Verteporfin Photodynamic Therapy for neovascular age-related macular degeneration: cohort study for the UK

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 03/36/01. The contractual start date was in December 2003. The draft report began editorial review in December 2010 and was accepted for publication in June 2011. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

The National Institute for Health Research awarded a grant to the London School of Hygiene and Tropical Medicine to carry out this research. The Queen’s University Belfast, Royal Liverpool and Broadgreen University Hospitals Trust were reimbursed for some of the time of Professor Chakravarthy (chief investigator) and SP Harding (principal investigator). Professor Reeves (principal investigator) was employed by the London School of Hygiene and Tropical Medicine until 2005 and part of his salary was paid by the grant. Part or all of the salaries of other authors employed by the London School of Hygiene and Tropical Medicine institution were also paid by the grant for all or part of the duration of the study. The Central Angiographic Research Facility and the three reading centres were funded by a separate grant from the NHS, which disbursed funding to the institutions of Professors Chakravarthy and Harding and Dr Peto. Professor Chakravarthy has received funding from Novartis for her membership of an advisory board on dry age-related macular degeneration. Her institution has received funding from Bausch and Lomb, Novartis, Pfizer, Oraya Therapeutics and Bayer for participation in studies on age-related macular degeneration. Professor Harding has received funding from Pfizer (unrestricted educational grant) and Novartis (conference attendance). His institution has received funding from Novartis for participation in studies on agerelated macular degeneration.

Copyright statement

© Queen’s Printer and Controller of HMSO 2012. This work was produced by Reeves et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2012 Queen’s Printer and Controller of HMSO

Background and rationale

Macular degeneration (MD) is the commonest cause of visual impairment in the developed world. 1,2 It mainly affects central vision, which underpins the ability to do tasks that require fine detail to be resolved, for example reading, watching television and recognising faces. The commonest cause of MD is ageing, although there are many other less frequent causes. Age-related MD (AMD) causes visual loss principally through the development of new vessels, which form in the choroid under the retina and leak fluid, bleed and eventually fibrose (choroidal neovascularisation, CNV). A less frequent site for the development of new vessels is in the retina itself, and this is termed retinal angiomatous proliferation (RAP). The process of neovascularisation is often referred to as ‘wet’ or neovascular AMD (nAMD) to distinguish it from a less aggressive or ‘dry’ form. CNV is identified and categorised using fundus fluorescein angiography (FA) into ‘classic’ and ‘occult’ forms based on patterns of fluorescence and location under the retina. RAP is identifiable using FA but more readily using indocyanine green angiography. At the onset of nAMD, a progressive fall in vision in the affected eye generally occurs over weeks and months and is more rapid with classic than occult CNV.

Verteporfin is a light-sensitive drug which is used in combination with an infrared laser to treat abnormal blood vessels proliferating under or within the macular retina in patients with nAMD and, less frequently, in some other eye conditions. The drug is injected intravenously and binds selectively to endothelial cells in the abnormal retinal vessels. The drug is activated by low-energy laser radiation, causing the abnormal vessels to regress. These two steps, that is initial infusion of verteporfin followed by laser exposure, are known as verteporfin photodynamic therapy (VPDT). Over time the vessels frequently reopen and so treatment usually needs to be repeated 3-monthly.

Verteporfin photodynamic therapy has been shown in randomised clinical trials to be better than sham treatment in maintaining sight in patients with nAMD. In the clinical trials of VPDT carried out for licensing, the Treatment of Age-related macular degeneration with Photodynamic therapy (TAP) trials and the Verteporfin in Photodynamic therapy (VIP) trials,3–5 the overall difference between treatment and placebo groups was small except in a subgroup of patients in whom ≥ 50% of the area of CNV was classified as classic by FA; in this subgroup, the benefit was larger. 4

There are no established criteria for stopping a course of treatment. The treatment protocol used in the TAP trials required 3-monthly reviews of FA for 2 years with retreatment if active CNV was observed. 3

The cost of a course of therapy has been estimated to be between £6000 and £8000. This relatively high cost arises largely because of the frequency of retreatment episodes in research cohorts. Cost-effectiveness was not estimated alongside the clinical trials used for licensing. In 2003, the National Institute for Clinical Excellence (NICE; now the National Institute for Health and Clinical Excellence) recommended treatment in the UK NHS for patients with nAMD ‘who have a confirmed diagnosis of classic with no occult subfoveal choroidal neovascularisation’. 6 However, NICE did not recommend VPDT for people with predominantly classic CNV ‘except as part of on-going or new clinical studies that are designed to generate robust and relevant outcome data, including data on optimum treatment regimens, long-term outcomes, quality of life and costs’ (Box 1). 6

Uncertainties about the effectiveness of VPDT were highlighted by the NICE technology appraisal. 6 Around the time of publication of the NICE guidance, the NHS R&D Health Technology Assessment programme, on behalf of the Department of Health, approached potential investigators about carrying out further research on VPDT to address these uncertainties. The overall aim was to characterise the cohort of patients referred for and treated with VPDT and to collect data about their visual acuity outcomes, quality of life and use of health and social care resources.

Initially, it was envisaged that this research should focus on the group of patients referred to in paragraph 1.2 of the NICE guidance (see Box 1), that is patients with ‘predominantly classic’ CNV lesions. However, discussions with the regional NHS commissioners and the Royal College of Ophthalmologists led to the scope of the research being expanded to include all patients treated with VPDT, irrespective of the subtype of the CNV. The VPDT cohort study was set up to meet these objectives. It built on surveillance programmes and research proposals active at the time and allowed patients with nAMD to access VPDT through the NHS.

The VPDT cohort study investigators aimed to address a number of questions which were relevant to NICE and to the NHS which were unanswered at the time of its guidance in 2003. These are set out in Chapter 2. However, the study was also of interest to other stakeholders, for example commissioners of NHS care. There is a need to develop robust methods of managing the introduction of new technologies into the NHS, especially when these are expensive, and the VPDT cohort study is one model by which this might be achieved. At the outset, there was the ambition that establishing a treatment register would allow a new technology (in this case, VPDT) to be introduced to a pre-specified service standard ensuring best possible care for patients, its use to be monitored effectively by commissioners and uncommon/rare adverse events (AEs) not identified in pre-licensing trials to be detected. A further benefit might be training clinical sites in research methods and processes to facilitate future clinical trials and research relevant to the NHS.

Visual acuity and health-related quality of life

A limitation of many randomised controlled trials (RCTs) of ophthalmological interventions is that the researchers conventionally choose best-corrected monocular distance visual acuity (BCVA) as the primary outcome. Reporting the effect of a new treatment as the average BCVA benefit relative to a control group allows ophthalmologists to consider the probable value of the new treatment compared with the best existing treatment (or alternative treatments) for the same condition. However, the limitations of clinical measures of outcome are now widely appreciated and many governmental and non-governmental organisations emphasise the importance of patient-reported outcomes or health-related quality of life (HRQoL) for measuring treatment effectiveness and health-care performance. 7,8 Moreover, the benefits of ophthalmic interventions are difficult to compare with other health-care interventions without being able to describe them in a common currency, for example quality-adjusted life-years (QALYs, see below and The health and social service costs of nAMD and associated treatments). 9

Health-related quality of life is a complex concept. A spectrum of instruments have been designed to measure HRQoL, from ones focused closely on functional performance to those assessing broader domains and rating the importance to an individual patient of a perceived loss of physical, social or emotional function. 10 This spectrum can be investigated specifically with respect to the condition affecting a respondent (condition-specific instruments) or to his or her wider life experience (generic instruments). The National Eye Institute Visual Function Questionnaire (NEIVFQ), which lies towards the functional performance end of the HRQoL spectrum, is perhaps the most widely used vision-specific HRQoL instrument. 11

A subset of generic HRQoL instruments explicitly recognises underlying preferences for different health states. These preference-based measures, such as the European Quality of Life-5 Dimensions (EQ-5D) and the Short Form questionnaire-6 Dimensions (SF-6D),12,13 report HRQoL on a scale with ‘anchors’ at 0 for death and 1 for perfect health. Preference-based measures of HRQoL are important because they can be combined with the relative effects of interventions on life expectancy to report QALYs. 9 QALYs allow comparison of interventions that may improve HRQoL but not life expectancy (such as many ophthalmic interventions) with interventions in other disease areas that can improve life expectancy but have little effect on HRQoL (e.g. statins to prevent coronary heart disease). Such comparisons, using HRQoL measures that take preference weights (i.e. societal values) or utilities from the general population,14,15 underpin health policy in many publicly funded health systems. They enable policy-makers to decide the relative worth of a new treatment in a wider context, that is compared with the value of health-care treatments for all other conditions that compete for funding from a finite budget.

Many studies have examined cross-sectional associations between visual acuity and HRQoL using a variety of HRQoL instruments including preference-based measures. 16,17 However, few have examined these associations longitudinally. Also, several studies that have reported preference-based measures of HRQoL have reported utilities elicited directly from patients18 rather than by the recommended process of taking these preferences from the general population. 14,15 Previous studies that have attempted to use preference-based measures of HRQoL to assess the gains from ophthalmic interventions for policy-making purposes have highlighted the deficiencies in existing studies. 19

The VPDT cohort study collected clinical measures of vision, measures of HRQoL and measures of resource use to achieve its principal aims of estimating the effectiveness and cost-effectiveness of VPDT in routine clinical practice. 20,21 Understanding the relationships between visual function and vision-specific and generic HRQoL was central to achieving these aims. 22

The health and social service costs of neovascular (wet) age-related macular degeneration and associated treatments

Neovascular AMD is potentially associated with high costs to health services and society. 23–25 Interventions for nAMD may improve HRQoL and reduce the costs associated with declining vision. 26 Cost-effectiveness analysis (CEA) is a powerful tool to evaluate and prioritise health-care interventions according to their relative effectiveness and cost. In many publicly funded health systems, policy-makers require CEA to assess whether or not a new intervention has sufficient gain to justify additional costs before recommending adoption. Decision-makers in predominantly privately funded health systems have recently shown interest in using CEA. 27

Previous CEAs of VPDT for nAMD have been contradictory. Some studies have reported that VPDT is ‘highly cost-effective’ and others that it is ‘definitely not cost-effective’. 26,28 For CEAs to provide a sound basis for decision-making, they must meet certain methodological standards. 15,29 The previous CEA of VPDT did not meet these standards on three grounds. 19 Firstly, intervention costs were based on treatment frequencies reported in the TAP trial, which are higher than those for routine practice. 3,20,30 Secondly, the HRQoL measures used took inappropriate preference weightings from patients with nAMD rather than generic measures such as the SF-6D that take health-state preferences from the general population. 19,31 Because CEAs are used to compare health gain across disease areas, the HRQoL measures used should weight different health states according to valuations taken from the general population rather than any specific patient group, for example patients with AMD. Thirdly, costing studies have reported that, compared with the general population, patients with AMD are more likely to use residential care and social services and to take antidepressants. 32,33 However, previous CEAs either ignored costs associated with vision loss or relied on expert opinion, rather than collecting appropriate patient-level costs. 19,26,32,33 The VPDT cohort study was commissioned to address some of these limitations.

Study design

The VPDT cohort study was designed as a longitudinal treatment register (case series) of all patients treated with VPDT in the UK. The study received research ethics committee approval (reference MREC/03/11/103).

The manual of operations, describing the study design and methods, including standard protocols for all measurements, was prepared before recruitment started and was updated periodically. The final version of this manual (version 2.1, December 2005) is included as Appendix 1.

Setting

When the VPDT cohort study was conceived, VPDT was not widely used in the NHS. Local Specialised Commissioning Groups (SCGs) recommended that VPDT should be provided only in designated ophthalmology departments in NHS hospitals. The Royal College of Ophthalmologists and clinicians involved with the VPDT cohort study took responsibility for providing a training programme for ophthalmologists and other health-care staff in hospitals that were newly implementing the provision of VPDT. Thus, the setting for the study was all designated hospitals (DHs) providing VPDT in the NHS.

After the majority of DHs had been identified and their personnel had been trained, the study team organised meetings of investigators. These meetings described progress with the study, discussed aspects of compliance with the study protocol, data submission and quality, and provided additional training. This training covered:

• for participating ophthalmologists, retreatment decision-making and interpretation of angiograms

• for independent grading for ophthalmic photographers and technicians, acquisition of angiograms and their subsequent submission to the Network of Ophthalmic Reading Centres in the UK (NetwORC UK, see Network of Ophthalmic Reading Centres in the UK)

• for nurses, optometrists and site co-ordinators, assessment of visual and other study procedures.

Participants

The manual of operations described the criteria for eligibility for treatment according to the NICE guidance:

• Best corrected visual acuity in the eye being considered for treatment must be equal to or better than Snellen 6/60, approximately equivalent to seeing one or more letters on the line corresponding to a logarithm of the minimum angle or resolution (logMAR) of 1.0, or > 30 letters when measured with an Early Treatment Diabetic Retinopathy Study (ETDRS) distance visual acuity chart (see Outcomes).

• Choroidal neovascularisation must be wholly or predominantly classic (i.e. ≥ 50% of the entire lesion must consist of classic CNV).

• Patients with subfoveal CNV due to nAMD or any other disorder are eligible for inclusion in the VPDT study.

All patients referred for assessment at a VPDT clinic in a designated treatment centre, whether eligible or not, made up the reference population. As part of the assessment, the ophthalmologist in charge of the patient made a decision on eligibility for treatment (above). There were no a priori exclusion criteria for people in the reference population. Participating hospitals were asked to submit a full set of data at the screening visit for all ineligible patients seen in person at the VPDT clinic, together with the FA used for decision-making, irrespective of whether the FA was carried out by the participating centre or by a referring hospital.

The study population consisted of all patients treated with VPDT at participating centres irrespective of CNV aetiology. (The decision whether or not to include all patients treated with VPDT, irrespective of aetiology, was made by the SCGs.) Participants were asked to give written informed consent for the collection of data and use of these data for the research.

Treatment with verteporfin photodynamic therapy

Participating centres were requested to classify CNV as had been done in previous RCTs3,5,34 in order to decide whether or not patients were eligible for treatment (Table 1).

Participating centres were also requested to review patients at 3-month intervals, carrying out ophthalmological and angiographic examinations to determine whether or not repeat therapy was needed. Two algorithms to guide retreatment decisions were included in the study manual (Figure 1 and Table 2). Investigators were also referred to the retreatment criteria developed by an international expert consensus group, the Verteporfin Round Table. 37

FIGURE 1.

Example of flow chart for making retreatment decisions – Belfast retreatment criteria. VA, visual activity.

Outcomes

The primary outcome was defined as BCVA, measured on a logMAR scale using the ETDRS distance visual acuity chart. 38

Secondary outcomes included:

1. safety, that is adverse reactions (ARs) and AEs

2. contrast sensitivity (CS) measured with the Pelli–Robson chart at 1 m39

3. generic HRQoL measured using the Short Form questionnaire-36 items (SF-36),40 from which SF-6D scores were also derived13

4. vision-specific HRQoL measured using the NEIVFQ11

5. independently graded morphological changes in treated lesions, that is total lesion size, total CNV leakage, classic leakage and fibrosis

6. health and social services (HSS) resource use measured using a custom-designed questionnaire administered to patients at the time of hospital visits for treatment or review.

Collecting data to characterise ARs and AEs was an important objective of the study because, at the outset, there was concern that such events experienced in licensing trials of VPDT may not be representative of the events observed in usual practice. The manual of operations specified that all ARs (during or just after treatment) or AEs (between treatment visits) should be recorded in the database. Any AR or AE considered to be serious and possibly, probably or definitely associated with treatment had to be reported to the Data Management Centre within 24 hours in accordance with good clinical practice in clinical research.

Other predictors of visual function

Data were collected at baseline to characterise study participants. The data included variables that were considered important for describing the study population and potential predictors of BCVA that might confound associations investigated in the analyses:

• age

• gender

• baseline BCVA and CS

• CNV composition, that is lesion area and proportion of the lesion graded as classic and occult CNV.

The collection of additional potential predictors of visual function was instituted after the study started to recruit (see Chapter 4, Collection of additional predictors of visual function).

Network of Ophthalmic Reading Centres in the UK

The baseline morphological characteristics of CNV lesions were defined in the protocol as potentially important covariates/confounding factors (see Other predictors of visual function) with respect to the primary outcome of visual acuity (see Outcomes). For example, classic compared with occult CNV is associated with more rapid loss of vision but greater responsiveness to treatment. 4 Changes in the morphological characteristics of CNV lesions over time, with or without treatment, were also defined as secondary outcomes (see Outcomes).

Considerable training is required to distinguish the various components of CNV lesions reliably, and, although the training put in place by the Royal College of Ophthalmologists and the research team included a session on lesion composition, this was not considered sufficient for research purposes. When designing the study, the research team was concerned that judgements about CNV composition made by the ophthalmologists treating patients might be unreliable and potentially biased. Unreliable judgements about CNV composition would have biased observed associations to the null. More importantly, given that the study was to some degree the means by which patients with predominantly classic CNV lesions could access treatment, there was a possibility that ophthalmologists might tend to overdiagnose the presence of classic CNV in order to classify patients as eligible.

The research context was also important. The TAP trials that were reporting at the time the study was being designed had taken great care to assure the quality of photographic images and the grading of these images. All photographers were accredited at the outset and reaccredited annually. All photographic images (colour fundus and FA) were submitted to a central reading centre for independent grading. 3 The VPDT cohort study research team wanted to establish the same standards to ensure that the results of the study were credible.

Therefore, NetwORC UK was established with the capacity to carry out independent grading of, potentially, > 5000 angiograms per year. Three geographically distinct centres, in Belfast, London and Liverpool, with facilities to grade stereoscopic fundus colour images and FAs were combined into a single network with a management facility in Belfast (Central Angiographic Resource Facility; CARF) to co-ordinate the administrative and technical issues. CARF managed the collection and archiving of images from designated VPDT treatment centres, performed consistency checks, certification and training of photographers, and transmitted images electronically to the three reading centres using a customised software platform. Regular training and concordance exercises were organised to ensure consistency between the reading centre grading staff and minimise grading protocol discordance.

The large volume of FAs to be graded precluded a double grading. Therefore, quality assurance was built in as an integral feature of the grading process. One in every eight FAs was randomly selected for regrading by the same reading centre, and 1 in 80 FAs was randomly selected for regrading by one of the other reading centres. All graders were masked to whether a particular grading was the original grading or a regrading.

Stereoscopic colour images and FAs were graded by the three reading centres that made up NetwORC UK using previously published definitions and protocols. 35,36 Grading involved the delineation and measurement of the area of classic and occult CNV and other lesion components contiguous to CNV, for example fibrosis and haemorrhage.

Data collection and management

Table 3 shows the schedule of data collection at follow-up visits. A decision was taken to collect HRQoL and HSS data in a subset of centres because of the workload involved and because not all primary care trusts (PCTs) paid the full tariff covering the costs of data collection. There was a strong desire to ensure that these data were collected in a representative population. Therefore, the research team reviewed fully funded sites and their geographic disposition and recommended to the Steering Committee that 18 centres collect these data. The geographic distribution of all sites is shown in Figure 2, distinguishing between the sites which collected only the clinical data set and the sites which also collected HRQoL and HSS data.

FIGURE 2.

Map of the UK showing sites which participated in the VPDT cohort study. Sites shown as open circles were selected to collect the extended data set, including contrast sensitivity, HRQoL and resource use.

To meet its objectives, the VPDT study had to collect data from all of the hospitals designated for providing VPDT identified at the outset. These hospitals had been selected to form a network of specialist retinal practitioners. When planning the study, we estimated that these hospitals would enrol about 7000 patients each year and that each patient would make, on average, three clinic visits per year. The study was planned to run for 3 years so that, in total, data would be collected for 21,000 patients who, between them, would make up to 168,000 clinic visits.

The study aimed to collect data for all patients who were referred to a participating VPDT clinic and who gave consent for their data to be collected. This population included those who were subsequently found to be ineligible for VPDT treatment, so that an accurate estimate could be made of the proportion of all referrals who were subsequently found to be eligible for treatment. This proportion, and the way in which it changed over time, was of interest because of uncertainty about the ability of referring practitioners to diagnose CNV lesions that were eligible for treatment. With experience and training, we expected the proportion of referrals found to be eligible for treatment to increase.

At each patient visit, data had to be collected on BCVA, any VPDT treatment given, ARs resulting from VPDT treatment given at the current visit and any AE possibly resulting from VPDT treatment given at the previous visit. Demographic and referral data were also collected at the patient’s first clinic visit.

Imaging of the fundus of the eye was carried out at each clinic visit, yielding colour images and FAs. Grading of these images was co-ordinated by CARF (see Network of Ophthalmic Reading Centres in the UK) and the resulting data then had to be transmitted electronically to the London School of Hygiene and Tropical Medicine (LSHTM) and linked with all other data.

Based on a previous postmarketing surveillance study sponsored by the manufacturer of verteporfin (Novartis), the cohort study adopted a strategy of requiring ophthalmologists, optometrists, nurses and administrative staff to collect and enter data in the course of providing treatment at each PDT clinic. It was felt that this approach would maximise data quality, reduce the burden of data collection upon a small number of individuals and avoid the need for additional staff to be appointed.

The information technology (IT) strategy for data collection required a robust, networked method of electronic data entry at each site. When the strategy was planned in 2003, web-based methods of longitudinal data collection were not easily available at a reasonable cost and this, combined with the complexity of the clinical data being recorded, meant that a database installed on the local network within each hospital was the preferred solution to allow data entry at multiple points of care. Participating sites were provided with a copy of a uniform Microsoft Access (Microsoft Corporation, Redmond, WA, USA) database developed by a third party. This database was a modification of the database used for the previous postmarketing surveillance study. Where possible, the database was installed on the hospital network. A minority of the initial 30 DHs had taken part in this study and were familiar with the database.

The strategy was based on electronic data entry taking place at each site, in real time, during the course of each patient visit. Thus, records clerks and reception staff would enter initial demographic data, optometrists the BCVA measurements and ophthalmologists the data relating to disease assessment and treatment. The use of paper case record forms would be avoided, and at the end of each session data would be transmitted electronically to the data management centre at the LSHTM. As the number of data entered at each clinic grew, so it was hoped that the data sources would become administrative tools in their own right, and be used routinely to access and review patient information. We expected regular review and appraisal of data to help to improve the accuracy of data collected for the study.

Data transmitted electronically to the LSHTM were placed into a secure central database and were ‘queried’ extensively with respect to data ranges and consistency. The data queries that resulted from these checks were e-mailed as a report to the relevant centres, which were asked to make corrections to the data held in their own local database or confirm that the original data were indeed correct. Corrections were part of the next data transmission, and queries which had been successfully resolved were labelled as such in the central database and removed from the subsequent data monitoring report.

The majority of data being collected was information that we expected to be collected in the course of usual care, although investigations were required to be carried out according to the manual of operations. Additional data were required about potential predictors of visual function outcome – possible ARs and AEs that might otherwise have been considered ‘expected occurrences’ (e.g. back pain). Additional data collection (CS, HRQoL and resource use) was carried out in the subset of centres.

Although the data collection strategy strove to avoid using paper case record forms, there were two elements of the study in which these were to be used. The first was the recording of visual acuity onto paper data sheets by optometrists at each clinic. This paper record was intended to act as a validation for the BCVA data (the primary outcome), which were also being collected electronically, and the paper forms were sent directly to the LSHTM for data entry and comparison with the electronic records. Paper forms were also used to collect HRQoL and resource use by the subset of 18 participating centres.

Risk of biases

Risk of bias is described below for the main bias domains identified in the Cochrane Handbook for Systematic Reviews of Interventions41 when reviewing primary studies of the effectiveness of interventions.

Sample size considerations

Because the UK Specialised Services Commissioning Group originally intended that treatment in the UK should be conditional on recruitment and participation in the study, the study population was defined in the manual of operations simply as the number of patients recruited during the study period. Uncertainties (e.g. about the proportion of patients likely to be referred, the proportion of referred patients found to be ineligible, the proportions of eligible patients categorised as having different CNV subtypes and the precise ways in which control data were to be modelled) made it difficult to provide in advance a clear sample size justification.

For illustrative purposes, we considered a simple comparison of a continuously scaled outcome, that is BCVA, between two subgroups of patients with different types of CNV lesions. 6 The following assumptions were made for this illustration: (a) equal sample sizes for the two groups; (b) analysis adjusted for baseline BCVA; (c) standard deviation (SD) of changes in BCVA = 0.1 logMAR; (d) two-tailed significance level of 0.01; and (e) power = 0.95. Such a comparison would require only about 50 subjects in each group to detect a difference of 0.1 logMAR in the mean change between groups.

We acknowledged that other outcomes might have a larger SD and that subgroups might not have equal sample sizes. For example, comparing a continuously scaled outcome with SD = 0.3 in two subgroups with sample sizes as unequal as 4 : 1 would require a total of about 1200 (960 : 240) subjects. These simple illustrations did not take into account the added precision from the longitudinal nature of the data but also did not consider dependencies between patients treated by the same medical retina teams.

Plan of analysis

The manual of operations recognised that the data set for the VPDT cohort study would have a complex structure, with varying numbers of visits/duration of follow-up across patients up to about eight visits/3 years of follow-up. It was also recognised that patients would be ‘nested’ within groups of retinal specialists and DHs. Therefore, we planned to analyse the data set by mixed regression with multilevel modelling, an extension of conventional regression methods to take into account statistical dependency between observations that are ‘clustered’ in the data structure, for example observations within patients or patients within retinal teams.

Follow-up of patients throughout the study period allowed repeated measurements of outcomes and changes over time to be described in detail. The main outcomes (BCVA, CS, HRQoL and lesion characteristics) were continuously scaled and could be analysed by mixed regression with multilevel modelling. We also planned to use similar models to quantify associations between clinical outcomes and HRQoL. The analysis plan did not provide details of additional analyses but envisaged that outcomes might also be analysed in different ways, for example by dichotomising the change in BCVA to describe a deterioration of ≥ 15 ETDRS letters or not (a deterioration expected to occur in about 50% of participants) or using survival analysis to describe the cumulative probability of a deterioration of this degree with increasing duration of follow-up.

The analysis plan stated that, because of the complexity of the data set and the likelihood that the composition of the cohort would influence the nature of the analysis, a detailed plan of analyses would be written after carrying out preliminary descriptive analyses. The preliminary descriptive analyses would characterise baseline clinical and treatment characteristics of patients recruited to the cohort but not involve any comparative analyses. A number of baseline factors were expected to influence outcomes independently following photodynamic therapy, including BCVA at presentation, CNV composition and fellow eye comorbidities, and the analysis plan specified that analyses should take all of these factors into account.

Protocol amendments submitted to the Research Ethics Committee

One formal protocol amendment was approved by the Research Ethics Committee during the course of the study.

An amendment was submitted for approval in September 2005 requesting approval for five changes:

1. to obtain an anonymised minimum data set for all patients considered for VPDT

2. to include presenting binocular visual acuity as part of the data set

3. to adopt a modified patient information sheet (PIS)

4. to allow nested RCTs as a secondary objective

5. to approve one such trial (comparing combined triamcinolone and VPDT vs VPDT only), for which a detailed protocol was submitted.

The request for an amendment was rejected because of the amendments describing nested RCTs. The amendment was resubmitted without items 4 and 5 in December 2005 and was finally approved in February 2006.

The reasons for seeking these amendments were as follows. Patients receiving VPDT in the NHS had to give informed consent for their data to be included. We wanted to describe the characteristics of all patients considered for VPDT, by eligibility for treatment and by willingness to take part in the study, so that we could comment on the representativeness of the study population. We wanted to collect presenting binocular visual acuity (i.e. binocular visual acuity with a patient’s habitual spectacle correction, rather than BCVA) because we reasoned that this was likely to be the visual function parameter most strongly associated with a patient’s self-reported vision-specific HRQoL. We requested approval to adopt a much simpler PIS because patients reported to us that the PIS initially approved (which included possible side effects of having VPDT) was too complex and discouraged participation. The proposed simpler PIS distinguished procedural consent for treatment (independent of the study) from research consent to use the data collected in the course of treatment.

These amendments did not alter the overall study design, the setting, the eligible study population or the outcomes.

Changes to study procedures relating to participating sites

As the study progressed, more hospitals were designated to provide VPDT. At the start of the study, because of the novel and complex nature of the treatment, there was professional concern to restrict the provision of VPDT to designated centres in which specified doctors and other personnel had been accredited by attending appropriate training, provided by the Royal College of Ophthalmologists. As time passed, this policy became more difficult to sustain because of the inconvenience to patients in sparsely populated geographic areas. Some hospitals, together with their commissioning PCTs which wanted the hospitals to provide VPDT, were unwilling to participate in the study. Unfortunately, despite the fact that policy-makers provided a lot of the impetus to carry out the study, there was no NHS directive requiring centres to participate. It also became clear at quite an early stage that participating sites were not complying with the follow-up schedule; this required the principal investigators to adapt the planned statistical analyses.

Collection of additional predictors of visual function

The data set being collected for the study was reviewed after 12 months. Because of evidence about the influence of other factors on BCVA, the Steering Committee accepted the recommendation of the principal investigators to ask centres to collect information on additional potential confounding factors:

• smoking status46–48

• use of statins49

• family history of nAMD50

• visual status of fellow eye. 51

These factors were likely not to vary over time for a participant, and we asked centres to collect the data at the next visit for participants who had already been recruited. At the time of this change to the data set, there was only a small minority of existing participants who had already been discharged from treatment or lost to follow-up.

Smoking status was classified as current smoker, ex-smoker or never smoked. The visual status of the fellow eye (worse or better BCVA) was assigned based on BCVA data collected across the duration of the study. If the better-seeing eye varied across visits, that is both eyes had similar BCVA, the fellow eye status was classified as uncertain.

Network of Ophthalmic Reading Centres in the UK

Images from the VPDT cohort study were graded from 2004 to 2008 by NetwORC UK. Regular training and concordance exercises within the three designated reading centres ensured the reproducibility and reliability of grading outputs. During this period, improvements occurred in image acquisition systems which led to an expansion in the knowledge of the different phenotypes of nAMD. The expanded phenotypic spectrum was incorporated into the grading vocabulary, and the grading protocols were also appropriately amended.

Data collection and management

Changes in the data collection strategy

Several factors were responsible for the study recruiting substantially fewer people than anticipated. However, one was the failure of the original data collection strategy to function in the manner intended. The strategy failed for four main reasons:

• The database supplied by the third party could not be adapted in a satisfactory way to the needs of the study.

• One particular aspect of the inadequacy of the database was the perceived lack of security of electronic data submission. The study coincided with considerable investment in IT modernisation in the NHS and greater awareness among IT managers of national guidance about the confidentiality of patient data held in electronic form. 52

• Some sites refused to install the local database on their local computer networks.

• Despite the investment in modernising IT in the NHS, many participating sites did not have reliable local computer networks to support data collection at the multiple points of care involved in the management of patients being treated with VPDT. Also, some sites provided VPDT in clinics that were remote from the main ophthalmology department.

There were two main consequences of the failure of the strategy. Data collection at most sites was carried out on paper forms, using forms developed and recommended by the co-ordinating centre (Figure 3) or custom forms developed by a site. Using paper forms often represented duplication of the recording of most of the clinical data and required an unexpected time commitment locally for entry of data into the database. There was also a general reluctance to use the adapted database and difficulties in submitting data at some sites.

FIGURE 3.

Paper data collection forms recommended by the data management centre.

Concerns about the third-party database became sufficiently grave that, during the summer of 2005, the local database was completely rewritten by LSHTM staff, retaining only the table structure of the original so that data from old and new databases could be combined with relative ease. A new data transmission protocol was also developed by the LSHTM, in which data were transmitted to a secure web address and were, therefore, powerfully encrypted by Secure Socket Layer technology. This revised data transmission protocol met with the requirements of the NHS Information Authority, which the original database could not do, allowing the sites that had refused to submit data electronically to do so; it also persuaded some IT managers who had previously been reluctant to do so to install the database on the local computer network. The revised database and data transmission protocol also allowed implementation of submission of the anonymised minimum data set for patients who had treatment but from whom consent had not been obtained for participation in the cohort study (see Protocol amendments submitted to the Research Ethics Committee).

All centres were provided with the revised database. Clinics which had already collected data via the original system were able to retain the original data tables and have the revised database added as a new ‘front end’. Setting up the new databases required every site to receive a visit from a member of the LSHTM staff, during which the updated database was installed and staff were trained. The first of these site visits took place in August 2005, with the majority of upgrades taking place during the 12 months from September 2005. The database upgrade also required additional investment by the data management centre at the LSHTM, which had to recruit an extra full-time member of staff for 12 months.

The fundamentally different design of the revised clinical database was welcomed by the vast majority of clinics and overcame a lot of the reluctance to collect data. However, it could not overcome inadequacies in local computer networks or the refusal to install the database at some sites. In two cases the database could be installed only on a stand-alone personal computer with no network/internet connection. Other obstacles which made networking the database difficult included a virtual private network at one clinic, a complex arrangement of virtual servers at another, specialist optometry software which altered the configuration of dates and an unwillingness to install the software which the database required to operate. These cases highlighted that the hardware infrastructure at clinics was far from standard.

Centres generated a data report by executing a standard query on their local database and submitted the data periodically to the co-ordinating centre by the secure internet link, except for the two sites without an internet connection which submitted data by computer disk sent by registered post. The co-ordinating centre implemented data validation checks and sent back data queries to sites centres, as originally planned (see Chapter 3, Data collection and management).

Bias

Attrition was very much worse than expected in that many patients were not followed up as described in the protocol. Consequently, we were required to rethink the approach to the analysis plan (see Chapter 3, Plan of analysis). The revised analysis plan allowed us to include data for all of the observation time/documented visits in the analysis plan, irrespective of compliance with the schedule, but did not address the risk of attrition bias/informative censoring.

Sample size considerations

The sample size considerations remained the same as described in Chapter 3, Sample size considerations. However, we originally expected the study to document VPDT in about 20,000–25,000 patients over 3 years. Although the rate of recruitment increased substantially over the course of the study, it was quickly apparent that the difficulties in ensuring that DHs participated would mean that the actual sample recruited would be considerably smaller.

Detailed consideration of possible biases also led us to decide to exclude from the main analyses of BCVA patients who were within 1 year of their first treatment, unless their treatment episodes were completed (see Plan of analysis). We used this strategy because the TAP trials reported 1-year outcome and suggested that outcomes continue to improve with repeated treatment up to 1 year.

Despite these limitations, the study still had considerable precision (greater than in the TAP trials) when estimating treatment outcomes after 1 year by virtue of the continuously scaled outcomes of BCVA and HRQoL and their repeated measurement over time in the study. These attributes of the outcomes contrast with primary outcome in the TAP trials, namely the percentage of patients losing > 15 ETDRS letters at a particular time point.

Plan of analysis

The objectives of the study were reformulated prior to analytical comparisons as described in Chapter 2.

A detailed description of our approach to derivation of new variables for the analyses from the data collected and analysis plans to address each of the objectives of the study are described in the ensuing sections. These data management decisions and analysis plans were established by exploring the accumulating data descriptively and before the key analytic comparisons were carried out. Methods of fitting of final models could not be completely pre-specified, but evolved primarily to optimise the fit of the models given the limitations of the data set.

Data analysis decisions and definitions of derived analysis variables

Close of data collection for the data analyses

Centres were told in advance that recruitment to the study and documentation of study visits would stop for the data analyses on 14 September 2007. An exception to this rule was made for centres that did not submit their final data download on or after this date. For these centres, the cut-off date for calculating whether or not a patient had a completed treatment episode was the date of the last data submission.

Because the NICE guidance stated that patients with predominantly classic CNV lesions with some occult CNV should be treated only in a research study, the VPDT cohort study was funded to continue to collect data up to 31 March 2008. This allowed sites to continue to address data validation queries and missing data for visits that took place up to 14 September 2007. Data submitted for visits after 14 September 2007 were excluded from the analyses in this report. Few new patients were recruited, and few additional visits took place during this period for patients who were already recruited, because new treatments, primarily drugs that inhibited vascular endothelial growth factor (VEGF), were supplanting VPDT (Figure 4).

FIGURE 4.

Recruitment to the VPDT cohort study for the UK. (a) Monthly recruitment up to the end of June 2008. (b) Cumulative monthly recruitment up to the end of June 2008.

A: Is verteporfin photodynamic therapy in the NHS provided as in randomised trials?

We aimed to address objective A by describing the following aspects of VPDT provision:

1. distribution of the number of treatments received in years 1 and 2 by patients classified as having completed their treatment for year 1/2

2. time to stopping treatment among patients classified as having completed their treatment

3. rate of treatment (per eye) among patients classified as having completed their treatment (treatments/year)

4. reasons for loss to follow-up before 1 year.

Item 3 was subsequently omitted because of the substantial loss to follow-up in the first 2 years after starting treatment. Item 4 was omitted because reasons for loss to follow-up were frequently not reported by participating centres.

In order to investigate numbers of treatments administered, we had to distinguish clinical follow-up visits from visits solely for the purposes of the study. In addition to the criteria for a completed treatment episode described in Classification of treatment as active or completed, treatment was defined as complete (despite continuing follow-up) if > 150 days (approximately 5 months) had elapsed between subsequent visits, except when a gap of > 150 days occurred between consecutive treatment visits. This criterion allowed for slippage in a scheduled 3-month visit, or one missed 3-month visit, and classified a 6-month follow-up visit without treatment as follow-up for the purposes of the study in accordance with the data collection schedule.

For item 1, the primary outcome was the number of applications of VPDT in years 1 (≤ 350 days) and 2 (> 350 and ≤ 715 days). Treatment frequencies were cross-tabulated with TAP eligibility and tested for significance using chi-squared statistics. We also compared treatment frequencies in year 1 with treatment frequencies reported for the TAP trials. 30 For item 2, we calculated the time until the first treatment episode was completed (see Classification of treatment as active or completed) or 350 days, whichever was later. These times were described as a Kaplan–Meier curve (see Figure 6), estimating the duration of follow-up when 50% of participants had completed their treatment.

In order to make the comparison with the TAP trials, the cohort for this analysis was restricted to patients with a CNV lesion diagnosed as nAMD and who had completed their treatment or who had completed follow-up for 1 or 2 years after the first treatment. The analysis was also limited to one eye per patient.

B: Is ‘outcome’ the same in the NHS as in randomised trials?

Objective B focused on patients who would have been EFT. We aimed to address this objective by estimating BCVA 1 year after the first treatment in patients classified as having completed their treatment for year 1. We fitted a mixed regression model to estimate the BCVA trajectory during the first year, using data up to 2 years where available. This method of analysis allowed all visit data for an eligible eye to be included irrespective of adherence to the data collection schedule. The duration of follow-up (‘time’) was a covariate in the model; interactions of other covariates with time represented non-parallel trajectories.

A single model was used to answer objectives B and C and included the following covariates: age, gender, baseline BCVA, TAP eligibility, CNV composition, smoking status and whether or not the fellow eye was the better-seeing eye. Coefficients from the model were used to estimate BCVA at 1 year for the EFT [objective (B)], IFT [objective (C)] and UNC subgroups. Inclusion of covariates was necessary because they were potential confounding factors when comparing outcome across the EFT, IFT and UNC subgroups. The influence of the covariates in such a large cohort was also intrinsically of interest; inclusion of the UNC subgroup increased the precision of the analysis with respect to estimating the influence of the covariates.

Because of substantial loss to follow-up in year 2, we again restricted our main analysis to estimating BCVA at 12 months for the cohort of patients described above for objective A (see A: Is verteporfin photodynamic therapy in the NHS provided as in randomised trials?).

C: Is ‘outcome’ the same for patients ineligible from randomised trials?

Objective C focused on patients who would have been IFT. A single model was used to address objectives B and C (see B: Is ‘outcome’ the same in the NHS as in randomised trials?).

D: Is verteporfin photodynamic therapy safe when provided in the NHS?

Adverse reactions and AEs were not classified as required for good clinical practice, although such events were promptly notified to the Data Management Centre at the LSHTM in accordance with good clinical practice. Attribution of ocular AEs to VPDT is difficult because such events may occur as part of the natural history of nAMD. An AR was defined as an ocular or systemic reaction at the time of treatment which was recorded on the same day as the treatment with other clinical data for that visit. An AE was defined as any other ocular or systemic AE reported at the next visit after a treatment or retreatment visit. The association of an AE with the previous treatment visit was coded during data management and, therefore, was associated with the corresponding treatment visit, not the visit on which it was reported.

The probability of a treatment visit giving rise to an AR or AE by site and visit was estimated using a logistic regression model, fitting participating site as a random effect. The distributions of centre-specific probabilities were examined carefully because of concern about the extent to which sites had adhered to the instructions for collecting data about ARs and AEs. To contextualise the overall probability of an AR or AE, we also described the probabilities for a site which had the largest number of treated patients and which we believed had collected such data better than average. We also investigated whether any site had a site-specific upper 95% confidence limit below the lower 95% confidence limit for the entire cohort; where this was the case, a sensitivity analysis was rerun omitting the site.

E: How effective and cost-effective is verteporfin photodynamic therapy?

Different approaches to estimate effectiveness were proposed in the manual of operations (see Chapter 3, Estimating the effectiveness and cost-effectiveness of verteporfin photodynamic therapy). For reasons outside our control, we were able to use only the second of these methods, that is to investigate associations between the use of resources and visual function and other outcomes in the study. This method is described in more detail in How effective is verteporfin photodynamic therapy?, below. This method also underpinned the second element of objective E, that is estimation of the cost-effectiveness of VPDT.

The first method depended on obtaining individual patient data from other researchers, including the TAP triallists. We were able to obtain some data for studies which had academic or public sponsors, but were unable to obtain the data for the key RCTs of VPDT (TAP and VIP trials3,5), even though the manufacturer of verteporfin (Novartis) was represented on the Steering Committee. Without these data, we judged that the first method was not feasible.

The third method depended on being able to characterise an untreated control group in the cohort of patients recruited for the study. However, it quickly became apparent that we were not capturing adequate data for patients who were not treated (either by choice or because of ineligibility) and that untreated patients represented in the database were not similar across sites because of the varied arrangements in place for triaging patients before referral to VPDT clinics.

Participating centres

The number of participating sites increased over time. We originally planned for 30 but, by the time that data collection started, 48 DHs had been identified by SCGs. A few more DHs were identified as the study progressed; this number is imprecise because, if DHs refused to register for the study, we sometimes did not receive definitive information about whether or not a prospective DH actually started to provide VPDT. Additional DHs joined the study up until May 2007. A total of 49 sites registered to take part in the VPDT service but only 47 contributed data to the study (see Figure 2). The first participating site gained the necessary approval to enrol patients on 21 May 2004. Approval for other sites progressed steadily throughout 2004–5. Twenty-one sites were submitting data by May 2005, 38 by May 2006 and 45 by May 2007. Some sites submitted data early during the course of the study but did not continue to do so throughout.

Study population

The first patient was enrolled on 3 June 2004. The rate of recruitment increased to a peak in April 2006 and then declined steadily (see Figure 4). The numbers of patients recruited by each site are shown in Table 5. The numbers ranged from 5 to 593 for all patients treated at any time, but only from 3 to 351 for patients with CNV caused by nAMD, baseline BCVA and at least one follow-up BCVA assessment and not under active treatment or > 1 year since the first treatment (i.e. patients included in the main analyses; Figure 5 and Chapter 4, Definition of ‘TAP eligibility’ and B: Is ‘outcome’ the same in the NHS as in randomised trials?).

FIGURE 5.

Consolidated Standards of Reporting Trials style diagram showing the patients and eyes treated in the VPDT cohort study.

The flow of recruited patients in the study with respect to consent, treatment, inclusion of one or both eyes and whether or not an eye was considered under active treatment when the database was locked is shown in Figure 5. Between June 2004 and September 2007, data on 11,727 patients were submitted. A total of 7748 patients were recorded as having been treated at any time; 575 patients were treated and contributed data for both eyes, giving a total of 8323 eyes. Data were submitted for 31,640 clinic visits in these 7748 patients. The referral mechanisms adopted by sites varied considerably; for example, some had systems for initial triaging of patients with respect to criteria determining eligibility for treatment. Therefore, the data on patients found to be ineligible when attending VPDT clinics could not be interpreted and were not analysed further.

Missing BCVA for 1527 patients resulted in the exclusion of 1676 eyes (142 missing at baseline and 1534 at follow-up). The characteristics of the remaining 6221 patients are shown in Table 6. Their median age was 78 years (interquartile range from 72 to 83 years). The majority were female (3620/6202, 58.4%). The majority (55.4%) were current (832/5282, 15.8%) or ex-smokers (2092/5282, 39.6%).

Of the 6221 patients, 426 (6.8%) were treated and contributed data for both eyes, giving a total of 6647 eyes treated at least once with valid BCVA data at baseline and at least one follow-up assessment; 1728 eyes in 1655 patients had been first treated ≤ 350 days (study definition equivalent to < 1 year) before the database was locked and the study was closed, leaving 4919 eyes in 4566 patients. Patients with CNV caused by nAMD made up 88.5% of these. The characteristics of the 6647 eyes are summarised in Table 7. Very few had had prior laser photocoagulation. The mean BCVA at baseline was 50.4 letters and the mean CS was 22.7 letters. The characteristics of the treated CNV lesions that were independently graded are summarised in Table 8. The majority of lesions had a greatest linear dimension of fewer than three disc areas (1892/2957, 64.0%), were subfoveal or juxtafoveal (2459/2756, 89.2%), were composed of > 50% classic CNV (1943/2777, 70.0%) and were without occult CNV (2243/2773, 80.9%).

Subgroup of patients with choroidal neovascularisation caused by neovascular (wet) age-related macular degeneration

As described previously (see Classification of treatment as active or completed), patients still under active treatment when the study closed were not included in the main analyses because of the risk of bias. It was challenging to include second eyes in the main analyses because most second eyes developed nAMD and received a first treatment at varying times after the first. Thus, a second eye was often still under active treatment or it was ≤ 350 days since the first eye had completed treatment or underwent a first treatment > 350 days before the study was closed. Restricting the analysis to one eye per patient excluded a further 8% of treated eyes. Given the marketing authorisation for verteporfin, CNV lesions caused by nAMD were of particular interest; after excluding treated eyes with non-AMD aetiology, a total of 4043 eyes remained.

The baseline characteristics of this subset of patients and eyes are shown in Table 9, which breaks down the data according to whether patients/eyes were classified as EFT, IFT or UNC. In the EFT group, predominantly classic CNV was present in 86.7% (1064/1227) and minimally classic in 13.3% (163/1227). In the IFT group, predominantly classic CNV was present in 52.9% (628/1187) and minimally classic in 47.1% (559/187). The mean baseline logMAR BCVA was 50 letters (20/100) in the treated eye, which was very similar to study eyes of patients randomised in the TAP trials (53 letters). 3

A: Is verteporfin photodynamic therapy in the NHS provided as in randomised trials?

The analysis for objective A focused on provision of VPDT in the year following the first treatment. Given that TAP trials required prospective participants to have CNV from nAMD, the analysis was based on only the subgroup of patients with CNV from nAMD who were classified as having completed their treatment (n = 4043), with one eye per patient.

Is treatment duration the same as in the randomised trials?

As described in Chapter 3, Data collection and management, the VPDT manual of operations set out a schedule for follow-up and retreatment. This schedule was intended to approximate the follow-up and retreatment guidance provided in the TAP trials, that is patients should be expected to be observed over a period of 2 years with retreatment every 3 months if required. The treatment frequencies described in Is treatment administered at the same frequency as in the randomised trials? show that much less treatment was administered in the study than would have been expected on the basis of the TAP trials.

Based on our definition of a completed treatment episode (see Chapter 4, Classification of treatment as active or completed), we constructed a Kaplan–Meier curve describing ‘survival’ until completion of the treatment episode for the 4566 patients who had data for one eye starting treatment > 350 days before the close of the study, shown in Figure 6. This figure shows that just over 50% of eyes completed the treatment episode in < 1 year. Thus, not only were fewer treatments administered in the study than in the TAP trials, but also the duration of review of patients’ CNV status was shorter than in the TAP trials for the majority of treatment episodes.

FIGURE 6.

Time to completion of treatment episode; Kaplan–Meier graph showing the cumulative proportion of eyes completing the first treatment episode.

B: Is ‘outcome’ the same in the NHS as in randomised trials?

This analysis included 4043 eyes of the 4043 patients who had a diagnosis of nAMD and a first treatment > 350 days before the close of the study. As described in Chapter 4, Definition of ‘TAP’ eligibility, eyes were classified as EFT, IFT or UNC. BCVA outcome in the EFT subgroup was of primary interest in addressing objective B. In order to maximise the power of the analysis, all three subgroups were included in one mixed regression analysis that included the TAP eligibility subgroup and baseline lesion classification (predominantly classic, minimally classic, occult only; Table 12). The possibility of a differing gradient of change over time by subgroup was tested by fitting an interaction of TAP eligibility subgroup and time, but this was found to be non-significant and was excluded from the final model.

We compared descriptively the change in BCVA in the two angiographic subtypes in our study with those previously reported for treatment and sham treatment arms in the TAP trials (Figure 7). 4,30 Both the fitted trajectory and the absolute changes in BCVA over time for patients with predominantly classic lesions in the EFT group were similar to those for patients in the TAP treatment arm. For eyes with minimally classic lesions in the EFT group, the trajectory of BCVA was parallel to that observed for the minimally classic subgroup of the TAP treatment arm but showed less absolute loss of BCVA (Figure 8). 4,30

FIGURE 7.

Change in BCVA from baseline in eyes classified as EFT with predominantly classic CNV. Confidence intervals are not shown for TAP groups because they were not reported. 30

FIGURE 8.

Change in BCVA from baseline in eyes classified as EFT with minimally classic CNV. Confidence intervals are not shown for TAP groups because they were not reported. 30

Baseline lesion classification influenced outcome. Within the EFT group, eyes with minimally classic CNV had better BCVA at baseline (+1.13 letters; see Table 12) and deteriorated more slowly than eyes with predominantly classic CNV (+1.13 + 2.08 = +3.2 letters at 1 year; see Table 12).

The influences of several covariates on BCVA were also investigated, partly because the covariates could have confounded the influences of the TAP eligibility subgroup and baseline lesion classification, and partly because their possible influences were of interest in their own right. The coefficients from the final mixed regression model, shown in Table 12, show that a number of baseline covariates did indeed influence BCVA. None of these statistically significant covariates interacted with the TAP eligibility subgroup, so they can be considered to apply equally to the EFT, IFT and UNC subgroups.

The rate of deterioration of BCVA was influenced by baseline acuity, with faster decline in BCVA over time in eyes with better starting acuity; a patient who read five letters more than average at baseline read only 2.7 letters more at 1 year. BCVA deteriorated faster in older patients; after 1 year of follow-up, BCVA was two letters worse for a person 10 years older than average (88 vs 78 years). Women presented with better baseline BCVA and maintained this difference during follow-up (+1.8 letters). Ex-smokers and those who had never smoked presented with better baseline BCVA (+1.6 and +1.8 letters respectively) and deteriorated more slowly than current smokers. The decrease in BCVA by 1 year was one letter fewer in treated eyes of ex-smokers, and three letters fewer in treated eyes of never smokers, than in treated eyes of current smokers. If the fellow eye had better BCVA than the treated eye, BCVA in the treated eye was worse at baseline (+2.6 letters) and deteriorated faster; the decrease in BCVA by 1 year was five letters worse than if the treated eye was classified as the better-seeing eye. The decrease in BCVA over 1 year was 8 to 16 letters depending on patients’ characteristics and lesion factors (see C: Is ‘outcome’ the same for patients excluded from randomised trials? below).

C: Is ‘outcome’ the same for patients ineligible from randomised trials?

The mixed regression analysis addressing this objective was the same as the one used to address objective B given that we did not find an interaction of TAP eligibility subgroup and time. BCVA outcome in the IFT subgroup, and to a lesser extent the UNC subgroup, was of primary interest in addressing objective C.

Eyes classified as IFT presented with better BCVA (+1.4 letters; see Table 12) than EFT eyes with predominantly or minimally classic CNV; UNC eyes also presented with better BCVA (+1.5 letters) than EFT eyes and deteriorated more slowly (+1.50 + 0.77 = +2.3 letters at 1 year; see Table 12); these eyes were UNC because there were no corresponding independently graded FA findings, and so, by definition, they had unknown lesion percentages classified as classic and occult. Eyes with occult only lesions (a subgroup of IFT eyes) were better by +3.4 letters at baseline and were observed to deteriorate more slowly (+3.43 + 2.79 = +6.2 letters at 1 year).

D: Is verteporfin photodynamic therapy safe when provided in the NHS?

The analysis was carried out on the cohort of patients who were treated at any time, that is n = 7748 (see Figure 5). This larger cohort was used to address objective D because ARs and AEs were collected at all visits and did not require patients to have achieved a particular duration of follow-up. If a patient had only one treatment, then an AR documented at the time of treatment or at a subsequent follow-up visit was considered relevant. Despite defined fields in the database and on the data collection form provided (see Figure 2), and a stated requirement in the manual of operations that these fields should be completed at all visits, information about ARs and AEs was often missing. For example, question 8 (‘Has there been an AE since the last visit or an AR at this visit?’) was often not completed or, when one of these fields was ticked ‘yes’, the specific AR/AE form was not completed.

The analysis of safety was complex because both ARs (at the time of treatment) and AEs (documented at the subsequent visit and attributed to the previous treatment) could in principle be either ‘systemic’, affecting the whole patient, or ‘ocular’, affecting a particular eye. However, ARs tended to be systemic (e.g. systemic AR = back pain, a known side effect of VPDT and an AR that was documented in the TAP trials) and AEs tended to be ocular (e.g. ocular AE = drop in BCVA > 20 letters within 7 days of treatment or intraretinal haemorrhage).

The cohort of 7748 patients had a total of 31,640 visits documented. Treatment was administered at 17,809 (56.3%) visits. ARs and AEs were reported infrequently, with total numbers of 216 (1.2%) and 253 (1.4%) respectively. The distributions of ARs and AEs by visit (with AEs attributed to the previous treatment visit) are shown in Figures 9 and 10. The high frequency of ARs and AEs associated with the first treatment is, to some extent, explained by the fact that the number of patients having a first treatment was higher than the number of patients having a subsequent visit. However, the proportion of treated visits at which ARs and AEs were reported/attributed declined from a maximum of 1.4% and 2.0%, respectively, for visit 1 to 0.3 and 0.9%, respectively, by visit 4 with no ARs or AEs being reported beyond visit 8. Thus, an AR or AE was much more likely to be reported or attributed to visit 1 than to subsequent visits. Therefore, analysis focused only on ARs and AEs reported at visit 1 because treatments on visit 1 were most numerous and most likely to be representative of the reference population of patients who were eligible for VPDT.

FIGURE 9.

Distribution of all ARs over time since eyes were first treated.

FIGURE 10.

Distribution of all AEs over time since eyes were first treated.

The proportion of first treatment visits at which an AR was reported varied from 0% to 7.8% across centres (Figure 11); the overall proportion was 1.4% [exact 95% confidence interval (CI) 1.2% to 1.6%]. Back pain was documented in 86 of 7748 first treatments (1.1%); 58% of these reactions were mild, 27% moderate and 15% severe. Pain at the injection site (n = 2) and extravasation (n = 1) were rarely reported. Fifteen centres (albeit with relatively few documented treatment visits) reported no AR.

FIGURE 11.

Probability of an AR at the time of the first VPDT administration. Each line represents one participating site, ordered by the mean site-specific probability. The vertical dashed lines represent the overall mean probability and 95% confidence interval across all participating sites. The solid vertical line represents the mean probability for the site which submitted data for the most treated patients and which was judged to have complied relatively well with the instructions for recording AEs.

We were concerned about the compliance of centres with reporting ARs and, consequently, the danger of underestimating the risk of an AR. For comparison, the probability of an AR in the centre with the most documented treatment visits (1603) was 2.6% (exact 95% CI 1.8% to 3.5%). ARs were reported less frequently than in the TAP trials in which, for example, back pain was documented in association with 2.2% of VPDT treatment administrations, photosensitivity reactions with 3.0% and ‘adverse events at the site of injections’3 with 13.4% (compared with 3.4% for sham treatments).

Ocular AEs were also reported infrequently. The proportion of first treatment visits at which an AE was reported varied from 0% to 14.3% across centres (Figure 12); the overall proportion was 2.0% (exact 95% CI 1.7% to 2.2%). AEs included a sudden fall in vision reported by the patient or a documented loss of ≥ 20 letters within 7 days of treatment in 25 of 7748 first treatments (0.3%), a tear of the retinal pigment epithelium in 5 (0.1%) and diverse other AEs in 121 (1.6%). AEs, like ARs, were reported less frequently than in the TAP trials, in which ‘visual disturbance’ was documented in association with 17.7% of VPDT treatment administrations compared with 11.6% of sham treatments, and vitreous haemorrhage with 1.0% (compared with 0.5% of sham treatments). 3

FIGURE 12.

Probability of an AE associated with the first VPDT administration. Each line represents one participating site, ordered by the mean site-specific probability. The vertical dashed lines represent the overall mean probability and 95% CI across all participating sites. The solid vertical line represents the mean probability for the site which submitted data for the most treated patients and which was judged to have complied relatively well with the instructions for recording AEs.

The variation between centres in the proportion of treatment visits in which an AE was reported was greater than expected; 70% (33/47; 95% CI 55% to 83%) had a centre-specific proportion below the overall proportion. We attributed this to poor compliance in reporting AEs that would have biased downwards the overall estimate of the probability of an AE. In order to estimate a more representative overall proportion, we excluded data for two centres which reported a proportion for visit 1 with a centre-specific upper 95% confidence limit below the lower 95% confidence limit for the entire cohort. The overall proportion increased to 2.1% (exact 95% CI 1.9% to 2.4%). For comparison, the probability of an AE in the centre with the most documented treatment visits (1603) was 4.6% (exact 95% CI 3.6% to 5.8%).

How effective is verteporfin photodynamic therapy with respect to best-corrected monocular distance visual activity?

Chapter 6 reports outcomes of VPDT in comparison with the TAP trials, that is the change in BCVA over time. As the VPDT cohort study did not have a control group, effectiveness could be estimated only indirectly. Although we considered three ways to do this, we were able to apply only the second method (see Chapter 4, E: How effective and cost-effective is verteporfin photodynamic therapy?).

In terms of BCVA, this method was, in effect, the comparison in change in BCVA over time reported in Chapter 6. Both the fitted trajectory and the average absolute change in BCVA over time for patients with predominantly classic lesions in the EFT group were similar to those for patients in the TAP treatment arm. For eyes with minimally classic lesions in the EFT group, the trajectory of BCVA was parallel to that observed for the minimally classic subgroup of the TAP treatment arm, but showed less absolute loss of BCVA (see Figures 7 and 8).

For HRQoL outcomes, no similar comparison could be made because HRQoL data were not reported for participants in the TAP trials.

How effective is verteporfin photodynamic therapy with respect to health-related quality of life?

Absence of HRQoL outcome in the TAP trials was a major limitation with respect to the NICE technology appraisal of VPDT. Therefore, describing the change in HRQoL with treatment was a key objective of the study. We aimed to do this by quantifying the extent to which BCVA predicted HRQoL.

The subgroup of 18 centres collected and submitted BCVA, CS and HRQoL data for 3262 visits for 1829 patients (Tables 13 and 14). Most data were available for visits at 0, 6 and 12 months, as planned, but data for many patients were also available for 3 and 9 months; 53% of patients had data for two or more visits [one visit, 47%; two visits, 33%; three visits, 16%; more than three visits (maximum six), 4%].

Generic health-related quality of life

Best-corrected monocular distance visual acuity: in the better-seeing eye strongly predicted SF-6D, physical component score (PCS) and mental component score (MCS) (p < 0.0001 for all three HRQoL measures). For each HRQoL measure, the best-fitting models were linear. No evidence was found to support the hypothesis of a sigmoid relationship. We also did not observe any tendency at all for gradients to decrease with the duration of follow-up. The relationship between BCVA in the better-seeing eye and the SF-6D utility score is shown in Figure 13, with the fitted regression superimposed on a scatterplot of the raw data. Predicted changes in SF-6D, PCS and MCS for 5- and 100-letter reductions in BCVA are shown in Table 15.

FIGURE 13.

Scatterplot showing SF-6D preference-based measure of health (equivalent to utility score) compared with better-seeing eye BCVA. The fitted regression line (see Table 15) is superimposed on a scatterplot of the raw data. Values of BCVA < 0 (–10 and –20) represent ‘counting fingers’ and ‘hand movements’ levels of vision respectively.

TABLE 15 - Relationships between visual function and SF-6D/SF-36

Estimated HRQoL change assumes the change in BCVA CS occurs in the better-seeing eye.

PCS and MCS: scored on a scale with a mean of 50 and an SD of 10. PCS and MCS are also normalised, i.e. 50 represents the mean for the reference (i.e. ‘normal’) population.

Change in HRQoL estimated for a three-letter contrast sensitivity triad from 35 to 32 letters for relationships.

HRQoL score	HRQoL scale	Linear regression coefficient (95% CI)	Quadratic regression coefficient (95% CI)	HRQoL change per five letters (≡ one chart line)a	HRQoL change per 100 letters (i.e. whole chart range)a
BCVA
SF-6D	0–1	0.0012 (0.0009 to 0.0014)	–	0.0058	0.116
SF-36 PCSb	Mean = 50	0.049 (0.025 to 0.073)	–	0.245	4.906
SF-36 MCSb	Mean = 50	0.109 (0.078 to 0.140)	–	0.546	10.920
				HRQoL change per three letters (i.e. one contrast sensitivity triad)a,c	HRQoL change per 48 letters (i.e. whole chart range)a
CS
SF-6D	0–1	–0.0016 (–0.0041 to 0.0009)	0.0001 (0.00003 to 0.00015)	0.014	~0.14
SF-36 PCS	Mean = 50	–0.269 (–0.476 to –0.062)	0.008 (0.003 to 0.013)	0.792	~7.7
SF-36 MCS	Mean = 50	–0.120 (–0.382 to 0.143)	0.008 (0.002 to 0.016)	1.155	~12.1

Contrast sensitivity in the better-seeing eye predicted SF-6D, PCS and MCS (p < 0.01 for all three HRQoL measures) but less strongly. The relationship between CS and the SF-6D utility score is shown in Figure 14, with the fitted regression superimposed. For all HRQoL measures, the best-fitting models were positive and quadratic, with the fitted values tending to an asymptote when < 15 letters could be read (see Figure 14). As with BCVA, no evidence was found to support the prior hypothesis of a sigmoid relationship. Predicted changes in SF-6D, PCS and MCS for three-letter (one ‘triad’) and 48-letter reductions in BCVA are shown in Table 15. The latter predicted changes are described as approximate because the quadratic models sometimes caused fitted values to increase slightly when very few letters were read. The predicted changes reported are the fitted value for 48 letters minus the minimum fitted value.

FIGURE 14.

Scatterplot showing SF-6D preference-based measure of health (equivalent to utility score) compared with better-seeing eye CS. The fitted regression line (see Table 13) is superimposed on a scatterplot of the raw data.

For predominantly classic nAMD lesions, VPDT was observed to confer a net benefit of 11 ETDRS and 5 CS letters after 1 year. 4,30 Based on the best-fitting model for SF-6D, these visual function benefits ‘translate’ into utility differences of 0.013 and 0.022 respectively (on a scale of 0–1). In the VPDT cohort study, the net BCVA benefit compared with the TAP sham VPDT group was slightly smaller, at about nine letters (see Figure 7); this BCVA benefit ‘translates’ into a utility difference of about 0.011.

For minimally classic nAMD lesions, VPDT was observed to confer a net benefit of four ETDRS letters after 1 year. 4,30 Based on the best-fitting model for SF-6D, these visual function benefits ‘translate’ into a utility difference of 0.005. The net BCVA benefit observed in the study, of about five letters (see Figure 8), ‘translates’ into a utility difference of about 0.006.

Vision-specific health-related quality of life

In the better-seeing eye, BCVA also strongly predicted the composite total NEIVFQ score, and the distance and near activity subscales (p < 0.0001 for all three NEIVFQ scores). The relationship between BCVA in the better-seeing eye and NEIVFQ composite total score is shown in Figure 15; similar relationships were observed for distance and near NEIVFQ scores. Relationships were positive and quadratic and had steeper gradients (in relation to the range of BCVA) than those seen for SF-6D and SF-36 component scores. Gradients were steeper for the distance and near activities subscores than for the composite total score. Predicted changes in composite total, distance and near NEIVFQ scores for 5- and 100-letter reductions in BCVA are shown in Table 16. Because the functions were quadratic, the predicted NEIVFQ changes for 100-letter reductions are described as approximate.

FIGURE 15.

Scatterplot showing the NEIVFQ composite total score compared with better-seeing eye BCVA. The fitted regression line (see Table 16) is superimposed on a scatterplot of the raw data. Values of BCVA < 0 (–10 and –20) represent ‘counting fingers’ and ‘hand movements’ levels of vision respectively.

Similarly, CS in the better-seeing eye strongly predicted the composite total, distance and near activity NEIVFQ score (p < 0.0001 for all three HRQoL measures, but less strongly than BCVA). The relationship between CS and NEIVFQ core total is shown in the scatterplot in Figure 16; the relationships were similar for distance and near NEIVFQ scores. As for BCVA, the relationships were positive and quadratic and had steeper gradients than those observed for SF-6D, PCS and MCS scores. Predicted changes in composite total, distance and near NEIVFQ scores for 3-letter (one triad) and 48-letter reductions in CS are shown in Table 16. Because the functions were again quadratic, the predicted NEIVFQ changes for 48-letter reductions are again described as approximate.

FIGURE 16.

Scatterplot showing the NEIVFQ composite total score compared with better-seeing eye CS. The fitted regression line (see Table 16) is superimposed on a scatterplot of the raw data.

TABLE 16 - Relationships between visual function and NEIVFQ scores

Estimated HRQoL change assumes the change in BCVA occurs in the better-seeing eye.

Changes in HRQoL estimated for a 5-letter drop in BCVA from 70 to 65 letters, and a 3-letter drop in CS (one triad) from 35 to 32 letters.

HRQoL score	HRQoL scale	Linear regression coefficient (95% CI)	Quadratic regression coefficient (95% CI)	HRQoL change per five letters (≡ one chart line)a,b	HRQoL change per 100 letters (i.e. whole chart range)a
BCVA
NEIVFQ composite	0–100	–0.129 (–0.342 to 0.083)	0.0067 (0.0049 to 0.0085)	3.90	~55.0
NEIVFQ distance activities	0–100	0.00002 (–0.236 to 0.236)	0.0075 (0.0054 to 0.0096)	5.08	~72.5
NEIVFQ near activities	0–100	0.397 (–0.626 to –0.127)	0.0111 (0.0090 to 0.0131)	5.48	~72.8
				HRQoL change per three letters (≡ one contrast sensitivity triad)a,b	HRQoL change per 48 letters (i.e. whole chart range)a
CS
NEIVFQ composite	0–100	–1.285 (–1.646 to –0.924)	0.0540 (0.0454 to 0.0625)	6.99	~44.4
NEIVFQ distance activities	0–100	–1.737 (–2.181 to –1.292)	0.0694 (0.0589 to 0.0799)	8.74	~57.6
NEIVFQ near activities	0–100	–1.967 (–2.406 to –1.527)	0.0757 (0.0653 to 0.0860)	9.32	~64.7

Based on the best-fitting model for the NEIVFQ composite total score and an average presenting BCVA of 50 letters and CS of 23 letters, the differences in BCVA and CS observed in the TAP trials for predominantly classic nAMD lesions (11 ETDRS and 5 CS letters, respectively)4,30 ‘translate’ into differences in score of about 9 and 12 respectively (on a scale of 0 to 100). In the VPDT cohort study, the net BCVA benefit was slightly smaller, at about nine letters (see Figure 7); this BCVA benefit ‘translates’ into a difference in NEIVFQ composite total score of about 7.

For minimally classic nAMD lesions, the visual benefit observed in the TAP trials of four ETDRS letters after 1 year4,30 ‘translates’ into a difference in NEIVFQ composite total score of about 3. The larger net BCVA benefit observed in the study, of about five letters (see Figure 8), ‘translates’ into a difference in NEIVFQ composite total score of about 4.

How cost-effective is verteporfin photodynamic therapy?

Health and social services use and costs related to vision loss

Health and social services and BCVA costs were available for a total of 3435 visits in 1764 patients. As in the case of HRQoL data, most resource use questionnaires were completed in association with visits at 0, 6 and 12 months (Table 18). All visits with data were included in the analysis. Only about 10% of patients reported using a health service, such as an unscheduled low-vision appointment or seeing their GP, while < 0.5% of patients reported moving into a nursing home, residential home or sheltered accommodation (Table 19). The mean annual total costs related to the patients’ eye conditions were low (approximately £300) relative to the VPDT intervention costs. Although the highest cost item was social service costs, the mean cost for this item was driven by the 1% of patients who received high levels of support from social service home carers costing > £3500 per year.

TABLE 18 - Distribution of visits for which visual function and resource use data were available

Parameter	Duration of follow-up (months)						Total
	0	3	6	9	12	> 12
Resource use and BCVA	1350	136	848	171	528	402	3435
Resource use and CS	1276	96	764	127	466	356	3085

TABLE 19 - The percentage of patients using each HSS item, unit costs (£) and HSS costs (£)

NA, not applicable.

Item	% using service in last 3 months	Unit cost	Mean (SD) annual cost per patient (£)
Low vision appointment	9.96	83	51.2 (259.2)
Visits to GP	10.88	44	22.1 (70.5)
Social services	1.14	35	129.0 (1531)
Day centre	1.21	32	26.3 (256)
Nursing home stay	0.06	648	19.6 (812)
Residential care	0.12	443	26.8 (785)
Sheltered housing	0.20	140	14.8 (328)
Antidepressant use	0.52	10	0.64 (8.8)
Other	NA	NA	10.0 (53)
Total	NA	NA	297 (2078)

There was a negative relationship between BCVA and annual HSS cost. Figure 17 shows that the gradient of this relationship, like that for HRQoL, was shallow. For example, for a five-letter decrease in BCVA for patients with baseline of 50 letters, the predicted increase in mean annual costs was about £28. For those patients who used a service, a five-letter decrease in BCVA was associated with a statistically significant increase in mean annual costs of £111 (95% CI ∼£48 to ∼£174). This association between BCVA and SF-6D was combined with the differential decline in BCVA from baseline for the VPDT and placebo groups in TAP to derive differences in HRQoL between VPDT and BSC at 3-monthly intervals (Figure 18). 30

FIGURE 17.

Scatterplot showing the annual cost of HSS resource use compared with better-seeing eye BCVA. The regression line superimposed on a scatterplot of the raw data is derived from fitting the two-part model. Values of BCVA < 0 (–10 and –20) represent ‘counting fingers’ and ‘hand movements’ levels of vision respectively.

FIGURE 18.

Mean predicted change in HRQoL at 3-monthly time points for VPDT vs BSC for eyes with predominantly classic lesions that would have been EFT. Predictions combine VPDT map of HRQoL and visual acuity with visual acuity for VDPT treatment and placebo groups reported in the TAP study.

Sensitivity analyses

The sensitivity analyses found that the results were robust to the methodological assumptions and data sources used in the base-case analysis (Table 21). The probability that VPDT is cost-effective is zero unless the willingness to pay for a QALY gain exceeds £100,000 per QALY (Figure 19). The further sensitivity analyses showed that:

1. If the treatment frequency was taken from TAP rather than the VPDT cohort study, the incremental costs increased to £5946 and the cost per QALY rose to £288,000.

2. If the pre–post BCVA difference from the cohort study was used rather than the difference between arms in the TAP trials, then the estimated QALY gain was higher (0.0212 vs 0.0207), but the cost per QALY gained still exceeded £150,000.

3. If the relationships of cost and HRQoL with CS rather than BCVA were used, this led to a small increase in QALY gain compared with the base case (0.0220 vs 0.0207), but again the cost per QALY exceeded £150,000.

4. If the costs of BSC were 10-fold those assumed in the base case, the cost per QALY was £91,000.

5. If the difference in BCVA observed for VPDT compared with BSC at 2 years was maintained until 5 years, the cost per QALY was £94,000.

TABLE 21 - Sensitivity analyses on incremental cost (£) per QALY for VPDT vs BSC according to different assumptions

VA, visual acuity.

Values rounded to the nearest 1000.

Sensitivity analysis	Incremental QALY	Incremental cost	Incremental cost/QALYa
Base case	0.02066	3514	170,000
TAP study treatment frequency	0.02066	5946	288,000
VPDT cohort study VA results	0.02122	3511	165,000
Contrast and sensitivity map	0.02201	3412	155,000
10-fold increase in BSC costs	0.02066	1891	91,000

FIGURE 19.

Cost-effectiveness acceptability curve for VPDT vs BSC.

Key findings

The key findings from the VPDT cohort study are outlined below:

• The change in BCVA was similar to that observed in the treatment groups in the pivotal licensing trials, although the deterioration in BCVA over time may have been underestimated (see Strengths and weaknesses of the verteporfin photodynamic therapy cohort study).

• The BCVA benefit was achieved with fewer treatments.

• In addition, non-treatment-related baseline covariates influenced change in BCVA over the study period.

• In the better-seeing eye, BCVA and CS were highly significant predictors of SF-6D utility, SF-36 component scores and NEIVFQ scores. The change in utility for a unit change in BCVA was less than several previous estimates.

• Realistic estimates of the cost-effectiveness of VPDT were obtained and were consistent with higher previous estimates, although these high estimates are almost certainly too low because of the assumption (common to all CEAs) that the better-seeing eye is being treated.

Strengths and weaknesses of the verteporfin photodynamic therapy cohort study

Despite its observational nature, the VPDT cohort study has many strengths. These include its size, pragmatic nature, and systematic collection of standardised data on acuity and lesion characteristics. This was the largest study of its kind to date, giving new insights into the HRQoL of people with nAMD and new data that will be pivotal to future studies of effectiveness in this population. Many aspects of the data collected in this study were robust. Protocol-based BCVA and CS were measured and HRQoL instruments administered at multiple time points, which allowed us to investigate in detail the relationships between clinical measures of vision and HRQoL. The large sample gave us reasonable power to test secondary hypotheses about the shape of the relationships and adaptation to vision loss over a 2-year period, even though the proportion of patients with data for more than three visits was small.

The VPDT cohort study was the first with concurrent concurrent collection of data on HRQoL and patient-level data on HSS resource use, and, hence, first to have undertaken a systematic CEA using data acquired during treatment rather than data stipulated by a trial protocol. Our CEA also tackled three major methodological concerns not previously addressed in CEAs of interventions for nAMD. Thus, our work extends the literature on the costs and cost-effectiveness of interventions for nAMD, and its findings will assist future CEA.

Set against these strengths, there were a number of limitations. These include the observational nature of the study, the loss to follow-up of large proportions of the original sample, missing data and the lack of a BSC comparison group of current relevance. There were major logistical challenges in establishing the study, which were discussed at a project review meeting about 18 months after the contract for the study started; key observations and recommendations from this meeting are set out in Appendix 3.

Unlike the pivotal trials, in which almost all patients were followed up for 24 months,3,4,30 about half of the patients included in our analyses did not have 1-year follow-up. Because poor data quality is a well-recognised limitation of observational studies, we undertook computerised data validation checks on an on-going basis and when compiling the final data set. We checked whether or not data were missing for some visits by (a) matching records from paper and electronic systems for collecting BCVA and (b) requesting that centres should check explicitly whether or not additional visits had taken place for selected patients. The results from these checks implied that data had been submitted for > 95% of completed visits.

The exact reasons for loss to follow-up are not known. We attempted to collect information about reasons for completing a treatment episode early or for patients being lost to follow-up, but participating centres did not report reasons reliably. Anecdotally, we became aware that some hospitals had a policy of not rebooking appointments for patients who missed a visit, and some ophthalmologists were put under pressure to discharge patients who did not require active treatment rather than to continue to review them. The context for these policies was the extremely overstretched nature of macular clinics. It should be remembered that providing VPDT required hospitals to make regular appointments (up to four per year) to review a large number of patients who had previously had fewer than one; reviewing a patient required FA and, if treatment was required, administration of VPDT. Irrespective of whether or not funding was available to pay for these resources, the expert workforce needed to provide VPDT could not be expanded rapidly.

Patients who were not followed lost the opportunity to be retreated if reactivation occurred; this could have led to worse BCVA outcomes with treatment in everyday practice than with treatment in the licensing trials. However, the BCVA outcome in the VPDT cohort study was generally similar to that observed in the treatment arm of the TAP trials.

The loss to follow-up introduced uncertainty to the data analyses, which was taken into account by using a mixed regression model to predict BCVA at 1 year in different subgroups. This approach allows all of the available data to be modelled but does not prevent attrition bias. We observed that patients who were lost to follow-up tended to have poorer BCVA at baseline (data available from the authors). Because follow-up data were more likely to be missing with increasing duration after first treatment, and patients with a poor outcome were more likely to be lost to follow-up, the model may have tended to underestimate deterioration in BCVA over time. The regression model for BCVA trajectory assumed that BCVA deteriorated steadily (on the principles of parsimony and ‘best fit’); this assumption is unlikely to be valid when BCVA is poor because the neovascular process burns out, causing a ‘floor’ effect. Attrition bias and a floor effect would have affected the results in opposite directions. Consequently, it is uncertain whether or not the BCVA deterioration over time in the study was truly similar to that observed in the TAP trials or underestimated because of selective attrition.

Although loss to follow-up is a scientific limitation, our experience also demonstrates vividly the difficulties associated with follow-up when treatments requiring multiple visits over an extended period of time in an older age group are introduced into routine clinical practice. Such data are invaluable to health service planners and are rarely available. We believe that patients and ophthalmologists became disheartened with eyes that experienced deterioration of vision during treatment and follow-up, causing treatment to be discontinued before the recommended time point of 2 years. We did not attempt to predict outcome at 2 years because the data were sparse.

A further limitation of the study was the inability to classify 40% of the lesions at baseline with respect to TAP eligibility (eyes classified as UNC), either because an angiogram was not submitted or because the submitted angiogram could not be graded. This group was retained in the model and had parameter estimates which tended to lie between those for EFT and IFT groups and between those for predominantly and minimally classic lesions. Thus, there was no reason to believe that these eyes represented a biased selection with respect to eligibility for the TAP trials or their lesion composition.

In the CEA, we were unable to use a direct control group, instead we relied on the control group in the TAP trials. Extrapolating relative effects across different populations is potentially problematic; for example, trials often report different estimates of effect from those given by observational studies. 58 Nevertheless, in this cohort study, BCVA at first treatment and at 1 and 2 years were similar to the predominantly classic subgroup studied in the TAP trials.

Interpretation

Is the Short Form questionnaire-6 Dimensions an appropriate measure of generic health-related quality of life?

A generic HRQoL measure chosen for comparing health gain across disease areas should have a descriptive system that covers all the important dimensions of health. The SF-6D, like the EQ-5D, has a descriptive system that purports to meet the World Health Organization definition of health: ‘complete physical, mental and social well-being and not merely the absence of disease or infirmity’. 69 Utilities measured using the SF-6D are similar to those measured using the EQ-5D. 70 By contrast, the HUI3 is based on a narrower, ‘within the skin’ definition of health focusing on impairment and not on the social context of the impairment. 71 Thus, the HUI3 consists of items that tap self-reported functioning more directly than the EQ-5D and SF-6D. The HUI3 is, therefore, likely to be ‘more sensitive’71 to visual loss than the SF-6D. 64 However, the HUI3 has been criticised for using this relatively narrow description of health. 9

A further concern that has been previously expressed is that the utilities for the HUI3 have been derived from a power transformation of values from a visual analogue scale,72 rather than by direct valuation with choice-based methods such as the standard gamble (used for the SF-6D) or the time trade-off (used for the EQ-5D). Our approach of using the SF-6D follows the recommendations of policy-makers such as NICE. 15 Therefore, it is not surprising that our findings are consistent with previous studies that have used the EQ-5D. 64

In terms of an internationally recognised measure of HRQoL appropriately based on societal preferences, the gradient of the decrease in HRQoL with deteriorating visual function is small. The estimated gain in HRQoL (utility) from VPDT is about 0.02 (in terms of BCVA, a difference of about 11 letters after 2 years, assuming that only the best-seeing eye is being treated), and from ranibizumab is about 0.04 (a difference of about 21 letters after 2 years,73,74 under the same assumption). Gains in utility (over varying time horizons) for other common interventions for chronic conditions (Table 23) show that the utility gain associated with VPDT is relatively small compared with other competing interventions. 75,76 These utilities measured over the appropriate time horizon, and combined with relative effects on life years gained, translate into QALYs and inform health policy decisions.

Implications for practice

The main implications do not relate to the use of VPDT to treat nAMD because VPDT has now been superseded by the introduction of new treatments (see Current status of verteporfin photodynamic therapy and future research).

• The fact that a much smaller number of treatments was administered than in the TAP trials suggests that treatment regimens receiving marketing authorisation may overestimate the intensity of treatment required. This observation may apply to other new health technologies.

• Our ability to estimate effectiveness was limited by substantial loss to follow-up, which prevented comparisons with outcomes in RCTs and a systematic review of RCTs. This limitation should be carefully considered in the design of similar future studies if the effectiveness of an intervention is not dramatic and treatment or follow-up is scheduled over many months (conditions which we suspect may cause clinicians or patients not to adhere to the treatment regimen).

• Verteporfin photodynamic therapy is less effective than newer technologies in treating nAMD. Its use should be limited to circumstances in which these newer technologies are contraindicated or refused by patients, and to categories of AMD such as polypoidal choroidopathy or other diseases with neovascularisation arising from the choroid, for example high myopia.

• Licensing trials generally involve only one eye, and the benefit to a person from effective treatment for an eye will depend on whether or not the person’s visual function is limited by the vision in the treated eye. In terms of cost-effectiveness, an appraisal of a technology that benefits one eye should evaluate the benefit at the level of a person, not an eye.

• The gradients of the relationships (a) between BCVA and EQ-5D utility and (b) between BCVA and HSS resource use/cost were shallower than most previous estimates. The consequences of these relationships for the appraisal of other technologies to treat eye diseases that impair vision need to be considered carefully.

Current status of verteporfin photodynamic therapy and future research

Verteporfin photodynamic therapy as a monotherapy for nAMD has been superseded by the introduction of molecular biologicals that target VEGF, a key molecule that promotes neovascularisation in AMD. The latter technology was introduced in 2006 following demonstration that monthly intravitreal injections of ranibizumab, a monoclonal antibody that inhibits VEGF, is vastly superior to both BSC and PDT. 72,73 Anti-VEGF therapies have also been shown to result in discernible improvements in HRQoL, outcomes which were not demonstrable with VPDT. VPDT combined with anti-VEGF therapy has been investigated, but the results suggest only a marginally improved benefit in terms of fewer treatments and no benefit in terms of visual acuity for nAMD. 84

Research recommendations outside the context of nAMD are:

• Benefit from VPDT has been shown for visual acuity outcomes in specific nAMD variants such as polypoidal choroidopathy. VPDT continues to be used as first-line treatment in the management of other conditions such as neovascularisation due to myopia, inflammation and certain other choroidal diseases including central serous chorioretinopathy. Further studies are required to investigate the effectiveness of VPDT in these disease conditions. Similarly, the methods used in this study could be applied, cautiously, in order to estimate the cost-effectiveness of VPDT for these other eye conditions.

• We observed differences in the rate of change in BCVA over time for a number of covariates. It would be interesting to investigate whether or not the effectiveness of new interventions for nAMD also varies in a similar way in relation to these covariates.

• The study demonstrates the value of estimating the relationship between a common clinical outcome, that is BCVA, and other outcomes less often measured in RCTs but which are important for technology appraisals, for example HRQoL or HSS resource use. Because of the size of the study population, these relationships were estimated relatively precisely. Once established, relationships of this kind should be able to be applied to modelling of the effectiveness of other technologies, on the basis of estimates of effect from RCTs for the common clinical outcome. Further research could investigate how widely these relationships can be applied, for example to other diseases that reduce BCVA.

Contribution of authors

• Professor BC Reeves designed the study with Professors U Chakravarthy and SP Harding and Dr R Grieve. He led the Data Management Centre at the LSHTM and oversaw the data analyses and their interpretation from an epidemiological perspective. He drafted the report with Professors Chakravarthy and Harding.

• Professor SP Harding designed the study with Professors Chakravarthy and Reeves and Dr Grieve. He contributed to the initiation and recruitment of sites and the management of the project, including the training of research staff at sites. With colleagues, he led the establishment and management of NetwORC UK, developed specifically for this project. He contributed to the interpretation of the results and drafted the report with Professors Chakravarthy and Reeves.

• Ms J Langham was the study manager at the Data Management Centre. She liaised with sites about protocol or data queries and managed the study on a day-to-day basis. She contributed to data management and descriptive statistics about the study cohort.

• Dr R Grieve designed and conducted the CEA, interpreted the cost-effectiveness study and drafted this section of the report. He also contributed to the design of the overall study, and to the interpretation and reporting of the analysis of HRQoL.

• Mr K Tomlin designed and developed the databases used for VPDT data collection at each of the participating hospitals and provided technical and data management support to each of the clinical teams. He contributed to data management, especially linking the clinical and grading data from NetwORC UK, and descriptive statistics about the study cohort.

• Ms J Walker carried out the regression modelling of visual outcomes and the relationships between visual outcomes, HRQoL and resource use under the supervision of Dr Carpenter.

• Dr C Guerriero helped to design the CEA and to interpreting and drafting the sections of the report on the CEA.

• Dr J Carpenter advised on the statistical aspects of the study, at both the design and analysis stages, especially the regression modelling and the handling of missing data.

• Dr WP Patton contributed to the setting up of NetwORC UK and managed all of the data coming from the reading centres to the Data Management Centre. He worked on the quality assurance processes, training and validation exercises, and contributed to the analyses.

• Dr KA Muldrew contributed to the setting up of NetwORC UK. She designed the grading protocols with clinician colleagues, developed training and validation programmes and trained staff employed in the reading centres. She worked on the quality assurance processes and contributed to the analyses.

• Dr T Peto is Head of the Moorfields Eye Hospital Reading Centre, part of NetwORC UK. She was involved in the setting up, design and execution of the imaging and grading protocol. She also contributed to the training and certification of graders, and advised clinicians about image interpretation, together with the clinical leads from the Belfast and the Liverpool reading centres.

• Professor U Chakravarthy designed the study with Professors Harding and Reeves and Dr Grieve. She was the principal applicant in the effort to secure funding. She led the study and, with colleagues, the establishment and management of NetwORC UK, developed specifically for this project. She contributed to the interpretation of the results and drafted the report with Professors Harding and Reeves.

All authors contributed to the reporting of the study through this report and other peer-reviewed publications arising from the study.

Verteporfin photodynamic therapy cohort study contributors

Mr Douglas Newman, Addenbrookes Hospital, Cambridge.

Mr Martin Harris, General Hospital, Barnet, London.

Mr Richard Antcliffe, Royal United Hospital, Bath.

Ms Salwa Abugreen, Royal Blackburn Hospital, Blackburn.

Ms Clare Bailey, Bristol Eye Hospital, Bristol.

Professor Usha Chakravarthy, Royal Victoria Hospital, Belfast.

Mr Simon Morgan, Blackpool Victoria Hospital, Blackpool.

Mr Jonathon Gibson, Birmingham and Midlands Eye Hospital, Birmingham.

Mr Faruque Ganchi, Bradford Royal Infirmary, Bradford.

Mr Anthony Casswell, Sussex Eye Hospital, Brighton.

Mr Shafiq Rehman, Calderdale Royal Hospital, Calderdale, Yorkshire.

Mr Robert Johnston, Cheltenham General Hospital, Cheltenham.

Mr Sergio Pagliarini, University Hospitals Coventry and Warwickshire, Coventry.

Mr Hean-Choon Chen, Derbyshire Royal Infirmary, Derby.

Mrs Geeta Menon, Frimley Park Hospital, Frimley Park.

Mr Sakkaf Aftab and Mr Satish Kotta, Northern Lincolnshire and Goole NHS Trust, Goole.

Mr Gavin Walter, Harrogate District Hospital, Harrogate.

Mr Kevin Gregory-Evans, Western Eye Hospital, London.

Mr Nicholas Lee Hillingdon Hopsitals NHS Trust. Middlesex.

Mr Meen Zaman, Hull Royal Infirmary, Hull.

Mr Clive Edelsten, Ipswich Hospital NHS Trust, Ipswich.

Mr Victor Chong, King’s College Hospital, London.

Mr R J Pushpanathan, Queen Elizabeth Hospital, King’s Lynn.

Mr Simon Horgan, Kingston Hospital, Kingston.

Mr Martin McKibbin, Leeds Teaching Hospital, Leeds.

Professor Simon Harding, Royal Liverpool University Hospital, Liverpool.

Mr Frank Ahfat, Maidstone Hospital, Maidstone.

Mr Paulo Stanga, Spire Manchester Hospital, Manchester.

Mr Adnan Tufail, Moorfields Eye Hospital, Moorfields, London.

Mr James Talks, Newcastle Upon Tyne Hospitals NHS Trust, Newcastle.

Mr Andrew Glenn, Norfolk and Norwich University Hospital, Norwich.

Mr Alexander Foss, Queen’s Medical Centre, Nottingham.

Ms Susan Downes, Oxford Eye Hospital, Oxford.

Professor Bal Dhillon, Princess Alexandra Eye Pavillion, Edinburgh.

Ms Sarah-Lucie Watson, Royal Berkshire Hospital, Reading.

Mr Christopher Brand, Royal Hallamshire Hospital, Sheffield.

Mr Niral Karia, Southend Hospital, Southend.

Professor Andrew Lotery, Southampton University Hospital, Southampton.

Mr Myer Mark Yodaiken, Stepping Hill Hospital, Stockport.

Ms Consuela Moorman, Stoke Mandeville Hospital, Aylesbury.

Mr Michael Cole, Torbay Hospital, Torbay.

Mr Salim Natha, Leigh Infirmary, Wigan.

Mr Yit C Yang, Wolverhampton Eye Infirmary, Wolverhampton.

Mr Sal Rassam, Worthing Hospital, Worthing.

Mr Gavin Walters, Harrogate District Hospital, Harrogate.

Publications

Harding SP, Tomlin K, Reeves BC, Langham J, Walker J, Carpenter J, et al. Verteporfin photodynamic therapy cohort study: report 1: effectiveness and factors influencing outcomes. Ophthalmology 2009;116:e1–8.

Grieve R, Guerriero C, Walker J, Tomlin K, Langham J, Harding S, et al. Verteporfin photodynamic therapy cohort study: report 3: cost effectiveness and lessons for future evaluations. Ophthalmology 2009;116:2471–7.

Reeves BC, Langham J, Walker J, Grieve R, Chakravarthy U, Tomlin K, et al. Verteporfin photodynamic therapy cohort study: report 2: clinical measures of vision and health-related quality of life. Ophthalmology 2009;116:2463–70.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health

Key observations

1. This is an example of how a number of influential people and organisations anticipated that there could be major problems with the introduction of a new treatment and aimed to deal with the issues proactively. Despite this foresight and attempts to avoid problems arising, introducing both the treatment and the cohort study has been one of the most difficult exercises.

2. The model developed in the cohort study can be used to meet many of the needs of the NHS in the introduction of an expensive therapy by providing postmarketing surveillance, a managed introduction to standard treatment protocols, evidence of effectiveness in routine clinical practice and data for optimisation of treatment protocols.

3. Two important communications problems arose which complicated implementation:

   1. The NICE guidance did not explicitly refer to the cohort study.

   2. The Department of Health did not explicitly support the cohort study nor did it issue clear directions in relation to it. In particular, this left those tasked with setting up the cohort study without any mandate to implement it in the NHS.

   It is not clear why NICE and the DOH did not feel able to provide this explicit support.

4. It is not clear:

   1. Why NICE did not recommend a RCT and why the National Coordinating Centre for Health Technology Assessment was willing to support a cohort study instead of a RCT.

   2. Why the cohort study was seen as the best option.

5. There was initial hostility to the cohort study from many of the clinicians, in part because of their general reluctance to gather clinical data and in part because they perceived the cohort study to be a threat to their clinical freedom. Many, but not all, have since changed that view.

6. During the implementation of the cohort study, significant weaknesses in the capabilities of clinicians, clinical teams and trusts in understanding the principles and practicality of gathering systematic information were identified.

7. Commissioners were initially unsupportive of the cohort study. This probably stemmed from a lack of understanding of what it was trying to achieve. Some have remained indifferent to it. The lack of clarity about the aims and objectives of the cohort study meant that interests of commissioners were never made explicit.

8. Both patient groups and the manufacturer (Novartis) were also initially hostile to the cohort study, although patient groups were not initially averse to the idea of a RCT. From the manufacturer’s perspective, the findings of the cohort study could at best only confirm the status quo; at worst, the findings could seriously damage their commercial interest.

9. The size of the task was underestimated by all.

10. The designation processes undertaken locally were very confused and varied greatly.

Recommendations

1. The NHS needs to better anticipate the requirements needed to manage the introduction of new treatments particularly where establishing a new treatment is likely to be complex. (Note that a recommendation has already been made through the Specialised Commissioning Group. This recommendation proposes that NICE, for example, involves ‘on the ground’ NHS staff from both commissioning and provider organisations in discussion of the practical issues that may arise from specific pieces of guidance.)

2. Establishing postlicensing studies to evaluate the introduction and use of new treatments should not be considered without the explicit support of the Department of Health. This support needs to be transparent and public.

3. A clear set of aims and objectives must be published and made widely available before commencement of any cohort study to ensure that all stakeholders are fully informed and take ownership.

4. There needs to be better understanding of the role of cohort studies together with that of registries and databases.

5. Clinicians should be better informed of the need for robust, continuing postmarketing evaluation of clinical therapies after marketing authorisation in order to ensure that results from RCTs carried out for the purposes of licensing are generalizable in everyday clinical practice after the RCTs have ended. The importance of continuing to acquire data on the use of treatments in everyday practice must be stressed and clinicians strongly encouraged to collect these data particularly when introducing high cost medicines into the NHS.

6. The NHS could benefit from the development of good practice guidance in relation to designation and accreditation processes. (Some work on this has already begun in the West Midlands.)

7. The NHS could benefit from the development of good practice guidance on postlicensing collection of data on new treatments.

Overall, despite very serious obstacles, the cohort study has been established and has achieved the majority of its objectives with a large number of trusts providing good data. Wide geographical access is being provided to a high standard.

The review group strongly supports the development of on-going assessment of new treatments and for the health community and policy-makers to gain as much experience from the cohort study as possible. In spite of a complex set of problems that have been overcome to lesser or greater extent, the model that has been developed by the cohort study team provides a useful model for the future.

Current status of the cohort study

The study is running effectively with good rates of recruitment and data being submitted by most designated providers.

There has been slippage estimated at around 1 year at present. It was expected to reach a full recruitment rate by June 2004 but in reality only a handful of patients had been recruited by then. A year later, 1200–1500 patients had been recruited. Since the project review, a year’s extension to data collection has been provisionally funded. The study is expected to recruit until November 2007.

The study team continue to deal with issues around NHS implementation requiring a major time investment to sort out on-going problems.

Funding for treatment was not considered a major issue at the time of the review other than for the establishment of some dedicated clinics to promote rapid referral. However, since then it has become apparent that the current financial situation has resulted in clinical and nursing posts being frozen, waiting lists being allowed to develop and planned service developments being put on hold. In addition, the effects of national tariffs under Payment By Results are causing concern and generating uncertainties in the minds of managers in designated provider trusts. Indeed the whole issue of funding flows is confused. It is likely that there are problems at all levels: failure of some commissioners to fully fund the cohort study, failure of some trusts not to forward funds given for the cohort study and treatment to clinical teams and also some trusts receiving money for the cohort study but not entering patients into the study.

Compliance among clinicians with respect to the collection and submission of high-quality data continues to be an important problem. Some providers are receiving funding to provide photodynamic therapy but are not participating in the study. The Steering Group constantly monitor this situation.

Data collection is still being established at some sites.

A number of centres are reporting considerable delays in getting patients to treating centres – such that a significant percentage of referrals are deemed ineligible for treatment owing to the poor level of visual acuity in the eye being assessed, suggesting that earlier referral would have resulted in the patient receiving treatment.

Alternative treatments with antiangiogenic drugs with equivalent efficacy to VPDT and, potentially, wider application are due to be licensed for use in this condition during 2006–7. It can be anticipated that the implementation of these new treatments will present a major problem for the NHS as well as a threat to the cohort study.

Health Technology Assessment programme

1. Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

2. Professor of Dermato-Epidemiology, Centre of Evidence-Based Dermatology, University of Nottingham