Non-invasive testing for early detection of neovascular macular degeneration in unaffected second eyes of older adults: EDNA diagnostic accuracy study

The scale of the problem in the UK and the use of NHS resources

Neovascular age-related macular degeneration remains the most common cause of blindness and partial sight in the UK, despite improvements in treatments. 1,17 The incidence of AMD increases with age and, therefore, the burden is projected to rise steeply in future years as the population ages. Vision loss is associated with a profound impairment of quality of life, increased risk of falling, emotional distress, depression and an inability to care for self and for others. 18 Patients with bilateral vision loss may suffer from visual hallucinations (Charles Bonnet syndrome), poor sleep patterns and loss of confidence. Managing nAMD presents an enormous burden to the NHS. Ophthalmology accounts for 10% (5 million per year) of all outpatient attendances to the NHS and AMD accounts for 15% of all ophthalmology outpatient attendances. 1 This is because patients are typically seen every 2 months for up to 2 years after initiation of anti-VEGF therapy, and long-term studies from the UK show that some 50% of those who are commenced on treatment are still on active treatment or being followed up after 5 years. 19

Evidence for monitoring intervals/diagnostic performance

When active nAMD is confirmed by FFA, treatment with anti-VEGF therapy is initiated. 7,20,21 FFA is the gold-standard diagnostic test for nAMD. In the early phases of treatment (i.e. up to about 1 year), at each subsequent visit, which is usually on an 8-week cycle after the loading phase of 3-monthly dosing, patients are re-assessed to evaluate disease activity. Visual acuity, clinical biomicroscopic examination and OCT are the most commonly employed tests in the follow-up setting (see Figure 3). OCT-guided re-treatment decisions are the standard of care in almost all NHS units (see Figure 4); however, the combination of visual acuity, clinical examination and fluorescein angiography in selected cases are also used in the monitoring phase. 7,20 In the absence of disease activity on the OCT, treatment is withheld and a future review arranged, or the patient is treated and the next follow-up visit is extended.

Patients are also given an Amsler chart to self-monitor their disease and are advised to report immediately any signs of distortion or missing areas. The accuracy of this test compared with the above tests has not been systematically evaluated. Interrogation of the national ophthalmology data set, which is an amalgamation of the electronic records of some 14,500 patients who have received anti-VEGF treatments since 2009 (Usha Chakravarthy is a contributing member), shows that the average number of visits is 10 in year 1 and around eight in year 2. 22 The average interval between visits in year 1 is 35 days (± 10 days). The interval between visits increases in years 2 to 5. However, even in year 5, more than half of all people are on regular review and treatment. Thus, there is an opportunity to obtain information about unaffected fellow eyes of patients with nAMD in one eye to determine the optimum method of early detection of incipient nAMD.

New-onset nAMD responds extremely well to anti-VEGF therapy, with rapid disappearance of subretinal fluid and a more gradual resolution of fibrinous exudate and haemorrhage (see Figure 1). However, any delays can result in worsening of vision, which is mainly because of the development of fibrosis that involves the outer retina. The fibrous tissue replaces the normal cellular layers consisting of the choriocapillaris, the RPE and the photoreceptors. Furthermore, contraction of the fibrous tissue distorts both the adjacent outer retina and the overlying inner retina, changing the architecture and orientation of the cellular elements and resulting in severe visual loss. The initiation of treatment after the onset of fibrosis cannot reverse the vision loss but may stabilise visual acuity. Therefore, early detection of nAMD and treatment before the onset of fibrosis is clearly important.

In the example shown in Figure 3, the upper panels show a colour photograph of the fundus (Figure 3a), in which the site of the neovascular complex is indicated by an asterisk; blue light autofluorescence (Figure 3b); an early frame of the fluorescein angiogram (Figure 3c), which shows leakage and pooling of fluorescein dye; and ICGA (Figure 3d), in which the crescentic-shaped outline of the neovascular complex is visible. The OCT scan in Figure 3e) shows fibrinous exudation (shown by the arrow). Anti-VEGF treatment has resulted in the resolution of subretinal fluid (white arrow) and the hyper-reflective material has shrunk after 6 months, with the eye receiving monthly anti-VEGF (yellow arrow) (Figure 3f).

FIGURE 3.

The upper panels show typical features of nAMD using a variety of imaging techniques that are commonly used in clinical practice. (a) Colour fundus photography; the asterisk shows the site of the neovascular complex. (b) Blue light autofluorescence. (c) FFA image showing a distinct and well-delineated region of bright hyper-fluorescence at the site of the nAMD lesion. (d) ICGA image showing the nAMD lesion, but without leakage obscuring the vessel complex. (e) A highly actively leaking exudative lesion. (f) OCT image of the same eye taken 6 months later and showing resolution of the exudation. The bright RPE layer is draped over the shrinking neovascular complex.

Figure 4 shows an image of an eye in which treatment was initiated at a much later stage of evolution of the nAMD lesion. The normal organisation of the outer retina into distinct hyper-reflective bands consisting of Bruch’s membrane, the RPE and the photoreceptor layers has been lost. Instead, a band of horizontally oriented disorganised layers of hypo- and hyper-reflective material can be seen, which has been shown in previous studies to correspond to fibrous tissue which represents scarring. 23 Because layers of the outer retina have become incorporated into the scar, function is grossly diminished. The overlying inner retina has multiple cystic spaces and is distorted and thickened despite treatment with anti-VEGF agents. The visual acuity did not improve despite multiple treatments.

FIGURE 4.

OCT B-scan showing a band of hyper-reflective material that has replaced the outer retina and is indicative of a fibrotic scar. The inner retina contains many hyporeflective regions, which are cystic spaces.

As stated previously, there are many reasons why the reduction of visual acuity in one eye may not be detected by the sufferer. To unmask a change in vision owing to nAMD, which is usually seen as distortion in the central visual field, a test known as the Amsler chart has been used. This is a simple test that consists of a square grid with horizonal and vertical lines and a central dot as a fixation target printed onto a sheet of paper. Patients are asked to look at the central dot with only one eye (covering the fellow eye). In the absence of nAMD, there is no separation of the retinal layers and the lines on the chart are unaltered. When nAMD is present, the RPE separates from the outer retina, with fluid accumulation in the subretinal or intraretinal locations. This results in distortion of the grid lines and/or the appearance of scotomas (blank spots) in the Amsler chart. However, the Amsler test has previously been shown to have high false positive and false negative rates, which suggests that it may be a poor screening tool. Visual acuity checks have also been used to help detect nAMD, but there may be other causes for a reduction in vision in older age groups. OCT, on the other hand, provides non-invasive images of the retina and, with the reduction in price of such devices and their widespread adoption within the optometric community, this technology offers an alternative method of early detection. To date, to the best of our knowledge, the use of OCT to detect nAMD has not been tested against the gold standard of fluorescein angiography.

Aims and objectives of the study

The aim of the EDNA study was to identify the optimum non-invasive diagnostic test strategy that will robustly detect nAMD in fellow eyes during follow-up in secondary care of persons with nAMD in the first affected eye.

The primary objective was to determine the diagnostic monitoring performance (sensitivity and specificity) of five index tests for diagnosing nAMD against a primary reference standard of FFA determination of conversion to nAMD at the clinical site. The index tests considered were:

• Amsler – participants completing an Amsler chart

• fundus clinical examination – clinical evaluation of the fundus for signs of nAMD

• OCT – clinical assessment of images captured on OCT

• self-reported vision – participant’s subjective assessment of their vision

• visual acuity – ETDRS visual acuity.

The secondary objectives were to:

1. develop an economic model to identify an optimal monitoring regime

2. develop a risk prediction model using baseline characteristics to predict the development of nAMD in the study eye

3. create a cohort (including a biobank) for future studies

4. create a repository of opportunistically collected OCTA imaging data to explore their potential value in the detection of new-onset nAMD.

Blood collection and storage

Approximately 20 ml of blood was collected at the baseline and exit visit in participants who had consented to donating blood for future research studies. Blood was pseudonymised with the EDNA study number. The samples collected in this study were sent to Queen’s University Belfast for storage and will be stored indefinitely. Research using the samples will be conducted only after approval by a Research Ethics Committee.

Study visits

Participants were examined at 18 and 36 months if they had not been diagnosed with nAMD in the EDNA study eye by these time points. At the study visit, all five index tests were undertaken along with a FFA. These study visits could take place within a 2-month time window either side of the exact due date. Therefore, participants could be monitored for up to 38 months. Participants remained in the study unless they withdrew consent or they were unable to continue for a clinical reason. All data collected up to the point of complete withdrawal were included in the analysis, unless the participant requested this to be destroyed and excluded.

Safety: serious adverse events

Within the EDNA study, only serious adverse events (SAEs) relating to the collection of blood or FFA requested during involvement in the study were recorded. SAEs relating to FFAs conducted prior to recruitment to the study or resulting from treatment to the nAMD eye were not recorded.

Reading centre assessment

An ophthalmic reading centre (Central Angiographic Resource Facility), located at Queen’s University Belfast, independently graded anonymised images (FFA, OCT, fundus photography, autofluorescence – optional) that were acquired at baseline, when any index test was positive, and at study visits (18 months and 36 months/study exit). The data from the graded images were used to provide the baseline and conversion ocular variables for the description of the cohort (see Chapters 3 and 5), the secondary reference standard 2 data set (see Chapter 4) and ocular variables for evaluation in the prognostic model (see Chapter 6). The grading protocol is provided in Report Supplementary Material 1.

Amsler chart

The Amsler chart is a grid of horizontal and vertical lines that is used to monitor a person’s central visual field. It is a simple, inexpensive diagnostic tool that aids the detection of visual disturbances caused by changes in the retina, particularly the macula (e.g. macular degeneration). After covering one eye, the person looks with the eye to be tested, fixating at the small dot in the centre of the grid printed on the Amsler test sheet. Patients with macular fluid may see distortion of the straight lines or areas of the pattern may be missing. Patients should have a normal Amsler test in the study eye at baseline. In participants who had distortion on the Amsler chart at baseline, Amsler tests were not collected in the subsequent assessments.

Fundus clinical evaluation for signs of neovascular age-related macular degeneration

Examination of the macula can reveal fluid and/or lipid exudates (yellow deposition) and/or blood. Other features of AMD, such as drusen and pigmentary irregularities, may be observed. Sometimes these latter features are obscured by the exudative manifestations or may be absent in specific AMD phenotypes, such as polypoidal choroidal vasculopathy (PCV). The fundus could be assessed using slit-lamp biomicroscopy or fundus photography.

Signs suggestive of nAMD include the following:

• Subretinal or subretinal pigment epithelium (RPE) neovascularisation, which may be visible as a dark grey lesion. Occasionally the lesion will have a dark pigmented edge that is thought to be a result of proliferation of the RPE at the edge of the membrane.

• Serous detachment of the neurosensory retina.

• RPE detachment.

• Haemorrhages – subretinal pigment epithelial, subretinal, intraretinal or preretinal. Breakthrough bleeding into the vitreous may also occur, most often indicating the presence of PCV.

• Hard exudates (lipids) within the macular area related to any of the above and not related to other retinal vascular disease.

• Epiretinal, intraretinal, subretinal or sub-pigment epithelial scar/glial tissue or fibrin-like deposits.

• Retinal angiomatous proliferations – a form of intraretinal new vessel complexes arising de novo in the macular retina, which are a subtype of nAMD.

• Choroidal polyps – spherical lesions associated with choroidal vessels that cause the RPE to be focally elevated (PCV).

Optical coherence tomography clinical assessment of images

Optical coherence tomography is a light-wave-based technology that produces cross-sectional images of the retina with scan rates and resolution parameters that have greatly improved over the last 10 years. It is a non-invasive, non-contact visual test, which is rapidly and easily performed and takes < 5 minutes to assess both eyes. 33 Tomograms are acquired by trained medical photographers. The tomogram is a sequential collection of some 25,000 A scans (reflectivity profile in depth), which are sequentially incorporated into a cross-sectional image of the retina generating a B-scan. A series of B-scans are constructed across the macular region of the eye and, depending on the orientation of the scan, can be a rectangular raster or a star pattern. The density of the scan lines can be modified from widely to tightly spaced, with the latter providing more detailed information. The scans can be displayed in three-dimensional mode to provide information on the various retinal layers. Automated segmentation algorithms provided by the manufacturer generate averaged retinal thickness and volume measurements for regions (sectors) of the retina. These algorithms have been shown to provide consistent and reliable estimates in normal eyes. However, in the presence of disease with alterations in the retinal layer anatomy, the algorithms frequently fail, leading to considerable error and variability in the segmentations; therefore, the thickness and volume measurements are generally unreliable. At present, many groups, including one of the applicants, are exploring the use of automated segmentation algorithms on the imaging outputs. Promising results indicate that automated segmentation is a reality and that subjectivity of interpretation may be replaced in the future by objective computerised assessments.

The separation of the various retinal layers can be seen on a scan and there may be deviation in the interfaces between the layers and/or alterations in reflectivity. In nAMD, abnormal dilations and the growth of blood vessels in the retina and choroid can result in fluid and/or blood seeping into the various tissue spaces, changing the normal retinal architecture and/or altering the normal reflectivity. 30 These characteristics are noted and reported by clinicians experienced in the interpretation of OCT.

When abnormalities as a consequence of nAMD develop in the retinal and choroidal circulations (such as dilation of existing vessels or growth of new vessels), there is an accumulation of fluid within the macular tissue compartments with separation of the normal tissue interfaces. In addition, seepage of haemoglobin and other cellular and proteinaceous or lipid constituents of blood into the retina can cause alterations in the internal reflectivity and homogeneity of the retinal layers. These can take the form of areas of dense hyper-reflective material or foci. The appearance of abnormalities when previously there was none, as well as their spatial localisation and distribution, can alert the clinician to the onset of nAMD even when the signs are subtle and only just discernible.

For the purpose of this study, any OCT machine could be used for data collection. No validation was required for OCT imaging data collection. There were four manufacturers of OCT used in the study: Heidelberg (Heidelberg Engineering, Heidelberg, Germany), Zeiss (Carl Zeiss Meditec, Jena, Germany), Topcon [Topcon (Great Britain) Medical Limited, Newbury, UK] and Nidek (Nidek Co. Ltd, Aichi, Japan). The most commonly used OCT was Heidelberg (13 sites); four sites had Heidelberg and Topcon, two sites had Zeiss only, two sites had Topcon only, two sites had Heidelberg and Zeiss, and one site had Nidek.

Self-reported vision: participant’s subjective assessment of their vision

The onset of exudative AMD may be heralded by the appearance of central visual blurring and distortion. Patients may complain that straight lines appear crooked or wavy when the lesion involves the central macula.

At each follow-up visit, participants were asked the following question: ‘How is your vision in the (unaffected) eye compared with the last visit?’. There were four possible answers to this question: ‘about the same or better’, ‘a bit worse’, ‘worse’ or ‘much worse’.

Visual acuity

Patients with new-onset nAMD will usually have a decrease in best corrected visual acuity (BCVA). Visual acuity is a measure of the spatial resolution of the visual processing system. It is a psychophysical test requiring a response from the person to be tested. Usually high-contrast letters of diminishing size are displayed on a chart at a set distance. The most commonly used chart is the ETDRS chart, which is based on a geometric progression of letter sizes with five letters in each row. A three-line difference in either direction from any given line represents a halving or a doubling of the visual angle. BCVA provides a measure of resolution at the fovea. A change of five or more letters (one full line) on the ETDRS chart is considered to be within the limits of the reliability and reproducibility of the measurement. Therefore, a drop of ≥ 10 letters will be considered to be a true reduction in BCVA.

Classic choroidal neovascularization

Classic choroidal neovascularization is said to be present when an area of well-delineated hyperfluorescence appears in the early phases of the FFA. Most commonly, classic CNV represents new vessels that have breached the RPE and lie in the subretinal space. Sometimes a typical lacy pattern of hyperfluorescence is observed in the very early phase of the angiogram, which corresponds to the vascular profiles before the fluorescein has leaked out of these vessels and obscured the margins. Classic CNV also leaks aggressively and, hence, there is considerable pooling of fluorescein dye in the subretinal space in late frames of the angiogram. Multimodal imaging shows that these neovascular complexes lie between the RPE and the neurosensory retina and have a feeder vessel arising from the choroidal circulation.

Occult choroidal neovascularization

As the name suggests, occult choroidal neovascularization refers to the presence of leakage without clear evidence of neovascular profiles in the early angiographic images. Two types of occult leakage are recognised. The first is a characteristic stippled hyperfluorescence that occurs early and is located at the level of the RPE. The RPE layer is elevated and in the later phases of the angiogram there is increasing hyperfluorescence and pooling of dye in the subretinal pigment epithelial space. The pattern of leakage suggests new vessels between Bruch’s membrane and the RPE and it is, therefore, considered to be a fibrovascular pigment epithelial detachment (FPED). The second pattern of occult leakage is a more diffuse hyperfluorescence with poorly demarcated boundaries that occurs late in the angiographic phase, generally after 2 minutes have elapsed since the injection of dye. There is no corresponding hyperfluorescence in the early frames and there is shallow elevation of the RPE. This type of leakage is referred to as late leakage of indeterminate origin (LLIO). OCT has shed further light on these patterns of leakage and has revealed that the neovascular complexes of FPED and LLIO patterns are present in the subretinal pigment epithelial space causing irregular elevation of the RPE.

Retinal angiomatous proliferation

This type of neovascularisation consists of intraretinal telangiectatic blood vessels that are strongly associated with serous pigment epithelial detachments and a form of drusen known as reticular pseudodrusen.

A better scientific term for these types of drusen is subretinal drusenoid deposits because the material accumulates in the outer retina (in the zone containing the photoreceptor matrix). By contrast, classic drusen, which are the hallmark of AMD, accumulate as nodules or diffuse amounts of material between the RPE cells or between the cells and their basement membrane.

Polypoidal choroidal vasculopathy

Polyps are said to be present when individual choroidal vessels exhibit focal dilations. 34 On colour imaging, these appear as orange or red nodules. On dynamic contrast imaging (fluorescein and indocyanine green), polyps become visible, well-defined focal, round areas of hyperfluorescence or hypercyanescence, which can be pulsatile. Polyps are best visualised on ICGA because this dye is tightly bound to plasma proteins and, therefore, does not leak out into the choroidal vascular bed. Polyps are often associated with haemorrhage and lipid exudates. The presence of a haemorrhagic pigment epithelial detachment (PED) is highly suggestive of the presence of this phenotype. ICGA is recommended if the combination of FFA and OCT features suggest presence of PCV. 1

A more recent classification based on FFA and OCT combined has been proposed by an international panel of experts (Consensus on Neovascular AMD Nomenclature criteria), which included the chief investigator for the EDNA study. 35 The new definitions classify CNV into type 1 and 2 with type 1 having vessels that have not breached the subretinal space and type 2 with vessels that ramify in the subretinal space. Retinal angiomatous proliferation (RAP) lesions form category 3 and PCV has been included in type 1 sub-category aneurysmal. These criteria were applied to classify nAMD subtype by the reading centre.

Primary reference standard: fundus fluorescein angiography determination of conversion to neovascular age-related macular degeneration at the clinical site

The primary reference standard was the clinical interpretation of the FFA by the participant’s clinician. A positive reference standard (FFA) test was one that showed typical changes of nAMD, as described above, and as determined by an experienced ophthalmologist.

Secondary reference standard 1: clinical determination of conversion to neovascular age-related macular degeneration at the clinical site

In the event that the FFA was negative or inconclusive, the clinician diagnosis of nAMD was based on all of the available clinical information. Should the clinician give a positive diagnosis of nAMD during the 3-year follow-up, this was the end of EDNA study monitoring for the participant. At the EDNA study exit visit, if a FFA was not performed, the clinical diagnosis was assessed in the absence of the FFA.

Secondary reference standard 2: reading centre determination of conversion to neovascular age-related macular degeneration

All FFA results that were collected at the site were uploaded to the reading centre to independently determine the nAMD diagnosis. The methods for the grading and interpretation of the images are shown in Report Supplementary Material 1.

Primary outcome measure

The primary diagnostic performance outcome was the sensitivity and specificity of the index tests in the detection of nAMD in the study eye in a monitoring setting. The primary economic outcome was the incremental costs (to the health service) per quality-adjusted life-years (QALYs) gained.

Secondary outcome measures

The secondary diagnostic performance outcomes included the diagnostic odds ratio (DOR), likelihood ratio and proportion of indeterminate tests. The performance of combinations of tests was also evaluated. Other outcomes included the time gain of early detection; the visual acuity at diagnosis; the performance of a risk predictor algorithm according to baseline characteristics; the establishment of a well-characterised cohort of clinical and biological data for future research; and the creation of a repository of optical coherence tomography angiography (OCTA) data to explore its potential value in the detection of new onset nAMD.

Main analysis

Under the main analysis approach (analysis A1, see Table 3), repeated monitoring test results were collapsed over the whole monitoring period to give a single candidate test result (positive or negative) as shown in Table 2. This was compared with a single final FFA determination of conversion to nAMD at the clinical site from the participant. Several alternative approaches, in which the definitions for reference standard and index tests were varied, were predefined to test the robustness of the findings, and are described in Table 3.

To be included in the main analysis, the following criteria were applied for each of the five index tests separately. If the last FFA result for a participant showed no nAMD, the participant was included if there was at least one prior index test result during the follow-up period. If the last FFA result for a participant showed nAMD, and the index test result was within the previous 3 months, the participant was included. If a participant did not have an index test result within the previous 3 months of their last FFA, the last FFA was not included. In such cases, the previous FFA result from the participant obtained during EDNA monitoring was used if satisfying the two rules above. If no follow-up FFA was available that satisfied either criteria, the participant was excluded from the assessment of that index test. The valid follow-up period for the index tests was then defined as between baseline and the date of the last valid FFA (as defined above).

Index test result definition for the main analysis

For each index test, multiple test results were collapsed into a single test result. Any positive test result over the valid follow-up period was classed as an overall positive result. To be a negative index test result, all index test results must be negative. The classic 2 × 2 table for assessing diagnostic accuracy with our definitions is provided in Table 2.

Sensitivity and specificity were calculated with 95% CIs calculated using the Agresti-Coull method. 42 Positive and negative likelihood ratios were calculated with 95% CIs calculated using the method in Zhou 2002. 42 The positive and negative likelihood ratios quantify how much the estimated probability of an individual having the disease should increase or decrease given a positive or negative test result, respectively. Diagnostic odds ratios and the proportion of indeterminate tests were calculated with 95% CIs. A receiver operating characteristic curve was plotted and the area under the curve (AUC) was calculated using the trapezoidal rule for visual acuity tests. The standard error for the AUC was calculated using the method of DeLong et al. 43 and was used to form an asymptotic normal 95% CI.

Monitoring sensitivity and specificity of the tests were compared using McNemar’s statistical test (with 95% CIs produced using Newcombe’s method). 44

For comparing sensitivities under the primary analysis, McNemar’s 2 × 2 table was constructed using only participants who had a positive FFA result. They were classified as having a positive index test if the index test had any positive results during the valid follow-up period. They were classified as having a negative index test if all prior available tests were negative during their valid follow-up period.

For comparing specificities under the primary analysis, McNemar’s 2 × 2 table was constructed using only participants who had negative FFA results throughout using the same index test classification.

Alternative time periods for inclusion of the index test result

Two alternative, more restricted approaches to including test results in the analysis were predefined. Analysis A2 collapses the index tests as shown in Table 2, but used index tests from only the individual’s last 6 study months (not the entire study period). A positive index test result occurring outside the 6-month window, therefore, did not lead to a positive test result at the collapsed individual level. The individual diagnostic performance of the tests were calculated using this index test definition and the McNemar comparisons of sensitivity and specificity, as detailed above.

Analysis A3 used the index test at the individual’s last study visit only. If a participant developed nAMD according to the reference standard during the follow-up period, only the index test from the visit at which nAMD was diagnosed were used. If a participant did not develop nAMD during the follow-up period, index test data from the last visit for which their reference standard was available were used.

Secondary reference standards

Two secondary reference standards were evaluated:

1. Secondary reference standard 1 – clinical determination of conversion to nAMD at the clinical site.

   Reference standard is positive if the clinical interpretation of FFA indicates nAMD or a clinician diagnosis of nAMD at any time point. The reference standard is negative if the clinical interpretation of FFA is negative/inconclusive/not carried out and the clinician diagnosis is negative.

2. Secondary reference standard 2 – reading centre determination of conversion to nAMD.

   The reference standard is positive if the reading centre FFA indicates nAMD at any time point and negative if it indicates no nAMD throughout.

Combination of tests

Using the primary reference standard and all available index test results over the monitoring period, the impact of combining OCT with each of the other index tests was considered using two approaches: (1) both positive and (2) either positive. For approach (1), a visit at which both of the index tests were positive was defined as a positive combined result and a negative combined result was produced if there were no visits at which both of the index tests were positive. For approach (2), at least one positive index test result from either of the tests resulted in a positive combined result, and a negative combined result was produced if there were no positive index test results from both of the tests. Two post hoc test combinations were evaluated. One test combined Amsler, fundus clinical examination, visual acuity and self-reported vision. The other test combined Amsler, self-reported vision and visual acuity. Both combinations assumed that any positive test leads to a positive combination test result. These further combination tests were motivated by considerations of the preliminary findings for the health economic model.

Sensitivity analyses

Several sensitivity analyses were predefined using the primary reference standard, and are detailed in Table 3. These explored possible threshold effects for two of the index tests (visual acuity and self-reported vision) and inclusion of indeterminate reference standard results. A planned sensitivity analysis to use OCT test results as assessed from the reading centre as an alternative diagnostic test (rather than OCT assessed at the clinical site) was not possible because the OCT result was assessed in conjunction with an assessment of the respective FFA at the reading centre.

Subgroup analyses

We undertook a pre-planned subgroup evaluation according to the type of nAMD in the study eye (CNV and RAP) for the main analysis. These subgroup analyses were classified as exploratory and were evaluated at the two-sided 5% significance level, and the individual group sensitivity (with 95% CIs) was calculated. It was not possible to quantify the proportion of classic disease (for the CNV subtype only) as planned because the data could not be retrieved.

Secondary complex analysis utilising repeated test results

A GEE modelling approach was used to allow the simultaneous modelling of sensitivity and specificity in a regression framework and the use of multiple test results per participants over time. This had the advantage of allowing a flexible regression framework (with easy comparison between tests), allowing for clustering of observations by participants and incomplete data without requiring extensive distributional assumptions. The GEE modelling was applied using the primary reference standard.

It was assumed that at time points when index tests were available but the reference standard was not available, the reference standard result was negative until a positive reference standard finding occurred. An independent correlation structure with robust standard errors was adopted (xtgee Stata command).

Handling of missing data

The absence of a positive reference standard during a valid follow-up period was presumed to indicate a negative result. Indeterminate tests were treated as missing, but sensitivity analyses 6 and 7 looked at the impact of this assumption. For the primary analysis, if the last available reference standard was missing/indeterminate, the previous reference standard was used with index data curtailed accordingly. If no reference standards were carried out after the baseline reference standard, then that participant was excluded from the analysis. If the FFA result was indeterminate or missing, it was excluded unless a curtailed follow-up period with a valid FFA and index test result existed.

Sample size

The prognostic model was a substudy within the EDNA study and, therefore, its sample size was limited to the number of participants available for the diagnostic accuracy study.

Candidate predictors

The baseline variables that were considered for inclusion in the model were selected through discussion between the clinicians and the statisticians of the study. They included person-specific risk factors collected via a baseline case report form completed at the site (age, raised blood pressure, smoking history, cardiovascular disease, diabetes, sex, nutritional supplements, family history of AMD and body mass index); ocular variables in the EDNA study eye [previous cataract surgery, baseline visual acuity, drusen type, maximum size of drusen, most frequent size of drusen, presence of pigmentary abnormalities in the fundus, retinal thinning, choroid thinning, external limiting membrane (ELM) disruption and ellipsoid zone (EZ) disruption]; and ocular variables in the first-presenting eye (type of wet AMD, lesion size and total lesion area). Ocular variables were collected from reading centre baseline image grading, with the exception of previous cataract surgery and baseline visual acuity that were collected via a baseline case report form completed at the site. Fractional polynomials were used to explore the presence of non-linear relationships of the continuous predictors (age, baseline visual acuity and total lesion area). To assess the association between each candidate predictor and the outcome, we conducted univariate analysis using Cox regression and calculated the hazard ratio and respective 95% CIs.

Analysis and selection of candidate predictors

We used a multiple Cox regression to estimate hazard ratios and 95% CIs for the candidate predictors. We initially ran the multiple Cox regression including all candidate predictors (‘full model’). Owing to the large list of candidate predictors and given the limited sample size, a backwards elimination process was implemented using a p-value of 0.10 to select predictors for the model. The ‘final model’ was estimated by including the selected predictors from this process. As part of a sensitivity analysis, we varied the threshold (from p = 0.05 to p = 0.20) to assess its effect on the final model.

Model performance

The final model’s predictive performance was assessed in terms of discriminative ability using Harrell’s c-index and accompanying 95% CIs. As a post hoc sensitivity analysis, we assessed the impact of two variables on the c-statistic: type of drusen and presence of cataract. Calibration was assessed visually by plotting the Kaplan–Meier survival estimates of risk groups against the monitoring time. Four risk groups were calculated (divided by the 16th, 50th and 84th centiles of linear prediction risk). 45

A planned exploratory analysis of the effect on conversion of extending the model by including as a covariate CNV subtype (‘percentage classic’) was not conducted because the data could not be retrieved.

Handling of missing data in the prognostic model

Two approaches were used to handle missing data in the development of the prognostic model: (1) complete case analysis, in which participants with all candidate predictors available were included in the selection process; and (2) multiple imputation, in which missing values were replaced in candidate predictors prior to implementing the selection process. Through this process, we assessed the impact of multiple imputation in the selection of predictors in the final model. In both instances, we excluded candidate predictors with a large number of missing data (> 20%).

Baseline measurements

At baseline, data collection included participant demographics, risk factors, whole blood (separated into white cells, serum and plasma) and baseline index test results. Where baseline measures were already documented in clinical case notes within 6 weeks prior to consent, these measures were used for entry into the baseline data collection form and they did not need to be repeated. FFA at diagnosis that confirmed nAMD in the first eye and a second eye free of nAMD were sent to the reading centre for grading along with baseline OCT and colour fundus photographs. Ethnicity data were collected from the medical records using ethnic categories defined on the Medisoft (Medisoft Ltd, Leeds, UK) EMR software.

Follow-up measurements

Data collected at all routine clinic visits after enrolment into the EDNA study for up to 3 years were extracted from the medical records. These included the index test results and, if a FFA was triggered based on positive test results, FFA results and clinical diagnosis. At 18 months and 36 months, all index tests were collected along with a FFA, unless the participant refused or there was a clinical reason to avoid FFA.

Post conversion case note review

Following conversion to nAMD, additional data about the study eye were extracted, and included number of visits, visual acuity and number of anti-VEGF treatments. These data were used to inform the health economic analysis.

Oversight of the study

A panel of Macular Society patient members (initially one, later three) and the chief executive of the Macular Society served as members of the EDNA Study Steering Committee, contributing both their individual perspectives of macular disease and the perspectives of macular disease of the wider community. As integral members of the Study Steering Committee, they attended regular investigator meetings in addition to the regular steering committee meetings, and received updates on study progress between meetings. The Study Steering Committee reviewed and commented on the study design, protocol and all study documentation, including patient-facing documents that were sent to potential and recruited participants in the EDNA study. In addition, the PPI partners co-produced a patient newsletter for EDNA participants and a patient diary to help to monitor eye health, designed with visually impaired people in mind, and used their own experiences of having their eye health monitored. Having three patient representatives with different stages of disease and visual impairment enabled a wide cross-section of input and perspectives.

Dissemination

The PPI partners have been actively involved in discussions of the study results with the Study Steering Committee and investigators, and have contributed towards the preparation of the Plain English summary. They will continue to be involved in dissemination to participants and academic papers. We anticipate that the PPI partners on the Study Steering Committee will comment on the participant results newsletter/audio. It is also anticipated that the publication of the study results will be co-ordinated with press releases from the participating academic/NHS institutions and the Macular Society.

At the end of the study, the PPI partners reflected on their input and made suggestions for future research, which is included in the discussion.

Primary reference standard: fundus fluorescein angiography determination of conversion to neovascular age-related macular degeneration at the clinical site

This was undertaken by the clinician at the clinical site. We chose this as the primary reference standard because this is the recognised method of determining the onset of nAMD at the clinic. This represents the most appropriate and pragmatic standard of care.

Secondary reference standard 1: clinical determination of conversion to neovascular age-related macular degeneration at the clinical site

We included this alternative reference standard to allow for clinical interpretation of disease status based on negative or indeterminate FFA results, or when a FFA was refused by participants or deemed unsafe owing to patient characteristics. Where a FFA was unavailable, a clinical diagnosis of nAMD was made based on all available clinical information. We included this secondary reference standard because we recognised that a proportion of participants would not be put through a FFA. This arose because of two reasons: (1) although enrolled participants had to have had a FFA, some had experienced mild allergic response to fluorescein and, therefore, subsequent FFA was deemed to be unsafe; and (2) participants sometimes declined a FFA because the procedure was found to be uncomfortable, inducing nausea in some susceptible cases.

Secondary reference standard 2: reading centre determination of conversion to neovascular age-related macular degeneration

This was carried out by the reading centre after submission of the FFA images to netwORC UK (Belfast, UK). It is known that there is disagreement between clinician determination and reading centre determination of conversion to nAMD. 61 Reading centres are known to produce data with higher reliability and consistency than clinicians and, hence, have been used as the gold standard in clinical trials and studies. Therefore, we included reading centre determination as a further reference standard to confirm whether or not FFA shows conversion from the non-neovascular stage to neovascular AMD.

We further presented the results of each of the three reference standards by three timeline-specific definitions for index tests. Under each specific timeline definition of the index test, additional analyses were conducted as follows:

1. Primary reference standard – FFA determination of conversion to nAMD at the clinical site –

   1. index test results cover the entire period of follow-up (main analysis):

      1. – main analysis and paired comparison

      2. – combining index test results

      3. – subgroup analysis based on nAMD subtype

      4. – sensitivity analyses varying the index test definitions for self-reported vision and severity of visual acuity loss

      5. – sensitivity analyses varying the definition of an inconclusive FFA result

      6. – modelling diagnostic accuracy by GEE

   2. index test results cover the last 6 months of participant follow-up

   3. index test results are based on the last visit available

2. Secondary reference standard 1 – clinical determination of conversion to nAMD at the clinical site –

   1. index test results cover the entire period of follow-up

3. Secondary reference standard 2 – reading centre determination of conversion to nAMD –

   1. index test results cover the entire period of follow-up

   2. index test results cover the last 6 months of follow-up

   3. index test results are from the last visit available.

Figure 8 shows a flow chart describing the cohort, the number of participants who were included within the analyses pertaining to each reference standard and the number of participants deemed to have converted from the non-exudative state to exudative AMD in the EDNA study eye under each reference standard. Of the 552 eligible participants in the EDNA cohort with a diagnosis of nAMD in the first-presenting eye (fellow eye) based on a FFA at baseline, nine participants had no follow-up data and were, therefore, excluded from all subsequent analyses. Of the 543 participants with at least one follow-up visit, 464 participants had at least one additional FFA and constituted the cohort included in the analysis using the primary reference standard for all index tests. Given that it was prespecified that participants with a positive Amsler test at baseline would be excluded, we had a smaller sample who could be included in the assessment of the performance of the Amsler test.

FIGURE 8.

Participants’ status at the end of the study per reference standard. a, Participants have at least one index test available during follow-up.

Secondary reference standard 1 represents the broadest definition of conversion to nAMD and included 489 out of the 552 eligible participants. In this cohort we had an additional 25 participants who did not have a FFA, but had a clinical determination of conversion to nAMD at the study site. In the remaining 54 participants, there was insufficient follow-up precluding use of their data in the analysis. The sample included in secondary reference standard 2 consisted of data from 460 participants whose FFA was graded at the reading centre. The differences in the reference standards defined above result in differences in the number of EDNA study eyes that were deemed to have converted to nAMD under each analysis. For a summary of the comparison between the primary and the two secondary reference standards, see Tables 25 and 26.

Time to conversion to neovascular age-related macular degeneration

Of the 456 participants who were included in the primary reference standard analysis, eight participants were excluded because the FFA was reported as inconclusive by the site clinician (see Figure 8). The average number of FFAs carried out during follow-up was 1.7 FFAs (standard deviation 1.1 FFAs). In participants who developed nAMD during follow-up, the average number of FFAs carried out was 1.5 FFAs (standard deviation 0.82 FFAs) compared with 1.7 FFAs (standard deviation 0.82 FFAs) in those whose study eye did not convert to nAMD. The maximum number of follow-up FFAs that was recorded in any one participant was six.

During follow-up, 145 conversions to nAMD were detected by clinicians using clinical information and any available imaging modality. Of these conversions, 120 had a FFA performed at the time of conversion and read by the clinician, who confirmed conversion to nAMD based on this imaging modality. This yielded a crude conversion rate of 26% (95% CI 22.3% to 30.6%), with a median follow-up time of 33 months (range 0.8 to 38.5 months).

The Kaplan–Meier curve for the conversion to nAMD in the EDNA study eye, using the time of FFA confirmation of conversion (primary reference standard) as the time of conversion to nAMD, is shown in Figure 9.

FIGURE 9.

Kaplan–Meier survival estimate: months from consent to conversion to nAMD or exit from the study (primary reference standard). Raw data: the time of conversion to nAMD is the date that the FFA is conducted. Reproduced with permission from Sivaprasad et al. 52 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

Reproduced with permission from Sivaprasad et al. 52 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

The rate of conversion to nAMD in the EDNA study eye over the follow-up period was uniform. The sharp decline seen in the Kaplan–Meier plot at 36 months reflects the small cluster of conversion events from the reduced number of participants still at risk at this time point. The revised Kaplan–Meier curve (Figure 10), in which the time of conversion was recalculated as the mid-point between the date of FFA confirmation of nAMD conversion and the last (negative) FFA, does not show this dip. In Figure 10, the curve flattens after month 12, whereas the curve in Figure 9 appears to show a more uniform risk over the first 24 months.

FIGURE 10.

Kaplan–Meier survival estimate: months from consent to conversion to nAMD or exit from the study (primary reference standard). Interpolated results: assumes that time of conversion to nAMD is midway between the last FFA without disease and the positive FFA.

Given that the study design required an exit visit with an assessment, which would differ from usual clinical practice, we specifically examined this time point. Of the 265 participants who were at risk and who did not have nAMD detected in the EDNA study eye by month 30, 18 participants were found to have converted to nAMD after 30 months, of whom 11 were detected at the planned study exit visit and seven prior to the planned exit visit. Among these 18 study eyes, none reported all index tests as negative in the period between 30 months in the study and the planned exit visit. One participant had a negative OCT and a positive FFA result.

Diagnostic performance of the index tests: index test acquired during the entire follow-up period (main analysis)

The following results relate to the primary reference standard and analyses for which the index test results cover the entire period of follow-up.

Table 10 shows the time in days between the last index tests and the reference standard, grouped by conversion to nAMD status. In participants who converted to nAMD, the mean interval between the index test and the reference standard ranged between 30.3 days for Amsler and 6.9 days for OCT. The median interval between a positive index test and a positive reference standard is zero, which shows that, in the majority of participants who converted to nAMD, the positive index test triggered the reference standard measurement on the same day. In participants who did not develop nAMD in their EDNA study eye during follow-up, the interval between the last index test and the final FFA ranged from an average of 8.4 days for self-reported vision to 117 days for visual acuity. Similar tables for the secondary reference standards are presented in Appendix 1.

The sensitivity and specificity of the index tests compared with the primary reference standard (FFA determination of conversion to nAMD at the clinical site) are shown in Table 11. The maximum available sample size, as shown in Figure 8, is 464 participants. However, data were missing for a few participants in each of the index tests. More information on missing data is available in Appendix 2. OCT had the highest sensitivity at 91.7%, followed by fundus clinical examination (53.8%). All remaining index tests, including Amsler, visual acuity and self-reported vision, had sensitivities that fell below 50%. The fundus clinical examination had the highest specificity (97.9%), followed by self-reported vision (97%) and OCT (87.8%).

Changing the reference standard to include FFA results that were inconclusive, by classifying them as all positive or all negative, had little impact on the sensitivity and specificity of the index tests (see Appendix 1).

Table 12 shows the likelihood ratios (with 95% CIs) that indicate the probability of conversion to nAMD, with values that are furthest from 1 being the most informative; a likelihood ratio of 1 means that the test result is equally likely to be positive or negative in participants with and without nAMD. OCT, fundus clinical examination and Amsler CIs for the positive likelihood ratio did not include 1. OCT and fundus clinical examination CIs for the negative likelihood ratio did not include 1. OCT had a high positive likelihood ratio and a very low negative likelihood ratio.

By contrast, the fundus clinical examination has a very high positive likelihood ratio, but the negative likelihood ratio is five-times worse than for OCT. Diagnostic odds ratios varied from 0.8 (visual acuity) to 79.5 (OCT).

Paired comparisons between index tests (main analysis)

Table 13 shows the paired differences (with 95% CIs) in sensitivity and specificity for detecting nAMD between two index tests along with the corresponding McNemar’s test p-values for analysis A1. Corresponding McNemar’s tests for the paired comparisons are also given. The sensitivity of the test pairs differed significantly from each other (see Table 13) except for the pair constituted by Amsler and visual acuity for which the difference in sensitivity was 9.2% and did not reach statistical significance. Differences in sensitivity between index tests varied from –16.3% (Amsler vs. fundus clinical examination) to 87.4% (OCT vs. self-reported vision). The differences in sensitivity between OCT and all of the other tests were statistically significant, with these differences ranging from 38% to 87%. The magnitude of differences in specificity between OCT and other tests was smaller; fundus clinical examination and self-reported vision specificity were marginally but significantly higher than that for OCT.

Combining test results (main analysis)

Table 14 shows the sensitivity and specificity of combinations of index test to detect nAMD, where a positive combination is defined as a positive OCT or a positive index test selected from any one of the other index tests. This definition specifies that both tests have to be negative for the combination to be deemed negative. The combination of index tests shows a marginal improvement in sensitivity compared with conducting OCT alone (sensitivity 91.7%, 95% CI 85.2% to 95.6%). Specificities are lower for a combination of index tests than for using OCT alone (specificity 87.8%, 95% CI 83.8% to 90.9%).

In Table 15, the definition of the combination has been changed, and the combination test is positive only if both the OCT and the index test results are positive. With the combination defined in this manner, the sensitivity drops but the specificity improves for all of the combinations tested. See Appendix 1 for the results of two further post hoc combination tests of Amsler, fundus clinical examination, visual acuity and self-reported vision, and Amsler, visual acuity and self-reported vision; the results similarly indicated that OCT alone was a better monitoring strategy.

Subgroup analysis based on neovascular age-related macular degeneration subtype

A planned subgroup analysis was conducted based on the nAMD subtype [CNV (type 1 and 2) vs. RAP (type 3)] in the study eye at conversion to nAMD. Figure 11 shows the participants included in this analysis. To be eligible for inclusion, the reading centre nAMD phenotype in the EDNA study eye at conversion to nAMD had to be available.

FIGURE 11.

Flow of participants included in the nAMD subtype subgroup analysis.

Table 16 presents the sensitivity of each index test to detect nAMD conversion in the study eye by nAMD-type subgroup. Differences in the sensitivity of detection of conversion to nAMD were seen by nAMD subtype. The Amsler test was numerically better in the RAP subgroup than in the CNV subgroup. In addition, the sensitivity of the detection of conversion to nAMD by OCT was better in the RAP subgroup at 100% than in the CNV subgroup at 91.7%.

Sensitivity analyses I: varying the definition of index tests

The definitions of two of the index tests (self-reported vision and visual acuity) were changed to estimate their impact on the sensitivity and specificity of detection of conversion to nAMD in the study eye.

On changing the test positive self-reported vision to either ‘worse’ or ‘much worse’ since the last appointment, Table 17 shows a higher sensitivity and lower specificity than the definitions used in the primary analysis (see Table 11).

For visual acuity, the definition of a positive test in the main analysis was a drop of ≥ 10 letters from baseline. Two changes in this test definition were made: (1) a drop of ≥ 10 letters since the last FFA and (2) a drop of ≥ 20 letters since baseline.

Both sensitivity analyses of visual acuity resulted in a lower sensitivity but a substantially higher specificity than that of the definitions used in the primary analysis (see Table 11).

Sensitivity analyses II: index test results cover the last 6 months of participants’ follow-up

Table 18 shows the sensitivity and specificity (with 95% CI) of the index tests in the last 6 months of the participants’ follow-up for detecting conversion to nAMD. The sensitivity and specificity are similar to that of the main analysis (see Table 11). The likelihood ratios (with 95% CIs) are shown in Table 19.

Table 20 shows the paired differences (with 95% CIs) in sensitivity and specificity for detecting nAMD between pairs of index tests along with the corresponding McNemar’s test for significance. Index test sensitivities were statistically significantly different in these comparisons, except for the comparison between Amsler and visual acuity. Statistically significant differences in sensitivity varied from –22.0% (self-reported vision vs. visual acuity) to 87.3% (OCT vs. self-reported vision) in magnitude.

For specificity, the differences were significant for all comparisons, except for OCT versus Amsler and fundus clinical examination versus self-reported vision. Statistically significant differences in specificity between index tests varied from –6.7% (Amsler vs. fundus clinical examination) to 15.2% (fundus clinical examination vs. visual acuity) in magnitude. The specificity of the detection of conversion to nAMD for the fundus clinical examination and self-reported vision was generally higher than that for self-reported vision and fall in visual acuity.

Sensitivity analyses III: index test results are based on the last visit available

Table 21 shows the sensitivity and specificity of index tests (with 95% CIs) for detecting conversion to nAMD using data from the last visit only. Results are similar to that of the main analysis (see Table 11). OCT has the best sensitivity, with 90.0%; all of the other index tests have a sensitivity below 50%. Fundus clinical evaluation has the highest specificity (99.7%), followed by self-reported vision (98.5%), OCT (96.4%), Amsler (93.7) and visual acuity (88.4%). Likelihood ratios (with 95% CIs) and diagnostic odds ratios are shown in Table 22.

Table 23 shows the paired differences (with 95% CIs) in sensitivity and specificity for detecting nAMD between pairs of index tests along with the corresponding McNemar’s test p-values for analysis A3. The index test pairs differed significantly from each other, except for Amsler versus visual acuity. The differences in sensitivity between index tests varied from –20.3 (self-reported vision vs. visual acuity) to 85.6 (OCT vs. self-reported vision) in magnitude.

The specificity of the index test pairs also differed significantly from each other, except for the following pairs: OCT versus Amsler, OCT versus self-reported vision and fundus clinical examination versus self-reported vision.

Statistically significant differences in specificity between index tests varied from –5.6% (Amsler vs. fundus clinical examination and Amsler vs. self-reported vision) to 11.3% (fundus clinical examination vs. visual acuity) in magnitude.

Modelling diagnostic accuracy by generalised estimating equations

The sensitivity and specificity of the five index tests against FFA as the reference standard using the GEE model also confirmed that OCT had high sensitivity (90%) and specificity (97.4%) (Table 24), which was consistent with all other analyses.

Fundus fluorescein angiography lesion characteristics in fellow eyes at study entry

At study enrolment (baseline), the reading centre classified 535 out of 548 available FFA images of fellow eyes as exhibiting nAMD (see Table 8). Of the remaining 13 eyes, eight eyes were classified by the reading centre as not exhibiting nAMD. The remaining five eyes were ungradable. The lesion characteristics of fellow eyes at initial presentation are shown in Table 8, and the reading centre grading confirmed neovascular complexes in 517 (94.3%) out of the 535 fellow eyes. The size of the lesion, as measured by the greatest linear dimension, was 5.9 mm and the area was, on average, 7.7 mm². Two-thirds of the lesions had type 1 neovascularisation, one-fifth had type 2 and the remainder had RAP. RAP lesions were more likely to be extrafoveal, whereas type 1 were most likely to be subfoveal. Fibrosis and atrophy were present in around 10% of contralateral eyes at presentation.

Fundus fluorescein angiography lesion characteristics in EDNA study eyes at conversion

A total of 119 EDNA study eyes with a FFA were transmitted to the reading centre for grading at the visit, at which conversion to nAMD was detected by the clinician examination or because of a trigger arising from one of the test technologies. The reading centre confirmed 99% of EDNA study eyes to have nAMD based on FFA at the conversion visit. Table 27 shows the reading centre-graded lesion characteristics at baseline of fellow eyes for the 119 participants constituting the subset who developed nAMD in the EDNA study eye. The characteristics of the fellow eyes at first presentation in this subset were highly similar to those of the full sample for frequency of nAMD subtype. In fellow eyes, the diameter and area of the total lesion and that of the active neovascular complex were similar between the subset that converted to nAMD and the full sample. On comparing fellow eyes with EDNA study eyes, the proportion with type 1, 2 and 3 lesions was similar between fellow eyes at enrolment and EDNA study eyes at detection of conversion to exudative nAMD. On comparing corresponding study eyes with fellow eyes in the 119 eyes that converted to nAMD, the area dimensions of the total lesion and active neovascular complex were markedly lower in the former than in the latter (see Table 27). Lesions were also more likely to be extrafoveal or juxtafoveal for all three lesion types in the EDNA study eye than in corresponding fellow eyes. In type 1 lesions, 50% were subfoveal at detection of nAMD in the EDNA study eye compared with > 80% that were subfoveal in the fellow eye.

In type 2 lesions, 10% were subfoveal in the EDNA study eye compared with 50% in corresponding fellow eyes. Of lesions classified as RAP, none was subfoveal in the EDNA study eye at conversion to exudative nAMD, with these lesions equally distributed between extra and juxtafoveal locations.

Fibrosis was rare in nAMD lesions at conversion in the EDNA study eye compared with fellow eyes at initial presentation (2% vs. 12%). Atrophy was present in similar proportions in the EDNA study eye at nAMD conversion compared with fellow eyes at initial presentation.

Optical coherence tomography findings in the entire cohort and in the subset of participants who converted to neovascular age-related macular degeneration in the EDNA study eye

In the fellow eyes, the average central foveal thickness was just under 400 µm at initial presentation in the 548 participants with gradable OCT images available to the reading centre (Table 28). In the 119 participants who converted to nAMD, the mean central foveal thickness in the fellow eye at presentation was 437 µm.

External limiting membrane and EZ disruption was graded as present in around one-quarter and one-third of contralateral eyes, respectively, and focal atrophy in one-fifth of eyes (Table 29).

In study eyes, at conversion to exudative nAMD the central foveal thickness was just over half of the average thickness of corresponding fellow eyes at 274 µm (see Table 28). The proportion of eyes with ELM and EZ disruption, which was similar between fellow eyes and study eyes at enrolment, had doubled in study eyes at nAMD conversion (see Table 29).

Discrepancies between fundus fluorescein angiography determination of conversion to neovascular age-related macular degeneration at clinical site and reading centre determination of conversion to neovascular age-related macular degeneration

A proportion of EDNA study eyes was determined to have nAMD at the clinical site, but did not have this confirmed by the reading centre (20 out of the 119 eyes with nAMD according to the primary reference standard). Conversely, 15 out of 333 cases that were classed as not having converted to nAMD by the clinician were determined by the reading centre to have converted to nAMD (see Table 26). This represents a kappa statistic of 0.91. In the case illustrated in Figure 16, the OCT was determined to be positive but the FFA was considered negative by the clinic, while the reading centre determined the case to have converted to nAMD both on OCT and on FFA. In this case, leakage was seen in the late frames of the angiogram in an area remote to the fovea (shown by red arrow). A further case (not shown) had a type 1 CNV that was indolent and minimal fluorescein leakage visible only on stereoscopic examination of the late angiographic frames. In this case, the clinic determined absence of conversion to nAMD whereas the reading centre determined that exudation was present.

FIGURE 16.

The OCT scan shows a shallow area of subretinal fluid. The fluorescein was deemed by the clinician to be negative. The FFA late frames show the presence of leakage (red arrow) indicating presence of nAMD.

Discrepancies between fundus fluorescein angiography diagnosis and optical coherence tomography diagnosis of neovascular age-related macular degeneration at the reading centre

In total, there were 12 instances for which the FFA was positive and the OCT was negative. There were also 12 instances for which the OCT was positive and the FFA was negative. Most discrepancies had equivocal findings in either OCT or FFA, or both. In the case shown in Figure 17, the OCT was determined to be negative with no leak, but the FFA was determined to be positive. In the case shown in Figure 18, the OCT was determined to be positive, with a small focus of hyporeflectivity seen in the subretinal space in the foveal B-scan. The early frames of the FFA show some hyperfluorescence at the site corresponding to the region of presumed fluid on the OCT scan. This hyperfluorescence fades over the angiographic run, which is not thought to be a result of a true leak but staining of the RPE.

FIGURE 17.

Disagreement between OCT- and FFA-detected nAMD: example 1. (a) The OCT shows no leakage, i.e. no intra or subretinal fluid. There is a drusenoid elevation of the RPE that is non-exudative. (b) Corresponding FFA shows increasing leakage over the course of the angiogram.

FIGURE 18.

Disagreement between OCT- and FFA-detected nAMD: example 2. (a) OCT: there is a small pocket of subretinal fluid that is located extrafoveally. (b) FFA: the early frames of the FFA show no leakage. The late frames show speckled hyperfluorescence. Additional late frames show no intensification of the speckled hyperfluorescence nor blurring of the margins of the speckles suggesting that these are just RPE abnormalities.

Comparison of visual acuity outcomes in study eyes and fellow eyes

In the overall group of participants, the mean visual acuity in fellow eyes at enrolment was 56.6 [standard deviation (SD) 15.5] letters. In the subset of 119 participants, the mean visual acuity in fellow eyes at enrolment was 54 letters and improved to 58 letters when conversion to nAMD was detected in the corresponding EDNA study eyes (Table 30). In the 119 eyes, the EDNA study eye mean visual acuity was 78 letters at enrolment and fell by four letters at nAMD detection.

Table 31 shows the visual acuity distribution by nAMD status at study exit. Participants who converted to nAMD had lower mean visual acuity than those who did not convert to nAMD (73.1 vs. 77.5 letters, respectively); however, there was a lot of variability in both groups.

Figure 19 shows the mean difference in visual acuity distribution by nAMD status at study exit. More participants who did not convert to nAMD had no change in visual acuity than those who did convert to nAMD. However, there was a lot of variability in both groups.

FIGURE 19.

Distribution of change in visual acuity (number of letters) from baseline at study exit by conversion to nAMD (primary reference standard). (a) Converted to nAMD; (b) did not convert to nAMD.

Impact of including or excluding specific candidate predictors in the c-statistic

Varying the selection threshold for variable inclusion

We explored the impact of varying the selection threshold that was used to decide the variables to include in the model from p = 0.05 to p = 0.20 (Table 36). This resulted in similar included variables when compared with the main analysis using a selection threshold of p = 0.10 (see Table 35). The most stringent threshold (p = 0.05) led to the inclusion of only two variables as predictors: type of drusen (study eye) and smoker. The least stringent threshold (p = 0.20) selected the main model variables with two additional predictors: study eye visual acuity and retinal thinning.

Multiple imputation

Variables with multiple categories were combined into binary variables to overcome the software conversion problems in running the multiple imputation model. We then applied a backwards selection process with a threshold of 0.10 to select the predictors following the same process as in the main analysis. The resulting model variables after imputation are shown in Table 37. The three predictors that were selected for the model after imputation differ from the four predictors that were selected for the model without multiple imputation. Specifically, most frequent size of drusen and study eye visual acuity are newly included in the model after multiple imputation, and type of drusen in the study eye, smoking status and family history of nAMD are all excluded from the model after multiple imputation.

Model structure

The economic model was structured around disease status (no nAMD or nAMD), diagnosis status (undetected or detected) and treatment status (untreated or treated). A simplified schematic of the structure is provided in Figure 21. The change in visual acuity was modelled at the individual level in each model cycle and there was a simplifying assumption that patients maintain their baseline visual acuity in the EDNA study eye (second eye) until it converts to nAMD. This assumption was considered reasonable by clinical experts within the study team. Visual acuity was updated as a continuous variable in the model, taking a value between 0 and 100 letters (ETDRS). The individual’s current acuity was tracked and linked to a health-related quality-of-life weight using a published equation (see Valuation of health effects). It was assumed that visual acuity in the second eye was representative of visual acuity in the best seeing eye throughout the modelled time horizon. This was justified by the baseline visual acuity in the EDNA study eye being better or within 10 letters of visual acuity in the treated eye in 93% of EDNA study participants. Given that eyes being treated for nAMD still tend to lose vision over the long-term period, it was deemed reasonable to assume that visual acuity in the second eye will remain, on average, higher than visual acuity in the first eye over time. The modelling adopted a 25-year time horizon, with a mean starting age of 77 years (mean age in the EDNA study cohort).

FIGURE 21.

Simplified schematic of the model structure. FN, false negative; FP, false positive; Ref, reference standard; TN, true negative; TP, true positive.

Population

The analysis was conducted for a simulated cohort of individuals with characteristics matching those of the EDNA study cohort. The baseline characteristics of simulated individuals were drawn at random from a table containing the baseline characteristics of each EDNA study participant. This ensured that any correlations between baseline characteristics were maintained in simulated individuals. Given that the EDNA study was embedded in routine NHS practice across 18 study sites, the simulated cohort can be considered representative of the NHS patient population. The average age was 77.4 years, the average baseline visual acuity in the second eye was 79 letters compared with 56.6 letters in the first eye and 57.2% of patients were female.

Comparators

The economic analysis considered each of the EDNA study monitoring tests individually. The base-case analysis followed the EDNA study protocol in assuming that positive tests would be confirmed or refuted by fluorescence angiography.

Clinical effectiveness parameters

Time to conversion to neovascular age-related macular degeneration

Conversion to nAMD in the second eye was based on the observed time to conversion to nAMD data from the EDNA study, using the primary reference standard (FFA determination of conversion to nAMD at the clinical site) to define conversion to nAMD (events). Individuals were censored at the time of their last observed FFA if no conversion to nAMD was observed. Standard parametric curves were fitted to the observed Kaplan–Meier data. The parametric distributions considered were exponential, Weibull and log-normal (Figure 22). Based on Akaike information criterion (AIC) and Bayesian information criterion (BIC), the exponential curve was identified as having the best statistical fit to the observed data (Table 38). This assumed a constant hazard of conversion to nAMD over time. Clinical opinion within the team was split between whether or not the hazard of conversion to nAMD would be expected to increase over time and whether or not the hazard of conversion to nAMD would be expected to decrease over time; therefore, it was agreed that the exponential was appropriate for application in the base case. The Weibull distribution (predicting an increasing hazard of conversion to nAMD over time) and the log-normal distribution (predicting a decreasing hazard) were tested in sensitivity analysis. The estimated rate parameter (λ) of the exponential function was used to derive monthly probabilities of conversion to nAMD in the model (transition probability) using the following equation, where t is time in months and u is the Markov cycle of one month:

⁽¹⁾ transition probability ( t u ) = 1 – exp { λ ( t – u ) – λ t } .

The curve was extrapolated beyond the observation period of the EDNA study over the lifetime of simulated individuals.

FIGURE 22.

Kaplan–Meier curve of time to conversion to nAMD overlaid with fitted parametric survival curves.

TABLE 38 - Statistical fit of alternative parametric curves for time to conversion to nAMD

Distribution	AIC	BIC
Exponential	758.21	762.33
Weibull	759.31	767.55
Log-normal	757.80	766.05

Pre-treatment change in visual acuity

There is a paucity of published data on the rate of vision loss in the early stages of nAMD immediately following conversion to nAMD. It is known that the overall trajectory of vision loss is not linear, with a rapid decline expected when nAMD first supervenes before the slope flattens somewhat. The available natural history data come from an era in which patients presented with nAMD late; therefore, reported rates of visual acuity loss will relate to the flatter part of the curve following the steeper initial decline after conversion to nAMD.

Nevertheless, a meta-analysis of older natural history data still found that > 50% of patients can expect to lose ≥ 3 lines (≥ 15 letters) over a period of 12 months from this later starting point, with 27% expected to lose ≥ 6 lines (≥ 30 letters). 2

To better inform the expected rate of visual acuity decline from the point of conversion to nAMD, we used survival analysis methods to estimate the time to losing ≥ 10 letters from visual acuity before conversion to nAMD, using post conversion to nAMD visual acuity data for the 145 patients who were judged to have developed nAMD in their EDNA study eye. This analysis relied on the fact that some patients had already experienced visual acuity loss by the time of detection, while other patients were not treated immediately. For those not treated immediately, visual acuity data at time of treatment initiation and at 1 year after conversion to nAMD were obtained from the Observing Fibrosis, Macular Atrophy and Sub retinal Highly Reflective Material – Before and After Intervention with Anti-VEGF Treatment (FASBAT) study (ISRCTN 76889861)84 for those EDNA participants who were enrolled or were otherwise extracted from the EDNA participant’s case notes. The time of conversion to nAMD was taken as the mid-point between the visit at which nAMD was detected and the preceding visit (when no nAMD was detected). The time at risk was taken as the time from conversion to nAMD to losing ≥ 10 letters or treatment initiation. If no treatment was initiated, time at risk was censored at the last available follow-up point at which visual acuity data were available. The data were used to generate a Kaplan–Meier curve, to which various parametric survival curves were fitted (Figure 23).

FIGURE 23.

Time to losing ≥ 10 letters from the point of conversion to nAMD, overlaid with the three best fitting parametric distributions.

It can be noted that the data are heavily affected by censoring from an early stage owing to the tendency for early-detected patients to be treated prior to losing ≥ 10 letters, which may bias the curves upwards. Furthermore, a lack of observation time points beyond the point of conversion to nAMD made the exact timing of later visual loss events and the shape of the Kaplan–Meier curve uncertain. Nevertheless, extrapolated curves suggest that 50% of patients can expect to lose ≥ 10 letters by 150–200 days (4.9 to 6.6 months) after conversion to nAMD without treatment. Given the limitations in estimating the exact timing of events and the shape of the distribution, we selected the exponential distribution for the base-case extrapolation. This function was also most consistent with the long-term rate of post-treatment visual acuity decline applied in the model (see below) because it avoids the chance of visual acuity loss in untreated patients dropping below the rate of visual acuity decline in treated patients.

To inform the expected rate of transition to higher levels of visual acuity loss (≥ 30 letters) among those losing ≥ 10 letters, we assumed that the fellow eye prior to the development of nAMD would have had a similar visual acuity to the unaffected EDNA study eye at study baseline. We, therefore, used the difference in visual acuity between the fellow eye and the EDNA study eye at baseline to estimate the proportion of nAMD eyes that could be expected to have lost ≥ 10 or ≥ 30 letters between conversion to nAMD and the time of treatment initiation: 78.4% and 29.2%, respectively. We then assumed that 29.2% of eyes would be expected to lose ≥ 30 letters by the time that the chosen exponential extrapolation (see Figure 23) predicted that 78.4% would have lost ≥ 10 letters from conversion to nAMD (15.15 months). This was used to calculate monthly probabilities of losing ≥ 30 letters from the time of conversion to nAMD (Figure 24).

FIGURE 24.

Extrapolated time to losing ≥ 10 and ≥ 30 letters from the point of conversion to nAMD.

For those modelled to fall into each category of vision loss, the exact number of letters lost was drawn from conditional distributions informed by the observed difference in visual acuity between the fellow eye and the EDNA study eye at baseline. For those losing 10–29 letters, this suggested a uniform distribution (minimum 10 letters, maximum 29 letters). For those losing ≥ 30 letters, a gamma-distribution was applied to account for the skewed nature of the data (mean 42 letters, standard deviation 9.27 letters, minimum 30 letters).

Post-treatment change in visual acuity

Post-treatment visual acuity changes were informed using evidence from several sources: (1) short-term visual acuity outcomes for patients who converted to nAMD in the EDNA study and then commenced treatment; (2) randomised controlled trials of anti-VEGF treatments conducted in the UK NHS; and (3) real-world cohort studies assessing the visual acuity changes in second treated eyes with good visual acuity (> 70 letters) at baseline.

Short-term visual acuity outcomes for participants who converted to neovascular age-related macular degeneration in the EDNA study

Visual acuity outcomes at conversion to nAMD, treatment initiation, post load and 12 months after conversion to nAMD were collected, where possible, for EDNA study participants who converted to nAMD within the study follow-up period. For EDNA study participants who were recruited to the FASBAT study,84 these data were obtained with appropriate permissions from the FASBAT database. For those not recruited to FASBAT, the visual acuity data were extracted from routine case notes.

Based on the above, we were able to obtain usable post-treatment visual acuity data on 93 out of 145 study eyes judged to have converted to nAMD during the EDNA study. The post-treatment changes in visual acuity were categorised by whether or not the individual had lost ≥ 10 letters from baseline at the time of treatment initiation (Figure 25). Among those who lost ≥ 10 letters prior to treatment initiation, mean visual acuity improved by 6.47 (95% CI 2.66 to 10.28) letters following the loading phase. For those treated prior to losing ≥ 10 letters, visual acuity was 1.47 (95% CI –0.22 to 3.17) letters lower after loading. These data were used to inform the loading phase visual acuity changes applied in the model. Following the loading phase, data were available for 23 patients who had lost ≥ 10 letters prior to treatment initiation and 39 who had not. The mean rate of deterioration following the loading phase was very similar between the groups (0.34 and 0.37 letters per month, respectively).

FIGURE 25.

Short-term mean visual acuity outcomes in the EDNA study eye of participants who converted to nAMD during the study.

Visual acuity outcomes in randomised controlled trials

Relevant randomised controlled trials were identified from existing reviews and considered with respect to their applicability to the NHS setting. ^1,85 Although several treatment protocols exist for anti-VEGF injections, systematic reviews have reported broadly comparable efficacy. 86 There is some evidence to suggest that treat-and-extend regimens (treatment interval gradually extended as long as the macula remains dry) are comparable to continuous monthly injections, and slightly superior to pro re nata (when needed) regimens. 86 With respect to medicinal products, there are two licenced anti-VEGF treatments available in Europe: ranibizumab and aflibercept. In addition, there is some limited use of off-license treatment with bevacizumab, which is a larger molecule but otherwise structurally similar to ranibizumab. Several studies have found comparable efficacy between these. 87,88

Given the above, visual acuity changes observed for patients treated with ranibizumab in the NHS-based Inhibit VEGF in Age-related choroidal Neovascularisation (IVAN) trial87,89 were used to help inform the modelling. The ranibizumab arm of the IVAN trial included individuals randomised to continuous monthly injections and discontinuous (as needed) treatment following a pro re nata protocol, with three monthly injections given on re-initiation of treatment. 87,89 For the purpose of the current model, the visual acuity results are considered broadly generalisable to patients treated with any anti-VEGF treatment in routine NHS practice. A further benefit of the IVAN trial was the availability of long-term follow-up data for 537 participants who completed the original 24 months of trial follow-up. 9

With respect to model implementation of post-treatment visual acuity change, we followed the approach described by Claxton et al. ,90 whereby IVAN trial data were used to derive time-dependent normal distributions for visual acuity change per month. The monthly visual acuity change for each simulated individual was determined by taking a random draw from the assigned change distributions. The approach of Claxton et al. 90 was simplified by applying average monthly letter changes across months 3 to 12 and 13 to 24 (Table 41). In addition, given that the IVAN trial included a mixture of eyes with baseline visual acuity above and below 70 letters, we adjusted the loading phase change distributions to yield the loading phase changes in mean visual acuity observed for patients who converted to nAMD during the EDNA study: + 6.47 letters for those who had lost ≥ 10 letters prior to treatment initiation and –1.47 letters for those who had lost < 10 letters prior to treatment initiation.

TABLE 41 - Visual acuity change for treated individuals from treatment initiation to 24 months

Variable	Point estimate (SD)	Source
Monthly visual acuity change (months 1–3) for eyes with visual acuity loss of ≥ 10 letters	2.1567 (2.89)	Chakravarthy et al.87 and Claxton et al.90
Monthly visual acuity change (months 3–12) for eyes with visual acuity loss of ≥ 10 letters	0.00 (2.25)	Chakravarthy et al.87 and Claxton et al.90
Monthly visual acuity change (months 13–24) for eyes with visual acuity loss of ≥ 10 letters	–0.0917 (2.17)	Chakravarthy et al.87 and Claxton et al.90

Longer-term visual acuity changes by visual acuity at treatment initiation

Given that most treated eyes in IVAN were symptomatic at baseline (mean visual acuity 61.4 letters), further data were required to inform the modelling for patients detected early and treated prior to significant vision loss. To do this, we reviewed studies reporting treatment outcomes for second and early treated eyes with good visual acuity at baseline. 14,16 In both studies,14,16 second-treated eyes with good visual acuity at baseline did not show the gain in vision that symptomatic first eyes show following treatment loading, confirming the observation in EDNA study participants. By contrast, these eyes remained stable over the loading phase before declining gradually at a similar rate to the first treated eyes after the loading phase. Vision in the second treated eyes remained better than vision in first treated eyes at all time points up to the end of the observed follow-up periods of 216 and 314 years. Thus, for second eyes treated prior to significant vision loss, this trajectory was assigned in the model by capping visual acuity at the baseline value and calibrating the visual acuity change distributions to yield output in line with the above observations.

Beyond 24 months, we specified a monthly visual acuity change distribution based on the average annual rate of decline observed during the long-term follow-up of IVAN participants: 4.3 letters per year (95% CI 3.7 to 4.9 letters). This was applied to eyes that initiated treatment prior to and after losing ≥ 10 letters. The modelled mean trajectories for eyes treated before and after significant vision loss are provided in Model validation below.

Valuation of health effects

The available health state utility data for quality-adjusting survival by visual acuity status were identified from searches of the published literature. Searches identified numerous potential sources. 93–101 An emphasis was placed on identifying values for visual acuity states based on the preferences of the UK general population. Given that there was some evidence to suggest that the EuroQol-5 Dimensions lack sensitivity to change in visual acuity,102 directly elicited values for specific visual acuity states were considered where these had been elicited from a sample of the UK general population using an appropriate valuation technique.

Based on consideration of the available evidence, we chose to base the visual acuity health state utilities on time trade-off values reported by Czoski-Murray et al. 101 In this study, contact lenses were used to simulate visual impairment in participants from the UK general population. Participants valued the simulated states using the time trade-off methodology, providing utility weights on a zero (death) to one (full health) scale. Czoski-Murray et al. 101 also reported a regression analysis of the health state utility values by visual acuity in the better seeing eye, with participant age included as a covariate. In line with the National Institute for Health and Care Excellence (NICE) appraisals of ranibizumab103 and aflibercept104 for the treatment of nAMD, we used this published equation to assign health state utility weights by visual acuity in the model, where SE is standard error and MAR is minimum angle of resolution:

Mortality

In line with the modelling used to support NICE guidance on the use of ranibizumab and aflibercept in the NHS, mortality was based on UK general population life tables, with risk ratios of 1.1 (95% CI 1.01 to 1.19) and 1.17 (95% CI 1.07 to 1.27) applied for visual acuity of < 75 letters and visual acuity of < 60 letters, respectively. 105 Thus, strategies that maintain higher visual acuity were modelled through evidence of association to provide a small survival benefit. The impact of removing these risks ratios was assessed in a sensitivity analysis.

Resource use and costs

The model included costs of testing and monitoring conversion to nAMD, costs of monitoring and treatment post conversion to nAMD, and health-care and personal social care costs associated with severe visual loss. All costs were expressed in 2018/19 Great British pounds.

Model validation

Prior to running the analysis, the modelled visual acuity outcomes were checked against the input sources. Figure 26 shows the modelled mean visual acuity trajectories for typical patients commencing treatment before (early treatment) and after (late treatment) losing ≥ 10 letters. It indicates the benefit of treating eyes prior to vision loss, with better visual acuity maintained over time than that for eyes treated after a significant amount of vision loss. 14,16 The trajectories are for patients who continue treatment and, therefore, do not account for those who are discharged with stable vision. Therefore, the modelled reduction in visual acuity for the treated cohort was slightly lower than that shown in Figure 26.

FIGURE 26.

Model output: visual acuity of continually monitored/treated patients by time since treatment initiation (EDNA model). Note that with application of a visual acuity cap in the model, there is a downwards pressure on visual acuity change between month 3 and month 24, which slightly overestimates the expected visual acuity during this time-period compared with the IVAN trial.

Model analysis

The analysis used first-order Monte Carlo simulation to propagate the passage of individuals through the model one at a time. A total of 200,000 individual trials were found to produce stable results. In line with the NICE reference case, future costs and QALYs were discounted at a rate of 3.5% beyond year 1. 112

Incremental cost-effectiveness ratios (ICERs) that expressed the additional cost per QALY gained were estimated by comparing each monitoring strategy with the next less costly strategy (excluding those strategies that are more costly and less effective than an alternative option). In the UK NHS, a threshold value of £20,000 to £30,000 is normally used to make judgements on cost-effectiveness of health technologies.

Probabilistic sensitivity analysis was conducted to characterise the joint uncertainty around the model outputs (incremental costs and QALYs) arising from the uncertainty around the model input parameters. Probability distributions were assigned based on the mean and the measure of variance reported for each parameter in the preceding sections (see Clinical effectiveness parameters, Valuation of health effects, Mortality and Resource use and costs). Beta-distributions were assigned to sensitivity and specificity parameters, gamma-distributions were used for costs of testing and the cost of blindness (assuming a standard error of 10% of the mean), and normal distributions were applied for regression coefficients (including the utility increment associated with change in visual acuity and age, and the exponential rate parameters for time to conversion to nAMD and time to visual acuity loss of ≥ 10 letters). Normal distributions were also used for the mean change in visual acuity during the loading phase, the long-term change in visual acuity per year for treated individuals and the probabilities of treatment discontinuation and treatment re-initiation. Second-order Monte Carlo simulation with an inner loop of first-order simulation (n = 5000) was then used to analyse the model for 1000 random draws from the assigned input probability distributions. The output from this analysis provided the probability of each monitoring test being the preferred strategy by increased cost-effectiveness thresholds. 113 Further deterministic scenario analyses were also undertaken to assess the impact on findings of uncertainty surrounding key model input parameters and structural assumptions (see Results).

Base-case assumptions

Several key assumptions of the base-case analysis are as follows:

• Time to conversion to nAMD follows an exponential distribution, that is a constant probability of conversion over time.

• Time to losing ≥ 10 letters from time of conversion to nAMD to treatment initiation is informed by exponential extrapolation of the EDNA study Kaplan–Meier data.

• All positive diagnostic monitoring test results are confirmed or refuted by FFA.

• New visual acuity loss of > 30 letters in undetected nAMD triggers self-referral and detection regardless of the testing strategy.

• New visual acuity loss of 10–29 letters in undetected nAMD triggers self-referral and detection in 50% of individuals.

• Undetected nAMD with a new visual acuity drop of ≥ 10 letters is detected by the visual acuity testing strategy at the next monitoring visit.

• Long-term monitoring of second eyes that do not develop nAMD continues indefinitely beyond 36 months (twice per year from 5 years onward).

• Long-term monitoring in patients who do not develop nAMD in the second eye is driven by the second eye beyond 5 years in the model.

• Following conversion to nAMD and detection, patients are treated immediately with anti-VEGF injections.

• Individuals treated prior to experiencing a significant visual acuity loss (≥ 10 letters) experience no gain in vision during the loading phase.

• Individuals treated after experiencing a significant visual acuity loss (≥ 10 letters) experience a loading phase gain in line with the gain observed in EDNA study participants who converted to nAMD during study follow-up.

• A percentage of patients discontinue treatment with stable disease over time but can reinitiate treatment.

Base-case results

The results of the base-case analysis are provided in Figure 27 and Table 46. They indicate that more sensitive monitoring strategies generate increased health benefits at a lower cost, with the visual acuity change test being least effective and most costly and OCT being most effective and least costly. It can be noted from Figure 27 that OCT lies at the bottom right of the plane (lowest cost and highest QALYs), dominating all other test strategies.

FIGURE 27.

Cost-effectiveness results: base case.

Table 47 disaggregates the total costs accrued in the model under each testing strategy. The index tests with lower sensitivity and specificity accrue moderately higher prediagnosis costs than more sensitive/specific strategies as a result of more visits prior to detection and increased chances of a false positive result. More sensitive index tests accumulate higher post-diagnosis monitoring and treatment costs owing to earlier detection. However, the increased costs of earlier treatment are countered by reduced costs associated with visual impairment and blindness. It is the lower costs of blindness resulting from earlier treatment with OCT that makes it the dominant strategy in the base case.

Table 48 further illustrates the benefits of OCT compared with the other strategies in terms of the expected time from conversion to nAMD to detection and treatment, and the mean visual acuity at time of treatment initiation. The modelling suggests that, compared with using visual acuity alone, OCT monitoring will bring detection forward by approximately 7.5 months, for a gain in visual acuity at time of treatment initiation of approximately 16 letters. It is this earlier initiation of treatment and maintenance of better visual acuity that drives the QALY gains compared with the other strategies.

Sensitivity analysis

To assess the sensitivity of the model findings to key assumptions and input parameters, a series of scenario analyses were undertaken. These are outlined as follows:

1. The Weibull distribution applied to model time to conversion to nAMD, giving an increasing hazard of conversion over time.

2. The log-normal distribution applied to model time to conversion to nAMD, giving a decreasing hazard of conversion over time.

3. The log-normal distribution applied to model time from conversion to nAMD to significant vision loss (≥ 10 letters).

4. The generalised gamma-distribution applied to model time from conversion to nAMD to significant vision loss (≥ 10 letters).

5. Pre-treatment visual acuity loss conditional on dropping 10–29 and ≥ 30 letters assumed to be skewed towards the lower end of the visual acuity loss ranges: mean of 12 and mean of 32 letters, respectively.

6. A reduced longer-term rate of post-treatment visual acuity decline applied: 3.1 letters per year in line with the average rate of decline estimated for younger patients (aged 70 years) in the IVAN long-term follow-up study.

7. A reduced longer-term rate of post-treatment visual acuity decline applied: 2 letters per year in line with the average rate of decline estimated for younger patients (aged 60 years) in the IVAN long-term follow-up study.

8. Distribution for long-term post-treatment rate of visual acuity loss per year assumed to be right-skewed and constrained by 0; gamma distribution with mean of 4.3 and standard deviation of 7.

9. A reduced rate of treatment discontinuation for stable vision applied (0.0092), based on data from Chandra et al. 91 (applied independent of visual acuity outcome at 24 months post treatment).

10. Removal of treatment discontinuation for stable vision (an extreme scenario to assess the impact of this uncertain parameter).

11. An increased rate of treatment re-initiation (approximately 2.9% per month) following discontinuation for stable disease, in line with data reported by Madhusudhana et al. 92

12. 100% test sensitivity applied for all tests at the subsequent monitoring visit for those who lose 15–29 letters as a result of nAMD.

13. Removal of excess mortality associated with visual impairment.

14. Removal of costs of blindness.

15. Wait-to-treat policy, which assumes that visual acuity must drop below the threshold specified for ranibizumab and aflibercept in NICE guidance (≤ 70 letters).

16. Treatment instigated following OCT positive findings, without confirmation with FFA – this assumes that any patient receiving a false positive OCT result incurs 12 months’ worth of anti-VEGF treatment inappropriately, before being identified as morphologically unchanged and treatment withdrawn.

17. Increased test monitoring costs as per the increased overhead scenario outlined in Appendix 4.

The results of these scenarios are provided in Table 49. They indicate that OCT remains dominant in all but five scenarios (4, 5, 14, 15 and 16). Scenario 4 results in a significant proportion of patients with nAMD (≈ 10%) maintaining stable vision without treatment to 10 years post conversion to nAMD. Under this scenario, OCT results in the early treatment of these patients at increased cost but with no benefit over the visual acuity strategy. This undermines the cost savings and QALY gains. Nevertheless, the ICER for OCT (£9040 per QALY gained) remains well below the accepted thresholds for cost-effectiveness. Scenario 5 minimises the penalty of missing cases with less-sensitive strategies. Again, the ICER for OCT becomes positive in this case but well below the accepted thresholds. When costs of blindness are removed (scenario 14), OCT becomes the costliest strategy but, again, its ICER remains favourable against the accepted thresholds even in this extreme scenario. In scenario 15, assuming that clinicians would wait for vision to drop below 70 letters before treating, the QALY gain associated with OCT is greatly diminished and its ICER increases to £19,488. It can be noted from this scenario that the costs of all testing strategies are higher and the QALYs lower than those for the base case, which assumes that treatment is initiated on detection. This indicates that early treatment dominates delayed treatment. Scenario 16, assessing the impact of initiating treatment following an OCT positive finding and forgoing a FFA, results in the ICER increasing to £17,256. Again, the base-case OCT strategy, with confirmatory FFA, dominates this alternative approach.

It should be noted that for all of these scenarios list prices for ranibizumab and aflibercept were applied. It is known that price discounts are available to the NHS via agreed patient access schemes for both of these drugs. The discounted price is confidential and, therefore, could not be applied in the model. However, a lower unit cost for ranibizumab and/or aflibercept would further improve the cost-effectiveness of OCT.

Probabilistic sensitivity analysis

Results of the probabilistic sensitivity analysis are provided in Table 50. They show consistency with the base-case deterministic results, with OCT dominating all other monitoring tests. The cost-effectiveness acceptability curve (Figure 28) indicates that OCT has a very high chance of being the preferred strategy across the range of cost-effectiveness thresholds applied by NHS decision-making bodies.

FIGURE 28.

Cost-effectiveness acceptability curve.

Secondary analysis using test combinations

In a secondary analysis, we assessed the cost-effectiveness when including two combination strategies: OCT or fundus clinical evaluation (test accuracy reported in Table 14), and a combination of all other tests apart from OCT (fundus clinical evaluation or Amsler or self-reported vision or visual acuity: test accuracy reported in Appendix 1). As for the single index tests, the specificity of each of the combination tests was also adjusted to account for the lack of independence across multiple observations within individuals to 0.9915 and 0.9673, respectively. The results are provided in Table 51. Under the base-case model specification, the OCT or fundus clinical evaluation combination resulted in a very small gain in QALYs compared with OCT alone. However, the increased costs resulted in the ICER being above the accepted thresholds for cost-effectiveness compared with OCT alone. The combination of all other tests, excluding OCT, resulted in higher costs and lower QALYs than OCT alone and the combination of OCT or fundus clinical evaluation.

Strengths and limitations

The modelling reported in this chapter has several strengths. By adopting an individual simulation approach, individual visual acuity could be modelled as a continuous variable, which increased the sensitivity of the model to capture the impact of changes in visual acuity on health-related quality of life. Key inputs around time to conversion to nAMD and diagnostic accuracy were informed by the prospective EDNA study data, which supports the internal and external validity of the model findings. Furthermore, pre-detection testing costs and post-detection treatment costs were also based on the resource use that was reported by EDNA study participating centres and pragmatic NHS-based trials, ensuring generalisability of costs across the NHS. Finally, post-treatment visual acuity changes were carefully informed by a range of sources applicable to NHS routine practice and capture the expected differences in visual acuity trajectories by the degree of visual loss in the second eye prior to treatment initiation, ensuring that expected visual acuity gains from early identification and treatment were accurately captured.

All of the modelled strategies assumed a confirmatory FFA after a positive index test and before treatment. For this reason, the OCT strategy reflects the current NICE diagnosis guidelines to offer OCT to ‘people with suspected late AMD (wet active)’ and FFA to confirm diagnosis if OCT does not exclude neovascular disease. 11 Discussions within the EDNA project team revealed that clinical practice might vary towards initiating treatment based on OCT results without a confirmatory FFA. A scenario reflecting this was modelled. Although a strategy with OCT and without confirmatory FFA would be cost-effective at the usual UK cost-effectiveness thresholds,112 this strategy is no longer cost-saving. These results suggest that it may be more cost-effective to first confirm the diagnosis of nAMD in an OCT positive case using FFA, rather than initiating treatment for all OCT positive cases immediately. This is because of the chance of inappropriately treating small numbers of false positive patients with expensive anti-VEGF drugs for some considerable time. However, it may be that clinical judgement can be applied in practice to efficiently circumvent the need for FFA in some OCT positive cases without risking inappropriate overtreatment of false positive patients.

With respect to limitations, there is a paucity of available data on the rate of visual acuity loss in the immediate period following conversion to nAMD. Although the EDNA study was able to provide a reasonable estimate of the proportion expected to lose ≥ 10 letters within 3–6 months of conversion to nAMD, the increasing tendency to treat patients early (prior to significant vision loss) and limited observations on the few patients who experienced delayed treatment results in a degree of uncertainty around this important parameter. However, more conservative extrapolations were explored and the ICER for OCT remained favourable under these. It would take a relatively large proportion of nAMD cases to remain stable without treatment to undermine the cost-effectiveness of an OCT monitoring and immediate treatment strategy. Nevertheless, the rate of visual acuity deterioration in the immediate period following conversion to asymptomatic nAMD, prior to initiation of treatment, is one of the most important parameters in the economic model. This could benefit from further analysis of routine observational cohort data, where a significant proportion of early identified patients have undergone a period of monitoring prior to treatment initiation.

To ensure that the impact of visual loss in the EDNA study eye (second eye) was not underestimated, the second eye was assumed to represent the better seeing eye over the model time horizon. Although this will probably hold true for the majority of patients, there will be some patients whose visual acuity in the second eye drops below that of the first eye. Thus, the model may slightly overestimate the health-related quality-of-life benefits and cost savings of early detection and treatment. Given that a relatively small proportion of patients lost ≥ 10 letters prior to detection of nAMD in the EDNA study, there is remaining uncertainty relating to the sensitivity of the different diagnostic tests following substantial visual acuity loss owing to nAMD. Therefore, several assumptions were required to model the expected time to diagnosis for such patients. This included an assumption that the visual acuity test would detect all patients at the next monitoring visit if visual acuity dropped ≥ 10 letters from baseline (the trigger for the test as used in EDNA). This is a favourable assumption for the visual acuity strategy because it assumes that no such cases would be missed as a result of measurement error. Despite this, the model suggests that the visual acuity strategy results in the longest delay from conversion to detection and treatment and that it generates the lowest QALYs of all the strategies. A further assumption related to the rates of self-presentation between scheduled monitoring visits in the event of significant visual loss occurring. For the base case, it was assumed that anyone with a drop in visual acuity of > 30 letters would present immediately, while 50% of those with a drop in visual acuity of 10–30 letters would also present immediately. However, the model findings were not found to be particularly sensitive to assuming that the remaining 50% of the latter group also present more rapidly. Nevertheless, the degree of visual loss that has occurred when nAMD in the second eye is identified by changes in visual acuity, or subjective tests of visual function, remains an area of uncertainty that could benefit from further data collection. Again, the analysis of existing observational data sets could help to inform this parameter. Finally, there were limited data available to inform the rates of treatment discontinuation for stable disease and long-term rates of treatment re-initiation with contemporary anti-VEGF maintenance regimens.

Comparison with other studies

We are not aware of any other studies that have explicitly assessed the cost-effectiveness of alternative diagnostic monitoring strategies for nAMD in the second eye of patients being treated for nAMD in the first eye. A previous health technology assessment assessing the cost-effectiveness of OCT for the diagnosis and/or treatment monitoring of individuals with nAMD reported a high degree of uncertainty related to the optimal use of OCT in the nAMD monitoring pathway. 111 This was because of identified uncertainty around the diagnostic accuracy of OCT. The current study has provided robust evidence for the high diagnostic accuracy of OCT as a monitoring test for the early detection of nAMD in the second eye, and consequently provides evidence that when coupled with early treatment it is likely to offer a highly cost-effective strategy in this context.

With respect to treatment, many studies have evaluated the cost-effectiveness of anti-VEGF injections, which are widely accepted as a cost-effective use of NHS resources under agreed confidential patient access schemes. However, the majority of studies have assessed the cost-effectiveness of treating symptomatic individuals with visual acuity of 70 letters or below. We are aware of only one study that has assessed the cost-effectiveness of immediate ranibizumab treatment in patients with visual acuity better than 70 letters at detection compared with waiting for visual acuity to drop below 70 letters. 114 Using data from the nAMD UK database on patients with visual acuity above 70 letters at detection, Butt et al. 114 were able to estimate time to dropping below 70 letters without treatment and transition probabilities between visual acuity states for treated patients. Owing to the rapid decline in visual acuity in the early period after conversion to nAMD, their modelling predicted that immediate treatment would maintain better visual acuity and offer a cost-effective use of NHS resources in the short to medium term (2–10 years). Our results are consistent with respect to the finding that early detection with OCT followed by immediate treatment offers a highly cost-effective approach compared with relying on a drop in visual acuity to detect and treat nAMD in the second eye. With a 25-year horizon and the inclusion of costs associated with severe visual impairment (not included in the modelling reported by Butt et al. 114) we found that this strategy dominated the other alternatives assessed.

Policy implications

The data and modelling reported in this chapter strongly suggest that the use of OCT, compared with other available diagnostic tests, can lead to substantial reductions in the time to diagnosis and treatment of nAMD in the second eye of patients being treated for nAMD in their first eye. The early initiation of treatment in the second eye, based on FFA-confirmed OCT-positive findings, can be expected to maintain better visual acuity and health-related quality of life over time than the less sensitive monitoring strategies. Moreover, our modelling suggests that this strategy may be cost-saving in the long term compared with less-sensitive monitoring strategies that result in greater visual acuity loss occurring before treatment is initiated.

Diagnostic accuracy

The primary reference standard was the diagnosis of the onset of nAMD in the study eye based on interpretation of the FFA by the clinician at the clinical site. Against this reference standard, the sensitivity of OCT in the detection of new-onset nAMD was 91.7% (95% CI 85.2% to 95.6%) and the specificity was 87.8% (95% CI 83.8% to 90.9%). Two alternative reference standards that we prespecified were (1) the inclusion of an additional set of patients who did not have a FFA but for whom the clinician made the diagnosis based on any other available modality, and (2) a parallel determination of nAMD confirmed at a specialised reading centre. In this latter reference standard, no other information apart from images were available to the graders and, therefore, no associated cues, such as clinical history, could be used. We also tested the sensitivity and specificity of the index tests in three specified periods of follow-up: (1) the entire length of the study, (2) index tests limited to the last 6 months of follow-up and (3) limited to the last visit.

Regardless of the reference standards and the time periods over which index tests results were used, OCT had the highest sensitivity (82–92%) and was consistently and significantly superior to all other index tests. Although OCT performed extremely well for specificity with a range of 88–96%, fundus clinical examination and self-reported vision reached comparable levels. However, neither of these two tests performed well for sensitivity, which was significantly lower and not at an acceptable level (lower than 60% for fundus clinical examination and 10% for self-reported vision).

We also examined the sensitivity and specificity of pair-wise combinations of OCT with each of the other index tests. When either test was positive, sensitivity increased for most, with a best performance of 96% for the combination of OCT and visual acuity. However, specificity decreased for all combinations except for OCT combined with fundus clinical examination. This was not surprising given that the index tests apart from OCT were affected by a large number of false positives. When both of the pair-wise combinations of tests were positive, sensitivity was markedly reduced for all combinations compared with OCT by itself. With this approach, specificity was seen to increase for all combinations. We conclude from the analysis of combinations of index tests that OCT alone is sufficiently sensitive and specific for the detection of new-onset nAMD. The combination of all alternative tests against OCT further emphasised the value of OCT in a monitoring setting over the other tests.

Because the nAMD phenotype results in different morphological changes in the retinal tissue compartments, there can be differential impact on function. Notably, the RAP phenotype results in fluid pockets within the retina and, therefore, is more likely to result in distortion and worsening function as soon as nAMD develops. Therefore, we prespecified a subgroup analysis comparing CNV with the RAP subtype. OCT achieved 100% sensitivity in the detection of participants who developed RAP nAMD. However, none of the other index tests showed the same sensitivity or better specificity for RAP. Additional planned sensitivity analyses using two different reference standards for conversion to nAMD – (1) based on clinical examination without a FFA and (2) reading centre determination – also yielded results that were highly similar to that observed in the main analysis.

Patient-reported tests (Amsler/self-reported vision)

Although OCT allows for a non-invasive appreciation of macular morphology, it requires appropriate instrumentation and technical expertise for acquisition and expert clinical opinion for interpretation. Therefore, we also included two index tests that are easily self-performed, that reflect retinal function and are widely used in practice. Participants undertook the Amsler test to determine the onset of distortion and, using a questionnaire, reported any reduction in vision in the EDNA study eye. 28,31 It was notable that neither of these tests achieved even moderate levels of sensitivity or specificity. To some extent, the finding that the Amsler test was insensitive to new-onset neovascularisation is not surprising. Despite ongoing use by the optometric community and ophthalmologists, the inadequacy of the Amsler test for detecting onset of nAMD is extensively recorded in the literature, with sensitivities as low as 34%. 115 A systematic review and meta-analysis evaluating the diagnostic accuracy of the Amsler chart produced a pooled sensitivity of 78% (96% CI 64% to 87%). 116 The Amsler tests’ failure is likely because of perceptual completion, a well-recognised visual phenomenon in which surrounding features provide sufficient contextual information that the scotoma or blind spot is ‘filled in’ in the absence of neural input. 117

We were surprised at the poor sensitivity of self-reported symptoms of visual change as a marker for the onset of nAMD in the EDNA study eye. The definition of a positive test was a self-reported recognition that vision in the study eye was ‘much worse’ since the previous visit. We anticipated that the better functional status of the study eye than the fellow eye would have resulted in a greater awareness of the onset of distortion owing to nAMD. Changing the definition to include a status of either ‘worse’ or ‘much worse’ increased sensitivity only from 4.2% to 15.3%. Our finding of a poor sensitivity for self-reported vision change even in the better seeing eye is, therefore, of interest because studies investigating patient-reported outcomes during AMD treatments have consistently observed greater concordance with disease severity in the better seeing eye than that of the worse seeing eye. 118 Our data are at variance with the results from the Age-related Eye Disease Study (AREDS),119 which evaluated the responsiveness of the National Eye Institute Visual Function Questionnaire (NEI-VFQ) to progression to advanced AMD. 25 The overall score was responsive (t = 14.0; p < 0.001) and showed moderate discrimination (AUC 0.74), and the subscales most sensitive to progression to either nAMD or geographic atrophy included general vision (AUC 0.65), near activities (AUC 0.73) and driving (AUC 0.72). One possible explanation is that, similar to the insensitivity of the Amsler in the detection of nAMD onset, fluid accumulation and leakage, which is more likely to be associated with metamorphopsia, is counteracted by perceptual filling-in rather than photoreceptor loss. However, it is also possible that the wording of our question was suboptimal for ascertaining vision changes. Our findings suggest that, although patient-reported outcomes are important in intervention trials to assess improvements in vision-related quality of life, they should not be used as tools for the detection of progression to nAMD. In this context, our findings also highlight the importance of regular eye examinations in the most at-risk age groups because relying on detection based on self-reported symptoms will almost certainly result in a lower detection rate and presentation at a more advanced stage of the disease.

Visual acuity

Visual acuity is another easily performed test and reflects central macular function. A visual acuity drop of 10 letters, representing a fall of two lines on the ETDRS chart, has been recognised as a clinically important change in function; in the EDNA study this was selected as the trigger for a FFA. However, the change in visual acuity in the EDNA study eye proved to be a poor marker for conversion to nAMD. visual acuity measurements were available at frequent time intervals in EDNA participants because fellow eyes were on active treatment for nAMD, which necessitated regular visits to clinics for assessment and intravitreal injections. Changes in visual acuity can occur because of medial opacities (cataract) and/or from the development of focal areas of outer retinal atrophy in the macular tissues. Degeneration of drusen leaving behind focal patches of atrophy is a recognised component of the pathogenetic pathways leading to nAMD. In fact, we observed that around half of all EDNA study eyes were graded as exhibiting foci of macular atrophy at the time of conversion to nAMD. This type of atrophy may be accompanied by imperceptible slow reduction in visual acuity, which is not recognised by the patient. It is also of note that the area of the neural sub-retina serving high-contrast distance visual acuity is small (< 1 °) and in some cases the location of the neovascularisation falls outside and in such cases visual acuity may be unaffected. 120 Changing the definition of a positive visual acuity test to a drop of 15 and 30 letters did not alter the sensitivity or specificity of the test.

Morphology

We performed detailed imaging in the EDNA study participants at initial presentation. Thus, we were able to compare visual acuity and morphological features of the nAMD lesion in the first eye with those at detection of nAMD in the EDNA study eye. We expected that first-presenting eyes would have only moderate impairment of vision and that the nAMD lesions would not be advanced because (1) we selected participants only at their very first presentation with nAMD and (2) of the increasing awareness of AMD in the UK population with the introduction of anti-VEGF treatments, which, in turn, resulted in a recognition at the primary-care level of urgent referral pathways into specialist retinal clinics. 1,121,122 In EDNA study eyes at nAMD detection, the average visual acuity was 74 letters compared with 54 letters in corresponding fellow eyes at initial presentation. Furthermore, first-eye lesions even at initial presentation often had severe exudative features. Comparing the features of nAMD lesions in study eyes at conversion with the features of fellow eyes at initial presentation, central subfield thickness was on average 150 µm less, lesions were half to one-third smaller and more likely to be juxta foveal or extra foveal, explaining the large differences in visual acuity between these groups. In addition, and of note, was the presence of atrophy in the macula in some 10% of eyes indicating that neural cell loss had already occurred. However, there was a low frequency of fibrosis in the nAMD lesion confirming that our selection criteria eliminated patients with long-standing disease. A high proportion of advanced nAMD lesions develop fibrosis and macular atrophy despite anti-VEGF therapy and, therefore, treating them while smaller, and prior to involvement of the fovea, may indeed prevent permanent foveal damage. 10,23 Vision lost to nAMD cannot be restored to normal levels even with intensive anti-VEGF therapy, as shown by studies that have examined large repositories of data collected in the real world. 9

The foregoing observations highlight key novel points justifying the need for early diagnosis and treatment of nAMD. Data from electronic medical records show that presenting visual acuity is a predictor of final visual acuity. Early diagnosis results in an average acuity that can be as much as 20 letters better than that of even a marginally later diagnosis. The institution of treatment at a stage when neither fibrosis nor atrophy has supervened and when lesions are small and less exudative can allow the preservation of visual acuity at good levels. 14

Prediction model

There were 145 conversions to nAMD (using secondary reference standard 1) out of 489 participants in the EDNA study, with the data available on their outcome used to develop the prediction model. Over the median follow-up duration of 34 months, we observed an approximate conversion rate of 30%. This rate was in keeping with the expected conversion rates observed in previous reports. For example, the risk of conversion to nAMD in the unaffected eye in individuals with unilateral nAMD is around 10% in the first year and some 50% by 5 years. 68

The predictors that were significant and used in the final model included lower visual acuity, cigarette smoking, family history of AMD, thin choroid, retinal thinning, presence of retinal pigmentary abnormalities, higher category of the most frequent or maximum size of drusen, and presence of nodular drusen or both nodular and reticular pseudodrusen. Age was not observed as a significant predictor. The mean age of this cohort was 77 years, which is no different from most other studies on nAMD. It is notable that age has not been shown to be a consistent risk factor. The availability of longitudinal data sets from clinical trials that have enrolled patients with nAMD has been exploited and used in secondary analyses as these repositories are generally robust with high quality information from both eyes. An analysis of the Harbor study, which analysed risk factors for the development of nAMD in fellow eyes of patients with nAMD in the first eye only, did not include age in the final model. 69,70 By contrast, a recent analysis of the VIEW (VEGF trap-eye: investigation of efficacy and safety in wet AMD) 1 and 2 studies showed that increasing age, female sex and lesion size in the nAMD eye were among the predictors of conversion to nAMD in unaffected fellow eyes. 71

It has been recognised for many years that the presence of nAMD in the first eye is by itself a strong risk factor. The simplified AREDS scale used only macular phenotype to predict progression from the non-neovascular stage to nAMD. In this scale, 2 points were assigned if one eye had nAMD and additional points could be given if the fellow eye had large drusen (1 point) and retinal pigmentary irregularities (1 point), underscoring the strength of nAMD in one eye as a key risk factor. 73 It was shown that having nAMD in one eye carries a two- to three-fold greater risk for conversion to nAMD than that of a state of bilateral intermediate AMD. 73

The progression from unilateral to bilateral nAMD in the AREDS cohort was seen in 35.4% of patients over 5 years if the fellow eye had additional risk factors of the presence of large drusen and/or pigment abnormalities. This figure increased to 76.8% at 10 years. Therefore, large drusen and RPE abnormalities are considered strong risk factors for conversion to nAMD. 75–79 Where information from OCT was available, hyper-reflective foci and drusen were the strongest contributors to predictive models. 70,80 Ferrara et al. 117 compared 40 patients who did not progress from unilateral to bilateral nAMD with another 40 patients who did progress. These patients were matched on an AMD baseline grade and follow-up interval and the comparison showed that OCT features, including total retinal thickness, pigmentary hyper-reflective material, nascent geographic atrophy and choroidal vessel abnormalities, were independently associated with progression to either form of advanced AMD (geographic atrophy or nAMD). 117 Given that both forms of advanced AMD were included in this small study, it is difficult to decipher the specific risks of nAMD. For Submacular Surgery Trials participants who had a FFA-confirmed absence of nAMD in one eye at baseline, drusen size, focal hyperpigmentation and non-foveal geographic atrophy were significantly and independently related to progression to nAMD within the 2-year follow-up. 123 No non-ocular characteristic (age, sex, history of hypertension or smoking) or ocular feature of the nAMD eye at baseline were predictive of progression to nAMD. An artificial intelligence approach to OCT feature analysis highlighted drusen volume and presence of hyper-reflective foci as significant predictors of progression to nAMD using spectral-domain OCT images from the HARBOR study. 124 Although reticular pseudodrusen are known risk factors for nAMD, they ranked only 11th out of 15 OCT risk features for conversion to nAMD in the fellow eye of the HARBOR participants when the model was developed using machine learning. 70 This may be because providing the ground truth for reticular pseudodrusen is challenging and they can disappear over time but the eye will retain the high risk for conversion to nAMD.

The ocular predictors in our model included nodular drusen with or without reticular pseudodrusen. The presence of nodular drusen in fellow eyes increased the risk of conversion to nAMD by six-fold compared with the absence of drusen, whereas the combination of nodular drusen and reticular pseudodrusen almost doubled this risk, highlighting that the presence of both drusen types was the strongest predictor for conversion. In a small study that observed 83 patients over a mean follow-up time of 2.8 years, Nathoo et al. 115 showed that drusen load (area and volume) was associated with progression to nAMD.

Decreasing visual acuity was also a predictor in our model, in keeping with previous studies. 78,81 When considering non-ocular risk factors, our models contained smoking status and family history of AMD. We did not have data on genetic risk. Seddon et al. 72 included a combination of demographic, genetic and environmental factors, and macular phenotype on colour photographs to develop and validate a model to predict progressors compared with non-progressors in the AREDS cohort. However, progressors were defined as both bilateral early or intermediate AMD and those with advanced AMD in one eye at baseline who progressed to advanced AMD.

Health economic analysis

The results of the base-case economic analysis show that OCT is expected to generate the greatest number of QALYs per patient (OCT, 5.830; fundus, 5.787; Amsler chart, 5.736, self-reported vision, 5.630; and visual acuity, 5.600) for the lowest health-care and social care costs (OCT, £19,406; fundus, £19,649; Amsler chart, £19,751; self-reported vision, £20,198; and visual acuity, £20,444) over the lifetime of the simulated cohort. The increased treatment costs associated with earlier detection with OCT were more than offset by the reductions in costs associated with severe visual impairment, compared with less-sensitive tests that resulted in greater visual acuity loss prior to detection and treatment. OCT was found to dominate the other tests or to have an ICER below accepted cost-effectiveness thresholds across the range of scenarios analyses explored. The probabilistic sensitivity analysis indicated a high probability of OCT being cost-effective across a range of cost-effectiveness thresholds typically applied by NHS decision-making bodies.

Patient perspectives of monitoring and diagnosis of neovascular age-related macular degeneration

Three patient representatives within our steering committee who had nAMD and experience of being monitored for disease in the second eye fed back their opinions and perspectives on the diagnosis of and monitoring for nAMD. Throughout the study, they highlighted a number of concerns and suggestions from a patient perspective.

Of particular concern was the poor performance of the patient-reported index tests (Amsler, visual acuity and patient-reported vision), especially considering that all three patient partners felt that self-monitoring had been very important to detect changes in their own eyesight. A highlighted challenge for patients (with and without nAMD) is how to monitor their vision consistently and the need for patient tools to support consistent self-monitoring so that it can be effective. Our PPI partners regularly highlighted a need for better education for the public on self-awareness of their eye health and understanding the importance of self-monitoring within the health-care setting. Considering the average age of the participants (aged > 70 years), the difficulties that some older people face in doing routine tasks and/or in following processes, and the possibility that vision change may be perceived to be just because of old age, may in some way explain the poor performance of the patient-reported index test results. A further challenge for patients is to understand which changes in vision are significant, especially because vision may be variable. Any well-meaning feedback from clinical teams reassuring the patient that their perceived deterioration is not clinically significant may reduce the incentive for the patient to monitor and report further deterioration. Further research to identify patient tools to help with self-monitoring and awareness in those patients not being routinely monitored was felt to be of benefit.

Diagnostic study

A number of strengths can be highlighted for the diagnostic study. The EDNA study was a large, multicentre, prospective, comparative, diagnostic accuracy study that evaluated routinely used diagnostic tests in a monitoring setting under standard of care. The benefit of the large sample size is reflected in the precision with which sensitivity and specificity were estimated, and the large number of centres gives confidence in the generalisability of findings to the NHS.

It was a strength that the interpretation of the index tests was undertaken by investigators who were unaware of the results of the reference standard. The compliance with conducting index and reference standard tests in accordance with the study protocol was extremely high.

We used a reference standard that represents standard care, that is FFA interpreted by an expert clinician. We also explored the performance of the index tests using a reading centre to interpret the FFA results as an enhanced reference standard. It was notable that the discordance between the clinician diagnosis of nAMD and the reading centre diagnosis of nAMD was extremely low.

Among the limitations, we recognise that the population enrolled in the EDNA study consisted of subjects at high risk of developing nAMD in the second eye because they presented with nAMD in one eye, and precluded patients with moderate or poor vision from inclusion in the cohort. Furthermore, we did not have a cohort of older patients with bilateral large drusen in whom the diagnosis might have been confounded with other common exudative maculopathies, such as diabetic macular oedema and vein occlusions. Therefore, the diagnostic performance reported here is pertinent to the characteristics of the included cohort and may be different in an unselected population without any history of nAMD.

The constraints of clinical practice, including variations in monitoring practice between sites, variable gaps for individuals between visits and study-specific methodology (utilisation of non-triggered 18 month and ‘exit’ FFAs), all affected the ability to estimate the survival curve (time to conversion) accurately. In particular, estimated survival beyond 30 months should be interpreted cautiously. Although we do not know what the true conversion point of patients is, only that they converted at some point between FFA results, the sensitivity analyses did not change test conclusions.

We were not able to rigorously assess the reading centre OCT assessment owing to the concurrent use of the FFA assessment in the reading centre. This would have helped inform about the inherent differences between OCT and FFA compared with differences between one-off individual assessment (i.e. clinical site vs. reading centre).

Health economic modelling

The modelling has several strengths. By adopting an individual simulation approach, this allowed for individual visual acuity to be modelled as a continuous variable, increasing the sensitivity of the model to capture the impact of changes in visual acuity on health-related quality of life. Key inputs around time to conversion and diagnostic accuracy were informed by the prospective EDNA study cohort, which supports the internal and external validity of the model findings. Furthermore, pre-detection testing costs and post-detection treatment costs were also based on resource use reported by EDNA participating centres and pragmatic NHS-based trials, ensuring generalisability of costs across the NHS. Finally, post-treatment visual acuity changes were also carefully informed by a range of sources applicable to NHS routine practice and capture expected differences in visual acuity trajectories by the degree of visual loss in the second eye prior to treatment initiation, ensuring that expected visual acuity gains from early identification and treatment were accurately captured.

With respect to limitations, there is a paucity of available data on the rate of visual acuity loss in the immediate period following conversion to nAMD. Although the EDNA study was able to provide a reasonable estimate of the proportion expected to lose ≥ 10 letters within 3–6 months of conversion, the increasing tendency to treat early (prior to significant vision loss) and limited observations on the few patients who experienced delayed treatment results in a degree of uncertainty around this important parameter. However, more conservative extrapolations were explored and the ICER for OCT remained favourable under these. It would take a relatively large proportion of nAMD cases to remain stable without treatment to undermine the cost-effectiveness of an OCT/immediate treatment strategy.

Prognostic model

The EDNA study enabled us to develop a predictive model of developing nAMD in the study eye because the EDNA study provided a well-standardised longitudinal cohort with a large number of potential risk factors ascertained using the standardised EDNA data capture forms. There were enough conversions to nAMD in the EDNA study to develop a prediction model. We used candidate predictors that are easily obtainable in routine ophthalmic practice and we had few missing data on predictors. The misclassification of ocular predictors is unlikely because the retinal images were graded by investigators at the study site, as well as by graders in the reading centre.

There were a number of limitations to this modelling work. The model applies only to second eyes of participants with nAMD in their first eye and, therefore, it is a selected population and the findings should be interpreted in that light. In particular, the lack of prognostic value of age is probably because of the effect of focusing on people who already have nAMD in one eye. Other variables not readily collected might have resulted in a better prognostic model. However, the known strong markers are all included in the model. A higher discriminatory performance had been anticipated because what we believed were important characteristics in the ocular environment were included. However, we did not have information on the genetic profile of participants, and the carriage of well-characterised genetic polymorphisms is known to be important in increasing risk and rapidity of progression of nAMD. This and additional OCT parameters of the fellow nAMD eyes may merit further investigation as candidate predictors; intraretinal fluid at baseline has been shown to be a predictor of conversion of unaffected eyes to nAMD. The c-statistic (discriminative ability) was only 0.66, so the performance does not suggest that this model would have clinical application at this point without further improvement. However, this model adds to the growing literature on risk predictions for conversion in the second eye and the predictors used in this model have been included in most models but not together.

We did not include genetic markers in our model; however, family history of AMD is a predictor in our model and may be an easier predictor to apply in clinical practice than genetic make-up. Including genetic risk in prediction models in nAMD remains a challenge owing to its multifactorial aetiology and gene–environment interactions. 74 Despite this, family history, our closest surrogate to genetic risk, remained significant in our final model. Blood samples collected during this study will enable the exploration of the predictive ability of genetic markers in the future. For example, a recent study involving a collaboration between DeepMind (DeepMind, London, UK) and Moorfields Eye Hospital has used deep learning to predict the conversion to nAMD from OCT within a 6-month window, with sensitivity of 80% at 55% specificity and sensitivity of 34% at 90% sensitivity. Higher conversion rates were seen in patients with larger drusen volume, FPED and hyper-reflective foci. 125

Belfast

Deirdre Burns, Guiseppe Casalino, Usha Chakravarthy (PI), Rebecca Denham, Lesley Doyle, Karen Gillvray, Katie Graham, Jonathan Keenan, Lisa Kelly, Nuala Jane Lavery, Georgios Mangioris, Orla McNally, Tunde Peto (PI), Louise Scullion, Vittorio Silvestri, Georgina Sterrett, Paul Wright and Graham Young.

Bradford

Helen Devonport, Faruque Ghanchi (PI), Nicci Hawes, Charlotte Hazel, Hayley Higgins, Zeid Madanat, Sarah Moss and Roopa Setty.

Bristol

Monalisa Bora, Claire Buckland, Colin Chu, Julie Cloake, Georgios Farantzos, Eleanor Hiscott, Sameer Hussein, Alice Johns, Danielle Lee, Rebecca Lunn, Stephanie Petrausluas, Adam Ross (PI), Serena Salvatore and Helen Talbot.

Cardiff

Ayad Al-Bermani, Sanjiv Banerjee (PI), Emma Dimond, Helen Hill, Kate Humphreys, Rhianon Reynolds, Pravin Sandanshiv, Chris Tetley, Hayley Westwood, Visith Sidath Wijetilleka, Caroline Young and Tafadzwa Young-Zvandasara.

Colchester

Giles Baggiony-Taylor, Elaine Chinery, Christopher Ellis, Judith Field, Alison Ghosh, Nicola Hopkins (PI), Tasneem Khatib and Jignesh Patel.

Edinburgh

Ali Al-Ani, Ana Maria Armbrecht, Shyamanga Borooah, Marion Brannan, Peter Cackett, Meg Das, Baljean Dhillon (PI), Margaret Frost, Mark Hope, Ash Khan, Margaret McDonald, Douglas Mitchell and Louise Ogilvie.

Frimley Park

Tayaba Akhter, Nadir Ali, Harbhajan Arora, Nadia Azad, Elizabeth Baker, Reynette Baroman, Leena Bhat, Manju Chandran, Maheswari Chekuri, Gloria Crawford, John Deligiannis, Mohamed El-Hefrim, Fernando Esposito, Sarah Gee, Petya Gencheva, Talat Gondal, Harinder Grewal, Jessica Haney, Ali Hassan, Wendy Hockney, Victoria Hodgson, Charlene Humphries, Psiyanta Kakai, Vijay Kakkar, Jaskiran Kaur Sandhu, Radhika Kotecha, Teena Kunnath, Stella Lee, Sally Mathew, Lesley McAllister, Brendan McIlhargey, Jessica McKean, Gail McKenzie, Lorraine McLoughlin, Geeta Menon (PI), Prashant Mistry, Narendran Nair, Karima Nesnas, Lorraine North, Rohina Patel, Ganga Pathinayake, Abigail Raguro, Darryl Ratiram, Rani Retnamma, Deana Robson, Anitha Sethumadhavan, Priyanka Srikanth, Kathryn Walters, Debbie Young, Fani Zacharaki and Louisa Zouita.

Gloucester

Fadi Alkherdahji, Jade Ayland, Liz Bristow, Victoria Edgeworth, Helen Fisher, Emily Fletcher, Rachel Healy, Katerina Ivanova, Seb James, Robert Johnston (PI), Vanessa Jones, Antonis Kaintatzis, Laura Lodge, Jennifer Mackenzie, Quresh Mohamed, Dawn Phillips, Jeffrey Pick, Claire Porter, Bridget Rosser, Peter Scanlon (PI) and Karen Townsend.

Harrogate

Bushra Al-Deiri, Tony Burton, Oana-Gabriela Graves, James Featherstone, Theodora Georgouli, Mamta Gupta, Natalie Hollingsworth, Sarah Mackenzie (PI), Karen Martin, Tracy Nixon, Kim Oses-Fred, Eve Panesar, Clare Stemp, Jo Varley and Gavin Walters.

Hillingdon

Muhammed Aliahmed, Balisee Beruessa, Doris Bilayon, Teresita Bolanos, Richard Cheong-Leen, Sheena George (PI), Charlene Hamid, Darren Hanumunthasu, Jason Ho, Neo Johnson, Kash Khuttan, Dhakshi Kumar, Ashley Long, Sonal Mehta, Rahila Naureen, Angelos Raytis, Anju Shrestha, Georgios Sotiropoulos, Denise Wheeler and Elizabeth Woodstock.

Hull

Seem Arora, Angela Atkinson, Atiq Babar, Helen Cook, Sally Cooper, Mark Costen, Louise Downey (PI), Kala Gopalakrishnan, Helen Hobman, Krishnappa Madhusudhana, Anish Mani, Christine Mayman, Muhammad Shaikh, Naomi Teal, Yvonne Waudby and Naeem Zaman.

James Paget University Hospital

Ben Burton (PI), Debbie Busby, Stephanie Cotton, Kelly Evers, Katherine MacKintosh, Helen Nutt, Lucy Palmer, Lesley Parsons, Loretta Poundall, Umair Qidwai, Mya So, Emma Stimpson, Thomas Webber and Celia Whitehouse.

Leeds

Samantha Bird, Frances Cassidy, James Cook, Swetha Devabhakthini, Mohammed Jama, Martin McKibbin (PI), Raj Mukherjee, Mike Stockton, Charmain Tidswell and Alice van Lare.

Leicester

Mandy Babbington, Kayleigh Brown, Sreekala Burgula, Patrina Christian, Melanie Chrystal, Sue Cockburn, Lee Daines, Theo Empeslidis (PI), Howard Fairey, Andrew Farmer, Konstantina Gorgoli, Vasileios Konidaris, Bhavesh Mistry, Eshmael Palmer, Michelle Rickerby, Lauren Rybicki and Konstantinos Tsaousis.

Liverpool

Julia Baxter, Alina Cordos, Karen Hawkins, Heinrich Heimann (PI), Vivian Ho, Oana Hotu, Samantha Jackson, Robert Jardine, Darina Koneva, Jae Ku, Pauline Lenfestey, Andrea Madden, Savitha Madhusudhan, Ira Mahac, Ian Pearce, Jayne Schumacher and Sandra Taylor.

Manchester

Mohammed Abid, Amy Aldridge, Tariq Aslam (PI), Aaron Bain, Konstantinos Balaskas, Rachel Bambrick, Goncalo Bento, Jay Brown, Ryan Carney, Rosalind Creer, Natalie Fox, Thomas Hamper, Mania Horani, Narcisa Ianopol, Binu John, Emma Linton, Katie McGuiness, Harriet Ndu-Melekwe, Jeremy Parkes, Laura Perry, Raisa Platt, Afeefa Rasheeth, Leanne Richards, Amy Stone (PI) and Daniel Todd.

Moorfields

Joynal Abadin, Nadine Abdelgalil, Muna Ali, Minara, Begim, Rawona Berinde, Kanom Bibi, Asma Burale, Dominic Carrington, Cristina Citu, Monica Clemo, Eleanor Connolly, Roxanne Crosby-Nwaobi, Diana Dabrickaite, Supeetha David, Jenny Elliot, Sherene Ettienne, Charlene Formento, Sash Jeetun, Layla Juma, Alexa King, Shamaine King, Tabassum Master, Aldrich Meneces, Deepthy Menon, April Neville, Luke Nicholson, Andrea Ogunyemi, Sweta Panchagnula, Namritha Patrao, Panayiotis Panayiotou, Samiul Rahman, Jayashree Ramu, Nafeesa Reece, Joshua Robinson, Vincent Rocco, Priyansha Sheel, Roy Schwartz, Anar Shaikh, Sobha Sivaprasad (PI), Hayley Thomas, Emerson Tingco and Marta Zola.

Rugby

Jeanette Allison, Susanne Armitage, Jenny Bhatti, Amritpal Chaggar, Krystal Gale, Jeanette Houston, Peter James, Charlotte Johnson, Obaid Kousha, Caitlin Lumsden, Marie McCauley, Parimala Nagarajan, Sergio Pagliarini (PI), Linzi Randle and Helen Shawl.

Sheffield

Nachiketa Acharya, Naadir Ansari, Christopher Brand (PI), Stephen Connell, Maria Edwards, Mary Freeman, Helen Pokora, Fahd Quhill, Lydia Reaney, Martin Rhodes and Katherine Wheeler.

Southampton

Hussein Almuhtaseb, Magdalena Ansari, Joanna Ballingall, Mercy Jeyeraj, Joanna Karkorz, Samir Khandhadia, Theodora Koutresi, Andrew Lotery (PI), Georgiana Matei, Marie Nelson, Thea Sass, Amanda Smith, Suresh Thulasidharan, George Tsokolas, Sirli Veetousme, Catrin Watkins and Isabela Wica.

Stoke Mandeville

Judith Abrams, Kavita Aggarwal, Ahmad Ahmado, Tayaba Akhtar, Mandeep Singh Bindra (PI), Nicola Cronbach, Amelia Davidson, Markus Groppe, Shreya Haldar, Christopher King, Katarina Manso, Libby McKerrow, Ruth Penn, Jamil Razzaque, Pavandeep Singh Sandhu and Shafak Toufeeq.

Sunderland

Steve Dodds, Louise Fairlie, Lauren Gardner, Casper Geenan, Maged Habib, Hugh Harris, Ajay Kotagiri (PI), Haifa Madi, Teresa Sandinha (PI), Gemma Sloanes, Jonathan Smith, David Steel, Nicolle Teasdale, Deepali Varma, Jade Ward, Michelle Young and Jill O’Brien.

Wolverhampton

Bhogal Singh Bhogal, Vina Bhundia, Nick Denyer, Gemma Edwards, Ibtesam Elaraoud, Imogen Hawthorne, Jennifer Howell, Sharon Hughes, Donna Jones, Susan Massey, José Maya, Nirodhini Narendran (PI), Jas Purewal, Habiba Saedon, Sara Simmons, Richard Webb, Diane Whistance-Smith and Yit Chiun Yang.

York

Archana Airody, Fiona Bailey, Robert Cann, Richard Gale (PI), Srilakshmi Gollapothu, Alison Grice-Holt, Reema Gupta, Richard Hanson, Greg Heath, Gary Lamont, Mike Pringle, Carol Sarginson, Helen Shaw, Sara Stringer, Nicola Topping, Gavin Walters and Joanne Wincup.