1.

Introduction

The normative profile of circumpapillary retinal nerve fiber layer thickness (cpRNFLT) is important to identify abnormal thinning of cpRNFLT, which is one of the hallmarks for glaucoma diagnosis and progression monitoring.^1,2 It is well known that the normative cpRNFLT decreases with increased age.^3,4 Moreover, studies have shown that the thinning of normative cpRNFLT with age is highly location-specific.^5,6 However, the location specificity of the age-related thinning varies among different studies. For example, Parikh et al.⁵ have shown the age-related thinning rate of cpRNFLT in superior and inferior quadrants to be the maximum and minimum among the four quadrants, respectively. By contrast, Leung et al.⁶ have shown the age-related thinning rate of cpRNFLT to be the maximum in inferior and minimum in nasal quadrants, respectively. The varying results of the quadrant specificity of age-related cpRNFLT thinning might be due to the relative small study populations in those studies, as all their results are based on less than 250 eyes. A second limitation of previously published cpRNFLT norms is their restriction on coarsely defined sectors, which neglects the large location specificity of cpRNFLT over the entire scan circle. One of the few studies that presented 12 instead of four to six sectors⁷ demonstrates this location-specific variability.

Besides the correlation between the normative cpRNFLT and age, studies have also shown the cpRNFLT to be related to factors that represent the eye anatomy, including ocular axial length and the spherical equivalent (SE) of the refractive error.^8,9 More specifically, it has been shown that thinner cpRNFLT is associated with larger axial length and more myopia.^10–12 Ocular magnification formulas have been proposed to estimate the actual scan diameter from axial length and SE.^13,14 However, to our best knowledge, none of the clinical ophthalmic optical coherence tomography (OCT) devices takes the true scan diameter into account. Typically, manufacturers report to apply a fixed size of measurement/scan circle, for instance, for Cirrus HD-OCT (Carl Zeiss Meditec, Jena, Germany), the cpRNFLT on the standard measurement circle with 3.46-mm diameter is extracted from the volume scans without considering the variation of individual eye size.¹⁵

To our best knowledge, none of the clinical retinal OCT devices measures the true circle diameter on the retina. Heidelberg Spectralis internally estimates the circle size in mm based on the focus settings adjusted by the device operator prior to the scan.¹⁴ This estimated diameter is not part of the printout and therefore not easily accessible to clinicians, but it can be electronically exported from the machine.

In this study, we will leverage a large population-based study to investigate the relationships between age, the scan diameter reported by the employed Spectralis OCT, and the full pointwise cpRNFLT profile in participants with healthy eyes.

2.

Methods

This investigation is part of the Leipzig Research Centre for Civilization Diseases (LIFE) adult study.¹⁶ The LIFE adult study was approved by the institutional ethics board of the Medical Faculty of Leipzig University and adheres to the Declaration of Helsinki. The written informed consent was obtained from all participants.

2.2.

Description of Scan Diameter

The Heidelberg Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany) device used in this study specifies a fixed circle diameter of 12 deg in the device settings, projected onto the retinal plane, as shown in Fig. 1(a), which corresponds to approximately a 3.5-mm diameter scan circle size on the retinal plane based on the standard eye model employed in the machine.^14,17 While the Spectralis machine always projects a fixed measurement circle into the eye, the size of the scan diameter on the retina varies with individual ocular parameters, particularly axial length and corneal curvature. For example, the actual scan diameter can be larger for eyes with larger axial length and smaller for eyes with shorter axial length, as illustrated in Fig. 1(a). Within the Spectralis OCT, conversion of this 12 deg circle scan into mm is achieved by calculations based on defocus correction and corneal curvature, as detailed in the work by Garway-Heath et al.¹⁴

Fig. 1

(a) Schematic illustration of scan diameter variation with respect to axial length of the eye and (b) schematic illustration of the cpRNFLT measurement circle around the ONH including the effect of varying circle diameters.

Figure 1(b) illustrates the cpRNFLT circle, the coordinate system used in this study, and the effect of varying circle diameters. Superimposed on the ONH-centered infrared fundus image of the right eye of a 42-year old participant, a typical cpRNFLT measurement circle (solid circle) is depicted and an illustrative larger circle (dashed circle) is overlaid. In addition, the schematic trajectories of the cpRNFLT peaks are illustrated (curves in superior/inferior temporal direction). These trajectories are correlated to the major temporal-superior/temporal-inferior arteries. As illustrated, the nonlinear curvy trajectories of the main nerve fiber bundles cause relative shifts of the angular positions of the cpRNFLT peak locations (illustrated by asterisks) if the measurement circle diameter changes. In addition, cpRNFLT is naturally thinner with increasing diameter for myopes due to ocular elongation.^10,11

In this study, we employed the scan diameter as a measure of ocular magnification. The scan diameter in mm measured by Spectralis OCT is based on the scan focus parameters as a surrogate measure of refractive error and standard corneal curvature, as detailed in the work by Garway-Heath et al.¹⁴ Within the current investigation, we deliberately employed only parameters routinely available from the Spectralis OCT device to ensure that our results can immediately be applied to all Spectralis OCT data without requiring additional tools or machines.

3.

Results

From 17,974 eyes of 9069 participants, 3329 participants were excluded in total due to any of the following exclusion criteria (or a combination of them): any clinically significant findings based on prior ophthalmic examination of fundus images and obtained OCT scans (2676 subjects), nongradable images due to poor image quality (2 subjects), a glaucoma diagnosis reported in the interview (613 subjects), the presence of glaucoma medication among all the medications the participant took at the time of the test (38 subjects). The 188 eyes from 94 participants were excluded due to unreliable measurements. After all data selection procedures, 5646 eyes from 5646 participants were selected for data analyses. The $\text{mean}\pm \text{standard deviation}$ of age was $54.9\pm 12.1\text{years}$. There were 2553 male and 3093 female participants.

The detailed ethnic proportion of the participants in LIFE study can be found in Table 1. The subjects of the current investigation are predominantly Caucasian (99.22%). The recruitment of participants in the LIFE study was performed by randomly selecting Leipzig residents from the city registration office. Citizens were sent invitation letters, and study participation depended on response. In the years of the data acquisition, the east German city of Leipzig had a low percentage of foreign residents (between 5% and 10%), most of them from European countries. Among the very small proportion of residents of non-European race, the response rate was extraordinarily low, particularly due to language issues. Therefore, the percentage of non-Caucasian participants in the current data sample is negligible (0.78%), and a race-specific data analysis was not feasible.

Table 1

The ethnic proportion of the participants in LIFE study.

Figures 2(a) and 2(b) show the distributions of age and scan diameter over all subjects, respectively. Since the LIFE adult study focuses on the age group of 40 to 75 years, as detailed above, there were fewer subjects with age under 40 years. The trend that the number of subjects decreased with increasing age was due to excluding participants with clinical signs on OCT or fundus images or nongradable images, as described above, or due to excluding patients with diagnosed glaucoma or possible optic neuropathy, as shown in Fig. 2(a), since there was a higher proportion of those cases for older subjects. As shown in Fig. 2(b), the scan diameter was normally distributed with peak at the 3.5-mm diameter, which was the scan circle size for the standard eye model used on the machine with corneal radius set to be 7.8 mm and scan focus to be an individual measure of refractive error.

Fig. 2

The distribution of (a) age and (b) scan diameter over all subjects. The LIFE adult study focuses on the age range of 40 to 75 years. While the Spectralis machine projects a measurement circle with a fixed radius of 12 deg into the eye, the size of the scan diameter on the retina varies with individual ocular parameters, particularly axial length and corneal curvature.

The scan diameter was mildly and inversely correlated to age ($r=-0.30$, $p<0.001$). The results of segmented regression from age to scan diameter suggested that the scan diameter was differently related to age for age group younger/older than 42.1 years old, which was detected as a break point for age by the segmented regression. For subjects under 42.1 years, there was no significant correlation between age and scan diameter ($r=-0.04$, $p=0.27$), whereas the correlation was $-0.28$ ($p<0.001$) for subjects over 42.1 years. In addition, the global RNFLT was inversely correlated to scan diameter ($r=-0.25$, $p<0.001$) as expected. In comparison, the global RNFLT was inversely correlated to age ($r=-0.12$, $p<0.001$).

Table 2 shows the age slopes and $p$ values obtained from (1) “univariate regression” from age to global and quadrant-specific cpRNFLTs compared to the slopes and $p$ values of age and scan diameter obtained by (2) “multivariate regression” from age and diameter to global and quadrant-specific cpRNFLTs. Global cpRNFLT increased with decreasing scan diameter. Without considering the effect of scan diameter, the age slope was substantially underestimated ($-0.09\mu \mathrm{m}/\text{year}$) compared the age slope with adjusted effect of scan diameter ($-0.17\mu \mathrm{m}/\text{year}$). Sectoral cpRNFLT decreased with increasing scan diameter except for the temporal cpRNFLT. Accordingly, the age slopes were mostly underestimated when not including the effect of scan diameter except for temporal cpRNFLT, for which the age slope was overestimated without including the effect of scan diameter. In addition, for the nasal sector, the age slope shifted from insignificant thickening to significant thinning after adjusting for the scan diameter effect.

Table 2

The age slopes and p values with (1) “univariate regression” from age to global and quadrant-specific cpRNFLTs in comparison to the slopes and p values of age and scan diameter with (2) “multivariate regression” from age and diameter to global and quadrant-specific cpRNFLTs. The table is to demonstrate how the ocular magnification can bias the estimation of age-related thinning rate of global and quadrant cpRNFLT as most previous works have not included ocular magnification in their cpRNFLT norm model.

Table 3 shows the age slopes and $p$ values obtained by univariate regression from age to clock-hour-specific RNFLTs compared to the slopes and $p$ values of age and scan diameter obtained by multivariate regression. Mean cpRNFLT for clock-hour 8, 9, 10, and 11 increased with decreased scan diameter while the mean cpRNFLT for all other clock-hour sectors decreased. The sector with the strongest scan diameter effect was clock-hour 6. In addition, there were three clock-hour sectors (clock-hour 3, 4, and 6) with positive age slopes when not adjusting for the scan diameter effect, which were counter-intuitive if we assume that there is no age-related thickening. After adjusting the effect of scan diameter, all age slopes were negative, as expected. The fastest age-related thinning was in clock-hour 8 and 7 before and after adjusting the scan diameter effect.

Table 3

The age slopes and p values with univariate regression from age to clock-hour-specific RNFLTs in comparison to the slopes and p values of age and scan diameter with multivariate regression from age and diameter to clock-hour-specific RNFLTs.

Figure 3 shows the spatial profile of the mean cpRNFLT for (a) younger (35 to 45 years) and older (55 to 65 years) age groups and (b) larger ($\ge 3.7\mathrm{mm}$) and smaller ($\le 3.4\mathrm{mm}$) scan diameter groups for subjects, respectively. To decouple the interactive effects between age and scan diameter on mean cpRNFLT: for age group comparison, we restricted the scan diameter to 3.5 mm, which was the most frequent diameter among the participants; for scan diameter group comparison, we restricted the subject age to be under 42.1 years, as we did not find a significant correlation between diameter and age for this age group (see above). Overall, the cpRNFLT peaks of the older group (555 subjects between 55-65 years) are more nasal than those of the younger group (454 subjects between 35 to 45 years). The superior and inferior peaks of the older group were located at 2.80 deg ($p=0.01$) and 2.83 deg ($p<0.001$) nasal to the younger group. The cpRNFLT peaks of the smaller scan diameter group were more nasal than the larger scan diameter group. The superior and inferior peaks of the smaller scan diameter group (97 subjects) were located at 9.77 deg ($p=0.006$) and 7.83 deg ($p=0.002$) nasal to the larger scan diameter group (194 subjects). There was no significant group age difference ($p=0.41$) between the smaller (37.1 years) and larger (36.2 years) scan diameter groups.

Fig. 3

The spatial profile of the mean cpRNFLT for (a) younger (454 subjects between 35 and 45 years) and older (555 subjects between 55 and 65 years) age groups for subjects with 3.5 mm scan diameter and (b) larger ($\ge 3.7\mathrm{mm}$) and smaller ($\le 3.4\mathrm{mm}$) scan diameter groups for subjects under 42.1 years, respectively.

Figures 4(a) and 4(b) show the spatial profile of age and scan diameter slopes to predict cpRNFLT in comparison to the spatial profile of mean RNFLT for all age subjects, respectively. The superior and inferior locations with largest age-related thinning rates were positioned at 9.76 deg and 3.78 deg ($p<0.001$ for both, bootstrapping) temporal to the mean superior and inferior RNLFT peaks. Conversely, the superior and inferior locations with cpRNFLT, which presented with the steepest change for a decrease in scan diameter were positioned at 15.32 deg and 30.65 deg ($p<0.001$ for both, bootstrapping) nasal to the mean superior and inferior RNLFT peaks.

Fig. 4

The spatial profile of (a) age slope and (b) scan diameter slope to predict cpRNFLT in comparison to the spatial profile of mean cpRNFLT for all subjects. This figure is the visualization of our pointwise cpRNFLT model based on age and scan diameter, which is described in detail in Table 4 in the Appendix.

This detailed cpRNFLT results based on over 5600 participants allow a comparison with the cpRNFLT norms of the Spectralis OCT device (based on only 201 eyes of 201 participants), which are employed for the standard clinical printout. Here, we illustratively perform this comparison for two cases. Figures 5(a) and 5(b) show the spatial profile of projected mean cpRNFLT norms based on our age and scan diameter model for a 47-year old subject with 3.5-mm scan diameter and a 75 years subject with 4.4-mm scan diameter in comparison to the Spectralis OCT machine norms, respectively. Those two exemplary cases were chosen to represent the variation of age and scan diameter. For the superior temporal region of the 47-year old participant with 3.5-mm scan diameter, our projected mean cpRNFLT was thinner than the machine norms, and the superior peak of our projected norms was more nasal than the machine norms, which might be explained by our findings in Fig. 4(a) that the fastest age-related thinning locations were temporal to the mean cpRNFLT peaks. The difference between our projected norms and machine norms might also be caused by the possibly higher resolution in our study, as we calculated pointwise models for all 768 measurement points. The considerably lower population size for the Spectralis norms might not have allowed such a pointwise approach. For the 75-year old subject with 4.4-mm scan diameter, our projected norms were substantially thinner than the machine norms and the mean cpRNFLT peak locations of our projected norms were more nasal compared to the machine norms, which might be mostly explained by the current findings in that larger scan diameter corresponding to thinner cpRNFLT and more temporal cpRNFLT peaks, as shown in Figs. 3(b) and 4(b).

Fig. 5

The spatial profile of age-specific cpRNFLT norms for (a) a 47-year-old subject with 3.5-mm scan diameter and (b) a 75-year-old subject with 4.4-mm scan diameter. For this graphical comparison, our projected norms (thick blue lines) were overlaid on the printouts of the two example participants. The thin black lines denote the actual measurement values of the subjects and are ignored here. Instead, our comparison focuses on the dark green lines, which denote the medians of the machine norms for these two ages.

To enhance comparisons with previously published age-specific cpRNFLT norms or the norms used by the different OCT machines, all of which ignored effects of scan diameter, we additionally provide the cpRNFLT quantiles of the study participants for different age decades, as shown in Fig. 6. Finally, this study provides the numerical coefficients for the multivariate regression models of all 768 measurement points in the Appendix, which may help scientists or manufacturers to apply both age and diameter corrections to cpRNFLT measurements.

Fig. 6

The spatial profile of cpRNFLT at different quantiles for different age groups: (a) 19 to 30 years old, (b) 30 to 40 years old, (c) 40 to 50 years old, (d) 50 to 60 years old, (e) 60 to 70 years old, and (f) 70 to 80 years old.

4.

Discussion

In this study, we not only verified the strong negative correlation between age and cpRNFLT at an unprecedented level of detail based on over 5600 participants of a population-based study but also demonstrated detailed location-specific effects of the scan diameter independent of as well as in relation to age. Our results with a spatial resolution of 768 measurement points show that both parameters are strongly location-specific, which could not be sufficiently described by previous studies that focused on global averages or coarsely defined sectors. In particular, increasing age and decreasing scan diameter are related to cpRNFLT profiles with more nasalized superior and inferior peaks. The very detailed graphical and numerical presentation of our results in the Appendix may help researchers and manufacturers to improve their normative cpRNFLT models.

A further central finding of this study is the significant “correlation between age and scan diameter” for participants of age 42.1 or older. As cpRNFLT decreases with age as well as with increasing scan diameter commonly reported in literature, the systematic decrease of the diameter with age for participants older than 42.1 years implies that the true age-specific cpRNFLT thinning effect is underestimated for elderly individuals if the values are not corrected for diameter. In addition, there are implications for clinical practice, as current disease progression models for optic neuropathies like glaucoma are typically based on cpRNFLT slopes over time. Our results demonstrate, for instance, that in the absence of diameter correction, which is current clinical practice, a slope that indicates glaucomatous nerve fiber layer thinning in a young patient will be steeper than the slope for an elderly patient, even if both patients suffer from exactly the same level of thinning, because the elderly patient is additionally subject to a decrease of the scan diameter over time.

Previous works typically explained differences in scan diameter with axial ametropia.^10,11 For axial myopia, the extension in ocular length yields larger diameters of the projection of the circle on the retina, and cpRNFLT correction models have been proposed based on ocular axial length.¹⁰ This effect, however, is unlikely to explain the relationship between diameter and age for elderly participants of the current study, as there are no indications that axial length systematically decreases at higher age. Instead, we assume this relationship to be caused by lens-related effects. Elderly individuals are not only subject to the well-known effect of presbyopia, which concerns near vision, but also to previously documented hyperopic shifts for distance vision.^20,21 These lens-related phenomena might explain the systematic decrease in scan diameter, which is otherwise typical for axial hyperopia in younger individuals.

Regarding the particular value of the results of the current investigation compared to the previous studies on ocular magnification and cpRNFLT,^10,11 two aspects need to be highlighted. First, our considerably larger sample size of over 5000 subjects substantially improves the reliability of the normative results and furthermore allows an unprecedented level of detail in terms of spatial resolution (measurements at 768 locations on the circle). In the previous studies, fewer than 300 subjects participated (e.g., 269 subjects in the work by Kang et al.¹⁰ and 45 subjects in the work by Savini et al).¹¹ Our large age-stratified sample allows us to cover a wide age range from 19 to 80 years, whereas the age range in the work by Kang et al.¹⁰ was restricted between 19 and 26 years. Note that optic neuropathies predominantly affect elderly people, which further stresses the additional benefit of the current results. Second, and more importantly, our study adjusts the effects of age and ocular magnification simultaneously. Only by this, we were able to show that age and scan diameter interact for elderly populations, so that adjusting the effects of axial length alone might not be sufficient for the generation of new cpRNFLT norms. This finding, which previous studies were unable to detect, is of special importance for scientists and OCT manufacturers alike.

As all participants who reported to have been diagnosed with glaucoma were excluded, regardless of whether they took glaucoma medication or had any clinically significant findings on their fundus/OCT images or nongradable images, participants who erroneously reported having glaucoma would have been excluded in our study as well. The strict exclusion criteria ensured to discard any subject with possible confounding diseases.

We did not exclude subjects based on cataract. First, cataract does not pathologically affect the cpRNFLT, and second, by excluding all cataract patients, we would unnecessarily bias our normative models. This is of particular importance, as due to the high cataract prevalence in elderly populations, a large proportion of patients presenting at clinical glaucoma services for cpRNFLT measurement will have cataracts, so that cpRNFLT norms should include cataract patients to account for the fact of its frequent occurrence in the total population to avert sampling bias.

This investigation also has limitations. First, the cross-sectional design of this study does not allow tracking changes of the scan diameter over age on an individual base. This will be made possible based on follow-up measurements on the same subjects in future. Second, the dataset contains an imbalanced gender ratio. Potential gender effects should be further investigated as part of future analyses. Third, the current study is based on subjects that are predominantly Caucasian, which restricts the applicability of extracted norms to Caucasian populations, and future studies are needed to investigate how generalizable the results are with respect to other ethnicities. However, it has to be noted that the current normative dataset of the Spectralis SD-OCT device approved by the US Food and Drug Administration is also solely based on a Caucasian population. Last, as the scan diameter reported by the Spectralis machine in mm for each subject is estimated based on the scan focus setting made by the operator and standard corneal curvature with a default value of 7.8 mm for mean anterior corneal radius, two further limitations of Spectralis scan diameter calculation also apply to the current dataset: (a) the manual setting of the focus is subjective and may vary between operators and (b) corneal radius is not measured on an individual basis in the current study.

Axial length was not measured in the course of the LIFE study due to capacity limitations, so it was not available for the current data analysis. The availability of axial length data would extend the scientific utility of the current work. However, none of the clinically used OCT machines currently measures axial length, so its inclusion, while interesting for the scientific community, would probably not be of immediate relevance to the improvement of cpRNFLT norms of current OCT devices. Instead, the current analysis employs scan diameter as the measure of ocular magnification, which is related to axial length and is immediately available from the OCT machine. This scan diameter is routinely calculated by the machine from the scan focus, based on a specific eye model.¹⁴ While the major part of scan diameter variance is explained by axial length, it additionally contains lens-related aspects not covered by axial length, and the current investigation indeed found a significant relationship of scan diameter and age, which could not have been revealed by including axial length alone.

The current work is of immediate relevance for scientists, engineers, and OCT manufacturers. The investigation of the impact of age and ocular magnification (scan diameter) on a clinically very important measurement protocol (cpRNFLT) has an unprecedentedly high spatial resolution (768 measurement locations on the scan circle), which can be achieved due to this extraordinarily large sample size (5646 subjects). While this study clearly has clinical utility by demonstrating the considerable impact of these parameters, the main consequence of this work targets research into the engineering of the OCT device itself and its inherent norms. The current work clearly demonstrates that: (a) the location specificity of existing age norms used in the Spectralis OCT machine, based on only 201 subjects,¹⁷ can be significantly improved by the presented detailed, high resolution model’ (b) the location specificity and the quantitative effect of the scan diameter is at least as substantial as the effect of age. Diameter effects are currently entirely neglected by OCT machines, and while they have previously been discussed in the literature, although only indirectly in the context of axial length, a location-specific quantification of the effect and the introduction of detailed normative distributions (detailed in Table 4 in the Appendix) are novel, to our best knowledge; (c) we demonstrate that scan diameter decreases with age, which is also, to our best knowledge, an entirely novel finding. One of the conclusions of this newly reported relationship, particularly important for OCT manufacturers, is that it is insufficient to base personalized cpRNFLT norms on axial length, as suggested by numerous previous works. Axial length remains constant in elderly subjects, but the scan diameter nevertheless systematically decreases, which indicates lens-related explanations; and (d) normative models are of particular importance for medical device manufactures with a need to be integrated into OCT machines. The quantitative presentation of our results and the particular level of numerical and statistical details is aimed at research scientists of ophthalmic companies and allows not only the support of the generation of novel cpRNFLT norms, which is the main purpose of our study, but also subsequently the implementation of diagnostic models for disease progression by comparing these normative cpRNFLT slopes to individual slopes of patients, etc.

To summarize, in this work, we study the interrelationship between cpRNFLT, age, and scan diameter at an unprecedented level of detail. These results may help researchers and manufacturers to improve their normative cpRNFLT models. Besides, we reveal a systematic relationship between age and diameter, which is restricted to elderly people based on our results. This finding likely has clinical relevance, as current disease progression methods to quantify cpRNFLT thinning over time are based on slopes, which are assumed to be independent of the baseline age of patients. Therefore, pathological cpRNFLT thinning might be underestimated in elderly patients compared to younger patients.

Table 4

The age slope, scan diameter slope, and intercept of the multivariate regression model with accompanying p values for all 768 measurement points on the scan circle with starting point on the horizontal line towards the temporal direction, as illustrated in Fig. 1.