithms are increasingly available to perform automated detection (particularly for the simple binary task of assessing presence/absence).^16–20

The purpose of this study was therefore to update the AREDS Simplified Severity Scale by making those two changes, to report the five-year risks of progression to late AMD at the person level, and to perform external validation of the updated scale using the AREDS2 dataset.

Methods

Age-Related Eye Disease Study: Study Procedures

The AREDS was a multicenter phase III randomized controlled clinical trial designed to assess the effects of nutritional supplements on AMD and cataract progression.²¹ The AREDS study design has been described previously.²¹ In brief, 4,757 participants aged 55 to 80 years were recruited between 1992 and 1998 at 11 retinal specialty clinics in the United States. The participants were randomly assigned to receive placebo, antioxidants, zinc, or the combination. The randomized clinical trial lasted five years and was followed by epidemiologic follow-up for another five years. At baseline and annual study visits, comprehensive eye examinations were performed, and CFP were obtained by certified study personnel, according to standardized protocols, and graded centrally at the Wisconsin Reading Center. The grading definitions of drusen, pigmentary abnormalities, GA, and neovascular AMD on CFP have been described previously.^22,23 The minimum size requirement to define GA was grading circle I-1 (1/8 disc diameter or 217 μm). Institutional review board approval was obtained at each clinical site and written informed consent for the research was obtained from all study participants. The research was conducted according to the tenets of the Declaration of Helsinki.

Deep Learning Algorithm Grading of Color Fundus Photographs for Reticular Pseudodrusen

Reading center grading for RPD presence on imaging other than CFP was not available in the AREDS, since CFP was the only imaging modality available in that study. Grades for RPD presence were therefore obtained by deep learning-based automated grading of the CFP. The algorithm and its performance metrics have been described previously.¹⁸ In brief, a deep learning algorithm was trained by exposing it to over 8000 AREDS2 CFP-fundus autofluorescence (FAF) image pairs from the AREDS2 ancillary FAF study. The ground truth labels for RPD presence/absence in each image pair came from reading center grading of the FAF images (i.e., label transfer). Multimodal multitask training was used, whereby the algorithm first underwent joint training with two other deep learning algorithms (an FAF algorithm and a CFP-FAF algorithm), using a representation shared between the three algorithms. This was followed by additional training separately from the two other algorithms, i.e., fine-tuning training suitable for grading from CFP alone. The benefits of multimodal multitask training are that what is learned by each algorithm from each image modality can improve the training of the other algorithms (by sharing features that are complementary between the image modalities), i.e., the grading from CFP alone benefits from paired FAF images having been present during training. In previous evaluation of the deep learning algorithm on an AREDS2 test set of CFP images, it achieved an area under the receiver operating characteristic (AUROC) of 0.832; in external validation using an independent test set (Rotterdam Study, Netherlands), it achieved an AUROC of 0.965.¹⁸

Statistical Methods: Modified Simplified Severity Scale

The unit of analysis was at the person level, owing to the person-based nature of the AREDS Simplified Severity Scale. The study population comprised AREDS participants without late AMD (defined as neovascular AMD or any GA) in either eye at baseline and with data available on progression to late AMD at five years. These participants were stratified on the basis of both (i) a modified Simplified Severity Scale (0–4) at baseline and (ii) RPD presence/absence at baseline. The modified Simplified Severity Scale was defined as follows, using information from both eyes: a score of 0–4 was recorded, based on the presence or absence of large drusen and AMD pigmentary abnormalities in the macula, with one point assigned for each feature in each eye (and, in the absence of large drusen in either eye, one point assigned if both eyes had medium-sized drusen). The modification from the original Simplified Severity Scale⁹ (in which non-central GA counted only as a pigmentary abnormality, rather than as part of the outcome) was made so that any GA, including non-central GA, could be considered as part of the outcome of late AMD, in line with modern practice. The proportions of participants with the outcome of progression to late AMD were analyzed according to the modified severity scale and RPD presence. The analyzes were not adjusted for any variables other than the modified Simplified Severity Scale and RPD presence. In addition, Kaplan-Meier curves of progression to late AMD were plotted according to the different levels of the updated severity scale. All analyses were conducted using SAS version 9.4 (SAS Inc, Cary, NC).

External Validation on the Age-Related Eye Disease Study 2

The AREDS2 was a multicenter phase III randomized controlled clinical trial designed to assess the effects of carotenoid and omega-3 fatty acid supplements on the course of AMD in people at moderate to high risk of progression to late AMD.²⁴ The AREDS2 study design has been described previously.²⁴ In brief, 4,203 participants aged 50 to 85 years were recruited between 2006 and 2008 at 82 retinal specialty clinics in the United States. Inclusion criteria at enrollment were the presence of either bilateral large drusen or advanced AMD in one eye and large drusen in the fellow eye. The participants were randomly assigned to receive placebo, lutein and zeaxanthin, docosahexaenoic acid and eicosapentaenoic acid, or the combination. The randomized clinical trial lasted five years. At baseline and annual study visits, comprehensive eye examinations were performed, and digital CFP were taken by certified study personnel, according to standardized protocols, and graded centrally at the Wisconsin Reading Center. The grading definitions of drusen, pigmentary abnormalities, and neovascular AMD on CFP were similar to those for the AREDS and have been described previously.²⁵ The minimum size requirement to define GA was grading circle I-2 (1/4 disc diameter or 433 μm), i.e., larger than in the AREDS, where the minimum size requirement was grading circle I-1 (1/8 disc diameter or 217 μm). Institutional review board approval was obtained at each clinical site and written informed consent for the research was obtained from all study participants. The research was conducted according to the tenets of the Declaration of Helsinki and complied with the Health Insurance Portability and Accountability Act.

Reading center grading for RPD presence was not available at the baseline study visit for many AREDS2 participants, since the AREDS2 ancillary FAF study started after the AREDS2 baseline visit for most participants. Therefore, and for consistency with the AREDS analyses, grades for RPD presence were obtained by deep learning-based automated grading of the CFP, using the same algorithm described above.¹⁸ Analyses similar to those performed for the AREDS were performed on the AREDS2 dataset. The study population comprised AREDS2 participants who met the AREDS2 eligibility criteria but did not have late AMD (defined as neovascular AMD or any GA) in either eye at baseline (i.e.., had bilateral large drusen) and had data available on progression to late AMD at five years. The modified Simplified Severity Scale was calculated in the same way. Again, the proportions of participants with the outcome of progression to late AMD were analyzed according to the modified severity scale and RPD presence. Similarly, Kaplan-Meier curves of progression to late AMD were plotted according to the different levels of the updated severity scale.

Results

Simplified Severity Scale with First Update, regarding Outcome Definition, in the Age-Related Eye Disease Study

The AREDS study population for these analyses comprised 2719 participants. Their demographic and clinical characteristics are shown in Table 1. The number of participants with RPD present at baseline (in either eye, by deep learning assessment of CFP) was 192 (7.1%).

Table 1.

Participant characteristics of the study population at baseline.

Participants: n	2719
Mean age (years), mean (SD)	69.0 (4.9)
Female	57.7%
Race
White	95.6%
Non-white	4.4%
Smoking status
Never	48.7%
Former	45.5%
Current	5.7%
RPD presence*
Neither eye	92.9%
One eye only	5.2%
Both eyes	1.9%
AMD severity†
0	44.3%
1	22.3%
2	15.7%
3	9.8%
4	7.9%
Drusen size‡
Small drusen, one or both eyes (or no drusen)	32.0%
Medium drusen, one eye only (no large drusen)	20.2%
Medium drusen, both eyes (no large drusen)	8.1%
Large drusen, one eye only	20.2%
Large drusen, both eyes	19.5%
Pigmentary abnormalities presence
Neither eye	66.4%
One eye only	19.9%
Both eyes	13.7%

^*by deep learning-based automated grading of color fundus photographs, at baseline

^†according to modified Simplified Severity Scale (levels 0–4), by person, at baseline

^‡small drusen are <63 μm; medium drusen ≥63 μm and <125 μm; large drusen are ≥125 μm

Abbreviations: AMD=age-related macular degeneration; AREDS=Age-Related Eye Disease Study; RPD=reticular pseudodrusen; SD=standard deviation

The first update was made to the Simplified Severity Scale by considering non-central GA as part of the outcome, rather than as a risk feature. With this update, in the study population, the number of participants that had progressed to late AMD (defined as neovascular AMD or any GA) in either eye at five years was 291 (10.7%). The proportions of participants that had progressed to late AMD, according to drusen size and pigmentary abnormality status, are shown in Table 2. In general, across all combinations, the proportions of participants who progressed to late AMD were higher according to (i) increasing drusen size (with dose-response associations), (ii) presence of pigmentary abnormalities, and (iii) bilaterality.

Table 2.

Number and percentage of participants progressing to late age-related macular degeneration in one or both eyes at or before the five-year follow-up visit, out of 2719 participants without late age-related macular degeneration in either eye at baseline, following updated definition of late age-related macular degeneration.

Drusen size* and number of eyes	Number of events/participants at risk: n (%)
	Pigmentary abnormalities
	None	One eye	Both eyes
Small, one or both eyes (or none)	1/788 (0.1%)	2/73 (2.7%)	0/8 (0.0%)
Medium, one eye (no large)	3/416 (0.7%)	6/103 (5.8%)	3/29 (10.3%)
Medium, both eyes (no large)	4/156 (2.6%)	3/42 (7.1%)	2/23 (8.7%)
Large, one eye	15/274 (5.5%)	22/177 (12.4%)	38/99 (38.4%)
Large, both eyes	27/172 (15.7%)	46/145 (31.7%)	119/214 (55.6%)

^*small drusen are <63 μm; medium drusen ≥63 μm and <125 μm; large drusen ≥125 μm

Number of risk factors (0–4, considering both eyes) is denoted by gradations in color, with white representing 0 and the darkest shade of blue representing 4

From these data, the Simplified Severity Scale with the first update was derived (Table 3 and Figure 1). The original Simplified Severity Scale⁹ is shown alongside it, for comparison (Table 3). The Kaplan-Meier curves of progression to late AMD are shown in Figure 2. As expected, the percentage progression rates increased according to the number of risk features present, with a dose-response association. Interestingly, the percentage progression rates in the updated version were relatively similar to those in the original version, although the rates were generally slightly higher in the updated version.

Table 3.

Age-related macular degeneration Simplified Severity Scale with first update, shown together with original Simplified Severity Scale, for comparison: five-year rates of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline).

Original Simplified Severity Scale			Simplified Severity Scale with first update
Risk factors*	Number of events/participants at risk: n	Number of events/participants at risk: %	Risk factors*	Number of events/participants at risk: n	Number of events/participants at risk: %
0	6/1466	0.4%	0	4/1204	0.3%
1	20/635	3.1%	1	27/606	4.5%
2	55/465	11.8%	2	55/428	12.9%
3	85/328	25.9%	3	86/267	32.2%
4	150/317	47.3%	4	119/214	55.6%

^*assign 1 risk factor for each eye with large drusen; assign 1 risk factor for each eye with pigmentary abnormalities; assign 1 risk factor if neither eye has large drusen but both eyes have medium drusen

Number of risk factors (0–4, considering both eyes) is denoted by gradations in color, with white representing 0 and the darkest shade of blue representing 4

Figure 1.

Age-related macular degeneration Simplified Severity Scale with first update: five-year rates of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline).

Figure 2.

Kaplan-Meier curves of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline), according to levels of the Simplified Severity Scale with the first update.

Simplified Severity Scale with Both Updates (Outcome Definition and Reticular Pseudodrusen Status), in the Age-Related Eye Disease Study

The second update was made to the Simplified Severity Scale by considering participants separately according to the presence or absence of RPD in either eye at baseline (while maintaining the first update). In participants without RPD at baseline, the number of participants that had progressed to late AMD in either eye at 5 years was 213 (8.4%). In participants with RPD at baseline, the equivalent number was 78 (40.6%). The proportions of participants that had progressed to late AMD, according to drusen size and pigmentary abnormality status, are shown in Table 4 (separately for those with and without RPD at baseline).

Table 4.

Number and percentage of participants developing late age-related macular degeneration in one or both eyes at or before the 5-year follow-up visit, for 2719 participants free of late AMD in both eyes at baseline, following both updates.

Drusen size* and number of eyes	Number of events/participants at risk: n (%)
	Reticular pseudodrusen absent†			Reticular pseudodrusen present†
	Pigmentary abnormalities			Pigmentary abnormalities
	None	One eye	Both eyes	None	One eye	Both eyes
Small, one or both eyes (or none)	1/762 (0.1%)	2/71 (2.8%)	0/7 (0.0%)	0/26 (0.0%)	0/2 (0.0%)	0/1 (0.0%)
Medium, one eye (no large)	2/406 (0.5%)	5/100 (5.0%)	3/29 (10.3%)	1/10 (10.0%)	1/3 (33.3%)	0/0
Medium, both eyes (no large)	3/152 (2.0%)	3/42 (7.1%)	2/22 (9.1%)	1/4 (25.0%)	0/0	0/1 (0.0%)
Large, one eye	15/258 (5.8%)	19/164 (11.6%)	25/80 (31.3%)	0/16 (0.0%)	3/13 (23.1%)	13/19 (68.4%)
Large, both eyes	21/155 (13.5%)	32/119 (26.9%)	80/160 (50.0%)	6/17 (35.3%)	14/26 (53.8%)	39/54 (72.2%)

^*small drusen are <63 μm; medium drusen ≥63 μm and <125 μm; large drusen ≥125 μm

^†defined as present if present in either eye

Number of risk factors (0–4, considering both eyes) is denoted by gradations in color, with white representing 0 and the darkest shade of blue representing 4

From these data, the final Simplified Severity Scale with both updates was derived (Table 5 and Figure 3). This consists of two scales: one for participants without RPD and another for those with RPD. The Kaplan-Meier curve of progression to late AMD is shown in Figure 4. For both scales, the percentage progression rates increased according to the number of risk features present, with a dose-response association. For the former (i.e., without RPD), the progression rates were relatively similar to those in the version with the first update only, though generally slightly lower. For the latter (i.e., with RPD), as expected, the progression rates were higher at each level of the scale. Several representative cases demonstrating use of the final Simplified Severity Scale are shown in Figure 5.

Table 5.

Age-related macular degeneration Simplified Severity Scale with both updates, shown together with original Simplified Severity Scale, for comparison: five-year rates of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline).

Original Simplified Severity Scale			Simplified Severity Scale with both updates
Irrespective of reticular pseudodrusen status			Reticular pseudodrusen absent			Reticular pseudodrusen present†
Risk factors*	Number of events/participants at risk: n	Number of events/participants at risk: %	Risk factors*	Number of events/participants at risk: n	Number of events/participants at risk: %	Risk factors*	Number of events/participants at risk: n	Number of events/participants at risk: %
0	6/1466	0.4%	0	3/1168	0.3%	0	1/36	2.8%
1	20/635	3.1%	1	25/581	4.3%	1	2/25	8.0%
2	55/465	11.8%	2	46/397	11.6%	2	9/31	29.0%
3	85/328	25.9%	3	59/221	26.7%	3	27/46	58.7%
4	150/317	47.3%	4	80/160	50.0%	4	39/54	72.2%

^*assign 1 risk factor for each eye with large drusen; assign 1 risk factor for each eye with pigmentary abnormalities; assign 1 risk factor if neither eye has large drusen but both eyes have medium drusen

^†defined as present if present in either eye

Number of risk factors (0–4, considering both eyes) is denoted by gradations in color, with white representing 0 and the darkest shade of blue representing 4

Figure 3.

Age-related macular degeneration Simplified Severity Scale with both updates: five-year rates of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline), according to reticular pseudodrusen presence (in either eye, at baseline).

Figure 4.

Kaplan-Meier curves of progression to late age-related macular degeneration (in one or both eyes, in participants with both eyes at risk at baseline), according to levels of the Simplified Severity Scale with both updates (i.e., including reticular pseudodrusen status).

Figure 5.

Four representative cases demonstrating the final age-related macular degeneration Simplified Severity Scale: five-year rates of progression to late age-related macular degeneration (in one or both eyes), based on the assumption of similar features in both eyes. Color fundus photographs are shown with the accompanying optical coherence tomography line scan. Abbreviations: reticular pseudodrusen (RPD); subretinal drusenoid deposits (SDD).

External Validation on the Age-Related Eye Disease Study 2

The study population for these analyses comprised 1472 participants. Their demographic and clinical characteristics are shown in Supplementary Table 6. The number of participants with RPD present at baseline (in either eye, by deep learning assessment of CFP) was 361 (24.5%).

The percentage progression rates with the first update were calculated (Supplementary Table 7). As expected, these increased according to the number of risk features present, with a dose-response association. For levels 2–4, the percentage progression rates were very similar to those obtained in the AREDS, at 16.7% (level 2), 31.1% (level 3) and 53.8% (level 4), compared to 12.9%, 32.2%, and 55.6%, respectively, in the AREDS. The Kaplan-Meier curves of progression to late AMD are shown in Supplementary Figure 6.

In AREDS2 participants without RPD at baseline, the number that had progressed to late AMD in either eye at 5 years was 387 (34.8%). In participants with RPD at baseline, the equivalent number was 227 (62.9%). The percentage progression rates with both updates were then calculated (Supplementary Table 8). For both participants without RPD and those with RPD, again, the progression rates increased according to the number of risk features present, with a dose-response association. In both cases, for levels 2–4, the rates were very similar to those obtained in the AREDS. For AREDS2 participants without RPD, the rates were 15.0% (level 2), 27.7% (level 3), and 45.7% (level 4), compared to 11.6%, 26.7%, and 50.0%, respectively, in the AREDS. For AREDS2 participants with RPD, the rates were 26.2% (level 2), 46.0% (level 3), and 73.0% (level 4), compared to 29.0%, 58.7%, and 72.2%, respectively, in the AREDS. The Kaplan-Meier curves of progression to late AMD are shown in Supplementary Figure 7.

Discussion

Main Findings, Implications, and Practical Use of the Updated Simplified Severity Scale

In recent analyses of both the AREDS and AREDS2 datasets, we demonstrated that RPD presence is a significant factor for increased risk of progression to late AMD.¹² This was true even when RPD status was considered in combination with the other macular risk features, with which RPD often co-exist; this is important, since RPD are increasingly common at higher AMD severity levels.^15,26 Similarly, this was also true when adjusted for other potential confounders, such as age and smoking status. However, in that study¹², we stopped short of developing an updated Simplified Severity Scale. We perform that important step here, by making two updates to the severity scale. First, we update the outcome to late AMD (defined as neovascular AMD or any GA), rather than advanced AMD (i.e., neovascular AMD or central GA only). As part of that update, we necessarily also remove non-central GA from being considered under the risk feature category of pigmentary abnormalities. Second, we report the severity scale separately for individuals with and without RPD.

Interestingly, the first update generates percentage progression rates that are very similar to those in the original scale. Expanding the progression outcome to include non-central GA would be expected to increase the percentage progression rates substantially, since incident GA comprises approximately 2/3 non-central GA and 1/3 central GA.¹⁰ However, this change is partially offset by the simultaneous narrowing in the definition of pigmentary abnormalities. In this way, many AREDS participants are effectively upgraded on the severity scale; all participants with non-central GA in one or both eyes at baseline were considered at risk of progression in the previous analyses but are now excluded from the current analyses, as already having late AMD. Hence, overall, moving non-central GA from part of a risk feature to part of the outcome has relatively little impact on the severity scale.

In the second update, RPD status is considered alongside the other features. For simplicity and accuracy, this was performed by generating two separate scales, rather than attempting to add an additional feature to the same scale (since that would assume equal weighting between RPD status and the other risk factors, which appears not to be justified¹²). In the new severity scale for individuals without RPD, the percentage progression rates were relatively similar to those in the version with the first update only, although they were generally slightly lower (particularly at the higher end of the scale). We would expect the rates to be lower, since all participants with the risk factor of RPD presence have been removed. However, the observed decrease is modest because of (i) RPD presence at baseline occurring with relatively low frequency (particularly at lower levels on the scale) and (ii) the increased risk associated with RPD presence being large but not overwhelming.¹² Indeed, the new scale for those without RPD is actually highly similar to the original Simplified Severity Scale. In both cases, the percentage progression rates at the 5 levels are approximately: <0.5%, 3–4%, 11–12%, ^~25%, ^~50%. This has the important practical advantage that, for individuals without RPD, physicians can essentially continue using the original Simplified Severity Scale. Hence, the original progression estimates remain valid for the new outcome of late AMD, provided that RPD are absent.

Regarding the new scale for individuals with RPD, the percentage progression rates are higher at each level than those of the scale for individuals without RPD. This is consistent with the results of recent analyses of RPD as a factor that conveys excess risk of late AMD.¹² Considered in terms of relative differences, the risk is approximately double for most positions on the scale (levels 1–3), though it is much higher at the bottom (level 0) and lower at the top (level 4). Considered in terms of absolute differences, the smallest differences are at the bottom of the scale (levels 0–1) and the largest differences at the top (levels 3–4). These findings have plausibility; overall (i.e., as a weighted average), RPD presence confers approximately double the risk of progression to late AMD. However, the relative risk is higher than double at the bottom of the scale, since individuals with 1 versus 0 risk features are being compared. The relative risk is lower than double at the top of the scale, since individuals with 5 versus 4 risk features are being compared (though the increase in absolute risk levels is high, to as much as 72% at 5 years). As a practical consideration, for individuals with RPD, physicians could either (i) use the new scale with the percentage progression rates themselves (i.e., 3%, 8%, ^~30%, ^~60%, and ^~70%) or (ii) use the original scale, double the percentage progression rates for levels 1–3 as approximations, but remember levels 0 and 4 separately.

External Validation on the Age-Related Eye Disease Study 2

The external validity of the original Simplified Severity Scale has previously been assessed on the independent AREDS2 dataset; the five-year progression rates were observed to be very similar.²⁷ In the current work, external validation of the scale on the AREDS2 dataset was evaluated for both the scale with the first update and the final scale with both updates. However, given that the AREDS2 inclusion criteria required all participants without late AMD to have large drusen in both eyes, the percentage progression rates in AREDS2 could be meaningfully compared only for levels 2–4. For the scale with the first update only, the AREDS2 progression rates are extremely similar to the equivalent rates for the AREDS. Similarly, for the final scale, the AREDS2 progression rates are very similar to the equivalent rates for the AREDS, for both the scale without RPD and that with RPD. This is despite multiple differences between the AREDS and AREDS2, including decade of study, genetic profile, dietary practices, oral supplementation, imaging practices, and GA minimum size definition.^{21,22,24,25,28,29} It therefore represents a high degree of replication and external validation of the updated Simplified Severity Scale on an independent dataset.

Comparison with Literature

Aside from the original AREDS Simplified Severity Scale, we are not aware of any severity scales or risk calculators for progression to late AMD that are designed to be simple for routine clinical use. Multiple more complex scales and risk calculators have been published; however, aside from the AMD Advanced Risk Calculator⁸ and the AREDS 9-Step Severity Scale⁷, all require detailed genetic data and most require comprehensive demographic and medical information (such as age, smoking status, educational level, diet, height, and weight).^28,30–32 Hence, they are different in design for the intended setting of the current scale. Some artificial intelligence-based approaches to predicting progression to late AMD have also been developed^33,34; again, that represents a different use case and lies outside the scope of the current work.

Similarly, we are aware of very few previous studies of large prospective datasets that have analyzed rates of progression to late AMD according to both traditional severity scales and RPD status. One study did examine this question but contained a limited baseline AMD severity spectrum and was relatively small in size, leading to very wide confidence intervals in its estimates for the excess risk of late AMD conferred by RPD presence.³⁵ One cohort study analyzed this in the specific setting of fellow eyes of patients with unilateral neovascular AMD, as described previously¹²; the results were consistent with increased risk of progression to late AMD conferred by RPD presence.³⁶

We are not aware of any AMD risk prediction algorithms derived from prospectively obtained data that have included RPD presence alongside the other typical macular features of soft drusen and pigmentary abnormalities. In one case, as discussed in detail previously¹², a retrospective review was performed of 138 patients with AMD undergoing routine clinical care at one medical center, with the large majority having unilateral neovascular AMD. This led to the proposal of an OCT risk prediction algorithm, which included RPD presence alongside three other binary variables: intraretinal hyperreflective foci, hyporeflective foci within a drusenoid lesion, and central drusen volume >0.03 mm³.³⁷ Unlike the AREDS Simplified Severity Scale, the algorithm was designed to work at the eye level (though using information from both eyes), rather than at the person level, and appears to be based on progression rates at two years, rather than five years. An automated OCT algorithm is required to calculate drusen volume. The algorithm was then applied to a cohort of 501 fellow eyes with early or intermediate AMD from patients with neovascular AMD enrolled in the HARBOR study.³⁸ The progression rates for categories 2, 3, and 4 were 5%, 33%, and 84% (compared to 14%, 48%, and 73% in the original description³⁷), i.e., similar for some but not necessarily all categories. Again, this represented the limited setting where one eye already had late AMD, specifically neovascular disease, such that the study population was not representative of the full spectrum of AMD. Overall, the OCT risk prediction algorithm has some similarities to the updated AREDS Simplified Severity Scale (since both include RPD status and some element of soft drusen quantification, and given some correspondence between intraretinal hyperreflective foci and hyperpigmentary abnormalities). However, the former works at the eye rather than the person level and operates over a different time period; it was developed predominantly and tested exclusively in the specific setting of the fellow eye of patients with neovascular AMD.

Strengths and Limitations

The strengths of this study include the use of the AREDS dataset, which comprises a large number of individuals with a very wide spectrum of baseline AMD severity recruited prospectively from multiple centers and followed for a long period, providing a high event rate for progression to late AMD. In this way, representative five-year progression rates could be calculated, even for participant groups with low baseline AMD severity. Additional advantages include the standardized collection of imaging and data at regular time-points and centralized reading center grading. The use of the AREDS2 as a second and independent dataset to validate the updated scale was an additional strength. For practical purposes, the finding that the original simplified severity scale is still valid for individuals without RPD is an important advantage, so that the scale remains simple for physicians to use. Hence, the balance between simplicity and predictive accuracy is not adversely affected.

Potential limitations are similar to those described previously.¹² These include the absence of reading center grading of RPD presence, owing to the lack of widespread multimodal imaging at that time in history. RPD grading was therefore performed by deep learning-based grading of the CFP. However, the ground truth of the RPD algorithm’s training was from reading center grading of FAF images (leading to high specificity of grading from CFP alone, in a previous study¹⁸). One possible implication of some false negative RPD cases in the analyses (given the imperfect sensitivity of deep learning-based RPD detection) is that the five-year progression rates might be slightly overestimated in the RPD-absent scale. By contrast, the high specificity of deep learning RPD detection means that the progression rates are likely to be accurate in the RPD+ scale. However, these levels of sensitivity and specificity might be close approximations for RPD assessment in routine clinical practice, where obvious cases of RPD presence will be identified but very mild cases might not. Other possible limitations include (i) low participant numbers in some subgroups (particularly those with RPD presence), leading to greater potential for some inaccuracy with the five-year progression rates, although external validation on the AREDS2 dataset argues against that idea, and (ii) a predominantly white study population, potentially limiting the generalizability of the scale estimates to other ethnic populations. We therefore recommend further validation in datasets and studies with participants from other ethnic groups. No adjustment was made for variables such as age, smoking, supplement use, diet, or genotype; this was deliberate, in order to maintain the simplicity of the scale. We consider that other more complex algorithms are available for those purposes.

Given that OCT can detect precursor or early atrophy features (particularly nascent GA, incomplete retinal pigment epithelial and outer retinal atrophy [iRORA], or small complete RORA [cRORA]) that may not be detected by CFP alone, the AREDS1/2 datasets likely included some eyes considered at baseline to have intermediate AMD but that did have one or more of these features. Eyes with these features are likely to progress to GA in a shorter time.^39,40 Hence, the 5-year progression rates reported in this study are likely slightly higher than they would have been in a study that had excluded eyes with these OCT features at baseline. However, the difference is likely relatively small and, since most physicians in clinical practice still use the Beckman⁴¹ or similar AMD classification schemes, which do not incorporate these features, the progression rates in this study appear appropriate for use in clinical practice.

In addition, CFP might have missed some cases of progression to very small GA lesions that might otherwise have been detected by OCT and/or FAF (i.e., even though the AREDS minimum size for defining GA, at 217 μm, was very similar to the cRORA threshold of 250 μm). Hence, the 5-year progression rates reported in this study are possibly slightly lower than they would have been in a study that used OCT and/or FAF to detect progression to GA. However, CFP does not miss progression to early GA completely in such cases; if anything, a possible delay in detection to the following annual study visit would have been most likely.⁴² Again, therefore, any difference is likely small and would apply equally (i.e., proportionally) across the scale.

Conclusions

In this work, we reconstruct a new AREDS AMD Simplified Severity Scale, with two important updates, by reporting the five-year rates of progression to late AMD from the AREDS. The first update comprises redefining non-central GA as part of the outcome rather than part of the risk features. The second update involves considering RPD status alongside the other risk features. The new scale for individuals without RPD has five-year progression rates (^~0.5%, 4%, 12%, ^~25%, ^~50%) that are extremely similar to those of the original scale. Therefore, physicians can continue to use these values in the setting of RPD absence. In the setting of RPD presence, the new scale has higher five-year progression rates (3%, 8%, ^~30%, ~60%, and ^~70%). Physicians could either use these values themselves or consider that the risk is approximately double for the three middle levels, but higher for level 0 and lower for level 4. The generalizability of the scale is demonstrated by successful external validation on the independent AREDS2 dataset. This updated severity scale fulfils the multiple aims of (i) being appropriate for modern definitions of late AMD (i.e., including any GA), (ii) increased prognostic accuracy, by incorporating RPD status, and (iii) remaining simple to use in routine clinical practice.