A randomised controlled trial to assess the clinical effectiveness and cost-effectiveness of alternative treatments to Inhibit VEGF in Age-related choroidal Neovascularisation (IVAN)

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 07/36/01. The contractual start date was in July 2007. The draft report began editorial review in January 2014 and was accepted for publication in November 2014. 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 reviewers 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

Usha Chakravarthy, Simon P Harding and Andrew J Lotery are principal investigators of trials sponsored by Novartis, the manufacturers of ranibizumab. Usha Chakravarthy has attended and been remunerated for attendance at advisory boards for Novartis, Bayer, Neovista, Oraya, Allergan, and Bausch and Lomb, and her employing institution has received payments from Novartis, Bayer, Neovista, Oraya, Alcon and Pfizer. Chris A Rogers has received an honorarium from Novartis for a lecture. The employing institutions of Susan Downes and Andrew J Lotery have received payments from Novartis. Susan Downes and Andrew J Lotery have received honoraria from Novartis for lectures. Andrew J Lotery has attended and been remunerated for attendance at advisory boards for Novartis and Bayer. Barnaby C Reeves has received a fee for teaching from Janssen-Cilag and is a member of the National Institute of Health Research (NIHR) Health Technology Assessment commissioning board and the NIHR Systematic Reviews Programme Advisory Group. James Raftery is a member of the NIHR Editorial Board and the NIHR Journals Library Editorial Group. He was previously Director of the Wessex Institute and Head of the NIHR Evaluation, Trials and Studies Coordinating Centre.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Chakravarthy et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) 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: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Background and rationale

Aims and objectives

The aim of the IVAN trial was to investigate alternative VEGF inhibition treatment regimens for the treatment of nAMD. We hypothesised that:

• Bevacizumab is not inferior to ranibizumab with respect to the benefits of VEGF inhibition in maintaining/improving VA in eyes with nAMD.

• Treatment with VEGF inhibition can be ‘safely’ withdrawn at 3 months with monthly review to detect reactivation, that is, criteria for restarting treatment can be prespecified to prevent any difference in average VA compared with continuing monthly treatment.

The IVAN trial was designed as a non-inferiority trial because it was intended to show that bevacizumab was not worse than the active control ranibizumab, as existing historical data suggested that both drugs resulted in similar functional outcomes. We did not hypothesise that bevacizumab would be more effective than ranibizumab with respect to VA. It was clearly not practicable to design an equivalence trial. We chose the non-inferiority margin based on evidence of the effect of the active control (namely ranibizumab), as well as existing data from another study. The drug effects are known to wear off over a 4-week period24 and our study design therefore mandated monthly visits for assessment and retreatment decision-making. Evidence from the pivotal trials and subsequent studies had demonstrated that average VA improved steadily from treatment inception over a period of 3 months. To maximise this benefit, we chose to have a loading phase of 3 months, after which we introduced the opportunity to discontinue treatment if the macula was judged to be fluid free.

The trial had three specific inter-related objectives:

• To estimate the:

  • relative effectiveness of two VEGF inhibitors, namely ranibizumab and bevacizumab, on visual outcome in patients with nAMD; existing evidence of the benefit of VEGF inhibitors compared with sham treatment precludes inclusion of a sham VEGF inhibition arm

  • effectiveness of more frequent compared with less frequent VEGF inhibition in improving or maintaining visual function, with stringent criteria for restarting treatment to prevent VA loss in patients receiving less frequent treatment

  • cost-effectiveness of the alternative treatment strategies outlined above.

In addition to estimating the effectiveness of the treatments on visual outcome, the trial was designed to estimate differences in lesion morphology, quality of life and safety as secondary outcomes.

Study design

The IVAN study was a multicentre, randomised controlled factorial trial. The trial is registered as ISRCTN92166560. The research objectives were addressed by randomising participants to one of four combinations of two treatment factors (Table 1). Participants, clinicians and trial personnel were masked to the VEGF inhibitor to which a participant was assigned. Pharmacies dispensed the appropriate drug to the ophthalmic clinic, as a pre-filled syringe for bevacizumab or in the commercially available vial for ranibizumab. We aimed to achieve masking of the drug by using unmasked study teams (ophthalmologists and nurses) who had no role in the outcome assessments in the trial. This method for masking had been used extensively in trials that involve invasive methods of treatment delivery, as it is ethically unjustified to inject a placebo into the eye because of the potential for serious adverse events (SAEs) related to the procedure itself. All assessments and treatment management decisions were made by other trial personnel who were masked to allocation to drug throughout the trial, and to allocation to treatment frequency until after the third treatment. We chose not to mask participants, clinicians and trial personnel to whether or not patients were allocated to continuous (i.e. monthly treatment) or discontinuous treatment (stop at 3 months with reinitiation of therapy if disease reactivation occurred).

Participants

Settings

Patients were recruited to the IVAN trial from 23 ophthalmic units in NHS hospital trusts.

Interventions

After obtaining written informed consent, participants were allocated to one of four combinations of the two treatment factors, intravitreal injections with ranibizumab or bevacizumab, on either ‘continuous’ or ‘discontinuous’ treatment regimens.

Drug doses were ranibizumab 0.5 mg or bevacizumab 1.25 mg. 8,11,25 Ranibizumab and bevacizumab were procured commercially. Bevacizumab was repackaged in pre-filled syringes in an aseptic manufacturing facility in the Liverpool University Hospitals pharmacy. The potency, sterility and stability of the repackaged bevacizumab aliquots were established prior to the start of the study.

The protocol required all participants to attend monthly (28- to 35-day interval) for clinical examination, OCT and fundus photography. All participants were treated at visits 0, 1 and 2. Participants randomised to the continuous regimen were treated monthly thereafter. Participants randomised to the discontinuous regimen were not retreated after visit 2 unless prespecified clinical and OCT criteria for active disease were met. 26 If retreatment was needed, a further cycle of three doses delivered monthly was required.

The criteria for retreatment were divided into three levels arranged hierarchically (Figure 1).

FIGURE 1.

Criteria for treatment failure/restarting treatment.

If any of these criteria were met, participants in the discontinuous treatment arm were considered to have failed stability criteria and treatment initiated. If treatment was reinitiated a further block of three treatments, given at monthly intervals, was mandated (Figure 2).

FIGURE 2.

Treatment over time by treatment regimen. In continuous groups, VEGF inhibitor is administered at each of 24 visits (from baseline, month 0, to month 23). In discontinuous groups, VEGF inhibitor is administered at each of three visits (from baseline, month 0, to month 2). The patient is then reviewed monthly, receiving a further cycle of three treatments (hatching indicates initiation of a further cycle of treatment) if the patient satisfies any of the criteria for ‘treatment failure’ (see Figure 1). Three different possible treatment courses over time are shown for patients in discontinuous groups. Asterisks (month 24) indicate stopping of treatment in the trial; at this time, the need for treatment will be assessed against guidance in the forthcoming National Institute for Health and Care Excellence (NICE) technology appraisal and evidence from the trial. This figure shows only three potential treatment regimens in the discontinuous treatment arm of the study, any combination from ‘no treatment after the initial 3-month cycle’ to ‘continuous treatment from 24 months’ is possible in this arm of the trial.

Outcomes

Changes to study outcomes after commencement of the study

Heart failure was added as a SAE at the request of the DMSC in February 2008, and RPE tear, which is part of the natural history of AMD and had been inadvertently missed, was added as an expected ocular adverse event (AE) in June 2008.

The lesion morphology variables and metrics from angiograms and OCTs were not specified explicitly in the study protocol, but the measures to be compared, their definition and derivation were defined prospectively, before any analyses of these measures were undertaken.

Prompted by the findings from the CATT at 2 years,22 colour and OCT images at baseline and most recent available follow-up were regraded specifically to identify any new GA lesions – in either the study or fellow eye – that had developed during the trial (Figure 3). This outcome was not specified in the study protocol.

FIGURE 3.

New GA assessment diagram.

Sample size

With respect to the primary outcome (BCVA), the trial is designed to answer non-inferiority questions.

Interim analyses

A formal prespecified interim analysis was undertaken when all participants had been followed up to 1 year. 26 No adjustment to the sample size and significance levels were made for this interim analysis.

Randomisation

Randomisation to one of the four treatment arms was stratified by centre, and blocked to ensure approximately equal numbers of participants per group within a centre. Allocations were generated by computer, in advance of starting the trial, by the coordinating centre, and concealed using an internet-based system provided by Sealed Envelope Ltd (London, UK). Staff in participating centres gained limited access to the system using a password. Information to identify a participant uniquely and to confirm eligibility had to be entered before the system assigned a study number and hence the randomised treatment allocation was determined. This study number was then written onto all patient case report forms (CRFs) and also on the prescription for study drug issued to the local pharmacy. The prescription for each participant, together with a printout from the randomisation system, was then sent to the local pharmacy. The local pharmacy staff took the prescription and looked up the prescription log sheet for that participant according to study number. The prescription log, which linked study number to drug allocation, was prepared by Bristol Clinical Trials and Evaluation Unit (CTEU) and sent to the local pharmacy as part of site initiation, before recruitment at the site started.

Investigators and patients were masked to treatment regimen (continuous/discontinuous) until all data for visit 2 had been submitted to Bristol CTEU. CRFs for visit 3 onwards were customised for a participant, depending on the allocated treatment regimen. Participant-specific CRFs folders were made up by the Bristol CTEU and sent to the centre at which a participant was being treated. Stickers were also added to the spine of the CRF folder, identifying the treatment regimen to which a patient had been allocated (green for continuous treatment, red for discontinuous regimen).

Masking

Participants, clinicians and trial personnel were masked to the VEGF inhibitor to which a participant was assigned throughout the trial, and to the allocation of treatment frequency until after the third treatment. We chose not to mask participants, clinicians and trial personnel to the continuous or discontinuous treatment regimen after the first 3 months, on grounds of pragmatism and practicability. We were interested in the effects of discontinuous treatment on treatment satisfaction. Masking treatment regimens would have required sham injections to have been ‘given’ to patients allocated to the discontinuous regimen on visits when treatment was not required.

Local pharmacies dispensed the appropriate drug to the ophthalmic clinic, as a pre-filled syringe for bevacizumab or in the commercially available vial for ranibizumab. We intended that drug allocation should be concealed by having separate masked assessment and unmasked treating teams. The study drug was transported to clinic in a concealed container, which was then opened and the drug prepared by the injecting ophthalmologist, out of sight of other study staff and participants. The injecting ophthalmologist was not involved in any aspect of the study, and the drug allocation was not revealed to the participant or to any other members of the study team. This system was achieved by 14 sites.

At the other nine sites, staffing levels could not support this system and an unmasked staff member (usually the pharmacist or a research nurse) prepared the injection. Ranibizumab was prepared in a syringe identical to those containing bevacizumab, and any drug labels were replaced with a label containing the participant’s IVAN study number. The unmasked staff member presented this to the masked injector, while ensuring that the original packaging and labels remained out of sight. The syringes were supplied by the IVAN trial, obtained from the same source supplying syringes to the manufacturing pharmacy preparing pre-filled syringes of bevacizumab. Those responsible for preparing the injections did not perform assessments.

To assess the adequacy of masking, ophthalmologists and participants stated at visits 3, 12 and 24 (and at exit visits if participants withdrew early), whether or not they knew the allocated drug (don’t know/Lucentis/Avastin).

Lesion morphology was assessed by independent graders, masked to drug and treatment regimen, in the UK Network of Ophthalmic Reading Centres. Because independent assessment of lesions could not be done immediately, some randomised participants were subsequently found to be ineligible.

Data collection

Data collection in the IVAN study was performed over a 2-year period, during which time patients attended 25 monthly visits. Most data were collected from patients using purpose-designed CRFs at these visits (Table 4 shows the schedule of data collection).

Blood samples (1 × 10-ml ethylenediaminetetraacetic acid tube) for genetic analysis were collected at visit 0 (baseline) and sent to the Gift of Sight Research Centre in Southampton for storage and subsequent analysis. Serum samples (2 × 6-ml serum separator tubes) were collected at visits 0, 1, 6, 11, 12, 18, 23 and 24; the samples at visits 6 and 18 were instituted only for participants recruited later in the trial after observing marked changes in mean VEGF levels between visits 1 and 11 in participants recruited earlier in the trial. 39 If the participant did not attend one of these visits, samples collected at the following visit were sent instead. These samples were sent to the Centre for Vision Science at Queen’s University Belfast for serological analysis. All samples were sent using the Royal Mail Safebox system (guaranteed next-day delivery).

Colour photographs and OCTs were captured every 3 months, and FFA was captured at baseline, 12 and 24 months. Images were submitted to the Central Angiographic Resource Facility (CARF) for analysis. If the criteria for treatment failure were met and a decision to restart treatment was made (for participants allocated to the discontinuous treatment regimen) at an intervening visit between the 3-monthly visits, the colour photographs and OCT images captured at the visit were submitted to CARF. The criteria for treatment failure were also applied to participants who were allocated to the continuous treatment regimen, even though the resulting data did not influence the decision to treat (as these participants were treated at every visit); the protocol required the colour photographs and OCT images captured for these participants at these visits also to be submitted to CARF. If a FFA was required to complete the assessment of criteria for retreatment, the FFA was also submitted to CARF. If a patient withdrew from the study during a study visit, an exit visit was undertaken. As part of this exit visit, OCT and FFA were taken and submitted to CARF.

Participants were asked to complete four HRQoL questionnaires and one questionnaire about symptoms since the last visit (to ascertain adverse effects and resource use relating to the symptoms). The EQ-5D, HUI3 and symptoms questionnaire were completed during the study visit. The EQ-5D and HUI3 were administered at visits 0, 3, 12 and 24, and when a SAE had occurred since the previous visit. Collection of participant responses to MacDQoL and MacTSQ questionnaires was coordinated centrally. A member of the coordinating centre study team contacted participants by telephone approximately 2 weeks after visits 3, 12 and 24.

A bespoke trial database was designed using Microsoft SQL Server 2008 (Microsoft Corporation, Redmond, WA, USA). The database was intended to act as both a data storage facility and a trial management resource. For example, the sites scheduled appointments using the database and the database issued reminders when the telephone questionnaires were due. Activity reports were generated based on data entered into the database, enabling the management of payments to sites (see Contractual and financial arrangements). The database also provided a facility for tracking the progress of SAE reporting.

Owing to the large sample size and number of scheduled visits, a considerable amount of data validation was applied to the database. The validation rules were determined as a result of detailed discussions between clinical trial coordinators, research nurses, statisticians and database developers working on the study, and were refined following any feedback from sites. Validation broadly included rules such as ensuring that:

1. the VA was recorded at all required time points and entered correctly; for example, the average value of VA was calculated at each time point and any value entered that deviated significantly from the average was queried to confirm it was correct

2. injections were given in accordance to the trial protocol; for example, a query would be raised if a patient on discontinuous treatment entered a 3-month treatment cycle and then failed to complete it

3. AEs and SAEs were raised when required and that the event was followed through to completion.

Automated links between the IVAN database and the CARF system were set up so that queries could be raised for any missing images and to check that images received at CARF were matched with the correct event.

Statistical methods

Analyses of the primary and secondary efficacy outcomes were carried out on the basis of treatment allocation, which was consistent with the analysis of CATT. 22,23 The analysis population (and the safety population) consisted of all randomised patients who received at least one injection. Patients who were not treated or withdrew and were unwilling for data already collected to be used were excluded. Reporting guidelines recommend that non-inferiority hypotheses and safety data are analysed by the treatment received. 40,41 As IVAN is a masked trial with respect to drug allocation, the drug treatment received should equal the treatment allocated. If the drug received differed by visit (i.e. the wrong treatment was given on one or more occasions), the patient was grouped according to the drug received with greatest frequency. Grouping patients according to the amount of treatment received was not straightforward, so for consistency of reporting, and on the recommendation of the IVAN DMSC, patients were grouped according to the allocated treatment frequency.

Most outcomes were collected for the study eye (e.g. BCVA, other measures of visual function, lesion morphology) but other outcomes were collected for the participant, notably data concerning systemic AEs/SAEs and HRQoL. The unit of analysis for the former was, therefore, study eye, and for the latter it was participant.

Continuous variables were summarised using the mean and SD [or median and interquartile range (IQR) if the distribution is skewed], and categorical data were summarised as a number and percentage. The primary outcome, BCVA, and other continuously scaled outcomes measured at multiple time points, were analysed using linear mixed-effects methods, incorporating parameter estimates for the mean baseline response where measured and for each treatment at each follow-up time (i.e. saturated model). Baseline and subsequent values were modelled jointly to avoid the necessity to exclude or impute values for cases with missing pre-treatment measures. Binary outcomes were compared using logistic regression, adjusted for baseline values where measured, with treatment estimates presented as odds ratios (ORs). Formal statistical comparisons were performed only if at least 10 participants in total experienced the outcome. Time from completion of the loading dose of three treatments to the first treatment failure was analysed using Cox proportional hazards regression, with treatment comparisons presented as hazard ratios (HRs) and described graphically using Kaplan–Meier plots. The Cox model was also adjusted for whether or not the patient had completed the full cycle in the loading phase (i.e. received three injections at visits 0, 1 and 2). Assumptions underpinning the statistical models were checked using standard methods (e.g. residual plots, tests for normality or for proportional hazards). If the assumptions were not satisfied then transformations were explored. Outcomes analysed on a logarithmic scale were transformed back to the original scale after analysis, and results presented as geometric mean ratios (GMRs). Outlying observations that meant models did not fit the data adequately were excluded from analyses.

For EQ-5D, lesion area and MacTSQ at 2 years, no suitable transformation could be found and so data were dichotomized (EQ-5D score, 1 vs. < 1; lesion present vs. absent; MacTSQ < median TSQ score over all time points vs. ≥ median TSQ score over all time points). For MacDQoL, the outcome was transformed from original scale of –9 to +3 to a scale of –3 to +9), and analysed using a log transformation. The GMR is, therefore, interpreted as the GMR of the MacDQoL score, and not of the MacDQoL score directly.

The interaction of VEGF inhibitor and treatment frequency was tested, and differences between ranibizumab and bevacizumab were to be reported separately for the continuous and discontinuous treatment arms only if the interaction term reached statistical significance (two sided) at the 5% level for outcomes for which the results from the CATT22 suggested possible interaction (i.e. OCT measures of total retinal thickness at the fovea and fluid, and presence of fluid on OCT) or at the 1% level for other outcomes (chosen to reduce the type I error rate); however, this level of significance was not reached in any of the models. As the interaction was not statistically significant, the main effects of ranibizumab vs. bevacizumab and of continuous vs. discontinuous treatment after 2 years were reported. Likelihood ratio tests were used in preference to Wald tests for hypothesis testing.

All treatment comparisons at 2 years are presented as effect sizes with 95% CIs. For tests of superiority, two-sided p-values of < 0.05 are considered to be statistically significant. For tests of non-inferiority, bevacizumab was considered inferior to ranibizumab, and discontinuous treatment inferior to continuous treatment if the lower limit of the 95% confidence interval (CI) for the difference between groups was < –3.5 letters (inferiority margin set at 3–4 letters). No formal adjustment was made for multiple testing. When interpreting the results, consideration has been given to the number of statistical tests performed.

The intention was to adjust all models for study centre. However, for some low-frequency outcomes (e.g. safety) it was anticipated that this would not be feasible, so, for consistency, analyses were adjusted for centre size, fitted as a fixed effect (see Appendix 5).

Best corrected distance visual acuity in the study eye, and several of the secondary outcomes, were measured at all visits (see Table 4). However, the analysis of BCVA included only the ‘main’ study visits, namely visits 0, 3, 6, 9, 12, 15, 18, 21 and 24. If the outcome was missing for one of these visits, but was recorded at the next scheduled visit (i.e. 1 month later), this measurement was used in place of the missing value. When this occurred, an indicator was included in the model (and retained, if statistically significant at the 5% level) to denote that the measurement was taken 1 month later than intended.

For BCVA, if a participant could not read any letters from the chart, the following scores were assigned;

• ‘counting fingers’ 0 (which equates to no letters read at 1 m)

• ‘hand movements’ –15

• ‘perception of light’ –30.

The scores of –15 and –30, which equate to a doubling of the visual angle, were chosen to allow the assignment of arbitrary points in deteriorating visual function to these categories.

The first step in assessing reading speed is to choose a Belfast chart that is appropriate for the patient’s near vision. On occasions the incorrect chart was used. When a chart one size larger or smaller than the correct size (based on the recorded NVA) was used, the data were used to calculate the reading index and were included in the analysis. If the chart was more than one size larger or smaller than the correct size, the data were treated as missing and the reading index was not calculated. Participants with a NVA logMAR of 1.6 have vision, which is too poor for their reading ability to be measured, and, therefore, in the majority of cases reading ability was not assessed. The reading index for these participants was imputed between 0.4 and 2.5 on the lognormal scale. This imputation method was chosen based on the distribution of values for participants with a NVA logMAR of 1.6, who did have their reading ability assessed.

Several morphological outcomes were derived from colour, FFA and/or OCT measurements. Details of how these measures were derived are described in Appendix 5. For all patient-reported outcome data, standard rules have been used to derive outcome measures (see Appendix 5). EQ-5D scores are reported in two sections: the first is the EQ-5D utility index, derived from responses to the five ordinal questions, and the second is the visual analogue (‘thermometer’) scale. Using the MacDQoL questionnaire, an average weighted impact score is derived from 22 out of the 23 questions (the question regarding work was excluded, as it was irrelevant to the majority of participants). The MacTSQ questionnaire was used to derive a single treatment satisfaction score comprising 12 questions.

Pre-planned subgroup analyses were defined as follows:

1. Baseline VA in study eye (< 55 vs. ≥ 55 letters read).

2. Baseline CNV size (< 6 vs. ≥ 6 disc areas).

3. Proportion of classic CNV (< 50% vs. ≥ 50%).

4. Presence of retinal angiomatous proliferation (RAP).

5. Fellow eye status (< 75 vs. ≥ 75 letters read). Differences between subgroups were tested by adding subgroup by treatment interaction terms to the model.

Two pre-planned sensitivity analyses of the primary outcome were carried out. First, we excluded measurements taken 1 month later, when the main study visit was missed. Second, we included data only for the study visits at which all functional outcomes were assessed (visits 0, 3, 6, 12, 18 and 24).

In all tables missing data are indicated by footnotes. Imputation for missing data, except as described above, was not used, as few data were < 20% as prespecified in the statistical analysis plan (SAP).

Fixed-effects meta-analyses were performed to combine the results from the IVAN trial with other head-to-head trials comparing bevacizumab and ranibizumab. There was no intention to perform a systematic review. These meta-analyses were carried out to place the IVAN trial findings in the context of the wider evidence base. Other head-to-head trials – namely the CATT,22 GEFAL (French Evaluation Group Avastin versus Lucenti),42 MANTA (Multicenter ANti-VEGF Trial in Austria),43 LUCAS (LUcentis Compared to Avastin Study), Subramanian et al. ,44 and BRAMD [comparison of bevacizumab (Avastin) and ranibizumab (Lucentis) in exudative age-related macular degeneration; unpublished data, presented at the 2013 meeting of European Society of Retina Specialists] trials – were identified from non-systematic searches during the course of the IVAN trial, through contacts with chief investigators of these trials and expert knowledge of the ophthalmologists in the research team.

Fixed-effects models were used because we judged that the populations (people/eyes with nAMD), the settings (specialist ophthalmology outpatient clinics) and the interventions (ranibizumab and bevacizumab) being studied across trials were relatively homogeneous. This was not the case for comparisons of treatment regimen, but, as only the IVAN trial and CATT22 contributed to these analyses, the fixed-/random-effects distinction is irrelevant. No sensitivity or subgroup analyses were planned within the meta-analyses.

The availability of outcome data differed across the trials. Where available, 2-year data have been used, otherwise data up to the 1-year point have been included. The following meta-analyses were performed: change in BCVA from baseline, safety outcomes (mortality, ATEs, one or more SAE and gastrointestinal disorders), change in total retinal thickness at the fovea from baseline, and new GA. Safety outcomes by treatment regimen were available only for the CATT22 up to 1 year (after this time point, participants allocated to monthly treatment in year 1 were rerandomised to monthly or pro re naba (prn) treatment in year 2).

All statistical models were fitted in SAS version 9.3 (SAS Institute Inc., Cary, NC, USA). All other analyses and data management were performed in Stata version 12.0 (StataCorp LP, College Station, TX, USA).

Health economics

Measurement of patient-reported health status and quality adjusted life-years

Two multiattribute utility measures were used in the IVAN trial: the three-level EQ-5D56,60,61 and HUI3. 33,62,63 EQ-5D was the main utility measure for the economic evaluation, as it has several advantages over HUI3. First, the EQ-5D tariff is based on ‘time trade-off’ valuations by around 3000 members of the UK general population. 60,64 By contrast, the HUI3 value set is based on a mixture of visual analogue and ‘standard gamble’ valuations by 256 members of the Canadian general population. 33,63 The HUI3 tariff is, therefore, less precise and less relevant to a UK setting than EQ-5D. Second, EQ-5D is recommended by NICE45 and is used more widely than HUI3. 65 ICERs calculated using the EQ-5D can, therefore, be directly compared with ICERs calculated in a large number of other UK economic evaluations to help decision-makers ensure that the most cost-effective treatments are provided. However, HUI3 was used in sensitivity analyses, as it includes questions specifically relating to vision and may be more sensitive or responsive to changes in eye disease than EQ-5D. 66,67

The EQ-5D and HUI3 were both completed at 0, 3, 12 and 24 months. In addition, to assess the impact of SAEs on utility, both instruments were administered at the next assessment attendance after any SAE or a reduction in BCVA ≥ 15 letters. EQ-5D and HUI3 were also completed at/after withdrawal if the patient opted to attend a full exit assessment. For EQ-5D, the published UK time trade-off valuation tariff was used to value each health state64 and calculate ‘utilities’. The Canadian valuation tariff set33,68,69 was used for HUI3. Missing utility data were imputed by using multiple imputation (see below).

Quality adjusted life-years were calculated as the area under the utility curve, i.e. EQ-5D utility multiplied by length of time spent at that utility. In the absence of SAEs, utility was assumed to change linearly between baseline and 3 months, between 3 and 12 months, and between 12 and 24 months (Figure 4), which is supported by the trends for BCVA (see Figure 15).

FIGURE 4.

How QALYs were calculated. (a) In the absence of SAEs, EQ-5D utility was assumed to change linearly between EQ-5D measurements. (b) As the EQ-5D measurement after this patient’s first set of SAEs is higher than the line joining the baseline and 3-month measurements, we assumed that utility rose linearly between baseline and the post-SAE measurement and between this measurement and 3 months. As utility is lower after SAE 2, we draw a line through the post-SAE 2 measurement with a gradient equal to the recovery rate estimated in the mixed model. This line was used to estimate the utility on the day SAE 2 starts, and the time and the utility at which the patient is expected to have recovered from SAE 2 and returned to the utility trend observed between visits three and 12. The patient died 5 days after SAE 3; the utility was, therefore, assumed to follow the linear trend observed between visit 12 and the value imputed at visit 24 up until the day before SAE 3. Utility was assumed then to fall linearly to 0 over the last 5 days of life.

As many patients had several different SAEs starting in a short space of time (e.g. angina, MI and bypass graft in the same admission), SAEs occurring less than 1 week apart were grouped into a set associated with a single post-SAE measurement and with a single onset date equal to the onset date of the first SAE in that set (see Appendix 4). Of the 183 trial participants with SAEs starting before they left the trial or attended visit 24, 129 had one set of SAEs, 37 had two, 15 had three and 2 had four.

For patients who experienced a SAE that reduced utility, the base-case analysis assumed that utility fell on the day of the SAE and rose linearly afterwards. Similar profiles have been used previously to model recovery from acute hepatitis70 and chronic obstructive pulmonary disease exacerbations. 71 We used linear profiles to simplify subsequent calculations and because models with quadratic terms did not fit as well. As post-SAE utility measurements were taken, on average, 56 (range 0–182) days after a SAE, we used mixed models to estimate the rate at which utility rose after each type of SAE (see Appendix 4). Coefficients estimated in the mixed models were used to estimate utility on the day of the SAE and identify the point at which utility returned to the level that would be expected from the utility measurements that were not taken after SAEs (see Figure 4). However, some post-SAE measurements were higher than would have been expected from other measurements for the patient (e.g. see Figure 4, SAE 1); in these cases, we assumed that utility changed linearly between the routine measurements (see Figure 4). For patients dying 1–7 days after the latest SAE, utility was assumed to fall linearly to 0 between the date of the SAE and the date of death. Full details of the methods are described in Appendix 4.

Patient and public involvement

The IVAN trial continued a long-standing association with the Macular Society (previously the Macular Disease Society). The Macular Society is ‘the national UK charity for anyone affected by central vision loss’ (www.macularsociety.org/How-we-help/About-us). Although the charity raises funding from corporate sponsors, it receives no government funding or donations from pharmaceutical companies. Professor Chakravarthy and other members of the research team approached the Macular Society in 2003 in relation to the VPDT for the UK. As the IVAN trial grew out of the VPDT, it was natural to continue the collaboration when we were seeking funding for the IVAN trial. The collaboration continued once funding was awarded, when the trial was being set up and throughout its conduct through the Society’s membership of the TSC. Mr Bremridge, the Society’s chief executive at the time, was a member of the TSC at inception. Following his retirement, Helen Jackman (the current chief executive) or Cathy Yelf, the Society’s head of communications and external relations, have represented the Society on the TSC.

We consulted with the Macular Society, directly with Tom Bremridge, Helen Jackman and Cathy Yelf, but also (through the Society) with some of their members, on several aspects of the trial:

1. the need for such a trial

2. the design of the trial

3. information given to patients at the outset who were being approached about the trial (construction of the patient information leaflet that was approved by a NHS research ethics committee) and subsequent revisions to this information

4. information about the first-year findings of the IVAN trial and those of the CATT35

5. information about possible risks that emerged during the trial and which were communicated to participants who had not completed the study

6. on study completion, letters to participants thanking them for their involvement and unmasking them to the drug assignment.

The advice of the Macular Society was most important with respect to (a), (b) and (d). The decision to use a discontinuous treatment regimen requiring cycles of three injections given monthly was substantially based on discussions with Mr Bremridge about the need to minimise the risk of participating in the trial. Mr Bremridge also advised in detail about how to describe the possible risks and benefits of treatment with each drug.

It was vital to have views about how to communicate the trial objectives fairly and to describe what the trial involved in the patient information leaflet. In this context, it should be remembered that the first few months of recruitment to the trial straddled NICE’s publication of its final appraisal determination for ranibizumab; ranibizumab was not widely available at the time that the trial started to recruit and, once ranibizumab was recommended by NICE, the possibility of being treated with bevacizumab was extremely controversial.

In the spring of 2012, the chairperson of the DMSC asked the trial statistician to combine the most up-to-date information from the IVAN trial with information available from the CATT. 35 This synthesis found a statistically significant increased risk of having a systemic SAE with bevacizumab compared with ranibizumab. At the next meeting, the committee reviewed the information and recommended that the trial should continue but the information should be disseminated to the minority of patients who were still continuing in the trial. This recommendation was discussed at length, during the following TSC, with Cathy Yelf, who concurred. The research team drafted a letter to patients (describing the information, the recommendation and giving the opportunity to withdraw) and sought edits and comments on the content from the Macular Society before finalising it. Given the sensitive nature of the information, it is testament to the quality of the patient and public involvement in the IVAN trial that none of the participants remaining in the trial asked to withdraw.

Management of investigational medicinal product

All drugs were purchased centrally by the Liverpool pharmacy, that is standard commercially available ranibizumab 0.23-ml vials and bevacizumab 4-ml vials. Ranibizumab vials as commercially supplied were labelled for use as investigational medicinal product (IMP) in the trial. Pre-filled bevacizumab syringes were manufactured in the aseptic manufacturing facility, packaged in a box and also labelled for use as IMP in the trial. Both drugs were distributed by courier to each centre. Drugs were dispatched bearing study labels approved by the Medicines and Healthcare products Regulatory Agency (MHRA). Each ranibizumab vial ordered and each bevacizumab syringe manufactured by the Liverpool pharmacy was logged on the appropriate product log sheet. The fate of each dose was recorded against each batch (e.g. which hospital they were dispatched to, used for quality control, etc.).

The central coordinating centre managed drug stock at each site. When drug supplies were required by a site, the coordinating centre communicated to the Liverpool pharmacy the drug type, quantity and destination to which they should be dispatched. The product log was completed, stating to which hospital the drug was to be dispatched and the date of dispatch. Drugs were packed for dispatch with cool packs to maintain a temperature of 2 °C to 8 °C. Packs included a temperature monitor to allow temperature breaches during transit to be identified. The Liverpool pharmacy confirmed that drugs had been dispatched via the trial database.

Each batch of study drug received at a site was logged in the IVAN study drug receipt log. On arrival the package was checked for temperature breaches and/or damage to study drugs; if either had occurred, the coordinating centre was contacted to arrange redelivery. All shipments were transferred immediately to a fridge (2 °C to 8 °C) for storage.

Any study medication that exceeded its expiry date before being used was logged and kept until it was checked by IVAN monitors. Expired drugs were recorded on the drug receipt log by pharmacies at local sites; if drugs expired prior to dispatch to site, this was recorded on the appropriate product log sheet by the Liverpool pharmacy.

On receipt of a study prescription, the correct drug was taken from a fridge and the expiry date checked to ensure that the drug was within shelf life. The IVAN study number and visit number were completed on the study label on the outside packaging of the study drug. The prescription log was completed, detailing the expiry date and batch number for each patient and visit number. The drug was concealed in an unmarked bag ready for collection by clinic staff/delivery to clinic.

Used vials and syringes were returned from clinic to pharmacy at the end of each clinic and checked and stored until after monitoring had taken place and accountability completed, when instructions to destroy them would be issued by the trial monitoring team. The trial monitoring team inspected returned packaging to check that the correct drug had been dispensed and that the drug had been given (i.e. that the vial/syringe was empty).

Contractual and financial arrangements

At the time funding was being sought and the trial was being set up, the costs of the trial had to be attributed as research, service support and treatment (including excess treatment) costs (www.gov.uk/government/news/attributing-the-costs-of-health-social-care-research-development-acord). The task was complex for the IVAN trial because trial set-up and initial recruitment spanned a period of time when (1) ranibizumab had not been recommended for use in the NHS91 and (2) the dispersion of service support costs to participating hospitals through the Comprehensive Research Network had not been completely implemented. Consequently, the attribution of the costs of the trial changed dramatically before and after NICE issued guidance.

Before NICE issued guidance, treatment with ranibizumab represented an estimated excess treatment cost of £15,000–30,000 per patient over the 2-year duration of follow-up (depending on whether a patient was allocated to continuous or discontinuous treatment); the corresponding range for bevacizumab was estimated to be £6000–7000 per patient. After NICE issued guidance, discontinuous ranibizumab approximately represented ‘usual care’, continuous ranibizumab represented an excess treatment cost of £15,000 per patient and bevacizumab a saving in treatment cost of about £8000–9000.

Service support costs (additional tests and other NHS resources used in the course of the trial that were related to the patient’s care but which were not required purely for the research) were estimated to represent a further £2830 per patient over 2 years. These costs were unaffected by the NICE guidance but still had to be reimbursed to hospitals to cover the full cost of the hospitals taking part.

The previous experience of several members of the research team including the chief investigator in evaluating an earlier technology for nAMD (VPDT)92 meant that, in the IVAN trial, the research team was determined to ensure that all costs for the trial were truly in place, namely research, service support and treatment costs, before starting to recruit. The team was also committed to ensuring that reimbursement to hospitals for trial activity related to reviewing and treating trial participants was conditional on carrying out the trial according to the protocol and providing valid data; these two conditions were written into contracts between the sponsor and participating hospitals.

Several unusual features of the trial followed from these decisions:

• There was a need to put in place service level agreements with commissioning organisations (63 primary care trusts in England and four health boards in Northern Ireland; some agreements were negotiated with consortia of commissioning organisations) to provide the treatment costs (as, at the time, these were excess treatment costs). These agreements required ‘up front’ payment (at the start of the financial year) to enable trial activity to happen.

• A mechanism had to be established for treatment costs to be paid to hospitals separately from the usual contracts with commissioners, to ensure that the funding for treatment in trial ‘followed the patient’, that is it was made available to the hospital ophthalmology department delivering the trial.

• A mechanism had to be established to ensure that hospitals did not ‘double count’ activity relating to trial participants, that is, charge their treatment to the trial and also to commissioners under the usual contract for nAMD services. This mechanism ensured that ‘savings’ from using bevacizumab were truly reflected in the funding commissioners paid for treatment of trial participants after ranibizumab was recommended by NICE.

• There was a need to establish a ‘bank’ to manage service level agreements with primary care trusts, hold their payments in advance on credit against future trial-related activity, and manage reimbursement for trial activity by participating trusts.

These features and requirements were further complicated by:

• The requirement for the IMP to be provided from a single, central, manufacturing pharmacy, which necessitated splitting the clinic/hospital costs from the drug costs. Contracts with participating hospitals to reimburse for trial activity were restricted to the clinic/hospital costs, with the IMP being provided to hospitals ‘free’ for use in the trial. The bank separately contracted with the central manufacturing pharmacy to supply the IMP to participating trusts. The central, manufacturing pharmacy was unwilling to bear the uncertainty of the costs of cold-chain transport (which were expensive, with cost per dose dependent on the number of doses being transported) in this contract, so the bank also had to pay separately for the transport costs.

• The fact that an element of the clinic/hospital costs represented research costs, which required the grant holder [Queen’s University Belfast (QUB)] to pay this sum to the bank, to be passed on to participating hospitals pro rata to trial activity.

• A research partnership consisting of four universities and two NHS trusts who were collaborating to deliver the research. The grant holder (QUB) was responsible for establishing non-commercial contracts with each partner, defining the role and responsibilities of each partner and the proportion of the research costs the partner would receive in return.

The NHS sponsor for the trial (Belfast Health and Social Care Trust) agreed to take on the risks and responsibilities (to administer income from primary care trusts and expenditure in relation to trial activity) associated with being the bank, in addition to its responsibilities for research governance, trial oversight and contracts with participating hospitals. The trial benefited from advance, in-depth discussions with the National Specialised Commissioning Group about the likelihood of NICE recommending ranibizumab and the future potential benefit to commissioners of having evidence about the cost-effectiveness of bevacizumab. These discussions were instrumental in persuading commissioning organisations to sign the service-level agreements required. The trial also benefited from the services of some very committed individuals involved in commissioning when drafting the service-level agreement and in negotiating these agreements on behalf of the trial with several primary care trusts. The contracting and monitoring arrangements, and the flows of finances and data in the trial, are shown in Figures 5–7.

The importance of this infrastructure to the ultimate success of the trial cannot be overemphasised. The trial team extends its gratitude in equal measure to the individuals and organisations that made the trial possible.

FIGURE 5.

Contracts and monitoring. TCC, Trial Coordinating Centre.

FIGURE 6.

Finance flow chart. HTA, Health Technology Assessment; ICH-GCP, International Conference on Harmonisation Good Clinical Practice guidelines.

FIGURE 7.

Data flow to and from the Trial Coordinating Centre (TCC). The image grading centre is shown for completeness. It does not appear in Figures 5 and 6 because the centre was located in Belfast, and financing and contractual arrangements were managed locally.

Screened patients

Patients were prescreened to exclude those with no interest in the trial; because of the different ways in which centres did this, data describing all patients approached were not available. Screening data were provided for 693 patients. Of these, 65 patients were excluded; 28 were found to be ineligible (reasons include BCVA of < 25 letters and presence of other ocular disease causing concurrent vision loss) (Figure 8) and the other 37 were excluded for unknown reasons. The remaining 628 patients were randomised into the trial.

FIGURE 8.

Recruitment rate. Recruitment is cumulative. Only participants included in the IVAN cohort are represented on this graph.

Recruitment

Patients were recruited to the IVAN trial between 27 March 2008 and 15 October 2010, and the last patient follow-up visit was on 7 November 2012. The 628 randomised participants were recruited from 23 centres (see Appendix 1 for details). Five of these 628 participants were randomised in error and a further 13 were not treated, leaving 610 cases who received at least one injection. These 610 participants make up the IVAN study cohort. Recruitment was steady throughout the course of the trial (Figure 9). The number of participants in the IVAN cohort recruited at each of the centres is shown in Figure 10.

FIGURE 9.

Number of participants recruited by centre. Only participants included in the IVAN cohort are represented on this graph.

FIGURE 10.

Flow of participants. a, More patients are likely to have been considered for the trial, and the exclusions section is therefore likely to be incomplete, because some sites did not enter full screening data. b, Some patients may be ineligible for more than one reason. c, Details of other withdrawals (i.e. patients that withdrew but data collection will still continue) will be given as footnotes. d, Of the patients who did not drop out, not all of them completed all three treatments. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Patient withdrawals

There were 60 participants who withdrew from follow-up after their first injection and before visit 24, that is the scheduled end of the trial. The rate of withdrawal was constant over the first 18 months of the follow-up period (Figure 11). The most common reason for withdrawal was that the patient was too ill to attend (participant-reported reason) or had experienced a SAE (clinician-reported reason) (Table 8).

FIGURE 11.

Time from randomisation to withdrawal (excluding deaths).

Protocol deviations

Protocol deviations in the IVAN trial included visits on which a participant did not receive the allocated drug, non-adherence to the treatment regimen, missing visits and visits outside the 28- to 35-day window. Table 9 shows the number of times each deviation occurred, and in how many participants. The wrong study drug was administered on 2 out of 12,761 follow-up visits. The allocated treatment regimen was not adhered to on 133 out of 12,761 visits (1.0%) (see Table 9). Further details on protocol deviations are given in Appendix 3. Overall, 57% of participants missed at least one visit, excluding visits missed because a participant died or withdrew early.

Patient follow-up

Of the 610 study participants, 525 completed the trial. Overall, 87% of all scheduled visits were attended (12,761 out of 14,640). Over 60% of patients attended 23 or more visits and 37% attended all 25 visits. Of the 525 participants who completed the trial, 518 attended the 24-month final visit; the other seven did not attend the 24-month visit, but did not withdraw, and survived beyond the scheduled 24-month visit date. The pattern of missed visits was similar for the two drugs and two treatment regimens (Figure 12). The distribution of number of visits attended, by drug and treatment frequency, is shown in Figure 13. Appendix 3 shows the number of participants at each centre, along with the breakdown of the number of visits attended, the number of injections administered and the number of missed visits at each centre.

FIGURE 12.

Distribution of the number of missed visits per visit (a) by drug and (b) by treatment regimen. Note that all patients attended visit 0, as this was a requirement for inclusion in the IVAN cohort.

FIGURE 13.

Distribution of the number of visits attended per participant (a) by drug and (b) by treatment regimen. The minimum number of visits attended was 1 and the maximum was 25. This does not include exit visits.

Numbers analysed

A total of 610 participants in the analysis population were included in tables of demographic characteristics, analyses of the primary outcome and secondary outcomes measured at multiple time points, time to treatment failure and of safety outcomes. The 525 participants who were classified as completing the trial (with BCVA at 2 years) are included in the descriptive outcome tables. Of the 525, 518 attended the month 24 exit visit; the remaining seven missed the month 24 exit visit (see Patient follow-up, above).

The 90 participants who withdrew (60 withdrawals) or who died (30 deaths) are reconciled with the 525 with data at 2 years as follows:

• One participant was recorded as having withdrawn, who then died within 105 days; this participant was classified both as a withdrawal and a death.

• Two participants completed visit 24 and died within 105 days; these participants are included in the 2-year cohort in the summary outcome tables and are also classified as deaths.

• Two participants missed visit 24 (but did not withdraw: see above) and died within 105 days of what would have been their visit-24 date; these participants are also included in the 2-year cohort and are classified as deaths.

Baseline data

Baseline characteristics of the trial cohort by drug and treatment regimen are shown in Table 10. The mean age was 77.7 years (SD 7.4 years) and 244 (40%) patients were male. Many (64%) were current or past smokers and almost one-fifth had a history of dyspnoea (19%). Characteristics were similar between the two groups, although slightly more participants in the bevacizumab group than the ranibizumab group had a history of angina (17% vs. 11%). Baseline visual function is described in Chapter 4. Baseline characteristics of nAMD lesions are described in Chapter 5.

Treatment received

Over 10,000 injections were given and, overall, participants received a median of 18 injections during the trial (IQR 12–23 injections). The number of injections received was similar by drug, the ranibizumab group receiving a median of 18 injections (IQR 11–23 injections) and the bevacizumab group receiving a median of 19 injections (IQR 12–23 injections). As anticipated, participants randomised to continuous treatment received a higher number of injections [median 23 injections (IQR 21–24 injections)] than those randomised to the discontinuous regimen [median 13 injections (IQR 8–17 injections); Figure 14].

FIGURE 14.

Distribution of the number of injections received per participant (a) by drug and (b) by treatment regimen. The minimum number of injections given was 1 and the maximum was 24 (patients were not given an injection at the 24-month visit). This does not include exit visits.

Success of masking

Regarding adequacy of masking, ophthalmologists reported not knowing which drug participants were receiving on 97.9% (555 out of 567) of visits 3, 98.7% (514 out of 521) of visits 12, 98.8% (506 out of 512) of visits 24, and 100% (22 out of 22) of exit visits. At visit 3, 4 out of 12 ophthalmologists who reported the treatment that they thought the patient received were incorrect, with 2 out of 7 incorrect and 2 out of 5 incorrect at visits 12 and 24, respectively.

Participants reported not knowing which drug they were receiving on 99.3% (560 out of 564) of visits 3, 98.7% (509 out of 516) of visits 12, 97.8% (499 out of 510) of visits 24, and 100% (21 out of 21) of exit visits. At visit 3, 2 out of 4 participants who reported the treatment that they thought they were receiving were incorrect, with 4 out of 7 incorrect and 5 out of 11 incorrect at visits 12 and 24, respectively.

Drug accountability and unmasking

Best corrected distance visual acuity

The mean baseline BCVA in the study eye was 61.4 letters (SD 15.3 letters). The BCVA was similar for ranibizumab and bevacizumab groups (mean 61.1 vs. 61.8 letters), but the average BCVA was almost 3 letters higher in the discontinuous group than in the continuous group (62.9 vs. 60.1 letters) (Table 11). For the majority of participants the study eye had poorer vision than the fellow eye at recruitment (see Table 11).

The BCVA in the study eye by treatment group and visit is shown in Figure 15. The mean BCVA at 1 year was 69.1 and 66.2 letters in ranibizumab and bevacizumab groups, respectively, and 66.9 and 68.5 in continuous and discontinuous groups. On average, there was little change in BCVA from 1 to 2 years; the mean BCVA at 2 years was 67.8 and 66.1 letters in ranibizumab and bevacizumab groups, respectively, and 66.6 and 67.3 letters in continuous and discontinuous groups. Change in BCVA in the study eye from baseline by treatment group and visit is shown in Figure 16. Bands of VA at baseline and 2 years by treatment group are shown in Appendix 3.

FIGURE 15.

Best corrected distance visual acuity in the study eye by visit (a) by drug and (b) by treatment regimen. The circles indicate the mean and the bars indicate one SD either side of the mean. The numbers in parentheses are the number of observations. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

FIGURE 16.

Change from baseline in BCVA in the study eye by visit (a) by drug and (b) by treatment regimen.

At 1 year, the estimated difference between drugs (bevacizumab minus ranibizumab, estimated from the mixed regression model) was –1.99 letters (95% CI –4.04 to +0.06 letters; p = 0.056) and between treatment regimens (discontinuous minus continuous) –0.35 letters (95% CI –2.40 to +1.70 letters; p = 0.74). 26 The comparison by drug was inconclusive; bevacizumab was neither inferior (because the CI includes zero) nor non-inferior (because the CI includes the non-inferiority margin) to ranibizumab using the 3.5-letter limit, but discontinuous treatment was equivalent to continuous treatment. At the primary outcome point of 2 years, mean BCVA did not differ significantly either by drug or by regimen. The difference between drugs (bevacizumab minus ranibizumab) was –1.37 letters (95% CI –3.75 to +1.01 letters; p = 0.26) and between treatment regimens (discontinuous minus continuous) –1.63 letters (95% CI – 4.01 to +0.75 letters: p = 0.18). Bevacizumab was neither inferior nor non-inferior to ranibizumab, and discontinuous was neither inferior nor non-inferior to continuous treatment, using the 3.5-letter non-inferiority margin (Figure 17).

FIGURE 17.

Difference in BCVA at 2 years. Negative values reflect a better mean VA in the ranibizumab or continuous groups. CIs extending beyond the non-inferiority limit of –3.5 letters indicate that the comparison of the two groups is inconclusive (ranibizumab vs. bevacizumab and continuous vs. discontinuous). Difference is estimated using data from visits 0, 3, 6, 9, 12, 15, 18, 21 and 24 adjusted for centre size. Circles = MD. 95% CIs are illustrated by horizontal bars. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

The sensitivity analyses outlined in Chapter 2 (see Statistical methods) gave results consistent with the main analysis.

In the analysis restricting visits to those during which other functional outcomes were assessed, the mean difference (MD) for the drug effect was estimated at –1.70 letters (95% CI –4.08 to +0.68 letters; p = 0.16). The analysis excluding deferred visits gave an estimate of –1.35 letters (95% CI –3.73 to +1.03 letters; p = 0.26). The corresponding estimated effects for the treatment regimen comparison are –1.59 letters (95% CI –3.97 to +0.79 letters; p = 0.19) for the sensitivity analysis restricting the visits included, and –1.60 letters (95% CI –3.98 to +0.78 letters; p = 0.19) for the analysis excluding deferred visits.

Secondary measures of visual function

Visual function at baseline was similar for the two drug groups, but, as was observed for BCVA, the discontinuous group had slightly better reading ability compared with the continuous group (40.0 vs. 33.1; Table 12).

Near visual acuity in the study eye by treatment group and visit is shown in Figure 18. Corresponding data for reading index and contrast sensitivity are shown in Figures 19 and 20. Binocular vision is shown in Appendix 3. At 1 year, NVA was 8% worse in the bevacizumab group (GMR 0.92, 95% CI 0.84 to 1.00; p = 0.058) but did not differ by treatment regimen. 26 By the end of year 2, the average difference by drug was no longer significant (GMR 0.94, 95% CI 0.85 to 1.04; p = 0.23). However, NVA was better when the treatment had been administered monthly for 2 years (GMR 0.90, 95% CI 0.82 to 0.99; p = 0.04). The reading index which takes into account the reading speed and the NVA threshold (by adjusting for size of the print being read) was, however, similar between both drug and regimen groups (see Table 12). Contrast sensitivity did not differ significantly by either drug or regimen groups at 1 year. However, although the drugs were similar, continuous treatment resulted in better outcomes at 2 years (MD –1.07, 95% CI –1.90 to –0.25; p = 0.011; Figure 21).

FIGURE 18.

Near visual acuity in the study eye by visit (a) by drug and (b) by treatment regimen. The direction of the logMAR scale for NVA is opposite to that for BCVA (letters), i.e. small logMAR values represent better vision. The circles indicate the median and the bars indicate the IQR. The numbers in parentheses are the number of observations.

FIGURE 19.

Reading index in the study eye by visit (a) by drug and (b) by treatment regimen. The circles indicate the median and the bars indicate the IQR. The numbers in parentheses are the number of observations.

FIGURE 20.

Contrast sensitivity in the study eye by visit (a) by drug and (b) by treatment regimen. The circles indicate the mean and the bars indicate one SD either side of the mean. The numbers in parentheses are the number of observations.

FIGURE 21.

Difference in secondary measures of visual function at 2 years (a) by drug and (b) by treatment regimen. Negative values of MD and GMR values of < 1 reflect a better outcome in the ranibizumab or continuous groups. Difference is estimated using data from visits 0, 3, 6, 12, 18 and 24 adjusted for centre size. Circles = MD or GMR. 95% CIs are illustrated by horizontal bars. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Treatment failure

Overall, 475 (78%) participants met the criteria for retreatment (treatment failure) during their participation in the trial. The median time from randomisation to first treatment failure was 5.1 months (IQR 3.7–16.8 months) in the ranibizumab group and 4.9 months (3.2 to 14.0 months) in the bevacizumab group. For those randomised to continuous treatment the median time to first treatment failure was 7.6 months (lower quartile 3.2 months) compared with 4.4 months in the discontinuous group (IQR 3.2–6.9 months; Figure 22).

FIGURE 22.

Time to first treatment failure (a) by drug and (b) by treatment regimen.

Comparing the time to first treatment failure from visit 2 (end of loading dose phase of the trial), no difference in the risk of treatment failure was found between the drugs [HR (bevacizumab–ranibizumab) 1.13, 95% CI 0.94 to 1.36; p = 0.18). In contrast, the risk of treatment failure was significantly higher for participants randomised to discontinuous treatment rather than continuous treatment [HR (discontinuous–continuous) 1.95, 95% CI 1.61 to 2.35; p < 0.001].

Subgroup analyses

Differences in BCVA by drug and treatment regimen in the five pre-planned subgroups are shown in Figure 23. No statistically significant differences by subgroup were found for the drug or treatment regimen comparisons, that is all tests for interaction were not significant (p ≥ 0.26). Two additional subgroups, which were not specified in the protocol, were also considered: (1) whether or not the vision in study eye was at least 5 letters better than in the fellow eye at baseline and (2) whether or not a haemorrhage was present at baseline. No statistically significant differences between these subgroups were found for the drug or treatment regimen comparisons (see Appendix 3).

FIGURE 23.

Best corrected distance visual acuity in the study eye at 2 years by subgroup (a) by drug and (b) by treatment regimen. Negative values of MD reflect a better mean VA in the ranibizumab or continuous groups. Difference is estimated using data from visits 0, 3, 6, 9, 12, 15, 18, 21 and 24, adjusted for centre size. Circles = MD. 95% CIs are illustrated by horizontal bars. Circle size is relative to the size of the subgroup.

Meta-analysis

As described in Chapter 2 (see Statistical methods), meta-analyses were carried out to place the IVAN trial findings in the context of the wider evidence base. There was no intention to carry out a formal systematic review and findings from these analyses are, therefore, not reported in line with PRISMA guidelines.

The meta-analysis of BCVA was complicated by the fact that triallists have used different ways to analyse BCVA data and to report their findings. With the aim of being as inclusive as possible, we have performed a single meta-analysis including trials that had 1- or 2-year follow-up, using the findings for the longest duration (as BCVA changes very little between 1 and 2 years). We used results from presentations at international conferences when final journal publications were not available.

Figure 24 shows meta-analyses of the trial findings for seven trials [including IVAN, CATT, BRAMD (presented at the 2013 meeting of EURETINA) GEFAL, LUCAS (presented at the 2013 meeting of the American Association for Ophthalmology), MANTA and Subramanian et al. ]22,42–44 for changes in distance BCVA by drug and of the CATT22 and IVAN trial findings for changes in distance BCVA by treatment regimen. The analyses use the 2-year findings for the CATT22 and IVAN, and 1-year findings for all other studies. The pooled BCVA point estimate shows that bevacizumab was statistically non-inferior to ranibizumab judged by the stricter IVAN margin [–0.38 letters; 95% CI –1.47 to +0.70 letters; p = 0.49; heterogeneity chi-squared (χ²) = 4.08; degrees of freedom (df) = 6; p = 0.666; I² = 0.0%]. Acknowledging the slight difference between the CATT22 and IVAN trial in their ‘as needed’ treatment regimens, discontinuous treatment was significantly inferior to continuous (–2.23 letters; 95% CI –3.93 to –0.53 letters; p = 0.010).

FIGURE 24.

Meta-analysis of change in BCVA from baseline (a) by drug and (b) by treatment regimen. Change in BCVA = 2 years minus baseline or 1 year minus baseline, as indicated in the figures. Negative values for differences reflect better outcomes in the ranibizumab or continuous groups. WMD, weighted mean difference. 95% CIs are given in parentheses and illustrated by horizontal bars. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Summary

The visual function at baseline was similar by drug, but the discontinuous group had slightly better BCVA and reading ability than the continuous group. For the majority of participants their vision was poorer in their study eye than their fellow eye. At the primary outcome point of 2 years, the BCVA was, on average, 1.37 letters worse with bevacizumab than ranibizumab, and 1.63 letters worse with discontinuous than continuous treatment. However, bevacizumab was neither inferior nor non-inferior to ranibizumab, and discontinuous was neither inferior nor non-inferior to continuous treatment, using the 3.5-letter non-inferiority margin. At 2 years, NVA and contrast sensitivity were similar by drug, but better in the group receiving continuous treatment. Reading index was similar by drug and treatment regimen at 2 years. Time to first treatment failure was similar for the two drugs, but the risk of failure was significantly higher with discontinuous treatment than with continuous treatment. No differences by subgroup were found. Meta-analyses of the IVAN data with other available trials comparing bevacizumab with ranibizumab show that acuity outcomes are equivalent for the two drugs, and that the confidence limits and point estimates of the observed difference lies well within the IVAN trial prespecified non-inferiority margin of 3.5 letters. Only the CATT22 and IVAN trials compared continuous and discontinuous treatment regimens; meta-analyses of the findings of these trials showed that discontinuous treatment is significantly inferior to continuous treatment.

Morphological characteristics of lesions

At baseline, 76% of participants had a lesion with foveal centre involvement and 7% had GA. The discontinuous group, as well as having better BCVA and reading ability than the continuous group (see Chapter 4) also had fewer participants with a lesion with foveal centre involvement (73% vs. 78%; Table 13). Other morphological characteristics were balanced across the groups at baseline, with the exception of fibrosis, which was more common in the ranibizumab group than in the bevacizumab group (29% vs. 24%; see Table 13).

There were no significant differences between drugs for dye leakage on FFA, a measure of lesion activity, at both years 1 and 2. However, dye leakage on FFA was observed less frequently in the continuous treatment group than in the discontinuous treatment group at 1 year (OR 0.51, 95% CI 0.33 to 0.78; p = 0.0017). 26 This difference was not maintained into year 2, when the OR increased to 0.74 (95% CI 0.51 to 1.07; p = 0.11; Figure 25). At 1 year, fewer participants had fluid on OCT in the ranibizumab group (52% vs. 57%), and at 2 years a similar pattern was observed (50% vs. 58%; p = 0.065). When compared by treatment regimen, significantly fewer participants had fluid on OCT in the continuous group at both 1 and 2 years (45% vs. 62%; OR 0.47, 95% CI 0.33 to 0.67; p < 0.001 at 2 years; see Figure 25).

FIGURE 25.

Secondary morphological outcomes at 2 years (a) by drug and (b) by treatment regimen. Ratios of < 1 reflect better outcomes in the ranibizumab or continuous groups. Circles = GMR or OR. 95% CIs are given in parentheses and illustrated by horizontal bars.

The choice of drug had no effect on the odds of an active neovascular lesion being present in year 1 or year 2. However, compared with discontinuous treatment, continuous treatment protected against an active lesion being present at both times (52% vs. 65% at 1 year, 51% vs. 61% at 2 years; OR 0.65, 95% CI 0.45 to 0.95; p = 0.024 at 2 years; see Figure 25). Despite this finding, the median neovascular lesion area – when an active lesion was judged to be present – was similar in the two groups (Table 14).

Several OCT metrics were measured and they are summarised in Table 15. Treatment effects were only estimated for total thickness at the fovea and retinal plus subfoveal fluid thickness at fovea (see Figure 25). Neither of these metrics differed by drug but both were statistically significantly better (smaller mean thickness) in the continuous group compared with discontinuous group at 2 years; they were, on average, 9% and 8% less thick for participants receiving continuous treatment (GMR = 0.91, 95% CI 0.85 to 0.97; p = 0.004; and GMR = 0.92, 95% CI 0.84 to 1.00; p = 0.046; see Figure 25).

At baseline, GA was present in 7% of participants and in 51% of cases it was within the lesion. All colour and FFA images were regraded (again, masked to treatment allocation) after the end of the study in order to identify whether or not new GA had developed during the course of the study. It was found that 30% of participants had developed new GA during the trial. There was no difference in the frequency of new GA between the drug groups (28% for ranibizumab vs. 31% for bevacizumab; OR 0.87, 95% CI 0.61 to 1.25; p = 0.46), but significantly more participants receiving continuous treatment developed new GA (34% vs. 26%; OR 1.47, 95% CI 1.03 to 2.11; p = 0.03).

Meta-analyses of lesion morphology characteristics

Figure 26 shows meta-analyses of the BRAMD, CATT,22 GEFAL42 and IVAN trial findings for changes in total retinal thickness at the fovea. The analyses use the 2-year findings for the CATT22 and IVAN trials and 1-year findings for the BRAMD (presented at the 2013 meeting of EURETINA) and GEFAL42 trials. All trials contributed to the analysis by drug; the pooled point estimate suggests a non-statistically significant difference in favour of ranibizumab (p = 0.12; heterogeneity χ² = 0.12, df = 3; p = 0.99; I² = 0.0%). Only the CATT22 and IVAN trials contributed to the analysis by treatment regimen, for which the pooled point estimate showed a statistically significant difference in favour of continuous treatment (p = 0.001; see Figure 26).

FIGURE 26.

Meta-analysis of change in total retinal thickness at the fovea from baseline (a) by drug and (b) by treatment regimen. Change in total retinal thickness at the fovea = 2 years minus baseline or 1 year minus baseline, as indicated in the figures. Negative values for differences reflect better outcomes in the ranibizumab or continuous groups. 95% CIs are given in parentheses and illustrated by horizontal bars. WMD, weighted mean difference. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Only the CATT22 and IVAN trials contributed to the meta-analysis of development of new GA over 2 years. The point estimates for the two trials were substantially different for the drug comparison (pooled OR 1.05, 95% CI 0.81 to 1.37; p = 0.70) but similar for the comparison by treatment regimen, showing that new GA developed significantly more often during follow-up among participants treated continuously (OR 1.56, 95% CI 1.20 to 2.03; p = 0.001; Figure 27).

FIGURE 27.

Meta-analysis of GA (a) by drug and (b) by treatment regimen. Ratios of ≤ 1 reflect better outcomes in the ranibizumab or continuous groups. 95% CIs are given in parentheses and illustrated by horizontal bars. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Summary

Drug comparisons showed that both ranibizumab and bevacizumab were similar with respect to morphological outcomes. Regimen comparisons found that OCT thickness measurements were lower and an active neovascular lesion was less likely to be present in the continuous group than in the discontinuous group at 2 years. These findings were unchanged in the meta-analysis of the CATT22 and IVAN results. In total, 30% of participants developed new GA during the trial, and this occurred more frequently when receiving continuous treatment, a finding that was also confirmed in the meta-analysis of the CATT22 and IVAN results.

Serious adverse events

Over the course of the trial, at least one SAE – defined as an event that is life-threatening, causes hospitalisation (initial or prolonged), disability or permanent damage, a congenital anomaly or birth defect, or death – was reported for 171 participants and, of these, 30 participants died. The frequency of the primary safety end point, an ATE or HF or death from a vascular cause, did not differ significantly between the drugs (OR 1.69, 95% CI 0.80 to 3.57; p = 0.16) and between the treatment regimens drugs (OR 0.56; 95% CI 0.27 to 1.19; p = 0.13) (Table 16 and Figure 28). Of the 30 deaths, 15 participants were the ranibizumab group (4.8%) and 15 in the bevacizumab group (5.1%). By treatment regimen, there were more deaths in the discontinuous treatment group [20 (6.6%) vs. 10 (3.2%), OR 0.47, 95% CI 0.22 to 1.03; p = 0.05] (see Figure 28). Causes of death, determined from the participant death certificates are summarised in Table 17.

FIGURE 28.

Safety outcomes at 2 years (a) by drug and (b) by treatment regimen. ORs of < 1 reflect fewer SAEs during the 2 years of follow-up in the ranibizumab or continuous groups. Circles indicate ORs and bars indicate 95% CIs. Vascular events include arteriothrombotic events, HF, deep-vein thrombosis, pulmonary embolism, transient ischaemic attack, hospitalisation for angina, and non-ocular haemorrhage. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

Serious adverse events grouped by MedDRA system organ class are shown in Table 18. The most commonly reported type of SAE was cardiac disorders. Tests of the frequency of SAEs for different organ systems (when there were more than 10 participant-specific events) showed a statistically significant difference in SAEs coded as general disorders and administration site conditions (which includes all deaths) by treatment regimen (continuous 3.2%; discontinuous 7.0%; p = 0.03) but not for any other organ system by drug or treatment regimen (Figure 29). SAEs coded as gastrointestinal disorders were more frequent with bevacizumab but the difference was less marked than at 1 year. 26 Serious ocular AEs in the study eye were extremely rare (14 participants; see Table 18), as were events in the non-study eye (five participants; Table 19).

FIGURE 29.

Safety outcomes classified by MedDRA system organ class at 2 years (a) by drug and (b) by treatment regimen. ORs of < 1 reflect fewer SAEs during the 2 years of follow-up in the ranibizumab or continuous groups. Circles indicate ORs and bars indicate 95% CIs. Source: © 2013 Chakravarthy et al. 93 Open Access article distributed under the terms of CC-BY-NC-SA (http://creativecommons.org/licenses/by-nc-sa/3.0/). Published by Elsevier Ltd.

The percentages of patients having any systemic SAE were very similar by drug (ranibizumab 26%; bevacizumab 27%; OR 0.96; 95% CI 0.66 to 1.39; p = 0.82) and by treatment regimen (continuous 24%; discontinuous 29%; OR 0.77; 95% CI 0.53 to 1.11; p = 0.16). In total, 39 non-ocular SAEs were classified as possibly, probably or definitely related to treatment (Table 20).