Abstract
Purpose: Currently, no outcome measures are clinically validated and accepted as clinical endpoints by regulatory agencies for drug development in intermediate age-related macular degeneration (iAMD). The MACUSTAR Consortium, a public-private research group funded by the European Innovative Medicines Initiative intends to close this gap. Procedures: Development of study protocol and statistical analysis plan including predictive modelling of multimodal endpoints based on a review of the literature and expert consensus. Results: This observational study consists of a cross-sectional and a longitudinal part. Functional outcome measures assessed under low contrast and low luminance have the potential to detect progression of visual deficit within iAMD and to late AMD. Structural outcome measures will be multimodal and investigate topographical relationships with function. Current patient-reported outcome measures (PROMs) are not acceptable to regulators and may not capture the functional deficit specific to iAMD with needed precision, justifying development of novel PROMs for iAMD. The total sample size will be n = 750, consisting mainly of subjects with iAMD (n = 600). Conclusions: As clinical endpoints currently accepted by regulators cannot detect functional loss or patient-relevant impact in iAMD, we will clinically validate novel candidate endpoints for iAMD.
Introduction
Age-related macular degeneration (AMD) affects almost 30% of the older population and progresses slowly from early AMD to intermediate AMD (iAMD) and ultimately late-stage AMD with severe and frequently irreversible visual loss over a decade [1]. Population aging will lead to a considerable increase in AMD prevalence. Today, late-stage AMD is the leading cause of blindness among the elderly in industrialized countries and affects more than 2.5 million patients in the EU [2-4] resulting in direct annual costs of over 2 billion Euros [5].
In order to reduce the significant burden of late AMD, novel interventions that stop or delay progression from iAMD to late AMD will need to be developed [6]. This has been highlighted by the number of failed studies to prevent the progression of geographic atrophy (GA) with different approaches such as inhibiting complement and visual cycle modulators. Parallels with Alz-heimer’s disease have been noted, where it is felt that part of the failure of clinical trials (CTs) to date has been the onset of treatment too late in the disease process [7, 8]. For CT clinical endpoints validated and accepted by regulatory agencies, health technology assessment (HTA) bodies, and payers are needed. Currently, these do not exist for iAMD CTs. The currently accepted endpoints like high-contrast, high-luminance best-corrected visual acuity (BCVA), are not appropriately reflecting the visual deficit in iAMD or only after progression to late AMD, which may require an observation period of up to several years. Therefore, BCVA is not an adequate endpoint for iAMD CTs.
Still, there is good evidence indicating that patients with iAMD experience impairment of visual function, yet it is unknown to what extent this impacts the patients’ life or how it can reliably be measured and quantified [9-11]. It is also not well known which the specific risk factors in the population of iAMD patients are to identify those with more rapid progression to late stages of the disease.
Therefore, the successful development of iAMD interventions requires validated functional, morphological, and patient-reported endpoints for CTs which are clinically meaningful and accepted by regulatory agencies. In addition, functional decline in iAMD and specific risk factors for iAMD progression to late AMD need to be better characterized to inform and improve conduct of future iAMD CTs.
Against this background, the major objective of the MACUSTAR consortium is to develop novel clinical endpoints for CTs with a regulatory and patient access intention in the area of functional, structural, and patient-reported outcome measures in patients with iAMD.
For this objective, it is not sufficient to test different endpoint categories in isolation, e.g. optimizing an imaging parameter without providing functional data to demonstrate that progression of disease shown by images is also reflected in functional measures and/or measures of patient-reported health. Specifically, this holistic and integrative approach for prospectively looking for correlation of candidate endpoints in one validation study is one of the unique strengths of the MACUSTAR project.
Using this approach, MACUSTAR will also characterize visual impairment in iAMD and its progression, as well as identify risk factors for progression to late-stage AMD. Herein, we report the process of selection and optimization of candidate outcome measures, protocol development, and considerations for statistical analyses for the MACUSTAR clinical study which will deliver the above-mentioned clinical endpoints.
Materials and Methods
Consortium Structure
The MACUSTAR consortium (online suppl. material; for all online suppl. material, see www.karger.com/doi/10.1159/491402) was awarded funding by the European Union Innovative Medicines Initiative 2 (IMI2) based on the Call 7 Topic 4 (No. IMI2-2015-07-04): “Dry Age-Related Macular Degeneration: Development of Novel Clinical Endpoints for Clinical Trials with a Regulatory and Patient Access Intention.”
The call topic text used the term “dry” AMD, which only separates AMD with choroidal neovascularization (CNV) (wet AMD) form all other AMD forms (dry AMD). In the MACUSTAR project (and as also suggested in the call topic text), the Beckman classification of AMD will be used as this much better reflects the progressive nature of AMD [12]. In line with the description of the call topic text, the MACUSTAR project will focus on assessment of patients with iAMD and the progression of iAMD to late-stage AMD (comprising neovascular wet AMD and atrophic AMD or GA). Patients with no AMD, early AMD, and neovascular AMD will be included in cross-sectional assessments to demonstrate the degree of discrimination of iAMD and its progression to late-stage AMD by the candidate endpoints.
IMI2 is the world’s largest public private partnership in life sciences, with the EU providing funding which is matched by the European Federation of Pharmaceutical Industries and Associations (EFPIA) and all projects include academic and industrial partners (EFPIA members and associated companies). Researchers from academia and industry work actively and jointly on all parts of research project. Decision bodies of IMI2 consortia have a 1: 1 distribution of academic and industry votes.
To optimally address the research question, the MACUSTAR consortium is structured in five work packages (WPs; see online suppl. material). These include conduct of the clinical study (WP1 – Clinical Trial), functional outcome measures (WP2 – Functional Endpoints), patient-reported outcome measures (WP3 – Patient-reported Heath and Health economics), structural outcome measure/imaging (WP4 – Anatomical and Imaging Endpoints), and project coordination (WP5 – Project Management, Communication, and Dissemination).
Protocol Development
Based on a review of the available literature up to September 2016 on PubMed as well as expert consensus within the consortium (see list of members) and with regulators, candidate outcome measures for clinical validation as endpoints were identified. Outcome measures were chosen to be included into the MACUSTAR clinical study if there were data supporting their use in iAMD, sufficient evidence to support adequate quality of the measurement obtained, testing could be standardized across multiple sites, and test burden would be compatible with a multisite setting and repetition of measurements for a follow-up of typically 2–3 years.
Study design was chosen to allow for the most comprehensive evaluation of chosen outcome measures, including assessments of reliability, responsiveness to change over time, structure-function correlation, and patient relevance.
Interaction with Regulators and HTA Bodies
The MACUSTAR consortium targets validation and acceptance of candidate endpoints by regulatory and HTA bodies. As such, it is central to early on align the validation strategy. Specifically, the design and analysis approach for clinical data acquisition and generation are developed with external input by stakeholders. For the MACUSTAR clinical study, feedback was implemented from patient organizations such as Retina International. In addition, the consortium initiated an official scientific advice procedure on biomarker qualification from regulatory agencies and HTA bodies early on in 2016/2017 before the active funding phase of the project. The scientific advice procedure including EMA, US FDA, and NICE was completed in April 2017 with a letter of support to the project issued by EMA (www.ema.europa.eu).
Discussions and feedback on general approaches for biomarker qualification in iAMD, on how to best correlate the different endpoint categories, and on the clinical study design and statistical analyses were integrated into the subsequent protocol development, sample size calculations, and project planning.
Two further follow-on joint scientific advices with EMA, US FDA, and NICE are targeted for biomarker qualification with (1) a data package from the first cross-sectional part, and (2) a final data package from the longitudinal part of the MACUSTAR clinical study (see below) at the end of the funding period.
Terminology Clarifications
All measures developed in this project are referred to as outcome measures, in lieu of a clinical endpoint, which requires acceptance by regulatory authorities and payers.
Results
Study Design
The outcome measures will be assessed in both parts of the MACUSTAR clinical study, the cross-sectional and the longitudinal part (Fig. 1).
These parts are operationally combined under one study protocol for synergies as they include identical assessments, identical functional and imaging devices, identical clinical sites and investigators, and for iAMD a partially overlapping patient sample (Fig. 1).
The two parts of the study are:
(1)Cross-sectional part. It consists of a comparative analysis of functional, structural, and patient-reported outcomes in age-matched subjects with normal ocular aging changes only (no AMD), and in patients with early AMD, iAMD, and late-stage AMD. For this part, two visits will be analyzed; the baseline visit (day 1) and a short-term follow-up visit (21 ± 7 days). The total number of patients in the cross-sectional part is 300. The data generated will be used for a technical evaluation including reliability of tests as well as ability to discriminate between the different AMD stages and controls.
(2)Longitudinal part. A 3-year longitudinal follow-up of patients with iAMD to observe change in candidate endpoints over time and especially with progression from iAMD to late-stage AMD (GA or nAMD). The longitudinal part will comprise all iAMD patients from the cross-sectional part (n = 150) and additional 450 iAMD patients as well as the 50 early AMD patients (total n = 650). The data generated will be used to assess the predictive power of endpoint candidates for progression of vision impairment in iAMD and progression to late AMD.
Inclusion and allocation to study groups will be based on the disease stage of both eyes (Table 1). Late-stage AMD will be defined as follows: GA will be defined as a retinal area with a severely reduced signal by fundus autofluorescence and with a minimum area size of 0.1 mm2 as quantified by RegionFinder software, correlating to loss of the outer nuclear layer and the occurrence of choroidal signal enhancement on spectral-domain optical coherence tomography (SD-OCT) and spatially confined to either complete or incomplete depigmentation or other AMD typical changes (e.g., hyperpigmentation, crystalline deposits) on color fundus photography (CFP), provided any hemorrhages or exudates are absent. Finally, no leakage must be present on fluorescence angiogram (FA) (examination at the discretion of the investigator). CNV will be defined based on CFP, SD-OCT, and/or FA (optional). Evidence of CNV must include at least 2 of the following within a radius of 3,000 μm of the fovea: (1) serous detachment of the sensory retina; (2) subretinal/ retinal hemorrhage; (3) pigment epithelial detachment (excluding drusenoid pigment epithelial detachment); (4) fibrous tissue; (5) hard exudates; (6) disciform scar. Through the use of multimodal functional testing and imaging, subclinical stages of late AMD are likely to be identified during the study. As the MACUSTAR study (amongst others) aims at finding early markers of disease progression in patients with iAMD, results indicating possible progression will not replace the diagnostic criteria of late-stage AMD described above.
Note that the Beckman classification of AMD defines disease stage per the individual eye. In MACUSTAR, the same disease stage in both eyes of one subject is generally required. Two exceptions have been defined in the study protocol. First, small GA areas up to a maximum lesion size of 1.25 mm2 may be present in the fellow eye for the iAMD group, provided extrafoveal location is present. Second, no distinction between GA and neovascular AMD is applied in the late AMD group.
For each patient, one eye is selected as study eye. If both eyes are eligible for the study based on inclusion criteria (Table 2), the eye with better BCVA will be selected as the study eye. In cases in which both eyes have the same visual acuity, the study eye will be selected by the investigator. Before enrollment, morphological inclusion and exclusion criteria need to be confirmed by the central reading center.
As the examinations per visit are comprehensive, patient acceptance was assessed accurately beforehand. The complete visit duration was 3 to 4 h, which was not reported to be inacceptable by any patients.
The MACUSTAR clinical study will be carried out in 20 clinical sites (see online suppl. material) of the European Vision Clinical Research Network (EVICR.net), a dedicated network for multinational ophthalmology clinical research. These clinical sites will identify subjects with AMD as well as controls mainly from regular ophthalmologic consultations, inform them about the study and screen those subjects interested in participating.
Sample Size
The cross-sectional part of the MACUSTAR clinical study will recruit 50 normal, 50 early AMD, 50 late AMD, and 150 iAMD patients across the three age categories 55–64 years, 65–74 years, and ≥75 years (age limit 85 years at baseline). Although limited information is available on test-retest reliability and on the distributions of the outcome measures within the four disease stages (no disease, early, intermediate, and late AMD), these numbers are justified by the following considerations [13]: (a) If one considers a continuous outcome measure and assumes a moderately high effect size (Cohen’s f = 0.34) in one-way analysis of variance, 50 patients per disease stage will be sufficient to detect mean differences between the four groups with 80% power at the 5% alpha level of significance (overall F-test). In addition, the planned sample size will be sufficient to detect differences with 80% power at the Bonferroni-Holm-corrected 5% alpha level in post hoc tests comparing the mean outcome in the iAMD group to the respective means in the other patient groups. (b) The analysis of the test-retest reliability of continuous outcomes in the iAMD patient group will be based on intraclass correlation coefficients (ICCs). Assuming that there are two ratings per patient and that the minimum acceptable ICC is 0.75, based on available data for microperimetry, 150 iAMD patients will be sufficient to reject the null hypothesis “ICC ≤0.75” for true values of ICC > 0.85 with 80% power at the one-sided 2.5% alpha level [14, 15].
For the longitudinal part of the MACUSTAR clinical study, a total of 600 iAMD patients across the age categories mentioned above will be recruited and followed up 6-monthly over 3 years. Although currently available data on the change in structural, functional or patient-related outcome measures preceding progression from iAMD to late AMD are insufficient to accurately calculate a sample size, this number is justified by the following considerations: using available data on microperimetric changes in retinal sensitivity [10], which is one of the key candidate methodologies for functional biomarkers in iAMD, and assuming constant rates of change (dB/y), the average 3-year longitudinal changes in microperimetry were estimated to be –1.26 dB in eyes that show morphological AMD progression in terms of changes in drusen or pigmentation according to two experienced graders and –0.93 dB in eyes that do not. Due to a lack of data in the literature correlating pointwise sensitivity from microperimetry with progression of AMD, the outlined reference group was chosen as a substitute. The subjects progressing morphologically are assumed to have a noticeably increased risk for progression of the disease to late AMD. Based on data from a recent iAMD cohort that has been followed up for 3 years [16], approximately 23% of iAMD patients are expected to progress to late AMD within the 3-year study period. Assuming normally distributed change rates and a dropout rate of 20% after 3 years, an estimated number of n = 600 patients will be sufficient to detect a [(–1.26) – (–0.93)] dB = –0.33 dB difference in microperimetry change rates between possible progressors and nonprogressors (alpha = 5%). This estimation is based upon the very limited longitudinal data regarding mesopic microperimetry in individuals with AMD. At the time the study was planned, no relevant studies on scotopic microperimetry in AMD were available. Power will be > 80% if the standard deviations of the change rates are smaller than the absolute values of the mean change rates (i.e., |mean/SD| > 1, an assumption justified by the results by Wu et al. [10]). The sample size of 600 iAMD patients will be sufficient to develop a prediction model for progression to late AMD (combining 3-year changes in microperimetry, dark adaption, low luminance acuity [LLA] and other measures) and to estimate true- and false-positive rates at a high precision. For example, since the expected number of late AMD events will be approximately equal to 110 (assuming a dropout rate of 20% after 3 years and a progression rate of 23%), there will be approximately 25 events per variable in a four-variable prediction model in tenfold cross-validation and more than 20 events per variable in fivefold cross-validation [17]. Furthermore, assuming that the sensitivity and specificity rates of the prediction model range between 95 and 99%, a sample size of n = 600 patients will result in 95% confidence intervals for tenfold-cross-validated sensitivities and specificities that are narrower than 0.11 and that have lower bounds above 0.89 [18].
To avoid low data quality and missing data, the electronic case report form has built in data checks, avoids ambiguity, and has a user-friendly design. The central reading center, one of the project partners, the Association for Innovation and Biomedical Research on Light and Image (AIBILI), and clinical data monitors will perform ongoing data quality checks.
Outcome Measures
Based on available literature and expert consensus as outlined above, the following outcomes measures will be integrated as candidates for clinical validation in iAMD into the MACUSTAR clinical study, a natural history study in iAMD for validation of clinical endpoints.
Functional Outcome Measures
Patients with iAMD frequently report difficulties in dim illumination (e.g., reading menus in restaurants) and in low contrast situations (e.g., navigating staircases, driving at night, driving through tunnels, etc.) [11, 19, 20]. Against this background, a set of functional vision tests conducted at low luminance will be validated to reliably cover the functional impairment in iAMD. Also, the course of these functional outcomes during progression to late stage AMD will be assessed.
For all functional tests, we will assess functional changes meaningful to the participant – referred to as minimally important difference (MID). Combining functional test results with patient-related outcome (PRO) assessments will allow us to capture MIDs for all functional tests by systematically assessing whether a clinically assessed change in function translates to a patient-reported change in functioning, as well as patient preferences (i.e., utility).
The following tests of visual function will be performed in both parts of the MACUSTAR clinical study (cross-sectional and longitudinal) in addition to standard high-contrast BCVA.
(1) Fundus-controlled perimetry (microperimetry). Rubin and coworkers [21] developed the first scotopic microperimeter by adding neutral density filters to a commercial photopic instrument (Nidek MP1, Nidek Technologies, Italy) to reduce the luminance of background and test stimuli. In a recent, exploratory study of 20 patients with advanced AMD, they found scotopic microperimetry to be the best predictor of patient-reported difficulties under low light as measured by the Low Luminance Questionnaire (LLQ). In patients with iAMD, localized scotopic dysfunction is more pronounced compared to mesopic dysfunction in areas with reticular drusen, a risk factor for AMD progression [22]. Preparatory assessments showed that where early scotopic microperimeters had problems with testing accuracy, speed, and acceptability to patients, the scotopic MAIA (S-MAIA, Centervue, Italy) is easier to operate, with less operator-induced variability. Further, it appears to be less affected by ceiling and floor effects due to its extended range for both mesopic and scotopic testing [23, 24]. Therefore, the MACUSTAR consortium will use the S-MAIA microperimeter for this project. As persons with AMD are known to adapt to reduced light levels more slowly than people without the condition [25], preliminary data are currently collected to determine the optimal duration of dark adaptation for the mesopic and scotopic tests in this patient cohort. The aim is to ensure that all participants are at a steady state level of adaptation when the microperimetry tests are conducted.
(2) LLA. The Salisbury Eye Evaluation (SEE) is a simple test of vision at low luminance [26]. All that was required was a pair of dark sunshades which reduce luminance by 1.4 log units, i.e. a factor of 25. This moves the test into the mesopic range. Using a neutral density filter that reduced the luminance by 2 log units, Sunness et al.[27] showed that an LLA deficit at baseline was a good predictor of GA progression. Finger and colleagues[11] found that the LLA deficit predicts patient-reported difficulties with night vision in iAMD.
(3) Moorfields Acuity Test (MAT; Vanishing Optotypes). Shah et al.[28] have developed a new set of optotypes for measuring visual acuity that consist of pseudo high-pass letters. The authors have demonstrated that the new optotypes have better test-retest variability than traditional optotypes, and recent evidence suggests that they are more sensitive to vision loss in early AMD to iAMD [29].
(4) Dark adaptation. Impaired dark adaptation in AMD patients results in patient-reported difficulties with night vision, especially when moving from high to low ambient illumination [25]. In a study of 325 normal participants, abnormal dark adaptation at baseline was a predictor of AMD onset over 3 years [30]. Longitudinal data from a small number of eyes with different stages of AMD also suggest that disease progression is reflected in advancing delays in rod-mediated adaptation [9]. Traditional dark adaptation measurement is very lengthy as it is measured until absolute threshold is reached. A faster alternative is to monitor dark adaptation only until it reaches a criterion threshold. A recently developed and introduced commercial instrument, the AdaptDx (Maculogix, USA), will be used in the MACUSTAR clinical study to measure the so-called rod intercept time (time to reach a threshold located within the second component of rod recovery). Adaptation is assessed for up to 20 min at a single location just outside the macula [31]. The standard diagnostic protocol for the device is based on a pass/fail criterion, where an adaptation time falling outside of a specific window is considered to be “abnormal” [31]. For a longitudinal study, this is an insufficiently precise outcome. Thus, we will determine the optimal bleach intensity and retinal location for testing people with iAMD. The aim of this standardization process is to select testing parameters, which will allow an actual value for rod intercept time to be obtained for the majority of participants with iAMD within a 20-min recording period. In addition, the AdaptDx will be used to measure Absolute Rod Threshold, the minimum intensity of light detectable when fully dark adapted.
(5) Contrast sensitivity (CS). A number of small studies have evaluated the effect of AMD on CS to various spatial frequencies, and evidence suggests that a mid-high spatial frequency deficit is present [32]. However, there is a paucity of data evaluating the ability of CS to predict and monitor disease progression in individuals with iAMD. Given the strong relationship between reduced CS and self-reported problems performing activities of daily living [33], an assessment of CS has the potential to provide a clinically meaningful outcome measure, which is why a classic chart-based test of CS (Pelli Robson) will be included.
(6) Performance-based tests (PBTs): Reading Speed and Navigation Performance. PBTs are standardized tests of everyday activities. In MACUSTAR, we will use Reading Speed and Navigation Performance, both performed under low illumination, to help validate the outcome of functional assessments and PROs against activities of daily life (ADL). For the reading PBT, we will use the International Reading Speed Texts (IReST) developed by members of this consortium, translated and standardized for all major European languages [34].
Navigational PBTs will utilize two unique facilities – the Pedestrian Accessibility and Mobility Laboratory (PAMELA) in London and Streetscape in Paris, Fondation Voir et Entendre – FVE. Both facilities allow standardized illumination levels that can simulate walking through a residential street scene at night and are the only such facilities available in Europe. Conditions are comparable at both facilities, i.e. the same levels of illumination, etc. Navigational performance will be assessed in the cross-sectional part of the study.
Structural Outcomes Measures
To date, the state-of-the-art in retinal imaging of nonneovascular, atrophic late AMD progression is based on multimodal retinal imaging which combines several imaging approaches. These include:
CFP
SD-OCT more recently supplemented by OCT-angiography
Blue and green confocal scanning laser ophthalmoscopy (cSLO) fundus autofluorescence
Near-infrared cSLO reflectance imaging.
Image data acquisition will be carried out according to standard operational procedures, including electronic data transmission to the reading center and certification of study-site technicians and photographers. Trained and independent readers will independently and systematically analyze imaging data according to predefined grading parameter. This also includes final confirmation of development of late AMD in any eye during the study (for specific criteria see Table 1).
A large number of previous and current clinical studies assessing AMD progression still rely on CFP. However, this is not sufficiently sensitive or specific for structural assessment of the complex retinal abnormalities in iAMD. An optimized combination of these existing imaging technologies has not been determined and clinically validated but will be the basis for an in-depth assessment of structural abnormalities and their progression in iAMD.
As the latest technological advancements provide continuously growing number of imaging modalities, retinal imaging produces an increasing amount of data. These data provide unique high-resolution structural disease information and allow us to potentially measure an array of relevant imaging biomarkers. However, the current state-of-the-art in retinal imaging analysis uses visual inspection to extract these biomarkers. With stringent application of multimodal imaging approaches, there is a large volume of multimodal images to be analyzed. Humans are unable to efficiently exploit all the acquired information, and commonly only a limited number of imaging biomarkers are extracted. Moreover, accuracy and reliability of measurements vary within and between graders, leading to imprecise biomarker assessment and reduced prognostic power.
Automatic and semiautomatic computerized quantification of imaging biomarkers could substantially add to efficient exploitation of multimodal image information, obtaining faster, more accurate, and more reproducible measurements. Developed algorithms will be implemented in the central reading center for a validation against the current gold standard of manual grading. Currently available algorithms have several limitations. The most commonly available algorithms measure the thickness of the retina or sublayers within the retina, but these may be inaccurate in the presence of disease [35], and are not well correlated to functional deficits. Sanchez et al.[36] have been using deep learning approaches to improve the robustness of layer segmentations, which may help with both problems. Other features available commercially include measures of elevation of the RPE, which may be correlated to likelihood of progression [35]. It will be useful to see if these are also correlated to functional deficits. There are numerous other morphological features that are known to be associated with severity of disease or risk of progression such as reticular pseudodrusen, but these are not yet quantified by any commercial software. Sánchez et al. [36] and others [37-41] have previously designed computer-based algorithms for the automatic extraction of several morphological biomarkers based on one or several imaging modalities. Although promising, the performance of these algorithms needs to be further improved to detect small, characteristic changes in iAMD that can be correlated to function, patient reports, or risk of progression if they were to be utilized in future CTs. Also, the functional relevance of any of the structural candidate biomarkers needs to be validated against functional changes to support their clinical relevance for future clinical endpoints.
Patient-Reported Outcome Measures
As highlighted above, persons with iAMD often report difficulties with contrast and low luminance. We found that self-reported night vision symptoms are associated with low luminance deficit [11] and the degree of self-reported problems with night vision could be shown to be predictive of disease progression from early to late AMD (especially GA) as well as a loss of BCVA ≥3 lines over a period of 6 years [42, 43].
However, most currently available vision-related PRO measures (PROMs) are focused on assessing the impact of visual loss, and almost always have a ceiling effect (i.e., patients report no problems or tasks are too easy) in persons with good BCVA as is the case in iAMD. Thus, they are poorly targeted to the study population, and better targeted PROMs need to be developed. While there are instruments that measure different aspects of visual functioning and vision-related quality of life including subjective dysfunction in low luminance (e.g., Low Luminance Questionnaire, NEI Visual Functioning Questionnaire 25), no vision-related PRO has been accepted by regulatory authorities as a clinical endpoint since recent guidelines for development of such tools have not been systematically followed. To allow drug development, which will target the low-luminance and low-contrast visual impairment of patients with iAMD, a novel PRO needs to be developed in accordance with existing guidelines from regulatory authorities. Results from this new PRO will also need to be correlated with functional endpoints, and PBT to understand the relationship between measures of function and patient report.
In order to allow for economic assessments of vision treatments alongside iAMD CTs, a utility measure sensitive to changes in vision is needed. Conventional generic utility measures such as the EQ-5D have been shown to be inappropriate in vision [44]. The added value of a number of recent developments in the field including the EQ-5D vision “bolt-on” that provides an extra vision dimension [45] and Vision Function Questionnaire Utility Index (VFQ-UI) [46] remains unclear. Thus, in order to allow for optimal utility capture in iAMD CTs a health state utility index will be developed to allow utility values to be generated alongside the collection of PRO data to ensure that future CTs generate data on patient-relevant benefits that can be utilized in cost-effectiveness analyses by HTA and reimbursement authorities.
Statistical Analysis
Since the aim of the cross-sectional part of the clinical study is to evaluate the sensitivity, repeatability, and discriminant ability of the candidate endpoints described above, statistical analysis of these endpoints will be based on methods for measuring agreement and for analyzing differences between the disease groups. This will be done by computing intraclass correlation coefficients (ICCs) with 95% confidence intervals for continuous outcomes and Fleiss’ kappa coefficients with 95% confidence intervals for ordinal outcomes. Graphical analysis of agreement will be performed using scatterplots and Bland-Altman plots for continuous outcomes, and heat maps for ordinal outcomes.
In the longitudinal part of the study, multivariable regression and discrete time-to-event analyses will be conducted in order to assess the associations of changes in functional, structural and PRO measures with progression to late AMD. Both, progressions to nAMD and to GA will be considered as progression events. Based on these exploratory models, we will develop a predictive model using the rate of change in one or several functional measures, structural measures and/or PRO measures to predict progression to late-stage AMD. For this analysis, rates of change over time will be established for all functional, structural, and PRO measures, and “points of no return” will be assessed for structural measures. This will be done for (a) all progression events to late-stage AMD and for (b) nAMD and GA separately. Results will be compared to identify biomarkers suggested to be specific for progression to one or the other late-stage AMD type. For predictive modelling, we will consider logistic regression, but will also investigate the use of statistical learning methods such as random forests, artificial neural networks, and support vector machines. Predictive models will be derived and evaluated by repeated k-fold cross-validation [47]. Accuracy of the models will be evaluated on the test datasets using performance measures such as AUC and partial AUC [47]. Structural-functional correlations such as the correlation of localized retinal changes to quantitative results of fundus-controlled perimetry will be performed. Reporting of the various prediction models will follow the principles and guidelines of the TRIPOD Statement [48].
A combination of functional information from microperimetry measurements and structural information from multimodal retinal imaging may further improve the performance of these two individual biomarkers. These kinds of analyses have not been performed before in patients with iAMD. It is likely that clinically significant deterioration in functional measures may be partially masked by measurement imprecision resulting in limitation in the ability to identify disease progression by the functional measurement alone. In this situation, information on novel or progressive structural deficits from multimodal retinal imaging could improve the performance of the biomarkers. Analysis of structural parameter outcomes in the same model as outcome of the functional measures would mitigate measurement variability and increase the signal-to-noise ratio. An improved signal-to-noise ratio would allow disease deterioration to be identified, therefore increasing the power of a CT to identify treatment effects. Modelling of joint outcomes has shown potential in other diseases [49-51].
Simply put, each person would have a functional “chance” and a structural “chance” of changing; then, we combine these probabilities in a Bayesian framework. Crabb and colleagues[52] have successfully done this in people with ocular hypertension, when trying to predict progression to glaucoma, using measures from standard automated perimetry and neuroretinal rim area imaging. This particular prescription simply relies on combinations of single summary measures on rate of change from the two measurement modalities, combining measurements in “higher dimensions” retaining “layers” of information from each measurement – pixel/voxel intensity in images and single visual field points [53]. The same could easily be applied to regional changes from microperimetry and specific areas of change on imaging, taking advantage of modern statistical feature extraction methods [54].
Conclusions
Herein, we outline the protocol development of the MACUSTAR clinical study which will establish a toolbox of functional, structural, and patient-reported outcome measures for use in future iAMD CTs. Likely a combination of multiple outcome measures will be used to assess progression and may serve as a surrogate endpoint for progression to late AMD. As the functional deficit in iAMD is most pronounced under low luminance and low contrast conditions, this is what we will explore in MACUSTAR.
Strengths of our protocol development are integration of available data, comprehensive expertise including biostatisticians, as well as feedback from all relevant regulators and HTA bodies. Limitations are the lack of pilot data for the overall protocol at this stage, a possible selection bias (such as sampling bias due to the nonrandom sample out of subjects with iAMD, attrition bias), as well as the uncertainty around the sample size planning given the lack of necessary data around functional changes preceding progression from iAMD to late AMD. Considering this, our current sample size assessments are the best possible to date.
In conclusion, novel outcome measures are needed for iAMD CTs as conventional high-luminance, high-contrast functional testing, or existing PROMs are unlikely to detect functional loss or patient-relevant impact of the disease and thus cannot be used in future CTs for iAMD. A combination of candidate outcome measures assessing function under low-luminance and low-contrast conditions with structural outcomes which allow for a topographical correlation may improve precision of the measurement and should be explored.
Acknowledgments
We are very grateful for the continuous support provided by Nathalie Seigneuret at the IMI Office in Brussels. We also gratefully acknowledge the work put into this project by consortium members who have since moved on to work elsewhere.
Disclosure Statement
R.P. Finger: C Novartis, Bayer, Santen, Opthea, Novelion, RetinaImplant, Oxford Innovation; F Novartis.
S. Schmitz-Valckenberg: Alcon/Novartis C, F; Allergan F; Bayer F, R; Carl Zeiss Meditec F, R; Bioeq/Formycon F; Centervue F; Heidelberg Engineering F, R; Optos F; Genentech/Roche C, F, R.
F.G. Holz: C Acucela, Aoellis, Bayer, Boehringer-Ingelheim, Genentech/Roche, Heidelberg Engineering, Ionis, Novartis, Optos, Thea, Thrombogenics, Zeiss, F; Heidelberg Engineering, Zeiss; R Centervue, Heidelberg Engineering, Genentech/Roche, Nieghtstar.
D.P. Crabb: Allergan R; Roche R; Santen R; Centervue C.
J. Cunha-Vaz reports grants from Carl Zeiss Meditec, outside the submitted work and is consultant for Alimera Sciences, Allergan, Bayer, Gene Signal, Novartis, Pfizer, Precision Ocular Ltd., Roche, Sanofi-Aventis, Vifor Pharma, and Carl Zeiss Meditec.
F. Asmus is an employee of Bayer Pharma AG, Berlin, Germany.
M. Durbin is an employee of Carl Zeiss Meditec AG.
U.F.O. Luhmann is an employee of F. Hoffmann-La Roche Ltd.
P. Margaron, G. Normand, and Parisa Zamiri are employees of Novartis.
Funding Sources
This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No. 116076. This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation program and EFPIA and Carl Zeiss Meditec AG.
References
MACUSTAR Consortium Members: M. Berger, J. Bohner, M. Böttger, J.E. Brazier, T. Butt, C. Carapezzi, J. Carlton, M. Costa, A. Ferrao, C. Hoyng, F. Kersten, J. Krätzschmar, A. Lüning, C. Martinho, D. Rowen, J. Sahel, D. Sanches Fernandes, A. Skelly, C. Wojek.