Choroidal signal hypertransmission on optical coherence tomography imaging: Association with development of geographic atrophy in age‐related macular degeneration

To examine the association between large choroidal signal hypertransmission ≥250 μm (LHyperT) on optical coherence tomography (OCT) with the risk of developing geographic atrophy (GA) and compare this risk with those associated with nascent geographic atrophy (nGA).


| INTRODUCTION
A major challenge in the evaluation of new interventions to slow or prevent the development of late age-related macular degeneration (AMD) complications in its early stages is the need for large and lengthy trials given the slowly progressive nature of this disease. 1,2Biomarkers that could be used to identify high-risk individuals to target in such trials, and/or act as earlier surrogate endpoints for progression to the conventional, gold-standard clinical endpoint of geographic atrophy (GA), could greatly improve the feasibility of such early intervention trials.
4][5] One previous study reported that the presence of nGA at baseline was associated with a 56-fold increased risk of developing GA in a cohort with an average follow-up duration of approximately 5 years, 6 and we showed in a recent prospective study that nGA was associated with a 78-fold increased rate of developing colour fundus photography (CFP)-defined GA. 7 We also recently found that nGA accounted for the observed association between incomplete retinal pigment epithelium and outer retinal atrophy (iRORA)-another recently proposed OCT imaging biomarker for risk of atrophy development [8][9][10][11] -and GA development. 12Collectively, these findings suggested that nGA might be a useful surrogate endpoint for the traditional clinical endpoint of GA. 13 Another imaging-based biomarker that may be useful for these purposes is the presence of choroidal signal hypertransmission, which has been observed in the development of atrophic AMD. 3,7,9,14,15One recent study suggested that the presence of choroidal signal hypertransmission with a greatest linear dimension (GLD) ≥250 μm on an en face sub-retinal pigment epithelium (RPE) slab visualised on an OCT scan pattern with a high transverse resolution (12-μm between B-scans)-termed a 'persistent hypertransmission defect (hyperTD)'-was associated with a significantly increased risk of atrophy development. 14OCT volume scans in clinical studies and practice are also often obtained using a scan protocol with a lower transverse resolution (usually with B-scans spaced 60-to 120-μm apart), [16][17][18][19] as such a scan protocol focusses on obtaining a fewer number of high-quality horizontal B-scans with a higher lateral resolution and lower speckle noise (through frame averaging).It remains to be determined whether choroidal signal hypertransmission ≥250 μm detected on such an OCT volume scan protocol are also associated with a significantly increased risk of GA development, and how this risk compares with the specific features used to define nGA.It also remains to be determined whether choroidal signal hypertransmission ≥250 μm detected on horizontal B-scans using the above OCT volume scan protocol reliably captures the presence of 'persistent hyperTDs' seen on en face OCT images.This study thus sought to address these abovementioned key knowledge gaps.

| METHODS
As an overview, this study was conducted in two phases, with the first being a cross-sectional evaluation to examine if en face 'persistent hyperTDs' on swept-source OCT (SS-OCT) imaging could be reliably detected based on the presence of choroidal signal hypertransmission ≥250 μm in width on horizontal spectral-domain OCT (SD-OCT) B-scans (termed 'large choroidal signal hypertransmission' [LHyperT] in this study).The second phase involved a retrospective review of longitudinal imaging data from participants 20,21 in the sham treatment arm of the Laser Intervention in the Early Stages of AMD (LEAD) Study, 22,23 which included additional visits from those who were enrolled in an observational extension study to the LEAD Study. 24Institutional review board approval was obtained at all sites for both of these studies, which were also conducted in accordance with the International Conference on Harmonisation Guidelines for Good Clinical Practice and with the tenets of the Declaration of Helsinki.The full details of the study methodology are described in the Supplementary Materials, but its key aspects are described below:

| Cross-sectional cohort
The cross-sectional cohort included individuals aged 50 years or older that had at least one eye with at least one large druse (>125 μm) without late AMD.Macular-centered CFPs, SD-OCT and SS-OCT scans were obtained in these individuals.SD-OCT imaging was performed using the Spectralis HRA + OCT device (Heidelberg Engineering GmbH, Heidelberg, Germany) with a 20 Â 20 volume scan consisting of either 49 or 97 horizontal B-scans, each with 25 and 16 frames averaged respectively and 1024 A-scans.SS-OCT imaging was performed using the PLEX Elite 9000 device (Carl Zeiss Meditec; Dublin, CA), covering a 6 Â 6 mm region and consisting of 500 horizontal B-scans, each with two frames averaged and 500 A-scans.SS-OCT scans were graded for the presence of 'persistent hyperTDs' at the eye level-a region of choroidal signal hypertransmission on an en face sub-RPE slab (between 64 to 400 μm below Bruch's membrane) with a GLD ≥250 μm 14,[25][26][27] -following initial training.SD-OCT horizontal B-scans were then also graded for the presence of LHyperT at the eye level.Two examples of 'persistent hyperTDs' and LHyperT are shown in Figure 1.The between-method (SS-OCT vs. SD-OCT) inter-reader agreement for detecting hypertransmission ≥250 μm was compared to the inter-reader agreement for its detection on SS-OCT based on Gwet's first order coefficient (AC 1 ) 28 using a non-parametric bootstrap procedure.

| Longitudinal cohort
The longitudinal cohort included individuals aged 50 years or older that had at least one large druse (>125 μm) and without multimodal imaging (MMI)-defined late AMD (including nGA) in both eyes.Macular-centred CFPs and SD-OCT imaging was performed in this cohort, and SD-OCT volume scans consisted of 49 horizontal B-scans, each with 25 frames averaged and 1024 A-scans.Participants were seen at six-monthly intervals up to a 60-month period, and the presence of LHyperT, nGA, and GA at the eye level was graded at each visit.Another example of LHyperT, and an example of nGA, is shown in Figure 2. Time-to-event analyses were used to determine the incidence of LHyperT and nGA, and their progression to CFP-defined GA between 12-and 36-months (due to the limited number at risk beyond the 36-month timeframe).The incidence and progression of LHy-perT and nGA was also compared using a non-parametric bootstrap resampling approach.The proportion of variance explained in the time to develop GA (survival R 2 ) by LHyperT and nGA was also derived and compared using the same non-parametric bootstrap resampling approach above.

| Comparison of methods for detecting hypertransmission ≥250 μm on OCT
A total of 100 eyes with large drusen from 85 participants who had undergone imaging on both SD-OCT and SS-F I G U R E 1 Detection of 'persistent hypertransmission defects (hyperTDs)' on swept-source optical coherence tomography (SS-OCT) and large choroidal signal hypertransmission (≥250 μm; LHyperT) on spectral-domain optical coherence tomography (SD-OCT) imaging in two examples (A and B) in the cross-sectional cohort of this study.On SS-OCT, the en face sub-retinal pigment epithelium (RPE) slab (between 64 and 400 μm below Bruch's membrane; top of each example) was used to detect 'persistent hyperTDs' (having a greatest linear dimension of ≥250 μm; the red arrow indicates one such lesion in each example).On SD-OCT, horizontal B-scans (bottom of each example, with the vertical position of the B-scan in the retina indicated by the red arrow on the SS-OCT en face image) were used to detect LHyperT (the extent of the hypertransmission is outlined by the white double arrow in each example).
OCT were included in the cross-sectional cohort for a comparison of the methods for detecting choroidal signal hypertransmission ≥250 μm.These participants were on average 72 ± 8 years old (range, 51-89 years old) and 63 (74%) were female.Note that 67 (67%) and 33 (33%) eyes had undergone SD-OCT imaging with a volume scan protocol that consisted of 49 B-scans with ART averaging of 25 frames and 97 B-scans with ART averaging of 16 frames respectively.
The level of inter-reader agreement for the detection of hypertransmission ≥250 μm on SS-OCT en face images (as 'persistent hyperTDs') and SD-OCT B-scans (as LHyperT; AC 1 = 0.64) was similar and not significantly different from the level of inter-reader agreement observed for the detection of 'persistent hyperTDs' on SS-OCT en face images (AC 1 = 0.69; p = 0.297).These findings are summarised in Table 1, and they suggest that the detection of LHyperT on SD-OCT B-scans was comparable with the detection of 'persistent hyperTDs' on SS-OCT en face images.Two examples where both graders considered LHyperT on SD-OCT B-scans and 'persistent hyperTDs' on SS-OCT en face images to be present are shown in Figure 1, and two further examples are shown in Supplementary Figure 1.Two further examples where both graders considered LHyperT to be present on SD-OCT B-scans and 'persistent hyperTDs' to be absent on SS-OCT en face images are also shown in Supplementary Figure 1.Note that there were no cases in study where both graders considered 'persistent hyperTDs' to be present on SS-OCT en face images and LHyperT to be absent on SD-OCT B-scans.
There was also no significant difference in the level of inter-reader agreement between these two methods and the inter-reader agreement on SS-OCT en face images when evaluating only the subset of eyes that underwent SD-OCT imaging with a volume scan protocol consisting of 49 B-scans (AC 1 = 0.64 and 0.72, respectively; p = 0.184), indicating that the detection of LHyperT on SD-OCT volume scans with a 120-μm B-scan spacing was comparable with the detection of eyes with 'persistent hyperTDs' seen on SS-OCT en face images (with a 12 μm B-scan spacing).

| Longitudinal cohort characteristics
In the longitudinal cohort, 280 eyes from 140 participants with bilateral large drusen at baseline were included.These participants were on average 70 ± 8 years old (range, 51-89 years old) and 108 (77%) were female.A total of 134 (96%) participants were seen at the 3-year follow-up visit, and 73 (52%) participants were seen at the 5year follow-up visit.There were 17 (6%) eyes from 16 (11%) participants who developed nAMD during the follow-up they were censored at the day of its development.There were 27 (10%) eyes from 20 (14%) participants who developed GA over the follow-up period.

| Incidence of large hypertransmission on OCT
In this cohort without nGA at baseline (as this was part of the exclusion criteria of the longitudinal study), 13 (5%) eyes from 12 (9%) participants were graded as having LHyperT at baseline.The three-and five-year probabilities of developing LHyperT were 23% (95% confidence interval [CI] = 17%-29%) and 37% (95% CI = 29%-46%) respectively, and they were 16% (95% CI = 11%-22%) and 27% (95% CI = 20%-35%) respectively for nGA.These probabilities at annual intervals are also presented in Table 2, demonstrating a significantly higher probability of developing LHyperT than nGA at all time points (p ≤ 0.019).Amongst the 63 eyes that developed nGA over the follow-up, there were 38 (60%), 12 (19%) and 8 (13%) eyes where LHyperT was detected at a prior, the same, or a subsequent visit, respectively, and 5 (8%) eyes where LHyperT was not detected over the follow-up.Additionally, there were 35 (38%) eyes from the 93 eyes that had or developed LHyperT where nGA was not detected over the follow-up.

| Progression of large hypertransmission to geographic atrophy
Amongst the 27 eyes that developed GA, there were 26 (96%) eyes where LHyperT was detected at a prior visit and one (4%) eye where LHyperT was detected at the same visit (and thus all eyes that developed GA had LHy-perT at or prior to its development).Similarly, nGA was detected at a prior visit in 25 (93%) eyes and at the same visit in 1 (4%) eye.The only eye that developed GA without evidence of nGA at a prior or same visit has been presented previously, 7 and occurred under the fovea with the GA lesion corresponding to an unusual RPE detachment associated with choroidal signal hypertransmission.For the 25 eyes that developed GA where LHyperT and nGA were both detected at a prior visit, LHyperT was detected before nGA in 17 (68%) eyes, at the same visit in 5 (20%) eyes, and at a subsequent visit in the remaining 3 (12%) eyes.
The probability of progression to GA after 24 months in eyes with prevalent or incident LHyperT was 17% (95% CI = 8%-27%), which was significantly lower than the probability of 40% (95% CI = 23%-56%) for eyes with incident nGA (p = 0.002).These probabilities at sixmonthly intervals from 12-to 36-months are also presented in Table 2, and they show how the probability of progression to GA was significantly lower for LHyperT compared to nGA at all time points (p ≤ 0.002).

| DISCUSSION
This study showed that eyes with hypertransmission ≥250 μm, seen as 'persistent hyperTDs' detected on SS-OCT en face sub-RPE slab images, could also be detected by assessing SD-OCT horizontal B-scans for LHyperT, from a scan protocol with a lower transverse resolution.
Evaluating LHyperT from SD-OCT horizontal B-scans in a longitudinal AMD cohort seen up to a five-year period, LHyperT and nGA were both found to be significantly associated with a highly increased rate of progression to GA.However, the rate of progression to GA for eyes with LHyperT was significantly lower than eyes with nGA across multiple time points examined in this study (between 12 and 36 months), being 17% and 40%, respectively, after 24 months for example.LHyperT also explained a significantly lower proportion of variance in the time to develop GA when compared to nGA.These findings collectively suggest that both LHyperT and nGA are features that are strongly associated with GA development, but with nGA being a slightly more robust biomarker and potential surrogate endpoint for the clinical endpoint of GA in early intervention trials.This is the first study to our knowledge that has shown that eyes with hypertransmission ≥250 μm, detected as 'persistent hyperTDs' (or regions of choroidal signal hypertransmission with a GLD ≥250 μm on sub-RPE en face slab images) detected using an SS-OCT scan protocol with a high transverse resolution (12 μm spacing between B-scans), could also be detected based on the presence of LHyperT using an SD-OCT scan protocol with a lower transverse resolution (≤120 μm spacing between B-scans).This was determined by demonstrating that the level of inter-reader agreement for detecting eyes with hypertransmission ≥250 μm on SS-OCT en face images ('persistent hyperTDs') and SD-OCT B-scans (LHyperT) was not significantly different from the interreader agreement when using the former alone.Although not directly comparable, a recent study by Corvi and colleagues 29 also showed that 98% of complete RPE and outer retinal atrophy (cRORA) lesions identified on SD-OCT B-scans using a protocol with a transverse resolution of 60 μm and 9 frames averaged spatially co-localised with 'persistent hyperTDs' on SD-OCT en face images using a different device and a scan protocol with a transverse resolution of 47 μm and without frame averaging.Note that all cRORA lesions will have LHyperT, since cRORA is defined by the presence of hypertransmission ≥250 μm associated with RPE attenuation or disruption ≥250 μm and evidence of overlying photoreceptor degeneration (and lesions with LHyperT that do not meet the criteria for cRORA would have iRORA). 8In another recent study by Corvi and colleagues 30 of the same cohort, they observed that the inter-reader agreement for the detection of cRORA using SD-OCT B-scans with the two different scan protocols and devices described above (intraclass correlation [ICC] = 0.97-0.98)was comparable with the intra-reader, between-device and scan protocol agreement (ICC = 0.94-0.98).These findings broadly support the notion from the findings of this study that 'persistent hyperTDs' on en face SS-OCT images could be similarly detected on horizontal SD-OCT B-scans using a protocol with a lower transverse resolution.
The inter-reader agreement for detecting 'persistent hyperTDs' on en face SS-OCT images in this study (AC 1 = 0.69) was markedly lower than those observed in a recent study by Liu et al. 14 (AC 1 = 0.97).This difference is likely explained by the inclusion of only eyes with 'persistent hyperTDs' in the study by Liu and colleagues, 14 compared to the inclusion of an unselected cohort of eyes with large drusen in this study.The inclusion of only eyes with 'persistent hyperTDs' in the study by Liu et al. 14 would be subject to confirmation bias, likely resulting in an over-optimistic estimate of inter-reader agreement than what would be expected in an unselected clinical cohort.The inter-reader agreement for detecting eyes with LHyperT on SD-OCT B-scans (AC 1 = 0.65) was instead more comparable with those observed in another previous study where 12 readers across six reading centers evaluated SD-OCT B-scans in a cohort of 60 eyes (AC 1 = 0.71; 95% CI = 0.62-0.80[unpublished data]; which was also comparable with the inter-reader agreement of nGA [AC 1 = 0.75]). 31Another recent study by Nanegrungsunk and colleagues 32 reported near-perfect inter-reader agreement for the detection of any hypertransmission ≥63 μm at the lesion level (AC 1 = 0.89), but lower agreement when categorising the size of the hypertransmission (based on lesions that were 63-124, 125-249 and ≥ 250 μm; AC 1 = 0.57).Re-analysis of the data presented in this study showed even higher levels of agreement for the presence of LHyperT at the lesion level (AC 1 = 0.97), although LHyperT were only present in 7% of all the readings due to the exclusion of eyes with cRORA. 32In contrast, 'persistent hyperTDs' was present in 36% of the readings on SS-OCT scans and LHyperT was present in 44% of the readings on SD-OCT scans in the cross-sectional cohort in this study.Similar to the abovementioned limitation of the study by Liu and colleagues, 14 the inclusion of a cohort in the study by Nanegrungsunk and colleagues 32 of eyes without cRORA may also be subject to confirmation bias (of the absence of LHyperT) and thus also result in over-optimistic estimates of inter-reader agreement.
The findings of this study in the longitudinal cohort that the presence or development of LHyperT detected on SD-OCT B-scans is associated with a 107.9-fold increased rate of developing CFP-defined GA is consistent with the suggestion from a previous study that 'persistent hyperTDs' are a high-risk precursor for GA development. 14This suggestion was based on the observations that 'persistent hyperTDs' were associated with an 80.4-fold increased rate of developing OCT-defined 'GA', defined as hypertransmission ≥250um with complete RPE and photoreceptor loss (based on the loss of the ONL and EZ), with the extent of both parameters being undefined. 14The inability to perform a masked assessment of the predictor and outcome (i.e., 'persistent hyperTDs' and OCT-defined 'GA' respectively) in this previous study can also result in confirmation bias.These limitations were addressed in this current study through the masked grading of OCT and CFP separately, and by evaluating the traditional clinical endpoint of CFPdefined GA.
This study also showed in the same longitudinal cohort that whilst LHyperT and nGA were both strongly associated with GA development, this association was stronger for nGA than LHyperT, as evident by the significantly higher proportion of variance in time to develop GA explained by nGA compared to LHyperT.The rate of progression to GA was also significantly higher for eyes with nGA compared to those with LHyperT at the same timepoint.Note that the progression rate to a clinical endpoint for two potential surrogate endpoints with the same prognostic significance could differ if one is systematically detected earlier than the other (i.e., is an earlier precursor), but both would show the same proportion of variance explained in the time to develop the clinical endpoint.Taken together, these findings suggest that nGA carries a greater prognostic significance for more imminent GA development than LHyperT, and may thus be a more robust surrogate endpoint for GA.The higher rate of progression to GA for eyes with nGA compared to LHyperT, despite its lower incidence over time, presumably reflects how nGA is likely a more specific OCT biomarker for atrophy development.Previous studies have also reported that not all regions of hypertransmission ≥125 μm persist over time. 26,32Whilst the specific patho-mechanisms for the non-persistence of choroidal signal hypertransmission is unclear, this mechanism could contribute to the lower specificity of LHy-perT for the subsequent development of GA observed in this study.
The limitations of this study include the evaluation of only individuals with bilateral large drusen and without nGA at baseline, and thus the generalizability of these findings to others with the early stages of AMD is not known.This study is also limited in its sample size and number of CFP-defined GA endpoints reached (n = 27), but the long-term follow-up of more than half of the cohort up to a five-year timepoint (and follow-up of almost the entire cohort up to 3 years) is a key strength of this study.Whilst the transverse resolution of the OCT volume scans used in the longitudinal study (approximately 120 μm) may be a limitation for detecting small nGA lesions, this study showed that such OCT volume scans enabled a comparable detection of hypertransmission ≥250 μm as OCT scans with a transverse resolution that was 10 times higher.However, a strength of the OCT volume scans used in the longitudinal study was its high lateral resolution (approximately 5.6 μm) and low speckle noise (achieved through averaging 25 frames), which would have been ideal for visualising the features used to define nGA, and also LHyperT.
In conclusion, this study demonstrated that both LHyperT and nGA were both high-risk features for progression to GA.However, nGA explained a significantly greater proportion of the variance in the time to develop GA than LHyperT, and it exhibited a significantly higher rate of progression to GA, when evaluated in the same longitudinal cohort.These findings suggest that nGA may thus be a slightly more robust surrogate endpoint for the traditional clinical endpoint of CFP-defined GA than LHyperT in intervention trials in the early stages of AMD.

F
I G U R E 2 (A) Large choroidal signal hypertransmission ≥250 μm (LHyperT) seen on a horizontal optical coherence tomography (OCT) B-scan (the extent of the hypertransmission in this example is outlined by the white double arrow).(B) Nascent geographic atrophy, defined by the presence of subsidence of the outer plexiform layer and inner nuclear layer (indicated by the white vertical arrow in this example) and/or presence of a hyporeflective wedge-shaped band within Henle's fibre layer (two seen in this example, indicated by white diagonal arrows).T A B L E 1 Comparison of the inter-reader agreement within and between methods for detecting hypertransmission ≥250 μm.Gwet's AC 1 p-Value*Within SS-OCT En Face 0.69 (0.54-0.83) -

ACKNOWLEDGEMENT
Open access publishing facilitated by The University of Melbourne, as part of the Wiley -The University of Melbourne agreement via the Council of Australian University Librarians.FUNDING INFORMATION This study was supported by National Health & Medical Research Council of Australia (project grant no.: APP1027624 [RHG], and fellowship grant no.: GNT1194667 [RHG], #2008382 [Z.W.]), grants from the Macular Disease Foundation Australia (ZW and RHG), and the German Federal Ministry of Education and Research (BMBF)funded BM-AXIS program [JHT].ZEISS provided the PLEX Elite 9000 instrument used in part of this study.CERA receives operational infrastructure support from the Victorian Government.The sponsor or funding organisation had no role in the design or conduct of this research.