Optical coherence tomography for the diagnosis, monitoring, and guiding of treatment for neovascular age-related macular degeneration: a systematic review and economic evaluation

Brief statement describing the health problem

Neovascular age-related macular degeneration (nAMD) causes severe visual loss and is the most common cause of blindness in persons aged > 50 years in the Western world. In recent years, there have been significant advances in the clinical management of patients with nAMD. For example, there are now effective treatments, specifically antivascular endothelial growth factor (antiVEGF), and novel diagnostic technologies, including both imaging and functional tests. Patients who are being treated for nAMD with antiVEGF require frequent and long-term follow-up for treatment to be most effective.

The current reference standard for diagnosis of nAMD is fundus fluorescein angiography (FFA)1 which may also be used to monitor the activity of the disease after treatment. However, FFA is time-consuming, invasive and requires expert interpretation. Optical coherence tomography (OCT) is now widely used for the diagnosis and management of nAMD. OCT is non-invasive, safer and more straightforward to do and interpret than FFA. OCT may help clinicians to provide a more cost-effective service for people with nAMD by potentially replacing the current reference standard of FFA and helping to distinguish between those patients with active disease requiring treatment and those whose disease is not active at a particular point in time and who do not require treatment. OCT might also lead to efficiencies by allowing other categories of health professionals to become involved in the diagnosis and monitoring of patients.

Aetiology, pathology and prognosis

Neovascular age-related macular degeneration is a pathological process in which new blood vessels arising from the choroid breach the normal tissue barriers and come to lie within the subretinal pigment epithelium (subRPE) and/or subretinal spaces. These new vessels, commonly referred to as choroidal neovascularisation (CNV) or choroidal neovascular membrane (CNVM), leak fluid, lipids and blood, elicit an inflammatory response and, as part of their natural history, undergo a scarring process, all of which has a deleterious effect on the visual cells of the retina (photoreceptors), leading to central loss of vision. Besides CNV, there are two other recognised phenotypes of nAMD: (1) retinal angiomatous proliferation (RAP) in which vascular complex seems to arise de novo from the retinal circulation, or results from CNV anastomosing with the retinal circulation; and (2) intrachoroidal/subRPE aneurysmal dilatation(s) of the choroidal vasculature, known as idiopathic polypoidal choroidal vasculopathy (IPCV). 2 These phenotypes may occur in isolation or be mixed with other phenotypes. 3

The onset of nAMD results in progressive and unremitting loss of central vision in the affected eye, with rare exceptions in cases of IPCV in which spontaneous improvement may be observed. A number of studies have shown that extrafoveal CNV will grow towards the fovea. Once foveal involvement has occurred, CNV will expand and involve ever-increasing areas of the macula. Thus, the majority of eyes will experience acute visual loss, either moderate [defined as a doubling of the visual angle which equates to a three-line worsening on the Early Treatment Diabetic Retinopathy Study (ETDRS) visual acuity (VA) chart] or severe (defined as a quadrupling of the visual angle and which equates to a six-line worsening on the ETDRS VA chart). However, some patients with a fellow eye with good vision will not notice any such changes despite the onset of neovascularisation.

Neovascular age-related macular degeneration is now treated with repeated intraocular injections of drugs designed to antagonise vascular endothelial growth factor (antiVEGF). This will stabilise sight in most patients (≈90%) and will improve vision in a smaller group (≈30%) during the first 2 years of treatment. 1 Long-term (beyond 3–4 years) outcomes from randomised controlled trials (RCTs) using antiVEGF are, however, not available. These drugs are administered monthly (often with a mandated minimum of three injections for the first 3 months, and thereafter depending on whether or not active nAMD is present) as intraocular injections until the macula is rendered fluid free. When the disease becomes quiescent, treatment is stopped and patients are monitored for relapse, with treatment being restarted if needed, by monthly intraocular injections based on findings of VA checks, clinical examination and OCT. FFA is typically used to confirm the diagnosis of nAMD prior to initiating antiVEGF therapy, but it is used only in selected circumstances for monitoring activity of nAMD after treatment. Relapse of nAMD is unpredictable and can occur within weeks, months or even years after stopping treatment.

Epidemiology, incidence and prevalence

The prevalence of all forms of age-related macular degeneration (AMD) (including neovascular and atrophic AMD), which affects more than 600,000 people in the UK, is expected to rise by a quarter to nearly 756,000 by 2020. The estimated number of individuals with nAMD in the UK for 2011 is 368,000 and will increase substantially due to the ageing population. 4–6 Estimates of incidence of nAMD in the UK suggest that there are between 13,000 and 37,000 new cases annually. 5 The National Institute for Health and Care Excellence (NICE) guidance on ranibizumab [Lucentis^®, Genentech Inc. (USA)/Novartis Pharmaceutical Ltd] and pegaptanib (Macugen^®, Pfizer Ltd) for the treatment of age-related macular degeneration (AMD) (issued 2008 and modified 2012) estimated that there were about 26,000 new cases of nAMD in the UK each year. 7 Many of these individuals will require monthly monitoring and treatment for several years. Relevant risk factors include age, cigarette smoking, nutritional factors, cardiovascular diseases and genetic markers, including genes regulating complement, lipid, angiogenic and extracellular matrix pathways.

Impact of health problem

Measurement of disease

The spectrum of disease may be classified according to the reduction of VA (e.g. mild, moderate or severe). In addition to this spectrum of disease, during monitoring of patients undergoing treatment with antiVEGF drugs, it is important to determine whether or not the disease is active. Disease activity is typically determined with imaging technologies, mainly FFA and OCT.

Management of disease

Diagnosis of neovascular age-related macular degeneration and care pathway

Typically, patients with possible AMD present to primary care (optometrists or general practitioners) with non-specific symptoms (such as reduced, blurred and distorted vision). Some patients do not report symptoms and are diagnosed at routine eye examination. Clinical examination of the retina reveals typical changes associated with AMD such as drusen and irregularities in the appearance of the retinal pigment epithelium (RPE), most commonly in both eyes. However, the presence of a neovascular component may be difficult to detect clinically, especially early on in the course of its development. The diagnostic pathway for nAMD and the management of patients with known disease include imaging technologies (Figure 1).

FIGURE 1.

Current diagnostic pathway of nAMD.

According to current guidelines from the Royal College of Ophthalmologists (RCO),1 FFA interpreted by an ophthalmologist is the method of choice and reference standard test to diagnose nAMD. Occasionally, indocyanine green angiography (ICGA) is associated with FFA as part of the reference standard when particular phenotypes of nAMD are suspected, including RAP and IPCV (see above). FFA is an invasive and time-consuming procedure, entailing the injection of a dye into a peripheral vein by a nurse and a trained photographer to undertake the test (obtain the images of the CNV, RAP, IPCV lesions). In addition to FFA, current guidelines recommend using OCT at diagnosis. Owing to recent developments in technology, it is possible that in some cases OCT might be superior to FFA in detecting nAMD (Table 1).

TABLE 1 - Apparent features of OCT and FFA for nAMD

Features	OCT (index test)	FFA (reference standard)
Accuracy	High?	Reference standard
Invasiveness	Non-invasive	Invasive
Knowledge and skills needed to interpret	Moderate	High
Interpretable	Most tests	Nearly all tests
Cost	Low to moderate	Moderate
Side effects	None	Allergy (rarely anaphylactic shock)

Treatment and monitoring of neovascular age-related macular degeneration

When active nAMD is confirmed, treatment with antiVEGF therapy is initiated. 8,9 For all patients with nAMD it is common practice to use three consecutive (monthly) intravitreal injections of antiVEGF therapy, and then the patient is reassessed to evaluate whether or not the disease is active (i.e., neovascularisation leaking fluid/blood at the macula) or inactive (Figure 2). For this purpose, both FFA and OCT may be used, although the latter more often than the former, according to the guidelines of the RCO. 1 Studies that have a large influence in current practice used VA and OCT at monthly intervals and FFA at quarterly intervals to decide on the need for retreatment. In some units, OCT is the only test performed to determine activity of the neovascular process in clinical practice; in some centres FFA is performed in selected cases during the monitoring phase. Other technologies such as fundus autofluorescence (FAF) may also be used at baseline and at variable intervals during the follow-up of these patients as areas of atrophy in the RPE (difficult to detect clinically but easily observed on autofluorescence images) could be associated with fluid in the retina in the absence of active nAMD.

FIGURE 2.

Current monitoring pathway of nAMD.

If fluid is not seen intraretinally or subretinally, further treatment is not given and the patient is followed thereafter regularly. The timing of follow-up visits is variable, typically every 4 weeks for the first year, extending the intervals after the second year. Varying intervals have been proposed, such as ‘treat and extend’ strategy, where if there is no active disease, no treatment is given and the monitoring intervals are progressively extended. If the disease is judged to be active, further injections of antiVEGF are given. Either a single injection or three injections are administered if activity is detected on follow-up and then the patient returns to the monthly monitoring scheme. The possibility of using VA (without imaging tests) as the only test to guide treatment during monitoring (i.e. treatment would be given if there is a loss of five or more letters from best previously observed VA) has been modelled using data from published trials for nAMD. 10 The authors concluded that an individualised VA-guided regimen could sustain visual outcomes and improve cost-effectiveness compared with current regimes.

Current service cost

Table 2 shows an estimation of unit costs associated with current diagnosis and monitoring care pathways. A first referral visit to a hospital eye service will involve an eye examination and is costed at £106. In addition, OCT and FFA tests can be indicated, with the overall cost for the first visits ascending to £274.71. A follow-up monitoring visit can involve a face-to-face attendance with an ophthalmologist and an OCT test only (£131). However, if a FFA is indicated, the monitoring visit will cost £248.27. Without doubt, the major cost category is given by the treatment cost. There are two possible antiVEGF treatments: ranibizumab and bevacizumab (Avastin^®, Roche) at £742.17 and £50 per injection respectively. NICE guidelines advocate for the use of ranibizumab unless individual sight is heavily deteriorated. It should be noted that special cost arrangements are in place and a reduced cost for ranibizumab is agreed under a Patient Access Scheme negotiated between the manufacturer and the Department of Health. Under this agreement, the cost of ranibizumab to the UK NHS (confidential) is significantly lower than the list price given above. The cost of bevacizumab is based on that of a compounded product as supplied by different compounding pharmacies in the UK.

Variation in service and/or uncertainty about best practice

Once nAMD has been diagnosed, monotherapy with an antiVEGF drug (administered into the vitreous) is the current standard of care. Ranibizumab is highly effective and recommended by current guidelines. Bevacizumab remains unlicensed in the UK although its use worldwide reflects the fact that it is much cheaper than ranibizumab (as currently supplied for intravitreal administration) with similar efficacy. 8,9

Retinal imaging with OCT before and after intravitreal administration of antiVEGF therapy is regularly used. 13 Following antiVEGF therapy a reduction of intraretinal and subretinal fluid is typically observed, often with rapid unification of the retinal layers and improvement/restoration of the anatomical contours. This anatomical improvement is often accompanied by improvements in VA.

The ultimate treatment goal when nAMD has already developed is to achieve restoration of central vision and prevent visual loss with normal or near normal foveal and macular anatomy. Complete cessation of exudation can result in good unification of the tissue layers, but most patients report difficulty with reading small print and other visually demanding tasks, even when tissue contours have been apparently restored. High-resolution OCT scans obtained after antiVEGF treatment show persistent abnormalities of the outer retina even though the tissues appear to be fluid free. In cases where localised atrophy and fibrosis have already occurred, considerable impairment of central visual function can remain, despite the achievement of a fluid free macula.

Patients who have been treated with antiVEGF therapy should be examined at regular intervals. Although most clinicians will use OCT for monitoring patients with nAMD, there is probably large variability on the tests used (e.g. biomicroscopy of the fundus, FFA and fundus photography).

As explained above, patients treated with antiVEGF injection should receive injections monthly for the first 3 months and, thereafter, should be monitored monthly. If active nAMD is present, treatment should be continued, and if there is no active exudative AMD, observation at monthly intervals is recommended. The use of technologies, including OCT, FFA and FAF during the follow-up of these patients is variable as it depends on clinical findings, the judgement of the treating ophthalmologist and the clinical pathways established at different centres. The workload associated with such contemporary AMD services is significant and is expected to increase, as the best outcomes are achieved with monthly follow-up visits. It is expected that these follow-up visits may continue for as long as 4 years or longer. The pressure on resources and service delivery in the AMD clinics is expected to become even more intense as many patients cannot be discharged, and there is a need to accommodate new incident cases. The regular monthly follow-up for AMD patients under treatment, in order to maintain efficacy, is demanding. This situation is likely to be further aggravated by the impending treatments with intravitreal therapies of macular oedema secondary to diabetic retinopathy and retinal vein occlusion. As such, the problem seems more acute than was originally envisaged, and is expected to get worse. It has been suggested engaging non-medical staff (optometrists, nurses, technicians) to undertake some of the duties in the AMD clinic in order to increase capacity. Such roles include clinical assessments, especially retreatment decision-making.

Relevant national guidelines, including National Service Frameworks

Subsequent to the technology appraisal and issuing of guidance by NICE, ranibizumab has been widely adopted as the treatment of choice for subfoveal nAMD in the UK. 7 However, the high cost of ranibizumab, along with the positive clinical experience with bevacizumab, has stimulated a debate on whether or not bevacizumab could be used in practice.

In the UK, guidelines for the management and treatment of nAMD were published by the RCO in 2009 (and in 2013 were undergoing revision). 1 According to the RCO guidelines, FFA interpreted by an ophthalmologist is the method of choice and reference standard test to diagnose nAMD. Occasionally, ICGA is associated with FFA as part of the reference standard when particular phenotypes of nAMD are suspected, including RAP and IPCV. In addition to FFA, current guidelines recommend using OCT at diagnosis. During follow-up and monitoring of disease activity, after treatment the current guidelines recommend the use of OCT mainly, and FFA at the discretion of the clinician.

Reference standard: fundus fluorescein angiography

Fundus fluorescein angiography is currently the reference standard for diagnosing CNV in AMD. A fluorescein angiogram is a sequence of images captured of the fundus over a 10-minute period after injection of the non-toxic dye fluorescein isothiocyanate into a suitable peripheral vein.

Neovascular lesions are classified by their location with reference to the foveal avascular zone – extrafoveal, juxtafoveal or subfoveal. Lesions lying more than 200 μm from fixation are defined as extrafoveal and may also be described as juxtafoveal or subfoveal when immediately adjacent to or involving the geometric centre of the fovea respectively. Neovascular lesions located away from the macula are termed peripheral and those around the optic nerve juxtapapillary. A more refined classification of the neovascular lesion is obtained by describing the composition of the exudative lesion after stereoscopic review of the entire sequence of the angiogram. The exudative lesion is defined as the area occupied by the neovascular complex, any associated blood, thick exudate and pigment epithelial detachments (PEDs) that are contiguous to the neovascular complex and obscure its margins. The neovascular complex can, therefore, consist of RAP, CNV and IPCV.

The classification of nAMD lesions is based on the temporal and spatial features of the patterns of fluorescence as observed on the FFA. CNV lesions are classified according to their location relative to the fovea (see above), and pattern of fluorescein angiographic leakage. The majority of CNVs occur subfoveally.

Optical coherence tomography

Optical coherence tomography was developed at the Michigan Institute of Technology, MI, USA in 1991. It is a light-wave-based technology producing cross-sectional images of the retina with scan rates and resolution parameters that have greatly improved over the last 10 years. OCT is a non-invasive, non-contact visual test that requires around 5–10 minutes to assess both eyes. 14 From the investigator’s point of view, it is user friendly (e.g. OCT is easier to do than FFA), typically undertaken by trained medical photographers or ophthalmic imaging technicians, and interpreted by ophthalmologists. Automated analysis can also be used.

There are two main types of OCT system. The earlier time domain optical coherence tomography (TD-OCT) system, available from 1995, had an image rate of 100–400 scans per second and provided information for a limited view of the retina by taking six scans radially-oriented 30 degrees from each other with a resolution in the range of 10 to 20 µm. 14 The newer system, spectral domain optical coherence tomography (SD-OCT), has been available since 2006. Improvements with this system include (i) a faster scan speed of approximately 27,000 scans per second, (ii) the ability to scan larger areas of the retina by taking several horizontal line scans such that there are no ‘missed areas’, (iii) increased resolution at 5 µm, and (iv) ‘real time registration’, which was not previously available with TD-OCT. 14 The real-time registration feature enables the identification of specific anatomical locations on the retina, against which subsequent tests may be evaluated, which is of particular importance in the monitoring of patients. 14 Compared with TD-OCT, the faster scan speed of SD-OCT enables the collection of additional information on larger regions of the retina and eliminates image distortion arising from patient movement, while the improved resolution allows for a clearer and more distinguishable view of retinal layers, with the possibility of detecting earlier signs of disease. 14

Identification of important subgroups

There are different subgroups of patients with nAMD. They are diagnosed according to FFA findings and are described above. Subgroup classification depends on the location (extra-, juxta- and subfoveal) and type of neovascularisation (classic and occult CNV, RAP, and IPCV), which could be mixed in different combinations. Although the initial treatment is similar for all subgroups (with antiVEGF therapy), the natural history and progression after treatment are different. It is also possible that the performance of diagnostic technologies may be different among subtypes of nAMD. OCT is not currently used in isolation to identify subgroups.

Current usage in the NHS

Both FFA and OCT are currently used in the NHS to diagnose and monitor patients with nAMD. They are recommended technologies to provide standard care. FFA is essential for diagnosis of the condition. Regarding monitoring, FFA is less commonly used than OCT.

Anticipated costs associated with intervention

Table 3 presents an estimation of the number of visits in a lifetime of the population. Based on census, nAMD prevalence and Interim Life Table data, it is possible to estimate the number of visits for the population lifetime. Calculations in Table 3 are for England and Wales, based on 2011 data and assumed that every individual with nAMD would contact NHS services. This estimation resulted in 33.7 million visits. If OCT was conducted at every monitoring visit, this would result in an undiscounted lifetime cost of above £1.7B [i.e. £51.27 (see Table 2) multiplied by 33.7 million people].

Clinical evaluation (with slit-lamp biomicroscopy with or without use of diagnostic contact lens and evaluation of patients’ symptoms)

The onset of exudative AMD is heralded by the appearance of central visual blurring and distortion. Most patients will complain that straight lines appear crooked or wavy. Sometimes patients do not notice visual symptoms when the first eye is affected. When nAMD occurs in the second eye, patients suddenly become limited in their daily activities, for example reading, driving and seeing fine detail such as facial expressions.

Examination of the macula usually reveals fluid and/or lipid (yellow deposition) and/or blood. Other features of AMD such as drusen and pigmentary irregularities are most often present. Sometimes these latter features are not observed once exudative AMD has supervened or in certain phenotypes such as IPCV. However, the fellow eye would usually exhibit some or all of these AMD early clinical signs (drusen and RPE changes) and their presence is helpful in confirming that the neovascular lesion is due to AMD (again with the exception of IPCV where the fellow eye may also be normal). Following slit-lamp biomicroscopy (SLB) the presence or absence of the following signs should be noted:

• Subretinal or subRPE neovascularisation which may be visible as a dark grey lesion. Occasionally the lesion will have a dark pigmented edge which is thought to be due to 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, indicating most often the presence of IPCV.

• 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 subpigment epithelial scar/glial tissue or fibrin-like deposits.

• RAPs: red, round, extra- or juxtafoveal lesions located within the retina.

• Polyps: red, round lesions located underneath the RPE or protruding through the RPE layer.

Visual acuity (for monitoring)

Visual acuity is a measure of the spatial resolution of the visual processing system. VA is tested by requiring the person whose vision is being tested to identify characters (like letters and numbers) on a chart from a set distance. Chart characters are typically represented as black symbols against a white background (for maximum contrast). The distance between the person’s eyes and the testing chart is set at a sufficient distance to approximate infinity in the way the lens attempts to focus.

Amsler grid

The Amsler grid is a grid of horizontal and vertical lines used to monitor a person’s central visual field. It is a diagnostic tool that aids in the detection of visual disturbances caused by changes in the retina, particularly the macula (e.g. macular degeneration). In the test, the person looks with each eye separately at the small dot in the centre of the grid. Patients with macular disease may see wavy lines or some lines may be missing. Amsler grids are supplied by ophthalmologists, optometrists or from websites, and may be used to test one’s vision at home.

Colour fundus photographs

Colour fundus photography provides a record of the appearance of the macular retina. Stereoscopic images of the macula viewed appropriately can help localise pathology to the different tissue layers. For the purposes of recording macular pathology, stereoscopic pairs of images taken at 35 degrees centred on the macula are recommended. Red-free images (RFs) can help detect some features of the fundus associated with nAMD, such as haemorrhages.

Infrared reflectance

Confocal near-infrared fundus reflectance is a non-invasive en-face imaging technique using an 830-nm diode laser capable of visualising subretinal pathology. In contrast to visible wavelength illumination, fundus reflectance may be up to 10 times higher in the near-infrared wavelength and is then largely independent of melanin content, which advances the visibility of deep fundus structures.

Red-free images or blue reflectance

See Colour fundus photographs, above.

Fundus autofluorescence imaging or blue reflectance

This test can give an indication of the health of the RPE. The conventional FAF signal (obtained with 488 nm) originates, predominantly, from lipofuscin in RPE cells. The near-infrared autofluorescence (NIA) signal originates, predominantly, from melanin in the RPE, with some contribution from choroidal melanin. Increased FAF represents accumulation of lipofuscin and suggests that the RPE cells are beginning to fail. Absence of a FAF and NIA signal, which appears as black areas in FAF and NIA images, is due to loss of RPE cells. The finding of patches of absent autofluorescence may explain central scotoma patterns. Although different patterns have been described in early and late AMD, the exact diagnostic performance of autofluorescence is yet to be determined. The role of FAF may be more important in monitoring patients undergoing antiVEGF therapy to evaluate atrophy (e.g. for potential discontinuation of treatment).

Indocyanine green angiography, dynamic high speed or digital subtraction indocyanine green angiography

Indocyanine green (ICG) is an alternative dye to fluorescein which is used to visualise the choroidal circulation. This dye binds to plasma protein and hence does not egress easily through the fenestrae of the choroidal vessels, remaining within the vascular compartment. ICGA is obtained using longer wavelengths than FFA and, thus, can penetrate through areas of fluid/blood, permitting visualisation of pathology in circumstances where fluorescein may not. ICG also has some limitations and very thick blood or pigment can reduce or block transmission of the ICG infra-red wavelength and the emitted light is of lower intensity compared with that of fluorescein. The use of the SLO with video capture can, however, yield images of high resolution. Video ICGA also allows better imaging of RAP. As ICG dye does not leak into the subretinal and subpigment epithelial spaces to the same extent as fluorescein, the enhanced definition of the vascularised tissue as a hotspot is possible and a combination of FFA and ICGA can produce complementary information. A dose of 25 mg of ICG in aqueous solution is usually injected intravenously and images acquired for up to 30 minutes.

Preferential hyperacuity perimetry

Preferential hyperacuity perimetry (PHP) is a psychophysical test of macular function that exploits the ability of the human visual system to perceive even minute differences in the relative localisation of two objects in space; a phenomenon termed hyperacuity. When there is separation of the retinal layers through breakdown of the blood–retinal barrier or blood–RPE barrier, distorted vision is the consequence. Through presentation of lines with artificial distortions of different intensities on the PHP, the presence of a real distortion in the patient’s central visual field can be detected as the brain ignores the smaller deviation when a larger one is introduced.

In a PHP test, the macula is scanned with a succession of stimuli, each stimulus consisting of a series of dots arranged along a vertical or horizontal axis. In each stimulus, a small number of dots are misaligned, thereby creating an artificial distortion (bump or wave). The examinee’s task is to perceive these artificial distortions and mark their locations on the visual field. When a stimulus is projected on a healthy portion of the retina, the examinee identifies the artificial distortion and is likely to mark a correct location. If the stimulus is projected on a damaged region of the retina, a pathological distortion may be perceived instead of the artificial distortion, especially if the pathological distortion is more prominent than the artificial distortion. The examinee may then mark a location that is distant from the artificial distortion, indicating that a pathological distortion may have been perceived. By manipulating the amplitude of artificial distortions, the amplitude of the pathology in the area of interest can be quantified. At the end of the test, comparison of the set of erroneous responses against a normative data base is used to determine if test results are within normal limits.

Microperimetry

One conventional measure of vision is subjective visibility thresholds of small, short-duration stimuli as performed by conventional automated static perimetry. In conventional perimetry, retinal localisation of a stimulus is implied indirectly from the assumed retinal location of fixation. This approach can work well when fixation is stable and foveal. However, loss of fixation stability or foveal vision, such as occurs commonly in nAMD, complicates the measurement of macular function with conventional perimetry. Accurate correspondence between retinal structures and visual function requires simultaneous imaging of the fundus. Microperimetry includes real-time automated tracking of the fundus and appropriate compensation of the location of stimulus presentation at predefined retinal loci.

Index test(s)

The index test considered was OCT, either alone or in combination with alternative tests as described below. Both TD-OCT and SD-OCT were considered.

Population

The population considered was people with newly suspected nAMD or those previously diagnosed with the disease and under surveillance monitoring.

The setting considered was secondary care.

Relevant comparators

The alternative tests considered included the following examinations:

• clinical evaluation (with SLB, with or without use of diagnostic contact lens and evaluation of patients’ symptoms)

• VA (for monitoring)

• Amsler grid

• colour fundus photographs

• infrared reflectance (IR)

• RFs or blue reflectance

• FAF imaging

• ICGA, dynamic high-speed or digital subtraction indocyanine green angiography (DS-ICGA)

• PHP

• microperimetry.

Reference standard

The reference standard considered was ophthalmologist-interpreted FFA. FFA is generally acknowledged as being the recognised reference standard for detecting nAMD. The RCO states in its guidelines for management of AMD that FFA is currently the reference standard for diagnosing exudative disease. 1 However, as few studies reported individual ophthalmologist-interpreted FFA (rather than reading centre-interpreted FFA), studies using FFA as the reference standard but with unclear information about which type of health-care professionals interpreted the images were also considered.

Outcomes

The following outcomes were considered for the use of OCT at presentation and during follow-up of patients with nAMD:

• diagnostic accuracy [e.g. sensitivity, specificity, likelihood ratios (LRs), diagnostic odds ratio (DOR)]

• clinical effectiveness (e.g. VA, anatomical control of the disease, patient-reported outcomes)

• interpretability of the test – to be defined as in included studies, considering the ability to acquire a quality image that can be interpreted or analysed

• acceptability of the test – to be defined as in included studies, considering users and health-care providers’ perspective

• proportion of participants not able to receive the diagnostic test [due to an eye condition (e.g. lens or other media opacity), or personal circumstances (e.g. wheelchair bound)].

The evidence for the use of OCT was considered separately for the purposes of diagnosis and monitoring.

Key issues

The key issues to be addressed are:

• How good a test is OCT, when used either alone or in combination with alternative tests, in the diagnosis of people newly presenting with a suspicion of nAMD?

• How good a test is OCT, when used either alone or in combination with alternative tests, in detecting recurrent nAMD activity during surveillance monitoring of people previously diagnosed with the disease?

• Is SD-OCT a better test than TD-OCT?

• Could OCT images be interpreted by other health professionals in addition to ophthalmologists?

• Could OCT replace FFA in some situations in the diagnostic and/or monitoring pathways?

• How cost-effective are strategies involving OCT, both in the diagnostic and monitoring pathways?

Types of studies

The following types of studies were considered.

1. Diagnostic studies:

   • Direct (head-to head) comparisons in which the index test and comparator test(s) are evaluated in the same study population. These could be fully paired [all study participants receive the index test, comparator test(s) and the reference standard] or not fully paired (participants receive only a subset of the tests, e.g. a randomised direct comparison in which study participants are randomly allocated to receive the index test or the comparator and all receive the reference standard.

   • Indirect comparisons in which estimates of the accuracy of the respective tests are obtained in different study groups, for example two-gate or ‘case–control’ type studies where different sets of criteria are used for those with and without the target condition. Indirect comparisons were to be considered if there was insufficient evidence from direct comparisons.

2. Studies reporting clinical effectiveness:

   • RCTs evaluating outcomes when treatment was based on OCT compared with FFA findings.

3. Qualitative studies evaluating patients’ and/or clinicians’/health-care professionals’ acceptability and/or interpretability of the OCT tests.

Types of participants

The types of participants considered were people with newly suspected nAMD or those previously diagnosed with the disease and under surveillance monitoring.

The setting considered was secondary care.

Index tests

The index test considered was OCT, either alone or in combination with alternative tests as described below. Both TD-OCT and SD-OCT were considered.

Comparator tests

The alternative tests considered included the following examinations:

• clinical evaluation (with SLB, with or without use of diagnostic contact lens and evaluation of patients’ symptoms)

• VA (for monitoring)

• Amsler grid

• colour fundus photographs

• IR

• RFs or blue reflectance

• FAF imaging

• ICGA, dynamic high-speed or DS-ICGA

• PHP

• microperimetry.

Reference standard

The reference standard considered was ophthalmologist-interpreted FFA. FFA is generally acknowledged as being the recognised reference standard for detecting nAMD. The RCO states in its guidelines for management of AMD that FFA is currently the reference standard for diagnosing exudative (neovascular) AMD. 1 However, as few studies reported individual ophthalmologist-interpreted FFA (rather than reading centre interpreted FFA), studies using FFA as the reference standard but with unclear information about which type of health-care professionals interpreted the images were also considered.

Types of outcomes

The following outcomes were considered for the use of OCT at presentation and during follow-up of patients with nAMD:

• diagnostic accuracy (e.g. sensitivity, specificity, LRs, DOR)

• clinical effectiveness (e.g. VA, anatomical control of the disease, patient-reported outcomes)

• interpretability of the test – defined as in the included studies, considering the ability to acquire a quality image that can be interpreted or analysed

• acceptability of the test – defined as in the included studies, considering users and healthcare providers’ perspective;

• proportion of participants not able to receive the diagnostic test [due to an eye condition (e.g. lens or other media opacity), or personal circumstances (e.g. wheelchair bound)].

The evidence for the use of OCT was considered separately for the purposes of diagnosis and monitoring.

Number and type of studies included

Appendix 2 lists the 29 studies, published in 31 reports, that met the inclusion criteria for the review of diagnostic and monitoring studies. 23–53 There were two reports of the studies by Cachulo et al. 25,47 and Torron et al. 50,51 Figure 3 shows a flow diagram outlining the screening process, with reasons for exclusion of full-text papers.

FIGURE 3.

Flow diagram outlining the screening process.

Twenty-seven studies (29 reports in total as two studies each had two associated reports) were full-text papers and two studies were only available as abstracts. 34,42 Four studies (five reports) were non-English language, with one each in Japanese,29 Chinese,26 German37 and Spanish. 50,51 Of the 29 included studies, 22 (24 reports)24–27,29,31,33–42,44–51 were diagnostic studies involving people with suspected nAMD and eight23,28,30,32,43,45,52,53 were monitoring studies involving people previously diagnosed with nAMD and under follow-up surveillance. One study, by Salinas-Alaman et al. ,45 reported results for both diagnosis and monitoring.

Number and type of studies excluded

A list of full-text papers that were excluded along with the reasons for their exclusion is given in Appendix 3. These reports were excluded because they failed to meet one or more of the inclusion criteria in terms of the type of study, participants, test, reference standard or outcomes reported.

Characteristics of the included diagnostic studies

Appendix 4 (see Table 42) provides details of the individual study characteristics for the 22 diagnostic studies. Table 4 provides summary information for these studies. Of the 22 studies, nine were prospective24,25,27,33,39–41,45,46 and seven were retrospective. 34–36,38,39,49,51 Seven studies did not provide this information. 26,29,31,37,42,44,48 (The study by Loewenstein et al. 39 reported both a prospective and retrospective component.) In 10 studies, participant recruitment was consecutive. 33,34,38,39,42,44–46,48,49 The studies enrolled more than 2000 participants. Twenty-one studies reported eye as the unit of analysis (1754 eyes), whereas one42 reported patient as the unit of analysis (155 patients).

Seven studies were undertaken in the USA,27,34,36,38,40,41,44 three in the UK,33,46,49 two each in Japan,29,31 Austria37,48 and Spain,45,51 and one each in Portugal,25 Italy (involving eight centres),42 the Republic of Korea35 and China. 26 The remaining two studies were international, taking place in (a) seven centres in the USA, Germany, Israel, Austria and Portugal24 and (b) 15 centres in Israel and the USA. 39 Of the three UK-based studies, two took place at the Royal Victoria Infirmary, Newcastle upon Tyne,46,49 while the third took place at King’s College Hospital, London. 33 One of the UK-based studies, by Talks et al. , involved a nurse-led, fast-track screening clinic. 49

The largest study was by Kozak et al. ,36 which reported TD-OCT, was set in the USA and analysed 541 eyes, whereas the smallest was by Sulzbacher et al. ,48 reporting ICGA and included only 13 eyes.

Across 15 studies reporting the mean or median age of the participants,24–27,29,31,35–37,39,40,45,46,49,51 the median (range) of these values was 76 years (51.4–84.6 years). Fourteen studies involving 1633 participants provided information on gender, in which 742 (45.4%) participants were men and 891 (54.6%) were women. 24,25,27,29,31,35,36,39–41,45,46,49,51 The median (range) prevalence of nAMD across 13 studies where this information was available at participant level was 80.0% (17.2–100.0%). 24,25,27,33,35,38–41,44,45,49,51

In three studies, by Cachulo et al. ,25 Do et al. 27 and Padnick-Silver et al. ,40 the inclusion criteria specified that participants were required to have previously diagnosed nAMD in the non-study eye.

Thirteen studies reported OCT (12 TD-OCT;25,27,33–38,40,45,46,49 one SD-OCT). 41 The study by Kozak et al. ,36 reporting TD-OCT, included a subset of patients who underwent additional examination with SD-OCT. 36

Of the other tests reported, three studies reported PHP,24,27,39 one reported colour fundus photography,24 one Amsler grid,27 one FAF imaging25 and eight ICGA. 25,26,29,31,42,44,48,51 Of the studies reporting more than one test, Cachulo et al. 25 reported TD-OCT, ICGA and FAF, Do et al. 27 TD-OCT, Amsler grid and PHP, and Alster et al. 24 reported PHP and colour fundus photography. Two studies reported combinations of tests: Alster et al. 24 reported colour fundus photography plus VA, whereas Sandhu and Talks46 reported TD-OCT plus colour fundus photography.

The 13 studies reporting OCT analysed 1262 eyes; in eight studies one eye per patient was analysed (n = 479 eyes) (all TD-OCT). 25,27,33–35,38,40,49 Eight studies reported detection of nAMD phenotypes (predominantly classic, minimally classic, occult CNV). 25,33,34,37,38,41,46,49 Four of these studies also reported detection of RAP. 25,34,37,38

Of the eight studies reporting ICGA, seven used the eye as the unit of analysis (number of eyes analysed = 291). 25,26,29,31,44,48,51 In three of these studies, one eye per patient was analysed (n = 109 eyes). 25,31,44 Three studies only reported detection of nAMD phenotypes: IPCV;31 occult CNV;26 and type 2 CNV without an occult component. 48 The study by Parravano et al. ,42 with patient as the unit of analysis (n = 155 patients), also only reported detection of an nAMD phenotype – RAP.

The three studies reporting PHP analysed one eye per patient (n = 302 eyes),24,27,39 as did the studies reporting colour fundus photography (n = 120 eyes),24 Amsler grid (n = 46 eyes)27 and FAF (n = 50 eyes). 25

Risk of bias of the included diagnostic studies

All 20 full-text papers were assessed using a modified version of the QUADAS-2 tool containing 12 items. QUADAS-2 consists of four key domains covering (1) patient selection, (2) index test, (3) reference standard, and (4) flow of patients through the study and timing of the index test(s) and reference standard. Each domain is assessed in terms of the risk of bias and the first three domains are also assessed for concerns regarding their applicability in terms of whether or not they match the question being addressed by the review. Figure 4 presents a summary of the results for the QUADAS-2 risk of bias and applicability domains across the full-text diagnostic papers. Appendix 5 (see Table 44) presents the results of the risk of bias and applicability concerns for the individual studies.

FIGURE 4.

Summary of risk of bias and applicability domains (diagnostic studies).

No study was judged to have a low risk of bias across all domains; in three studies the risk of bias was judged to be unclear across all domains. 29,35,48 The domains in which the greatest number of studies were judged to be at high risk of bias were the patient selection domain (n = 11, 55%) and flow and timing domain (n = 8, 40%).

In the patient selection domain, only one study36 was judged to be at low risk of bias, whereas the majority were considered to have either a high (n = 11, 55%)24,27,31,37–41,44,45,49 or unclear (n = 8, 40%)25,26,29,33,35,46,48,51 risk of bias. Reasons for studies being judged to be at high risk of bias included not enrolling a consecutive sample of participants,27,37 not avoiding inappropriate exclusions24,31,38–41,44 and not avoiding pre-selection of participants. 24,27,31,39,40,44,45,49

In the index/comparator test domain, eight studies (40%) were judged to be at low risk of bias,24,27,33,37,38,41,46,49 two (10%) were considered high risk of bias44,51 and in half (n = 10, 50%) the risk of bias was considered to be unclear. 25,26,29,31,35,36,39,40,45,48 The reasons for the two studies being judged to be at high risk of bias were that the test (ICGA in both cases) was interpreted with knowledge of the results of the reference standard.

In the reference standard domain, five studies (25%) were judged to be at low risk of bias,24,27,33,37,46 three (15%) were considered high risk of bias44,49,51 and in the majority (n = 12, 60%) the risk of bias was considered to be unclear. 25,26,29,31,35,36,38–41,45,48 The reasons for the three studies being judged to be at high risk of bias were that the reference standard test was interpreted with knowledge of the results of the index test (TD-OCT)49 or comparator test (ICGA). 44,51

In the flow and timing domain, six studies (30%) were judged to be at low risk of bias,26,31,37,38,41,44 and the majority were considered to have either a high (n = 8, 40%)24,25,27,36,39,40,46,49 or unclear (n = 6, 30%)29,33,35,45,48,51 risk of bias. Reasons for studies being judged to be at high risk of bias included an interval of more than 1 week between the index/comparator test and reference standard,24,39 not all patients receiving the reference standard test,39 or not all patients being included in the analysis. 24,25,27,36,37,40,46,49

All 20 diagnostic studies were judged to have low concerns for applicability regarding the patient selection, index/comparator test and reference standard domains, in that the participants and setting, index/comparator test and target condition as defined by the reference standard were considered to match the question being addressed by the review.

Results: diagnostic accuracy

Individual study results are presented in Appendix 6 (see Table 46).

Single tests

Optical coherence test

Thirteen studies, analysing 1262 eyes, reported the diagnostic accuracy of OCT in detecting nAMD (12 TD-OCT;25,27,33–38,40,45,46,49 one SD-OCT41). In eight studies, one eye per patient was analysed (n = 479 eyes) (all TD-OCT). 25,27,33–35,38,40,49 Eight studies reported detection of nAMD phenotypes. 25,33,34,37,38,41,46,49

The median (range) prevalence of nAMD across nine OCT studies where this information was available at participant level was 100.0% (17.2–100.0%). 25,27,33,35,38,40,41,45,49

Figure 5 shows a forest plot of the sensitivity and specificity of the individual studies (excluding three where information was only available at phenotype level). 34,37,38 Across these 10 studies, the median (range) sensitivity and specificity values reported were 94.5% (36.0–100.0%) and 73.5% (66.0–94.0%) respectively. Only four studies (all TD-OCT) reported specificity. For TD-OCT, across the studies, the median (range) sensitivity values reported were 92.0% (36.0–100.0%) whereas the only SD-OCT study reported sensitivity of 100%.

FIGURE 5.

Individual study results for all OCT diagnostic studies reporting sensitivity and/or specificity.

The studies shown in Figure 5 demonstrate heterogeneity across the sensitivities reported. The lowest sensitivity reported was by Hughes et al. 33 (36%) and Do et al. 27 (40%). In the study by Hughes et al. ,33 set in the UK, 22 individuals were classed as nAMD by FFA, seven with classic and 15 with occult CNV. TD-OCT detected six of the seven classic CNVs but only 2 of the 15 occult CNVs, hence the low overall sensitivity. The overall prevalence of nAMD in this study was 100%. Do et al. ,27 using TD-OCT in a study set in the USA, reported two separate sets of results, one for when the reference standard was FFA graded as positive by the reading centre irrespective of treatment decision (sensitivity 40.0%, specificity 70.8%), and one for when the reference standard was FFA graded as positive by the reading centre and the clinician recommended treatment (sensitivity 69.2%, specificity 66.2%) (see also Appendix 6, Table 46). The former reference standard was considered closer to the one used in this review and therefore it was these results that were taken to represent the study. Of 87 eyes analysed by Do et al. ,27 15 were classed as nAMD by FFA, with 13 of the 15 CNVs described as occult with no classic. The overall prevalence of nAMD in this study was low at 17.2%. In theory, prevalence should not affect sensitivity, but if the low prevalence contained more people with phenotypes that were difficult to diagnose compared with studies with a higher prevalence of disease, then this might reduce the sensitivity of the test.

By far, the largest study was that by Kozak et al. 36 This retrospective study was set in the USA and involved the analysis of 1272 eyes of 654 participants with a diagnosis of confirmed or suspected macular oedema of various aetiologies; in 541 eyes (number of participants not reported) the aetiology was nAMD. In this study, no data were presented for TNs for the nAMD group and the total number of suspected nAMD classed by FFA as without disease was not reported; as such it was not possible to calculate specificity. The study stated that TD-OCT had detected nAMD in 13 eyes that had not been detected by FFA. As the reference standard of FFA, for the purposes of this review, was considered to have perfect sensitivity and specificity, these 13 cases were classed as TD-OCT FP (although not shown in Figure 5 in order to prevent a spurious specificity value of 0% being calculated based on 13 FPs and zero TNs).

Pigment epithelial detachments can be classified as serous (non-specific) or vascularised. The latter are characteristic of nAMD. A serous PED can occur as a result of retinal conditions other than nAMD, such as central serous chorioretinopathy, angioid streaks or others. The study by Sandhu and Talks,46 considered a serous PED to constitute presence of nAMD and on this basis reported sensitivity of 96.4% and specificity of 66.0%. However, as a serous PED did not fall within our definition of nAMD for diagnostic studies, cases with serous PED were classed as non-nAMD and the data from the study were recalculated accordingly, resulting in alternative values for sensitivity of 77.8% and specificity of 76.0% and it was these values that were taken to represent this study.

Four studies, all TD-OCT,27,40,46,49 reported both sensitivity and specificity, providing sufficient data for inclusion in a meta-analysis. One of the studies, by Talks et al. 49 was a retrospective audit on new patients referred with nAMD to a nurse-led, fast-track screening clinic. Figure 6 shows a forest plot of the sensitivity and specificity of the individual studies and a SROC curve for the four OCT studies. Table 5 shows the pooled estimates for the OCT studies. For all OCT studies, the pooled sensitivity and specificity (95% CI) was 88% (46% to 98%) and 78% (64% to 88%) respectively.

FIGURE 6.

All OCT diagnostic studies reporting sensitivity and specificity. (a) Individual study results; and (b) SROC curve.

TABLE 5 - Pooled estimates for the OCT diagnostic studies

Test	Number of studies	Number of eyes analysed	Pooled estimates (95% CI)
			Sensitivity, %	Specificity, %	LR+	LR–	DOR
All OCT	4	406	88 (46 to 98)	78 (64 to 88)	4.08 (2.37 to 7.04)	0.15 (0.02 to 0.98)	26.86 (3.36 to 214.81)

A LR describes how many times a person with disease is more likely to receive a positive (LR+) or negative (LR–) test result than a person without disease. It has been suggested that LR+s > 10 or LR−s < 0.1 can provide convincing diagnostic evidence, whereas those > 5 and < 0.2 demonstrate strong diagnostic evidence. 54 The LR+ did not exceed 5 for OCT.

The DOR is a single summary of diagnostic performance and describes the ratio of the odds of a positive test result in an individual with disease compared with someone without disease. It has been suggested that a DOR of 25 could provide strong diagnostic evidence and that a DOR of 100 could provide convincing diagnostic evidence. 20

The risk of bias assessment of the four OCT studies included in the meta-analysis is shown in Table 6. The domains in which most studies were judged to be at high risk of bias were the patient selection domain, for reasons such as not enrolling a consecutive sample of participants,27 not avoiding inappropriate exclusions40 and not avoiding pre-selection of participants,27,40,49 and the flow and timing domain, due to all patients not being included in the analysis (all four studies).

TABLE 6 - Risk of bias of the four OCT studies included in the meta-analysis

Study	Risk of bias domain
	Patient selection	Index/comparator test	Reference standard	Flow and timing
Do 201227	High	Low	Low	High
Padnick-Silver 201240	High	Unclear	Unclear	High
Sandhu 200546	Unclear	Low	Low	High
Talks 200749	High	Low	High	High

Eight studies25,33,34,37,38,41,46,49 reported the sensitivity of OCT in the detection of nAMD phenotypes (Table 7). The studies by Cachulo et al. 25 and Khondkaryan et al. ,34 and Talks et al. ,49 using TD-OCT, and Park et al. 41 using SD-OCT showed equally high sensitivity for the detection of classic CNV compared with occult CNV. On the other hand, the studies by Hughes et al. 33 (TD-OCT), Krebs et al. 37 (TD-OCT), Liakopoulos et al. 38 (TD-OCT), and Sandhu and Talks46 (TD-OCT) reported higher sensitivity for OCT in the detection of classic CNV compared with occult CNV.

TABLE 7 - Sensitivity of OCT in detecting nAMD phenotypes

Study ID	Test	Unit of analysis	nAMD phenotype	Number by FFA	OCT sensitivity, %
Cachulo 201125	TD-OCT	Eye	Predominantly classic	2	100.0
			Minimally classic	4	100.0
			Occult	6	100.0
			RAP	5	100.0
Hughes 200533	TD-OCT	Eye	Classic	7	85.7
			Occult	15	13.3
Khondkaryan 200934	TD-OCT	Eye	Classic	Not reported	80.9
			Occult		81.1
			RAP		57.1
Krebs 200737	TD-OCT	Eye	Primarily classic	5	100.0
			RAP	11	72.7
Liakopoulos 200838	TD-OCT	Eye	Subretinal fluid
			Predominantly classic	11	100.0
			Minimally classic	23	91.3
			Occult with no classic	24	79.2
			RAP stage III	8	50.0
			Cystoid oedema
			Predominantly classic	11	81.8
			Minimally classic	23	73.9
			Occult with no classic	24	58.3
			RAP stage III	8	100.0
Park 201041	SD-OCT	Eye	Classic	7	100.0
			Minimally classic	3	100.0
			Occult	11	100.0
Sandhu 200546	TD-OCT	Eye	Classic	56	78.6
			Occult	25	20.0
	TD-OCT + fundus photo	Eye	Classic	56	82.1
			Occult	25	12.0
Talks 200749	TD-OCT	Eye	Predominantly classic	22	100.0
			Minimally classic	6	100.0
			Occult	45	100.0

Amsler grid

One study, by Do et al. ,27 in an analysis of 46 eyes of 46 patients, reported sensitivity of 41.7% for the Amsler grid in detecting nAMD (specificity not reported and insufficient information to calculate prevalence of nAMD in this group). As this study also reported OCT, information on risk of bias is presented in that section. 27

Fundus autofluorescence imaging

One study, by Cachulo et al. ,25 in an analysis of 50 eyes of 50 patients, reported sensitivity of 93.3% and specificity of 37.1% for FAF in detecting nAMD. The prevalence of nAMD in this group was 30.0%. As this study also reported ICGA, information on risk of bias is presented in that section. 25

Colour fundus photography

One study, by Alster et al. ,24 in an analysis of 120 eyes of 120 patients, reported sensitivity of 70.0% and specificity of 95.0% for colour fundus photography in detecting nAMD. The prevalence of nAMD in this study was 53.3%. As this study also reported PHP, information on risk of bias is presented in that section. 24

Preferential hyperacuity perimetry

Three studies analysing 302 eyes of 302 patients reported the diagnostic accuracy of the PHP test in detecting nAMD. 24,27,39 Figure 7 shows a forest plot with the individual study results for sensitivity and specificity. The studies by Alster et al. 24 and Loewenstein et al. 39 reported similarly high sensitivity and specificity. However, it was not possible to calculate pooled estimates using HSROC methodology due to insufficient data. The study by Do et al. 27 reported lower sensitivity and did not report specificity. Across the studies the median (range) of sensitivity values reported was 82% (50–85%). The specificity values reported by Alster et al. 24 and Loewenstein et al. 39 were 88% and 85% respectively.

FIGURE 7.

Preferential hyperacuity perimetry studies: individual study results for sensitivity and specificity.

Across the three studies, the median (range) prevalence of nAMD was 50.4% (17.2–53.3%).

The risk of bias assessment of the three PHP studies is shown in Table 8. The domains in which most studies were judged to be at high risk of bias were the patient selection domain, for reasons such as inappropriate exclusions24,39 and pre-selection of participants,24,27,39 and the flow and timing domain, for reasons such as an interval of more than 1 week between the index test and reference standard,24,39 not all patients receiving the reference standard test39 and not all patients included in the analysis. 24,39

TABLE 8 - Risk of bias of the PHP studies

Study	Risk of bias domain
	Patient selection	Index/comparator test	Reference standard	Flow and timing
Alster 200524	High	Low	Low	High
Do 201227	High	Low	Low	High
Loewenstein 201039	High	Unclear	Unclear	High

Loewenstein et al. 39 also reported the ability of PHP in detecting nAMD phenotypes, with 90% (18/20) sensitivity for minimally or predominantly classic CNV and 82.6% (38/46) sensitivity for occult CNV.

Indocyanine green angiography

Eight studies reported the diagnostic accuracy of ICGA in detecting nAMD, of which seven25,26,29,31,44,48,51 reported the eye as the unit of analysis and one42 reported the patient as the unit of analysis. Four of these studies only reported detection of nAMD phenotypes: IPCV;31 occult CNV;26 type 2 CNV without an occult component;48 and RAP. 42

The median (range) prevalence of nAMD across three studies where this information was available at participant level (and excluding studies reporting results only at phenotype level) was 80.0% (32.7–100.0%). 25,44,51

Figure 8 shows a forest plot of the sensitivity and specificity of the individual studies (excluding the four that only reported detection of phenotypes). Across the studies, the median (range) sensitivity reported was high at 93% (85–100%). Only the study by Fujii et al. 29 reported specificity, which was low at 37%.

FIGURE 8.

Indocyanine green angiography sensitivity and specificity: individual study results.

In the study by Reichel et al. ,44 all participants were deemed to have nAMD (therefore there could be no TNs and it was not possible to calculate specificity). Only participants who were suspected to have a CNV obscured by haemorrhage were included in this study. The authors stated that ICGA had detected nAMD in four eyes that had not been detected by FFA. As the reference standard of FFA, for the purposes of this review, was considered to have perfect sensitivity and specificity, these four cases were classed as ICGA FPs (although not shown in Figure 8 in order to prevent a spurious specificity value of 0% being calculated based on four FPs and zero TNs).

The risk of bias assessment of the four ICGA studies is shown in Table 9. The domains in which most studies were judged to be at high risk of bias were the index/comparator test domain, due to the ICGA test being interpreted with knowledge of the FFA results, and the reference standard domain, due to FFA being interpreted with knowledge of the ICGA results. 44,51

TABLE 9 - Risk of bias of the four ICGA studies included in the forest plot

Study	Risk of bias domain
	Patient selection	Index/comparator test	Reference standard	Flow and timing
Cachulo 201125	Unclear	Unclear	Unclear	High
Fujii 199629	Unclear	Unclear	Unclear	Unclear
Reichel 199544	High	High	High	Low
Torron 200251	Unclear	High	High	Unclear

Four studies26,31,42,48 reported the sensitivity of ICGA in the detection of nAMD phenotypes, with each study reporting detection of a different phenotype (Table 10). Sensitivity was 100% for detection of IPCV31 and type 2 CNV without an occult component,48 high (85.1%) for detection of RAP42 but lower (62.9%) for detection of occult CNV. 26

TABLE 10 - Sensitivity of ICGA in detecting nAMD phenotypes

Study	Test	Unit of analysis	nAMD phenotype	Number by FFA	ICGA sensitivity, %
Chen 200326	ICGA	Eye	Occult CNV	35	62.9
Gomi 200731	ICGA	Eye	IPCV	37	100.0
Sulzbacher 201148	ICGA	Eye	Type 2 CNV without an occult component	13	100.0
Parravano 201242	ICGA	Patient	RAP	155	85.1

Assessment of other outcomes of interest

Characteristics of the included monitoring studies

Appendix 4 (see Table 43) provides details of the individual study characteristics for the eight monitoring studies. 23,28,30,32,43,45,52,53 Table 12 provides summary information for the studies. Of the eight monitoring studies, four were prospective,32,43,45,53 three were retrospective,23,28,30 and in the study by van de Moere et al. 52 this information was not reported. In five studies, the participants were a consecutive sample. 28,30,43,45,53 The eight studies enrolled 463 participants.

Five studies used the eye as the unit of analysis (363 eyes),23,28,30,52,53 whereas three used test examination as the unit of analysis (61 pairs of OCT and FFA examinations,32 176 pairs of OCT and FFA examinations45 and 54 pairs of ICGA and FFA examinations). 43

Two studies were undertaken in the USA23,43 and one each in Italy,30 Germany,32 the Netherlands,53 Spain45 and the UK (Royal Victoria Infirmary, Newcastle upon Tyne). 52 One study was international, taking place in two centres in the USA and Germany. 28

The largest study was by van de Moere et al. ,52 which reported TD-OCT, was set in the UK and analysed 121 eyes, while the smallest was by van Velthoven et al. ,53 reporting TD-OCT and analysing 30 eyes.

Across seven studies23,28,30,43,45,52,53 reporting the mean or median age of the participants, the median (range) of these values was 76.5 years (73.9–78.1 years). Six studies involving 378 participants provided information on gender,28,30,43,45,52,53 in which 177 (46.8%) participants were men and 201 (53.2%) women. The median (range) prevalence of active nAMD across five studies where this information was available at participant level was 57.9% (49.2–83.3%). 23,28,30,52,53

Seven studies reported OCT (six TD-OCT;23,28,32,45,52,53 and two SD-OCT). 23,30 (The study by Khurana et al. 23 reported both TD-OCT and SD-OCT.) One study, by Regillo et al. ,43 reported ICGA.

Of the seven studies reporting OCT, five used the eye as the unit of analysis (number of eyes analysed = 363). 23,28,30,52,53 In four of these studies, one eye per patient was analysed (n = 304 eyes). 28,30,52,53 Two studies reported examination as the unit of analysis (both TD-OCT). 32,45 Two studies reported detection of nAMD phenotype activity: classic and occult CNV;30 and PED and cystoid macular oedema. 52 The studies by Henschel et al. 32 and van de Moere et al. 52 also reported the performance of OCT in detecting intraretinal and subretinal fluid.

In two OCT monitoring studies,23,30 the participants had received antiVEGF therapy and in five28,32,45,52,53 the treatment was photodynamic therapy (PDT). In the study reporting ICGA,43 the participants had received laser photocoagulation treatment.

Risk of bias of the included monitoring studies

Figure 9 presents a summary of the results for the QUADAS-2 risk of bias and applicability domains across the eight full-text monitoring papers. Appendix 5 (see Table 45) presents the results of the risk of bias and applicability concerns for the individual studies.

FIGURE 9.

Summary of risk of bias and applicability domains (monitoring studies).

No study was judged to have a low risk of bias across all domains. More studies in the patient selection domain (n = 2, 25%) and the flow and timing domain (n = 2, 25%) were judged to be at high risk of bias than in the index/comparator test domain (n = 1, 12.5%) and reference standard domain (n = 1, 12.5%).

In the patient selection domain, three studies43,52,53 (37.5%) were judged to be at low risk of bias, two30,45 (25%) were considered to have a high risk of bias and in three23,28,32 (37.5%) the risk of bias was unclear. The study by Giani et al. 30 was judged to be at high risk of bias due to not avoiding inappropriate exclusions and pre-selection of participants, whereas the study by Salinas-Alaman et al. 45 was judged to be at high risk of bias due to not avoiding pre-selection of participants.

In the index/comparator test domain, three studies (37.5%) were judged to be at low risk of bias,28,30,32 one (12.5%) was considered high risk of bias43 and in the remaining four (50%) the risk of bias was considered to be unclear. 23,45,52,53 The reasons for the study by Regillo et al. 43 being judged to be at high risk of bias was that the test (ICGA) was interpreted with knowledge of the results of the reference standard.

In the reference standard domain, four studies (50%) were judged to be at low risk of bias,28,30,32,53 one (12.5%) was considered high risk of bias43 and in the remaining three (37.5%) the risk of bias was considered to be unclear. 23,45,52 The Regillo et al. 43 study was judged to be at high risk of bias as the reference standard test was interpreted with knowledge of the results of the comparator test (ICGA).

In the flow and timing domain, two studies (25%) were judged to be at low risk of bias,23,43 two45,52 (25%) were considered to have a high risk of bias and in the remaining four28,30,32,53 (50%) the risk of bias was considered to be unclear. The studies by Khurana et al. 23 and Regillo et al. 43 were judged to be at high risk of bias as not all patients were included in the analysis.

All eight studies were judged to have low concerns for applicability on the patient selection, index/comparator test and reference standard domains.

Results: detection of active neovascular age-related macular degeneration

Individual study results are presented in Appendix 6 (see Table 47).

Single tests

Optical coherence tomography

Seven studies reported the accuracy of OCT in detecting active nAMD, of which five reported TD-OCT,28,32,45,52,53 one reported SD-OCT30 and one reported both TD-OCT and SD-OCT. 23 In five studies the unit of analysis was the eye23,28,30,52,53 and in two the unit of analysis was pairs of OCT and FFA examinations. 32,45

The median (range) prevalence of active nAMD across five studies where this information was available at participant level was 57.9% (49.2–83.3%). 23,28,30,52,53

Three TD-OCT studies23,28,53 and two SD-OCT studies,23,30 with eye as the unit of analysis, reported both sensitivity and specificity, providing sufficient data for inclusion in a meta-analysis. Figure 10 shows forest plots of the sensitivity and specificity of the individual studies and SROC curves for (a) all of the OCT studies, (b) the three TD-OCT studies and (c) the two SD-OCT studies respectively. Table 13 shows the pooled estimates for these studies. As the study by Khurana et al. 23 reported both TD-OCT and SD-OCT for the same 59 eyes, we chose to display only the data for SD-OCT from this study in the forest plot of all OCT studies and to include only the SD-OCT data from this study in the pooled estimates for all OCT studies, in order to avoid double counting and on the basis that the SD-OCT data were the more appropriate to include in the pooled estimates for all OCT. The TD-OCT data from Khurana et al. 23 are included in the forest plot and SROC curve for TD-OCT in Figure 10 and were included in the pooled estimates for TD-OCT shown in Table 13. For all OCT studies, the pooled sensitivity and specificity (95% CI) was 85% (72% to 93%) and 48% (30% to 67%) respectively. For TD-OCT, the pooled sensitivity and specificity (95% CI) was 70% (56% to 80%) and 65% (48% to 79%) respectively. For both TD-OCT and the group of all four OCT studies, the LR and DOR values reported were below the level suggestive of strong diagnostic evidence.

FIGURE 10.

Optical coherence tomography monitoring studies reporting sensitivity and specificity for detection of nAMD activity: individual study results and SROC curves. (a) Sensitivity and specificity – individual study results – all OCT; (b) sensitivity and specificity – individual study results – TD-OCT; (c) sensitivity and specificity – individual study results – SD-OCT; (d) SROC curve – all OCT; (e) SROC curve – TD-OCT; and (f) SROC curve – SD-OCT.

TABLE 13 - Pooled estimates for the OCT monitoring studies

Khurana et al. 23 reported both TD-OCT and SD-OCT for the same 59 eyes of 56 patients analysed; only the SD-OCT data were included in the pooled estimates for ‘All OCT’ in order to avoid double counting.

Test	Number of studies	Number of eyes analysed	Pooled estimates (95% CI)
			Sensitivity, %	Specificity, %	LR+	LR–	DOR
All OCTa	4	242	85 (72 to 93)	48 (30 to 67)	1.64 (1.19 to 2.26)	0.31 (0.18 to 0.54)	5.33 (2.57 to 11.06)
TD-OCT	3	149	70 (56 to 80)	65 (48 to 79)	2.00 (1.19 to 3.36)	0.47 (0.28 to 0.78)	4.27 (1.58 to 11.53)
SD-OCT	2	152	Not calculable using HSROC methodology

It was not possible to calculate pooled estimates using HSROC methodology for the two SD-OCT studies due to insufficient data. These studies reported sensitivities of 94%30 and 90%23 and specificities of 27%30 and 47%,23 which suggests that SD-OCT has higher sensitivity than TD-OCT but lower specificity.

The risk of bias assessment of the four OCT studies included in the meta-analysis is shown in Table 14. The only judgement of high risk of bias was for the study by Giani et al. 30 for the patient selection domain (inappropriate exclusions and pre-selection of participants).

TABLE 14 - Risk of bias of the four OCT studies included in the meta-analysis

Study	Risk of bias domain
	Patient selection	Index/comparator test	Reference standard	Flow and timing
Eter 200528	Unclear	Low	Low	Unclear
Giani 201130	High	Low	Low	Unclear
Khurana 201023	Unclear	Unclear	Unclear	Low
van Velthoven 200653	Low	Unclear	Low	Unclear

Two studies used examination as the unit of analysis. Henschel et al. ,32 in an analysis of 61 pairs of TD-OCT and FFA examinations from 14 patients, reported sensitivity of 96.8% and specificity of 36.7% for CNV based on detection of intraretinal and/or subretinal fluid. Salinas-Alaman et al. ,45 in an analysis of 176 pairs of TD-OCT and FFA examinations (number of patients not stated), reported sensitivity of 95.7% and specificity of 59.0% based on detection of intraretinal or subretinal fluid.

Four studies23,30,32,52 reported the sensitivity of OCT in detecting active nAMD phenotypes or active nAMD based on detection of intraretinal/subretinal fluid (Table 15). The study by Giani et al. 30 reported high sensitivity for the detection by SD-OCT of both classic and occult CNV activity (90.9% and 100% respectively). In the studies by Henschel et al. 32 (unit of analysis: examination) and van de Moere et al. 52 (unit of analysis: eye) sensitivity was higher for nAMD activity based on detection of intraretinal fluid (90.3% and 82.9% respectively) compared with subretinal fluid (71.0% and 47.1% respectively). van de Moere et al. 52 also reported sensitivity of TD-OCT for detection of cystoid macular oedema and PED, both low at 22.9% and 5.7% respectively. In the study by Khurana et al. ,23 the sensitivity of SD-OCT was higher than that of TD-OCT for nAMD activity based on the detection of intraretinal fluid, retinal cystoid abnormalities or subretinal fluid.

TABLE 15 - Sensitivity of OCT in detecting nAMD phenotype activity

CMO, cystoid macular oedema.

Study	Unit of analysis	Detection of	Number by FFA	OCT sensitivity, %
Giani 201130 (SD-OCT)	Eye	Classic CNV	57	90.9
		Occult CNV	36	100.0
Khurana 201023 (TD-OCT)	Eye	Intraretinal fluid	29	37.9
		Retinal cystoid abnormalities	29	34.5
		Subretinal fluid	29	48.3
Khurana 201023 (SD-OCT)	Eye	Intraretinal fluid	29	65.5
		Retinal cystoid abnormalities	29	58.6
		Subretinal fluid	29	69.0
van de Moere 200652 (TD-OCT)	Eye	Intraretinal fluid	Not reported	82.9
		Subretinal fluid	Not reported	47.1
		CMO	Not reported	22.9
		PED	Not reported	5.7
Henschel 200932 (TD-OCT)	Exam	Intraretinal fluid	31	90.3
		Subretinal fluid	31	71.0

Indocyanine green angiography

One study, by Regillo et al. ,43 in an analysis of 54 pairs of ICG angiograms compared with fluorescein angiograms, obtained from 24 eyes of 21 patients, reported sensitivity of 75.9% and specificity of 88.0% in detecting nAMD activity. It was not possible to ascertain (at participant-level) the prevalence of nAMD. This study was judged as high risk of bias for the index/comparator test and reference standard domains, due to the ICGA–FFA pairs being analysed directly from the computer monitor (ICGA test results interpreted with knowledge of the FFA results, and vice versa) and low risk of bias for the other domains.

Assessment of other outcomes of interest

Diagnostic studies

Twenty-two diagnostic studies were included (20 full-text papers, two abstracts). 24–27,29,31,33–42,44–46,48,49,51 The full-text papers were assessed for risk of bias using the QUADAS-2 checklist. The domains in which the greatest number were judged to be at high risk of bias were the patient selection domain (55%, 11/20), for reasons such as inappropriate exclusions and pre-selection of participants, and flow and timing domain (40%, 8/20), for reasons such as the length of time between the index test and the reference standard, and not all participants being included in the analysis. The risk of bias in the index/comparator test and reference standard domains was judged to be unclear in 50% (10/20) and 60% (12/20) of studies respectively. All of the studies were judged to have low concerns in terms of their applicability to the question being addressed by the review.

A descriptive summary of the results of the diagnostic studies with eye as the unit of analysis is shown in Table 16 (excluding studies that only reported detection at phenotype level). Across the studies the median (range) sensitivity was high for OCT (94.5%, range 36.0–100.0%; 10 studies25,27,33,35,36,40,41,45,46,49). Sensitivity was also high for ICGA (93.2%, range 84.6%–100.0%; four studies25,29,44,51) and FAF (93.3%; one study25), followed by PHP (81.5%, range 50.0–84.8%; three studies24,27,29), colour fundus photography (70.0%; one study24) and lowest for Amsler grid (41.7%; one study27). The median (range) specificity for OCT was moderate (73.5%, range 66.0–94.0%; four studies27,40,46,49). Specificity was highest for colour fundus photography (95%; one study24), followed by PHP (84.6% and 87.7%; two studies24,39), and was low for FAF (37.1%; one study25) and ICGA (36.8%; one study29).

Two studies reported the diagnostic accuracy of combinations of tests. Sensitivity and specificity for TD-OCT plus colour fundus photography46 was 74.1% and 92.0%, respectively, whereas for colour fundus photography plus VA,24 sensitivity was lower at 53.0% but with similarly high specificity at 94.0%.

Four OCT diagnostic studies (all TD-OCT) provided sufficient data for inclusion in a meta-analysis (Table 17). 27,40,46,49 The pooled sensitivity and specificity (95% CI) for all four OCT studies was 88% (46% to 98%) and 78% (64% to 88%) respectively.

Eight diagnostic studies reported the sensitivity of OCT in the detection of specific nAMD phenotypes. 25,33,34,37,38,41,46,49 Four showed equally high sensitivity for the detection of classic CNV compared with occult CNV. 25,34,41,49 In four others, sensitivity for OCT was higher in the detection of classic CNV (range 79–100%) compared with occult CNV (range 13–79%). 33,37,38,46 Four studies reported the sensitivity of ICGA in the detection of specific nAMD phenotypes. 25,29,44,51 Each study reported detection of a different phenotype, with 100% sensitivity for detection of IPCV and type 2 CNV without an occult component, high sensitivity (85.1%) for detection of RAP but lower sensitivity (62.9%) for detection of occult CNV. 25,29,44,51

Monitoring studies

Eight monitoring studies were included (all full-text papers). 23,28,30,32,43,45,52,53 Seven reported OCT,23,28,30,32,45,52,53 five with eye as the unit of analysis23,28,30,52,53 (one of which only reported detection at phenotype level30) and two with test examination as the unit of analysis. 32,45 One study reported ICGA. 43 As with the diagnostic studies, the QUADAS-2 domains in which the greatest number of monitoring studies were judged to be at high risk of bias were the patient selection domain (25%, 2/8),30,45 for reasons such as inappropriate exclusions and pre-selection of participants, and flow and timing domain (25%, 2/8),30,45 for reasons such as the length of time between the index test and the reference standard, and not all participants being included in the analysis. The risk of bias in the index/comparator test and reference standard domains was judged to be unclear in 50% (4/8)23,45,52,53 and 37.5% (3/8)23,45,52 of studies respectively. All of the monitoring studies were judged to have low concerns in terms of their applicability to the question being addressed by the review.

Four OCT monitoring studies, with eye as the unit of analysis, provided sufficient data for inclusion in a meta-analysis (Table 18). The pooled sensitivity and specificity (95% CI) for all four OCT studies was 85% (72% to 93%) and 48% (30% to 67%) respectively. For TD-OCT, the pooled sensitivity and specificity was 70% (56% to 80%) and 65% (48% to 79%). It was not possible to calculate pooled estimates using HSROC methodology for the two SD-OCT studies due to insufficient data. These two studies reported sensitivities of 94% and 90% and specificities of 27% and 47%, which suggests that SD-OCT has higher sensitivity than TD-OCT but lower specificity.

Two OCT monitoring studies used test examination as the unit of analysis. The first, in an analysis of 61 pairs of TD-OCT and FFA examinations from 14 patients, reported high sensitivity of 96.8% but low specificity of 36.7%, for CNV based on detection of intraretinal and/or subretinal fluid. The second, in an analysis of 176 pairs of TD-OCT and FFA examinations (number of patients not stated), reported similarly high sensitivity of 95.7% and moderate specificity of 59.0% based on detection of intraretinal or subretinal fluid.

One ICGA monitoring study used test examination as the unit of analysis. In an analysis of 54 pairs of ICGAs compared with fluorescein angiograms, obtained from 24 eyes of 21 patients, sensitivity of 75.9% and specificity of 88.0% was reported for detecting nAMD activity.

Three studies reported OCT sensitivity in detecting activity of specific nAMD phenotypes or nAMD activity based on detection of intraretinal/subretinal fluid. SD-OCT sensitivity was high for the detection of both classic and occult CNV activity (90.9% and 100% respectively) (one study). 30 Sensitivity of TD-OCT for detection of cystoid macular oedema and PED was low (22.9% and 5.7% respectively) (one study). 52 In two studies, sensitivity was higher for detection of nAMD activity based on intraretinal fluid (90.3% and 82.9% respectively) compared with subretinal fluid (71.0% and 47.1% respectively). 32,52

Inclusion and exclusion criteria

Inclusion criteria required the studies to be full economic evaluations,55 that is, to consider cost and effects for more than one strategy, in order to be included in the review. No restrictions were imposed in the way cost and/or effects were calculated. In addition, at least one of the compared strategies for diagnosis or monitoring of nAMD had to include OCT. Finally, the studies were required to be performed in adults with nAMD.

Search strategy

Studies that reported both costs and outcomes in diagnosing nAMD using OCT were sought from a systematic review of the literature. No language restrictions or limitations to searches were imposed.

Databases searched were MEDLINE (1996–November Week 2 2012), EMBASE (1980–Week 45 2012), MEDLINE In-Process & Other Non-Indexed Citations (14 November 2012), NHS Economic Evaluation Database (inception to October 2012), HTA database (inception to October 2012), Health Management Information Consortium (1979–September 2012), Research Papers in Economics (September 2012) and ARVO meeting abstracts from April 2009. In addition, reference lists of all included studies were scanned to identify additional potentially relevant studies. Full details of the search strategies used are documented in Appendix 1.

Results

From the database searches, 473 hits (titles and abstracts) were retrieved; from these 44 studies were selected for full-text assessment. No studies fulfilled the inclusion criteria as none of these were diagnosis or monitoring interventions for individuals with nAMD.

Diagnosis strategies

1. (Stereoscopic) FFA interpreted by an ophthalmologist. If positive, treat and monitor; if negative, discharge.

2. OCT alone interpreted by an ophthalmologist. If positive, treat and monitor; if negative, discharge.

3. VA, OCT and SLB in all. If positive or unclear, then arrange for stereoscopic FFA. If negative, discharge. This is the strategy for diagnosis that best reflects standard practice.

Monitoring strategies

1. OCT alone (interpreted by an ophthalmologist). If positive, treat; if negative or unclear, review in 1 month’s time.

2. VA, SLB and OCT interpreted together by an ophthalmologist. If positive, treat; if negative, review in 1 month’s time. If unclear, then the ophthalmologist will arrange for a stereoscopic FFA. This is the monitoring strategy that best reflects standard practice.

3. VA and OCT interpreted by a technician or nurse. If negative, review in 1 month’s time. If positive or unclear, refer for ophthalmologist assessment (e.g. SLB and ophthalmologist’s own interpretation of VA and OCT test results). The ophthalmologist will make a decision: if positive, treat; if negative, review in 1 month’s time; if unclear, arrange for stereoscopic FFA.

Monitoring strategy c has been included in the monitoring stage in order to explore the cost-effectiveness of the option, for example, of virtual clinics involving other health-care professionals (e.g. nurses, technicians). Virtual clinics are increasingly used in NHS services for monitoring patients with nAMD. 56

Table 19 shows the final nine combined strategies incorporated into the decision model. All strategies considered monitoring on a monthly basis with a decision to treat when the disease was deemed active (i.e. retinal fluid on OCT). All monitoring strategies that relied on stereoscopic FFA as a final assessment step (e.g. monitoring strategies b and c) would treat if FFA positive, or review in a month’s time if FFA negative. Treatment consisted of one injection only (i.e. 0.5 mg ranibizumab) with review in 1 month’s time.

The economic model

A Markov model approach was selected for the decision-analytic model exercise. 57–60 Markov models have Markov states where individuals spend a period of time, named a ‘cycle’. At the end of each cycle the individuals can remain in their current Markov state or move to another state. The probabilities of moving to other Markov states or remaining in the current state are named ‘transition probabilities’. Individuals in the model would accrue costs and benefits (e.g. ‘life-years’) depending on the time spent in each Markov state and the interventions and/or events modelled within each Markov state. Markov models are particularly suitable to model recurrent issues and chronic diseases. They allow incorporating health states to reflect the movement of the patients during diagnosis and monitoring. In the current study, model states reflect the underlying condition (e.g. nAMD active or inactive) together with the decision on treatment (e.g. treated or untreated nAMD) and VA states of the individuals (Table 20). In all these models, an absorbing state is included where all individuals would end up if the model was run for a sufficiently long period of time (e.g. death state).

The present model incorporates a first diagnosis stage combined with a recurrent (monthly) monitoring phase.

The Markov models

This section presents a stepwise introduction to the Markov models used to compare the alternative strategies. Individuals’ VA status is set aside for the moment to focus on the other two issues and assumptions underpinning the movement of individuals throughout the model: (1) the underlying disease condition (e.g. if the disease is present or not and, if present, its active or inactive status) as well as (2) the diagnosis or monitoring test results on which the treatment decision will depend (i.e. a positive result will trigger a decision to treat and a negative results will trigger a decision not to treat). Figure 11 shows the schematic diagram of the final model used for the economic evaluation for this study.

FIGURE 11.

Markov model schematic diagram assuming imperfect information at diagnosis and monitoring stages. ‘Inactive’ = underlying condition regarded as inactive nAMD when the disease was actually not present.

This section presents three schematic diagrams for this model. The figures differ in the assumptions made with respect to the information retrieved from the diagnosis and/or monitoring test or assessments. Namely, if perfect information from the tests or assessments is assumed, then there would be no FP or FN results (i.e. equivalent to assuming that sensitivity and specificity are equal to 1). That is, the underlying condition is detected with certainty. When this assumption is relaxed, then the possibility of incorrect assessments appears.

Perfect information from diagnosis and monitoring tests

Figure 12 assumes perfect information at diagnosis and monitoring stages in the model. The whole modelled cohort starts at the black arrow on the left hand side of the figure (corresponding to an initial Markov model stage). The assumption of perfect information means that, at diagnosis stage, all individuals with the disease will have a positive result while all those without the disease will obtain a negative result. Individuals with a positive result will go to a monitoring scheme while those with a negative result will be discharged. Those individuals with the disease and positive results will start within a Markov model state with an ‘active disease and under treatment’ (e.g. ‘active/treated’ state). Note that ‘active’ refers to the underlying condition while ‘treated’ or not depends on the test or assessment result.

FIGURE 12.

Markov model schematic diagram assuming perfect information at diagnosis and monitoring stages.

Assuming monthly monitoring visits and assessments, a positive result at a monitoring visit means the individual’s disease is active (assuming, again, perfect information and no possibility of FP or FN results) and will therefore mean that the person remains in the ‘active/treated’ Markov state. If a negative result from the monitoring assessment is obtained, then it would mean that the individual’s disease has become inactive and the decision not to administer treatment will follow. In this case, the individual will move to the ‘inactive/untreated’ Markov state. At each Markov cycle (monthly) individuals can become active or inactive; this status would be detected at the next monitoring visit with a positive or negative result, and the individual will either move from or stay in the corresponding Markov model state with a consistent treatment decision.

Individuals without the disease at the moment of first diagnosis could develop the disease in the future (i.e. incident cases among the population). In the model (see Figures 11–13), it was assumed that these individuals would be correctly diagnosed within a second visit and eventually moved to be monitored within the ‘active/treated’ state. Finally, the ‘dead’ state is the absorbing state in this model (i.e. a state that individuals cannot move out of); individuals can move from any other Markov state into the ‘dead’ absorbing state.

Imperfect information from diagnosis test combined with perfect information from monitoring tests

Figure 13 shows a similar schematic diagram but in this case there is imperfect information at the moment of first diagnosis. After this initial diagnostic intervention, further diagnosis and/or monitoring assessments will be done with certainty (e.g. assuming perfect information). This opens the possibility of obtaining TP, TN as well as FP and FN results from the initial diagnosis test/s. Individuals with positive results, therefore, might not have nAMD whereas individuals with negative diagnostic test results might actually have the disease. This situation will have an effect on the Markov states the individuals will start at after diagnosis. Those with a TP result will start with their active disease being treated and eventually move to an inactive state (e.g. ‘inactive/untreated’) depending on the treatment effect. Individuals with a FP result will not have nAMD but will be treated and monitored. However, this treatment cannot be effective as these people did not have the disease. As this schematic diagram assumes perfect information at the monitoring phase, these individuals would be correctly assessed in their subsequent monitoring visits, moving to the ‘inactive/untreated’ state.

FIGURE 13.

Markov model schematic diagram assuming imperfect information at diagnosis and perfect information at monitoring stage. ’Inactive’ = underlying condition regarded as inactive nAMD when the disease was actually not present.

In addition, if the person has a negative result at diagnosis, this could be a TN or a FN result. In either case the individual would be discharged under the belief that nAMD was not present. If TN, meaning that the disease was not present, the individual will start at the ‘no disease’ state and will remain at that stage unless they develop nAMD. If FN (patients with the disease and negative test), the person will start within the ‘disease (active)’ state.

Finally, an identical assumption of using FFA for diagnosis for those presenting for a second time (rediagnosis) is followed for those with FN results at first diagnosis. These people will start to be monitored and moved to the ‘active/treated’ state after second presentation for diagnosis. A further assumption is used for this subgroup: based on expert opinion, these nAMD individuals that have been missed at first diagnosis will present for rediagnosis within 3 months. The rationale behind this was the natural history of the disease and the belief that nAMD would advance with VA deterioration making the individual return for a further eye check.

Markov model states and health status valuation link

The former diagrams show how individuals can move in the model according to their underlying condition and the result of the test/s or assessments. However, it is not possible to attach utility weights to these Markov states. In essence, individuals can experience alternative active or inactive disease but no difference in their reported health status. The economic model attaches utility weights according to, mainly, VA. Therefore, the effect on health status will come through the deterioration in VA, whereas VA deterioration will result from the fact of individuals being misdiagnosed (e.g. no nAMD when actually the disease was present) or misclassified as inactive when their true condition was active nAMD.

In terms of the presented diagrams, the number of Markov states is multiplied by the number of VA ranges considered by the model. Therefore, there is a trade-off between the number of VA ranges in order to reflect differences in VA – and patient-reported health status – and the model complexity. Utility differences between the alternative model strategies result from the different periods of time individuals are misclassified within each strategy. It was considered that five VA states (see Table 20) would give sufficient refinement for utility differences to be reflected. This approach has been used in other models in this area of health care. 61 Therefore, each strategy (i.e. each Markov model) has 32 Markov model states [e.g. four VA states multiplied by six monitoring states, plus four VA states multiplied by one nAMD undiagnosed state, plus profound visual loss/blindness, a ‘no disease’ state (normal VA only), the absorbing state ‘dead’, and an initial state for first diagnosis].

Figure 14 shows a Markov model schematic diagram for the VA states considered in the model. Arrows in the figure show the possible movements in the model in one cycle (e.g. 1 month). Individuals’ VA can remain the same, improve or deteriorate in one particular cycle. Individuals can have their VA improved and move one level up at the end of a cycle; however, their VA can deteriorate and move one or two levels down from their current VA state. Finally, the model considered that a VA deterioration of ≤ 3/60 (i.e. profound visual loss/blindness) was not reversible and the individual was referred to supportive care.

FIGURE 14.

Markov model schematic diagram for VA states.

Probabilities

Table 21 shows data on nAMD prevalence, incidence and VA at the start of the model run. Colquitt et al. 61 reviewed studies assessing the prevalence and incidence of AMD and nAMD. The setting for this economic evaluation was secondary care; therefore, the prevalence rate to inform the model should be that corresponding to the group of individuals referred to hospital eye services with a suspected nAMD diagnosis. The prevalence rate used was obtained from the literature retrieved by the systematic review of test accuracy and agreed within the project management and advisory groups. An overall incidence of 1% per year was used based on Mitchell et al. 65 These incidence figures, presented for Australia, were similar to the results by van Leeuwen et al. 66 for the Rotterdam study but were reported in a form that could be readily incorporated into the economic model. Mortality data were obtained from Interim Life Tables for England and Wales (2009–11). 16 No difference in mortality rates were found when comparing age-specific mortality rates form the Interim Life Tables and those from the Comparison of Age-Related Macular Degeneration Treatments Trials (CATT)8 and the Inhibit VEGF in Age-related choroidal Neovascularisation (IVAN)9 studies. Therefore, no excess mortality was included due to nAMD. 67,68 However, excess mortality risk was incorporated for the last disease VA stage (profound visual loss/blindness – VA ≤ 3/60).

Table 21 also shows probability distributions defined for the probabilistic sensitivity analysis. Uninformative uniform distributions were used for nAMD prevalence and profound visual loss/blindness excess mortality. Ranges for defining these were assumptions based on data from the literature if available (e.g. from the review of test accuracy). A gamma distribution was defined for nAMD incidence based on mean and standard deviation (e.g. 1/10 of the mean) using the tool provided by TreeAge (TreeAge Software, Inc., Williamstown, MA, 2013).

The cohort start age was set at 65 years as this is the age where particular changes are observed in the retina and macula (Dr Noemi Lois and Project Advisory Group, NHS Grampian, 2012, personal communication). In addition, mean VA at the start was set at between ≤ 6/12 to > 6/24 for those individuals with nAMD. This was agreed to be the most common VA at presentation by experts and also the mean VA at baseline in the CATT and IVAN studies. 8,9

Table 22 presents diagnostic test performance data. As mentioned above, three strategies were defined for diagnosis within the economic model. For each of these strategies, sensitivity and specificity data were needed, specifically for FFA, OCT, and ophthalmologist assessment (i.e. with VA test, SLB, and the results from the OCT). FFA interpreted by an ophthalmologist was stated as the reference standard for the diagnosis of nAMD; therefore, perfect information was assumed from this test, with sensitivity and specificity equal to 1. OCT sensitivity and specificity were obtained from the systematic review of diagnostic studies. These data correspond to OCT pooled estimates (four studies, number of eyes 406). 27,40,46,49 No studies were identified on the ophthalmologist assessment diagnostic performance. Hence, sensitivity and specificity estimates were derived from expert opinion.

Sensitivity and specificity data are bounded between 0 and 1. Therefore, beta distributions were defined for probabilistic sensitivity analysis. For OCT, these were obtained using mean values and standard deviation in order to obtain values within the 95% CI provided by the systematic review of diagnostic studies (see Chapter 4, Table 17). Probability distributions for ophthalmologist diagnosis assessment were obtained using the approximation tool provided by TreeAge, based on mean and standard deviation (e.g. 1/10 of mean).

Table 23 shows similar data to Table 22 but for monitoring of individuals with nAMD. FFA was also stated as the reference standard to detect disease activity; therefore, perfect information was assumed, with sensitivity and specificity defined as equal to 1. OCT monitoring sensitivity and specificity data were obtained from the systematic review of test performance (see Chapter 4). Pooled estimates (e.g. four studies, n = 242), were used. 23,28,30,53 No studies were identified reporting the diagnostic performance of nurse or technician assessment, or for ophthalmologist assessment. Therefore, estimates for the sensitivity and specificity of these strategies were derived from expert opinion.

A similar approach to the one used for the sensitivity and specificity of diagnostic tests was used for this information for monitoring tests. Beta probability distributions were approximated and defined using mean and standard deviation (e.g. 1/10 of the mean) values. The range of values of OCT used in the model did not exceed the 95% CI values obtained from the systematic review of monitoring studies (i.e. OCT range data in Table 23).

Disease progression in the model was defined in terms of VA changes. Gaining or losing three lines in the Snellen chart (approximately 15 letters in the ETDRS chart) was assumed to make individuals move from their current Markov model state to the next level (see Table 20 and Figure 14). Data for this were obtained from Rosenfeld et al. 69 [i.e. the Minimally Classic/Occult Trial of the Anti-VEGF Antibody Ranibizumab in the Treatment of Neovascular Age-Related Macular Degeneration (MARINA) study]. This study was based in the USA, involved 716 participants and compared monthly treatment with ranibizumab (0.3 mg, n = 238 or 0.5 mg, n = 240) against sham injection (n = 238). Data from treatment (0.5 mg) and control groups were used to calculate monthly progression probabilities for active treated and non-treated individuals respectively. No VA progression was assumed for nAMD inactive individuals as well as non-AMD individuals.

Beta distributions were attached to VA progression data for probabilistic sensitivity analysis (Table 24). Unfortunately, there were no data available to construct CIs around mean values used in the model. As such, probability distributions parameter values were developed using mean values and assuming 1/10 of mean values for standard errors.

Additional data were required on disease status, namely the probability of becoming active when the individual’s disease was inactive and under no treatment, as well as the probability of becoming inactive when the individual’s disease was active and under treatment. First year data for these were developed using data from the IVAN study (Dr Chris Rogers, University of Bristol, 12 June 2013, personal communication). The IVAN study was a 2 × 2 factorial design and adults with untreated nAMD were randomised into four groups: ranibizumab or bevacizumab, given either every month (continuous) or as needed (discontinuous). All individuals were reviewed on a monthly basis. Survival data for participants’ first treatment failure (e.g. subretinal fluid, increasing intraretinal fluid, or fresh blood) for the discontinuous arm (n = 302) were used to develop mean probability values. All individuals were active at baseline and 95% of these did not fail the retreatment criteria (i.e. did not need to be treated) at 3 months. This rate was used to obtain the monthly probability of becoming inactive when active and under treatment. 62 At month 6, 54% of individuals were still inactive. The difference between the proportion of inactive individuals at months 3 and 6 was used to develop the probability of becoming active when inactive and under no treatment (Table 25). Probability distributions were developed using the 95% CI from the IVAN study survivor function using Crystal Ball software (release 11.1.2.0.00, 2010, Oracle Corporation, Redwood Shores, CA, USA) (see Table 25).

Second year data for the probability of becoming inactive were developed using data from the CATT study. The inclusion criteria and the treatment group for the CATT study were similar to that of the IVAN study. However, within the IVAN study three monthly injections were administered when participants failed the disease inactive criteria. The CATT study administered one injection only and reviewed participants in 1 month’s time before making a further treatment decision. A monthly probability was sought in order to obtain the CATT study mean number of injections within the as needed arm at 2 years. A beta distribution was attached based on mean value and 1/10 of the mean value as standard error (TreeAge software).

Second year data for the probability of becoming active when participants were inactive and under no treatment was developed using data reported by Horster et al. 70 The authors reviewed data on all patients receiving intravitreal ranibizumab injections for nAMD at the University of Cologne, Germany. Eyes with at least two recurrences (i.e. reappearance of intraretinal or subretinal fluid on OCT, and/or leakage on angiography) were selected. The mean follow-up time (months) and number of recurrences were 28.8 and 2.8 respectively.

A number of individuals that were inactive at 3 months within the monthly treatment group in the IVAN study9 failed the no retreatment criteria (e.g. subretinal fluid, increasing intraretinal fluid, or fresh blood) in subsequent months. This means that, even under monthly treatment, inactive individuals could become active again. Based on this, half the probability of becoming active when inactive and under no treatment was assumed for the probability of becoming active when inactive and under treatment.

Diagnosis or monitoring strategies could result in over- or undertreatment; therefore, it was believed important to include adverse events as a result of treatment. Two recent studies8,9 report systemic and ocular adverse event rates. It was not clear from inspection of these data that systemic adverse events could be due to treatment of nAMD. Therefore, only ocular adverse events were included in the model. Table 26 shows monthly estimates for the proportion of individuals that were under treatment that experienced cataract, endophthalmitis, glaucoma, retinal detachment and uveitis.

Costs

Table 27 shows cost estimates used in the model. Prices are expressed in 2011–12 pounds sterling (£). Strategy assessment costs were a combination of the cost of a visit (e.g. ophthalmologist, nurse or technician) and the cost of a particular test used for the assessment (e.g. FFA or OCT). For instance, the diagnosis cost for strategies where diagnosis was conducted using FFA only was calculated adding up the cost of an ophthalmologist visit and the cost for an FFA (e.g. £79.74 + £117.26 = £197.00). NHS reference costs were used for all but the ranibizumab unit costs in Table 27, for which British National Formulary (BNF) data were used (£742.17). 12 The unit cost for face-to-face consultant-led follow-up attendance that resulted in non-admission for the ophthalmology service was used for the cost of a diagnosis or monitoring visit to the ophthalmologist (£79.74). Likewise, non-consultant led was used for the cost of a nurse or technician monitoring visit (£58.53). Minor vitreous retinal procedures cost category [Healthcare Resource Group (HRG) BZ23Z code] was used to cost FFA (£117.26). Finally, after consultation with clinical experts, an ultrasound scan (HRG RA23Z Ultrasound scan, less than 20 minutes) was deemed more likely to reflect the cost of an OCT test (£51.27).

Gamma probability distributions were defined for unit cost data for probabilistic sensitivity analysis as these are defined non-negative and provide a possibility of a right tail that could account for few very high unit cost cases. Ranges for reference cost based data are also reported in Table 27; these are lower and upper quartiles. These were used to tailor cost probability distributions.

The cost of profound visual loss/blindness from the NHS and Personal Social Services perspective was calculated following Colquitt et al. 61 The authors used proportion for service utilisation developed by Meads and Hyde71 (Table 28). The unit costs reported by Colquitt et al. 61 were updated using Hospital and Community Health Service specific price inflation index (base 2005 = 100) for March 2012 (e.g. £121.85). Using an alternative weekly cost figure of £497 for residential care (the item in the list with higher unit cost) reported by Curtis,72 results in an annual cost of £556 and £537 for the first and subsequent years, respectively, and these were used as the basis for deterministic sensitivity analysis.

Utility weights

Guidelines for economic evaluation of health-care technologies in the UK advocate the use of a preference-based measure of utility. 73 We conducted a focused search for these data for AMD individuals. It was confirmed that one group had the majority of studies in this area74,75 and data from Brown et al. 74 were included in the economic model. The study by Brown et al. 75 used the time trade-off approach on 72 consecutive patients seen at the Retina Vascular Unit at Wills Eye Hospital, Philadelphia, USA, to obtain utility weights for alternative VA scores. Table 29 presents utility weights used in the economic model according to the Markov model health state. CIs were also obtained from Brown et al. 74 Mean utility weights and CIs were used to define beta distributions (see Table 29) for probabilistic sensitivity analysis.

Utility decrements due to adverse events were retrieved from Brown et al. 75 The authors derived utility values from 233 patients with AMD and decrement values were obtained from individuals who experienced alternative adverse events. Table 29 shows the (monthly) utility decrements used within the model. These were applied to the proportion of individuals who experienced an adverse event from within those that were under treatment (see Table 26). Searches were conducted to retrieve information on the effect of treatment injections on the quality of life of patients with nAMD; however, no evidence was found. Moreover, from discussions within the project advisory group and clinical experts, anxiety seemed to be associated with the uncertainty of the disease condition (i.e. active or inactive) rather than the treatment itself. Adding a utility decrement for each monthly monitoring visit for all strategies would have had no effect on the final results. As such, no utility adjustments were conducted due to treatment injections.

Base-case and sensitivity analyses

The UK NICE guidelines of methods for technology appraisals were followed. 73 The model base-case analysis was run for a cohort of 65-year-old men for a time horizon of 35 years (lifetime). A 1-month cycle length was defined. The analysis was conducted from the NHS and Personal Social Services perspective. Costs were expressed in 2011–12 pounds sterling and effectiveness in quality-adjusted life-years (QALYs). Costs and QALYs were discounted at 3.5%. 73 Cost-effectiveness analysis results are reported using incremental cost-effectiveness ratios (ICERs). 55 ICERs are calculated as the ratio between the difference in average cost between two alternative strategies and the difference in average QALYs. This ratio measures the additional cost that would have to be paid in order to obtain an extra unit of effectiveness (i.e. an extra QALY). Probabilistic analysis results are reported using cost-effectiveness acceptability curves (CEACs). 76,77 CEACs show the probability of a particular strategy to be cost-effective at alternative values of willingness to pay for an extra QALY.

Results

Table 31 reports base-case analysis results for men for the nine compared strategies. Model strategies are ordered in terms of average cost in an ascending order. Diagnosis with FFA combined with the nurse or technician-led monitoring strategy (e.g. nurse or technician as first monitoring contact conducting a VA examination and interpreting OCT test results; if negative, discharge, if positive or unclear, refer to an ophthalmologist for further assessment) was the strategy with the lowest average total cost. The next non-dominated strategy (i.e. dominated strategy meaning a strategy with higher expected costs and lower expected QALYs) is diagnosis based on FFA only, followed by ophthalmologist-led monitoring. This strategy has higher total expected cost but also produces higher total expected QALYs. However, the incremental cost for an extra QALY (i.e. ICER) to adopt this strategy is above the often accepted cost-effectiveness threshold (i.e. £30,000). 73 All other strategies are dominated by either of the strategies that based diagnosis in FFA only followed by nurse-led or ophthalmologist-led monitoring. Diagnosis based only on OCT appears in third place combined with nurse-led monitoring. In terms of costs, the strategies’ order is driven mainly by the monitoring pathway, with the lowest average total costs coming from the nurse-led monitoring pathway (first to third places), then the ophthalmologist-led (fourth to sixth) and OCT only-based (seventh to ninth) monitoring pathways respectively. It should be noted, then, that the three model strategies that used OCT only as the basis for monitoring criteria were the strategies with higher average costs (see Table 31). This is due to the cost of treatment, that represents 76% of the total average cost within these strategies, the highest proportion for all compared strategies (e.g. average 65% and minimum 55%).

Figure 15 shows the cost-effectiveness plane for the base-case analysis and the nine diagnosis–monitoring combination strategies. For easier interpretation, data marker shapes relate to the diagnosis strategy and marker filling/colour relates to the monitoring strategy. Namely, square, circle and triangle shapes are used for FFA only, OCT only, and ophthalmologist stepwise diagnosis respectively. In addition, blue, green and none marker fillings correspond to ophthalmologist-led, nurse- or technician-led and OCT only-based monitoring respectively.

FIGURE 15.

Base-case cost-effectiveness results: men.

Three clusters can be seen in Figure 15 according to the monitoring strategy. As such, the ophthalmologist-led monitoring strategy cluster seems to produce higher expected QALYs and slightly higher expected costs than the nurse- or technician-led monitoring strategy. The OCT only monitoring strategy cluster results in a higher expected cost and lower expected QALYs than the other two monitoring strategies.

Within each of these clusters, the FFA diagnosis strategy dominates OCT only as well as the ophthalmologist stepwise diagnosis strategy (e.g. VA, OCT and SLB in all, followed by FFA if positive or unclear results). Also, to note is that the ophthalmologist diagnostic and FFA diagnostic pathways have very similar expected cost and QALYs within each cluster and, as such, data markers seem to overlap. This is due to the close values assumed for diagnosis sensitivity and specificity in these two diagnostic pathways.

Table 32 and Figure 16 show probabilistic sensitivity analysis for the base case. Diagnosing with FFA only followed by nurse- or technician-led monitoring has the highest probability of being cost-effective for up to £40,000 willingness to pay for an extra QALY. At higher threshold values (e.g. £50,000) diagnosing with FFA only followed by ophthalmologist-based monitoring has a higher probability of being cost-effective. Overall, diagnosis with FFA with either nurse- or ophthalmologist-led monitoring has more than a 70% chance of being cost-effective at willingness-to-pay values for an extra QALY of between £10,000 and £50,000. These strategies lose some ground against ophthalmologist-based diagnosis (e.g. ‘Ophthalmologist & Ophthalmologist’ and ‘Ophthalmologist & Nurse’) at high levels of willingness to pay for extra QALY threshold values (see Table 32 and Figure 16). At £30,000 willingness to pay for a QALY threshold value and regardless of the diagnosis pathways (e.g. FFA only, OCT only or ophthalmologist), nurse- or technician-led monitoring has a 61% probability of being cost-effective.

FIGURE 16.

Base case cost-effectiveness acceptability curves: men.

Figure 16 shows that when expanding this range up to £100,000, diagnosing with FFA only followed by the ophthalmologist-based monitoring strategy will have more than a 50% chance of being cost-effective. In addition, FFA only-based diagnosis strategies lose some ground against ophthalmologist-based diagnosis strategies (i.e. ‘Ophthalmologist & Ophthalmologist’ and ‘Ophthalmologist & Nurse’) at high levels of willingness to pay threshold values (see Table 32 and Figure 16).

Sensitivity analysis

Using mortality rate data for women

Table 33 and Figure 17 present cost-effectiveness results for women. As expected, all strategies produce more QALYs, incurring higher average costs. This is because of the longer life expectancy for women. This affects all of the model strategies in a similar manner. As such, there are no differences in the (average cost) order of the strategies or the general results compared with those for the base-case analysis for men (see Table 31). Diagnosing with FFA followed by nurse- or technician-led monitoring is still the strategy with the lowest average cost and dominates all other compared strategies, apart from diagnosis with FFA followed by ophthalmologist-led monitoring. However, the ICER for moving to the latter strategy is above the usually accepted cost-effectiveness threshold (i.e. £30,000). 73

TABLE 33 - Cost-effectiveness results: women

Strategy	Cost (£)	Incremental cost (£)	QALYs	Incremental QALYs	ICER (£)
(3) FFA & Nurse	44,099	0	11.604	0.000	0
(9) Ophthalmologist & Nurse	44,119	21	11.603	–0.001	–30,521
(6) OCT & Nurse	46,125	2026	11.595	–0.009	–226,433
(2) FFA & Ophthalmologist	49,527	5428	11.725	0.121	44,959
(8) Ophthalmologist & Ophthalmologist	49,547	20	11.724	–0.001	–28,491
(5) OCT & Ophthalmologist	52,262	2735	11.715	–0.009	–296,276
(1) FFA & OCT	69,712	20,185	11.576	–0.148	–136,016
(7) Ophthalmologist & OCT	69,731	20,204	11.576	–0.149	–135,517
(4) OCT & OCT	74,847	25,321	11.568	–0.157	–161,433

FIGURE 17.

Cost-effectiveness results: women.

Similar clusters can be observed in the cost-effectiveness results for men (see Figure 15) and women (see Figure 17), with the three clusters depending on the monitoring care pathway (i.e. OCT only, nurse-, technician-, or ophthalmologist-led monitoring). As was the case with Figure 15, the ophthalmologist diagnostic and FFA diagnostic pathways have very similar expected cost and QALYs within each cluster and, as such, data markers seem to overlap. The Table 33 and Figure 17 results indicate that no dramatic differences can be expected for the women and men model run results. Therefore, further sensitivity analyses were conducted only for the male cohort.

One-way sensitivity analyses

Extensive one-way sensitivity analyses were undertaken. This section reports a selected number of these, with full results presented in Appendix 8. All one-way sensitivity analyses show results moving in the expected direction (i.e. lower sensitivity or specificity for OCT would result in OCT-based strategies being less cost-effective). Tables 34–38 show one-way sensitivity analysis for OCT diagnostic sensitivity and specificity, OCT monitoring sensitivity and specificity and OCT unit cost respectively. The base-case analysis results seem robust. In all reported sensitivity analyses, diagnosis with FFA combined with nurse- or technician-led monitoring (based on VA and OCT with a referral to the ophthalmologist if positive or unclear) has the lowest total expected costs and dominates all others, apart from FFA for diagnosis with ophthalmologist-led monitoring. In a limited number of model runs, alternative strategies stop being dominated by diagnosis with FFA followed by nurse- or technician-led monitoring. However, in many of these cases the variable values used to run the analysis were extreme (see Tables 34 and 36 for OCT diagnostic and monitoring sensitivities equal to 1 respectively). Results are sensitive to the value of monitoring specificity for OCT. Table 37 suggests that OCT monitoring specificity above 80% could make diagnosis with FFA combined with monitoring with OCT only, a cost-effective strategy. However, this is to almost double the specificity values reported for monitoring in Chapter 4.

Summary and discussion

This chapter reported on a systematic review of economic evaluations and a model-based economic evaluation of alternative strategies for the diagnosis and monitoring of individuals with nAMD. No studies identified in the literature met the inclusion criteria for the systematic review.

Nine strategies (combinations of three different diagnostic and monitoring pathways) were considered within the economic model. The strategies used OCT for diagnosis and/or monitoring of nAMD individuals to a different extent. Extensive deterministic and probabilistic sensitivity analyses were conducted. The strategy that based its diagnosis decision on the results of FFA only, combined with VA and OCT interpreted together by a nurse or technician as the first monitoring step, with a referral to an ophthalmologist if the first monitoring assessment was positive or unclear (‘FFA & Nurse’), had the lowest expected total cost. This strategy dominated (i.e. lower expected costs and higher expected QALYs) all others apart from one: diagnosis with FFA only, combined with monitoring by an ophthalmologist (‘FFA & Ophthalmologist’). The ‘FFA & Nurse’ and ‘FFA & Ophthalmologist’ strategies had, respectively, a 46.5% and 29.8% probability of being cost-effective at the £30,000 threshold value of willingness to pay for an extra QALY. In addition, the ‘FFA & Nurse’ strategy dominated all others in the great majority of sensitivity analyses.

The strategies that used OCT only for their monitoring decisions were in almost every model run ordered last in terms of total expected cost and were often dominated by others. The strategy that used OCT only for both diagnosis and monitoring decisions was in almost every model run, the most costly strategy.

Scenario analysis was conducted in order to explore the conditions under which an OCT only strategy would become cost-effective. Three scenarios were developed using the best test performance data for OCT combined with a lower cost for OCT (£20.90) and a higher cost for FFA (£137). Scenario 2 added to this a lower unit cost for treatment (e.g. equivalent to the cost of bevacizumab, £50, instead of the £742 cost of ranibizumab considered for the base-case analysis). This scenario showed the ophthalmologist stepwise pathway for diagnosis combined with OCT only for monitoring, to be, on average, the least costly strategy. Alternative strategies were either dominated (i.e. more costly and produced fewer QALYs) or the resulted ICER was well above the usual threshold accepted for policy decisions. 73 This was an expected result. The low OCT specificity for monitoring in these scenarios and in the base case (0.61 and 0.44, respectively) meant that a high number of positive results would actually be FPs. The lower cost of treating individuals who do not need to be treated reduced the model penalisation for the OCT only-based strategies and therefore improved their cost-effectiveness.

Best practice guidelines were followed for this model-based economic evaluation exercise. 73,81 In spite of this, these results should be interpreted with caution. A considerable effort was made to retrieve the best available test or assessment performance data by conducting a systematic review of the literature. Other data were obtained from focused but reproducible searches. Nevertheless, there is an inherent problem with model-based economic evaluations that incorporate evidence from several sources, even when these data have been obtained systematically. The limitations of the SD-OCT performance data incorporated into the economic model have been mentioned in Chapter 4, with no SD-OCT studies contributing to the diagnosis performance data and only two SD-OCT studies23,30 contributing to the monitoring performance data. Moreover, although OCT diagnosis and monitoring performance data were retrieved from a systematic review of the literature, no such data were available for the strategies involving diagnosis or monitoring assessment by an ophthalmologist or monitoring assessment by a nurse or technician. Therefore, these data for the model were obtained from expert opinion. This constitutes a major caveat of the analysis and further research in this area is needed.

This economic model needed to consider individuals’ disease status (i.e. active or inactive nAMD) as well as test results on a monthly basis. In addition, these had to be combined with alternative VA states in order to incorporate utility weights into the model. It was felt that considering the effect of a fellow eye status (VA and nAMD status) would add major complexity to the model without a great deal of benefit from such incorporation. This is the most common approach used among economic models in this health area but constitutes a limitation of the current study. Utility weights used were obtained from nAMD individuals and grouped according to VA in the better-seeing eye. It is believed that this would better reflect individuals’ health status. However, the clear limitation of ‘one eye models’ is the underestimation of resources used. A proportion of monitored nAMD individuals will have this condition in both eyes instead of one, and, had the disease been active, would be receiving treatment injections in each eye. Intuitively, this would increase the treatment cost for those strategies with a higher number of TP and FP results (i.e. higher sensitivity and lower specificity) and hence would be unlikely to modify the overall conclusions of this economic evaluation.

The model did not consider effects on utility due to treatment injections. Anxiety in nAMD individuals was believed to occur at each monitoring visit mainly due to the uncertainty of the underlying condition (i.e. whether or not nAMD was active) and not the effects of the treatment injections. No evidence was obtained on this issue in spite of focused searches. Further research in this area is needed. Utility weight decrements from treatment adverse effects were included and this might partially overcome the above-mentioned potential limitation.

Limited evidence was available on the probability of nAMD active individuals becoming inactive when under treatment or inactive nAMD individuals becoming active. Data were retrieved from the literature and also from a UK-based RCT. 9 Survival data were received from the IVAN study (Dr Chris Rogers, personal communication) on first retreatment failure criteria (i.e. inactive individuals who needed to be retreated). These were used to develop model parameter values for the first year of the model run. There were no such data available for further failures and we had to rely on the available limited data from the literature70 or on expert opinion for year 2 onwards. In addition, progression data on VA were based on the 2-year follow-up MARINA study. 69 All of these were relatively short-term follow-up studies (around 2 years) but used to inform model parameters for a lifetime time horizon. These are clear limitations of the model and therefore its results should be interpreted with caution. Further research investigating individuals’ nAMD active/inactive status (e.g. probability of disease changing from inactive to active) would be desirable.

Conclusions

A strategy that based its diagnostic decision on the results of FFA only, combined with VA and OCT interpreted together by a nurse or technician as a first monitoring step, with a referral to an ophthalmologist if this first monitoring assessment was positive or unclear, had the lowest expected total cost. This strategy had a 46.5% probability of being cost-effective at a £30,000 threshold value of willingness to pay for an extra QALY. In addition, this strategy dominated all others apart from one (i.e. diagnosis with FFA combined with ophthalmologist-led monitoring) in the great majority of sensitivity analyses. Strategies that used OCT test results alone to make diagnosis or monitoring treatment decisions were unlikely to be a cost-effective use of resources. This result seemed to be driven by the OCT low specificity that resulted in a high number of FP results. The present analysis indicated that a further refinement of monitoring (i.e. a further monitoring step other than OCT alone) seemed desirable.

These results should be interpreted with caution. The economic model would benefit from further research to better inform a number of model parameter values. Studies that investigate the likelihood of nAMD individuals becoming active or inactive after subsequent treatments are desirable. In addition, a preference-based health status and process of care valuation study to explore the effects of treatment injections on individuals’ utility weights is needed. Finally, a comparative study to establish the performance of the ophthalmologist-based strategy compared with the nurse- or technician-based strategy for monitoring individuals with nAMD is required to inform future economic models in this area.

Estimating the numbers of patients with neovascular age-related macular degeneration

A summary of the epidemiology of nAMD has been described in this study. In brief, the prevalence and incidence of nAMD and the consequent burden to the NHS will increase over the next few decades because of the ageing population. By 2060, mean life expectancy will grow by 8.5 years for men (to 84.5 years) and 6.9 years for women (to 89.0 years). 82

Implications for service provision

The clinical workload associated with the frequent follow-up required for patients with nAMD is substantial. As more new patients are diagnosed and the population continues to age, the patient population will continue to increase. It is thus vital that clinical services continue to adapt so that they can provide a fast and efficient service for patients with nAMD.

There are still challenges and questions about whether or not ophthalmology departments have sufficient capacity and the means to offer relevant testing and treatments within adequate time scales. Local diagnostic pathways require updating and assessment to ensure compliance with national guidelines (e.g. to detect recurrence of active disease in these subjects). Occasional local disruptions may occur if OCT equipment suffers technical failures.

In 2012, Amoaku et al. 56 published a document entitled ‘Action on AMD’ that was developed by eye health-care professionals and patient representatives with the intention of highlighting the urgent and continuing need for change within nAMD services. This document also provided examples of good practice and service development, including the possibility of involving other health professionals and using OCT in the community.

Considerations regarding the performance of optical coherence tomography for diagnosis and monitoring

At the diagnostic stage, OCT is currently used in addition to FFA to provide a baseline that will be used for comparisons during the monitoring stage.

For monitoring, OCT has virtually replaced FFA in most NHS units. 83 During follow-up, monitoring also includes VA testing. There is larger variability in the adoption of other tests and perceived need for FFA during follow-up. The replacement of FFA is probably due to the convenience of OCT (e.g. non-invasive, user friendly, quick, efficient). However, expert clinicians recognise the difficulty of interpreting FFA and OCT in patients with previously treated nAMD who often develop atrophic changes. The low specificity of OCT observed in this study would suggest that OCT alone should not be used for monitoring.

Another consideration is the evolving technology. For example, theoretically an increased sensitivity and specificity of new versions or novel technologies (SD-OCT) would lead to more patients being correctly diagnosed with active nAMD, and fewer wrongly diagnosed as having no active disease. This review did not find sufficient evidence on the performance of SD-OCT and it is unclear if it is superior to TD-OCT.

Regarding cost implications, there will be little cost implications for procuring and maintaining OCT equipment because most centres already use this technology. Although many units will already have access to the new SD-OCT equipment, other centres may have to upgrade the current TD-OCT (e.g. purchase or lease new SD-OCT equipment).

There may be a need for training ophthalmology staff to ensure adequate technical skills to interpret the OCT scans. There is a learning curve to interpreting OCT images, especially in relation to those patients who are being monitored after treatment. Adequate quality control and quality assurance programmes would be needed in order to maintain high standards of interpretation.

Statement of principal findings

Strengths and limitations of the assessment

In terms of strengths, a comprehensive literature search was undertaken and non-English-language studies were included. Risk of bias was assessed using a modified QUADAS-2 questionnaire, tailored to the needs of this review. A HSROC model was used for the analysis, which takes account of the trade-off between TPs/FPs and models between-study heterogeneity. 84 The evidence for diagnosis and monitoring was considered separately. In addition to the pooled estimates for all OCT, separate pooled estimates were undertaken for TD-OCT (monitoring studies). It was not possible to undertake separate pooled estimates for SD-OCT as no SD-OCT studies were included in the diagnosis meta-analysis, and, in the monitoring meta-analysis, there were insufficient data from the two SD-OCT studies to use HSROC methods.

There was a very limited amount of evidence available for evaluating the performance of SD-OCT, both for diagnosis (one study) and for surveillance monitoring of those previous diagnosed with nAMD. There was also limited evidence for the performance of TD-OCT for surveillance monitoring. Although this review considered a number of alternative tests, only a few of these were reported by studies that met our inclusion criteria. There was insufficient information to address the questions of (1) the clinical effectiveness of OCT compared with FFA; (2) the acceptability of the tests; and (3) the performance of other health professionals compared with ophthalmologists in interpreting OCT findings.

Uncertainties

Other relevant factors

Statement of principal findings

No studies met the inclusion criteria for the systematic review of economic evaluations as none compared diagnostic or monitoring strategies for individuals with nAMD.

Nine strategies that used to a different extent OCT for diagnosis and/or monitoring of nAMD individuals were considered within the Markov cohort economic evaluation model. The strategy that based its diagnosis decision on the results of FFA only, combined with VA and OCT interpreted together by a nurse or technician as the first monitoring step, with a referral to an ophthalmologist if the first monitoring assessment was positive or unclear (‘FFA & Nurse’), had the lowest expected total cost. This strategy dominated (i.e. lower expected costs and higher expected QALYs) all others apart from one: diagnosis with FFA only, combined with monitoring by an ophthalmologist (‘FFA & Ophthalmologist’). The ‘FFA & Nurse’ and ‘FFA & Ophthalmologist’ strategies had, respectively, a 46.5% and 29.8% probability of being cost-effective at the £30,000 threshold value of willingness to pay for an extra QALY. In addition, the ‘FFA & Nurse’ strategy dominated all others in the great majority of sensitivity analyses.

The strategies that used OCT only for their monitoring decisions were, in almost every model run, ordered last in terms of ascending total expected cost and were often dominated by others. The strategy that used OCT as its only criteria for diagnosis and monitoring decisions was in almost every model run the most costly strategy.

Results were sensitive to the unit cost of treatment injections. A scenario with a lower unit cost for treatment (e.g. £50, equivalent to the cost of bevacizumab, instead of £742 considered for the base-case analysis) resulted in the FFA only for diagnosis combined with OCT only for monitoring strategy having the lowest total expected cost. Alternative strategies were either dominated or had an ICER well above the usual threshold stated for cost-effectiveness (i.e. £30,000).

Strengths and limitations of the economic assessment

The major strength of the economic evaluation is that it attempted to use the best available evidence with the compared strategies developed from extensive discussions within the project team and advisory group. Best practice guidelines were followed for this economic evaluation exercise. 73 For instance, test performance data were obtained from the systematic review of the literature with other data retrieved from focused but reproducible searches. There is, however, an inherent problem with model-based economic evaluations that incorporate evidence from several sources, even when these data have been retrieved systematically.

The economic model needed to consider individuals’ disease status (i.e. active or inactive nAMD) as well as test results on a monthly basis. In addition, these had to be combined with alternative VA states in order to incorporate utility weights into the model. It was felt that considering the effect of fellow eye status (VA and nAMD status) would add major complexity to the model without much benefit from this incorporation. A clear limitation of the so-called ‘one eye models’ is the underestimation of resources used. A proportion of nAMD individuals will have this condition in both eyes instead of one eye and would need treatment injections in each eye should the disease be active. In the current model this would increase the cost for those strategies with higher numbers of FPs (i.e. lower specificity) and therefore would be unlikely to modify the general conclusions of this study. A ‘one eye model’ has also been adopted by other teams involved in economic evaluations in this health area. 61

The model did not consider effects on utility due to treatment injections. Anxiety in nAMD individuals was believed to occur at each monitoring visit mainly due to the uncertainty of the underlying condition (i.e. active or inactive nAMD) and not the effects of the treatment injections. No evidence was obtained on this from the utility weight searches. However, utility weight decrements from adverse effects as a result of the treatment were included and this might partially overcome the above-mentioned potential limitation. The model did not consider factors relating to patient experience of alternative monitoring schemes. As such, there was no consideration of the process of care on patient preferences and only the effect of VA and the adverse effects of treatment on individual utility were incorporated into the model.

Limited evidence was available on the probability of nAMD active individuals becoming inactive when under treatment or inactive nAMD individuals becoming active. Data were retrieved from the literature, from a UK-based RCT (Dr Chris Rogers, personal communication) and expert opinion. In addition, progression data on VA were based on the 2-year follow-up MARINA study. 69 All these data were based on short follow-up but in a number of cases extrapolated to a lifetime time horizon. These clear limitations of the analysis indicate that its results should be interpreted with caution. Further research looking at the individual’s nAMD active/inactive status is desirable. A conditional or a retrospective analysis of existing data sets would be helpful in order to obtain data to inform future economic models.

The analysis was conducted from the NHS and Personal Social Services perspective, incorporating cost of visual impairment that considered, for instance, cost for community care and residential care. The model, however, did not take into account the cost for patients or their carers. For instance, as this is likely to be an elderly population, someone might accompany the patient for their monitoring visits. These costs have not been considered in the model.

Uncertainties of the economic analysis

Undoubtedly, the limitations of the data together with the assembly of key data of varied quality are of most concern. No SD-OCT studies contributed to the diagnosis performance data and only two SD-OCT studies23,30 contributed to the monitoring performance data in the economic model. Moreover, although OCT diagnosis and monitoring sensitivity and specificity data were retrieved from a systematic review of the literature, no such data were available for other tests proposed in alternative diagnosis or monitoring pathways (e.g. examination by the ophthalmologist or the monitoring assessment by a nurse or technician). Therefore, data for the model were obtained from expert opinion. These constitute major limitations of the analysis and further research in these areas is needed.

EMBASE, Ovid MEDLINE(R) and Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations

Searched: 1988 to 2013 Week 12 (EMBASE), 1946 to March Week 2 2013 [Ovid MEDLINE(R)] and 25 March 2013 [Ovid MEDLINE(R)] (In-Process & Other Non-Indexed Citations).

Ovid multifile search. URL: https://shibboleth.ovid.com/.

Date of search: 25 March 2013.

Science Citation Index and Bioscience Information Services

Searched: 1995–22 March 2013 (Science Citation Index) and 1995–22 March 2013 (Bioscience Information Services).

ISI Web of Knowledge. URL: http://wok.mimas.ac.uk/.

Date of search: 22 March 2013.

The Cochrane Library

Searched: Cochrane Database of Systematic Reviews Issue 2 2013; Cochrane Central Register of Controlled Trials Issue 1 2013.

URL: www3.interscience.wiley.com/.

Date of search: 22 March 2013.

Health Technology Assessment database/Database of Abstracts of Reviews of Effects

Searched: inception until March 2013.

Centre for Reviews and Dissemination. URL: www.york.ac.uk/inst/crd/index.htm.

Date of search: March 2013.

Medion

Searched: inception until March 2013.

URL: www.mediondatabase.nl/.

Date of search: March 2013.

ClinicalTrials.gov

Searched: inception until March 2013.

URL: http://clinicaltrials.gov/ct/gui/c/r.

Date of search: October 2014.

International Clinical Trials Registry Platform

Searched: inception until March 2013.

World Health Organization. URL: www.who.int/ictrp/en/.

Date of search: October 2012.

Association for Research In Vision and Ophthalmology

Searched: 2009–12.

URL: www.iovs.org/search?arvomtgsearch=true.

Date of search: March 2013.

American Association of Ophthalmology

Searched: 2009–12.

URL: http://aao.scientificposters.com/.

Date of search: October 2012.

European Association for Vision and Eye Research

Searched: 2009–12.

URL: www.ever.be/.

Date of search: October 2012.

EVER 2009, September 30–3 October 2009 Portoroz, Slovenia.

EVER 2010, October 6–9, Crete, Greece.

EVER 2011, October 5–8, Crete, Greece.

EVER 2012, October 10–13, Nice, France.

EMBASE Ovid MEDLINE(R), Ovid MEDLINE(R) and In-Process & Other Non-Indexed Citations

Searched: 1988 to 2013 Week 12 [EMBASED Ovid MEDLINE(R)], 1946 to March Week 2 2013 [Ovid MEDLINE(R)] and 25 March 2013 (In-Process & Other Non-Indexed Citations).

Ovid multifile search. URL: https://shibboleth.ovid.com/.

Date of search: 25 March 2013.

Applied Social Science Index and Abstracts (1995–23 March 2013)

Searched: 1995–23 March 2013.

ProQuest. URL: http://search.proquest.com/assia/.

Date of search: 23 March 2013.

PsycINFO

Searched: 1995–26 March 2013.

EBSCOhost. URL: http://web.ebscohost.com/ehost/.

Date of search: 26 March 2013.

EMBASE, Ovid MEDLINE(R) and Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations

Searched: 1980–2012 week 45 (EMBASE), 1996–November week 2 2012 [Ovid MEDLINE(R)] and 14 November 2012 (Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations)

Ovid Multifile Search URL: https://shibboleth.ovid.com/

Date searched: November 2012.

Health Technology Assessment/NHS Economic Evaluation Databases

Searched: inception until October 2012.

Centre for Reviews & Dissemination. URL: http://nhscrd.york.ac.uk/welcome.htm

Date searched: October 2012.

Health Management Information Consortium

Searched: 1979 September 2012.

Ovid URL: https://shibboleth.ovid.com/

Date of search: November 2012.

Research Papers in Economics

Searched: inception until September 2012.

URL: http://repec.org/

Date of search: September 2012.

Association for Research In Vision and Ophthalmology

Searched: January 2009 to January 2012.

URL: www.iovs.org/search?arvomtgsearch=true.

Date searched: January 2012.

EMBASE, Ovid MEDLINE(R) and Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations

Searched: 1980–2012 week 45 (Embase), 1946–November week 2 2012 [Ovid MEDLINE(R)], 14 November 2012 [Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations].

Ovid Multifile Search URL: https://shibboleth.ovid.com/

Date searched: November 2012.

Alster 2005

Alster Y, Bressler NM, Bressler SB, Brimacombe JA, Crompton RM, Duh YJ, et al. Preferential hyperacuity perimeter (PreView PHP) for detecting choroidal neovascularization study. Ophthalmology 2005;112:1758–65.

Cachulo 2011

Cachulo L, Silva R, Fonseca P, Pires I, Carvajal-Gonzalez S, Bernardes R, et al. Early markers of choroidal neovascularization in the fellow eye of patients with unilateral exudative age-related macular degeneration. Ophthalmologica 2011;225:144–9.

Silva R, Cachulo ML, Fonseca P, Bernardes R, Nunes S, Vilhena N, et al. Age-related macular degeneration and risk factors for the development of choroidal neovascularisation in the fellow eye: a 3-year follow-up study. Ophthalmologica 2011;226:110–18. (Secondary to Cachulo 2011.)

Chen 2003

Chen S, Han M, Wang L. Indocyanine green angiography of exudative age-related macular degeneration. Chin Ophthalmol Res 2003;21:428–30.

Do 2012

Do DV, Gower EW, Cassard SD, Boyer D, Bressler NM, Bressler SB, et al. Detection of new-onset choroidal neovascularization using optical coherence tomography: the AMD DOC Study. Ophthalmology 2012;119:771–8.

Fujii 1996

Fujii C, Inobe K, Sugimoto Y, Sugimoto A, Takahashi Y, Akagi Y. Indocyanine green angiographic findings in eyes with age-related macular degeneration. Folia Ophthalmol Jpn 1996;47:300–5.

Gomi 2007

Gomi F, Sawa M, Mitarai K, Tsujikawa M, Tano Y. Angiographic lesion of polypoidal choroidal vasculopathy on indocyanine green and fluorescein angiography. Graefes Arch Clin Exp Ophthalmol 2007;245:1421–7.

Hughes 2005

Hughes EH, Khan J, Patel N, Kashani S, Chong NV. In vivo demonstration of the anatomic differences between classic and occult choroidal neovascularization using optical coherence tomography. Am J Ophthalmol 2005;139:344–6.

Khondkaryan 2009

Khondkaryan A, Keane PA, Liakopoulos S, Walsh AC, Sadda SR. Comparison of optical coherence tomography and fluorescein angiography for the classification of neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci 2009;50:E-abstract 5259.

Kim 2003

Kim SG, Lee SC, Seong YS, Kim SW, Kwon OW. Choroidal neovascularization characteristics and its size in optical coherence tomography. Yonsei Med J 2003;44:821–7.

Kozak 2008

Kozak I, Morrison VL, Clark TM, Bartsch DU, Lee BR, Falkenstein I, et al. Discrepancy between fluorescein angiography and optical coherence tomography in detection of macular disease. Retina 2008;28:538–44.

Krebs 2007

Krebs I, Binder S, Stolba U, Krepler K, Zeiler F, Glittenberg C. The value of optical coherence tomography in diagnosis and therapy of age-related macular degeneration. Spektrum der Augenheilkunde 2007;21:33–8.

Liakopoulos 2008

Liakopoulos S, Ongchin S, Bansal A, Msutta S, Walsh AC, Updike PG, et al. Quantitative optical coherence tomography findings in various subtypes of neovascular age-related macular degeneration. Invest Ophthalmol Vis Sci 2008;49:5048–54.

Loewenstein 2010

Loewenstein A, Ferencz JR, Lang Y, Yeshurun I, Pollack A, Siegal R, et al. Toward earlier detection of choroidal neovascularization secondary to age-related macular degeneration: multicenter evaluation of a preferential hyperacuity perimeter designed as a home device. Retina 2010;30:1058–64.

Padnick-Silver 2012

Padnick-Silver L, Weinberg AB, Lafranco FP, MacSai MS. Pilot study for the detection of early exudative age-related macular degeneration with optical coherence tomography. Retina 2012;32:1045–56.

Park 2010

Park SS, Truong SN, Zawadzki RJ, Alam S, Choi SS, Telander DG, et al. High-resolution Fourier-domain optical coherence tomography of choroidal neovascular membranes associated with age-related macular degeneration. Invest Ophthalmol Vis Sci 2010;51:4200–6.

Parravano 2012

Parravano M, Varano M, Virgili G. Integrated imaging approach in RAP diagnosis. Acta Ophthalmol 2012;90:abstract 4426.

Reichel 1995

Reichel E, Duker JS, Puliafito CA. Indocyanine green angiography and choroidal neovascularization obscured by hemorrhage. Ophthalmology 1995;102:1871–6.

Salinas-Alaman 2005

Salinas-Alaman A, Garcia-Layana A, Maldonado MJ, Sainz-Gomez C, Alvarez-Vidal A. Using optical coherence tomography to monitor photodynamic therapy in age related macular degeneration. Am J Ophthalmol 2005;140:23–8.

Sandhu 2005

Sandhu SS, Talks SJ. Correlation of optical coherence tomography, with or without additional colour fundus photography, with stereo fundus fluorescein angiography in diagnosing choroidal neovascular membranes. Br J Ophthalmol 2005;89:967–70.

Sulzbacher 2011

Sulzbacher F, Kiss C, Munk M, Deak G, Sacu S, Schmidt-Erfurth U. Diagnostic evaluation of type 2 (classic) choroidal neovascularization: optical coherence tomography, indocyanine green angiography, and fluorescein angiography. Am J Ophthalmol 2011;152:799–806e1.

Talks 2007

Talks J, Koshy Z, Chatzinikolas K. Use of optical coherence tomography, fluorescein angiography and indocyanine green angiography in a screening clinic for wet age-related macular degeneration. Br J Ophthalmol 2007;91:600–1.

Torron 2002

Torron FB, Perez O, Melcon SF, Ferrer N, Ruiz-Moreno O, Honrubia L. Dynamic angiography in age related macular degeneration. Arch Soc Esp Oftalmol 2002;77:353–9.

Torron FB, Melcon SF, Ferrer N, Ruiz M, Honrubia L. Indocyanine green angiography and subretinal neovascularization. Patterns in age related macular degeneration. Archiv Soc Esp Oftalmol 2001;76:221–8. (Secondary to Torron 2002.)

Eter 2005

Eter N, Spaide RF. Comparison of fluorescein angiography and optical coherence tomography for patients with choroidal neovascularization after photodynamic therapy. Retina 2005;25:691–6.

Giani 2011

Giani A, Luiselli C, Esmaili DD, Salvetti P, Cigada M, Miller JW, et al. Spectral-domain optical coherence tomography as an indicator of fluorescein angiography leakage from choroidal neovascularization. Invest Ophthalmol Vis Sci 2011;52:5579–86.

Henschel 2009

Henschel A, Spital G, Lommatzsch A, Pauleikhoff D. Optical coherence tomography in neovascular age related macular degeneration compared to fluorescein angiography and visual acuity. Eur J Ophthalmol 2009;19:831–5.

Khurana 2010

Khurana RN, Dupas B, Bressler NM. Agreement of time-domain and spectral-domain optical coherence tomography with fluorescein leakage from choroidal neovascularization. Ophthalmology 2010;117:1376–80.

Regillo 1998

Regillo CD, Blade KA, Custis PH, O’Connell SR. Evaluating persistent and recurrent choroidal neovascularization. The role of indocyanine green angiography. Ophthalmology 1998;105:1821–6.

Salinas-Alaman 2005

Salinas-Alaman A, Garcia-Layana A, Maldonado MJ, Sainz-Gomez C, Alvarez-Vidal A. Using optical coherence tomography to monitor photodynamic therapy in age related macular degeneration. Am J Ophthalmol 2005;140:23–8.

van de Moere 2006

van de Moere A, Sandhu SS, Talks SJ. Correlation of optical coherence tomography and fundus fluorescein angiography following photodynamic therapy for choroidal neovascular membranes. Br J Ophthalmol 2006;90:304–6.

van Velthoven 2006

van Velthoven ME, de Smet MD, Schlingemann RO, Magnani M, Verbraak FD. Added value of OCT in evaluating the presence of leakage in patients with age-related macular degeneration treated with PDT. Graefes Arch Clin Exp Ophthalmol 2006;244:1119–23.