Aflibercept solution for injection for treating wet age‑related macular degeneration

Clinical effectiveness

3.1 The manufacturer performed a systematic literature review of the evidence on the clinical effectiveness of aflibercept. The review identified 2 studies that directly compared aflibercept with ranibizumab in people with wet age‑related macular degeneration: VIEW 1 (n=1217) and VIEW 2 (n=1240). Both studies were multicentre (VIEW 1: 154 centres in USA and Canada; VIEW 2: 172 centres in 26 countries, including the UK), active‑controlled, double‑blind, randomised trials that compared aflibercept with ranibizumab. Both studies were identical in design (except for location) so that data could be pooled. Only one eye per patient was included in both studies. If a patient needed treatment in the second eye during the study, the second eye was allowed to receive any approved treatment although it was not included in the study. In both studies patients were randomised on a 1:1:1:1 basis to receive either (i) aflibercept 2 mg every 8 weeks after 3 initial monthly loading doses, (ii) aflibercept 2 mg every 4 weeks, (iii) aflibercept 0.5 mg every 4 weeks, or (iv) ranibizumab 0.5 mg every 4 weeks. The manufacturer stated that both studies were designed primarily to test whether aflibercept at its recommended dose (2 mg every 8 weeks) was non‑inferior to ranibizumab (0.5 mg every 4 weeks). Therefore, the results reported here are limited to the treatment arms of both studies. The manufacturer stated that non‑inferiority margins and definitions were established in discussion with the US Food and Drugs Administration, European Medicines Agency and other regulatory agencies to be consistent with key trials of ranibizumab, including the MARINA study (2006), for treating wet age‑related macular degeneration. Both studies had 2 phases, including a primary phase (from randomisation to week 48) during which patients received treatment according to randomisation arm, with patients in the aflibercept 2 mg every 8 weeks arm receiving sham injections when no active treatment was due. In the follow‑up extension phase (up to 92 weeks), patients in all 4 treatment arms continued to be evaluated every 4 weeks and remained in their allocated treatment groups. The total duration of both studies was 96 weeks consisting of up to 92 weeks of treatment plus a screening period and a 4‑week safety follow‑up period.

3.2 For both studies, the manufacturer defined 3 populations for analysis. The full analysis set included all randomised patients who received any study drug and had a baseline and at least 1 post‑baseline assessment. The per protocol set included all patients in the full analysis set who received at least 9 injections of study drug or sham and attended at least 9 scheduled visits during the first 52 weeks, except for those who were excluded because of major protocol violations. The safety analysis set included all patients who received any study drug. The manufacturer stated that the per protocol set was used for primary analysis (statistical evaluation of non‑inferiority). A patient who withdrew from the study before week 36 because of treatment failure was considered a 'non‑responder'. The last observation carried forward approach was used to calculate missing data except for baseline values. Patients withdrawing before week 36 were not included in the primary analysis but were included in the secondary analysis (in the full analysis set).

3.3 The baseline demographics and disease characteristics were similar between the aflibercept 2 mg every 8 weeks and ranibizumab treatment arms in the VIEW 1 and 2 studies. In VIEW 1, the mean age was 78 years, 41% of patients were male, and 97% of patients were white. In VIEW 2, the mean age was 73–75 years, 45% of patients were male, and 73% of patients were white. The total mean baseline best‑corrected visual acuity score (defined by Early Treatment Diabetic Retinopathy Study [ETDRS] scale) ranged from 54 to 56 letters in VIEW 1 and from 52 to 54 letters in VIEW 2. In both studies, the distribution of occult, minimally classic and predominantly classic lesion types in the study eye was similar across both treatment arms.

3.4 The primary outcome of VIEW 1 and 2 was the proportion of patients who maintained vision at week 52, defined as losing fewer than 15 letters on the ETDRS scale compared with baseline. This outcome was also measured at week 96. In a pooled analysis of both studies, the proportion of patients treated with aflibercept who maintained vision at week 52 was 95.3% compared with 94.4% of patients treated with ranibizumab (difference −0.9%, 95% confidence interval [CI] −3.5 to 1.7). At week 96, the proportion of patients treated with aflibercept who maintained vision was 92.4% compared with 91.6% of patients treated with ranibizumab (difference −0.8%, 95% CI −3.8 to 2.3). The manufacturer stated that aflibercept showed non‑inferiority to ranibizumab at weeks 52 and 96 because the upper limits of the confidence intervals for the differences in proportions were consistently below the pre‑specified boundary of 10%. The manufacturer also evaluated the primary outcome for pre‑planned subgroup analyses in both studies by age, sex, race, renal function, hepatic impairment, baseline visual acuity, lesion size and type. The manufacturer stated that the results for all subgroups in both studies and in pooled analyses were consistent with the results in the whole study populations. However, the results of these subgroup analyses were not presented by the manufacturer.

3.5 Secondary outcomes in VIEW 1 and 2 included changes from baseline to week 52 for: best‑corrected visual acuity as measured by ETDRS letter score, proportion of patients gaining at least 15 letters, and choroidal neovascularisation area. For the outcome of best‑corrected visual acuity at week 52, mean ETDRS letter scores increased by approximately 7−11 letters in both treatment arms in VIEW 1 and by approximately 8–10 letters in VIEW 2. No statistically significant differences in change in best‑corrected visual acuity from baseline to week 52 were reported between aflibercept and ranibizumab in a pooled analysis of both studies (mean difference −0.32 letters, 95% CI −1.87 to 1.23). In VIEW 1 and 2, improvements in visual acuity observed at week 52 were largely maintained at week 96 in both treatment arms. No statistically significant differences in the proportion of patients who gained at least 15 ETDRS letters were reported between aflibercept and ranibizumab treatment arms at week 52 in a pooled analysis of both studies (30.97% and 32.44% respectively, p‑value not reported). Similar results were reported at week 96. In VIEW 1, the ranibizumab arm had a statistically significantly greater mean reduction in choroidal neovascularisation area at week 52 than the aflibercept arm (−4.2 mm² and −3.4 mm² respectively, p=0.017). No statistically significant differences in choroidal neovascularisation area at week 52 were reported between ranibizumab and aflibercept 2 mg every 8 weeks in VIEW 2 (−4.16 mm² and −5.16 mm² respectively, p=0.073). Similar results were also reported at week 96 in both studies.

3.6 Vision‑related quality of life was measured in VIEW 1 and 2 using the National Eye Institute Visual Function Questionnaire‑25 (NEI VFQ‑25), which includes 25 questions designed to measure the effect of visual impairment on daily functioning and quality of life. Improvements in the mean NEI VFQ‑25 total score from baseline to week 52 were similar in both the aflibercept and the ranibizumab treatment arms in a pooled analysis of both studies (5.0 points and 5.6 points respectively, p‑value not reported). These improvements in vision‑related quality of life were maintained at week 96 in both treatment arms. The VIEW 2 study also measured changes in health‑related quality of life using the EQ‑5D questionnaire, which were incorporated in the manufacturer's cost‑effectiveness analysis.

3.7 The manufacturer did not present a formal meta‑analysis of the VIEW 1 and 2 studies on the basis that both studies were similarly designed so that their data could be pooled directly. The manufacturer commented that, although the VIEW 1 and 2 studies used a fixed dosing regimen for ranibizumab (0.5 mg every 4 weeks), in clinical practice a 'treatment as needed' approach is used which involves monthly ranibizumab treatment until the patient's visual acuity is stable for 3 consecutive months, with re‑treatment in a similar way upon loss of visual acuity (with a minimum of 2 injections). Therefore, the manufacturer conducted a systematic literature review and mixed treatment comparison (network meta‑analysis) to compare aflibercept 2 mg every 8 weeks with ranibizumab 0.5 mg in a 'treatment as needed' regimen.

3.8 The manufacturer produced 3 networks at 6, 12 and 24 months. Because no data were available for aflibercept at 6 months, only networks for outcomes at 12 and 24 months were considered further by the manufacturer. The manufacturer assumed that 52‑week and 96‑week data from VIEW 1 and 2 corresponded with outcomes at 12 and 24 months respectively. Results were presented for 3 outcomes: maintained vision (defined as the proportion of patients losing 15 or fewer ETDRS letters), improved vision (defined as the proportion of patients gaining more than 15 ETDRS letters) and mean change from baseline in best‑corrected visual acuity. The network meta‑analysis of outcomes at 12 months incorporated up to 10 studies, depending on the outcome, and included the VIEW 1 and 2 studies. For the outcome of mean change from baseline in best‑corrected visual acuity, the manufacturer repeated the analysis after excluding one study (DETAIL study; London et al. 2009) because patients in the study responded differently to ranibizumab in a 'treatment as needed' regimen compared with other studies.

3.9 The manufacturer presented separate network meta‑analyses for outcomes at 12 months, using both frequentist methods, based on traditional statistical methods applied in making comparisons, and Bayesian methods, which combine the probability of the data as a function of the parameters with prior beliefs about possible values of those parameters. These analyses showed no statistically significant differences between aflibercept 2 mg every 8 weeks and ranibizumab 0.5 mg treatment as needed in the proportion of patients who maintained vision (frequentist method: odds ratio [OR] 1.44, 95% CI 0.68 to 3.09; Bayesian method: OR 1.51, 95% CI 0.42 to 5.94) or gained vision (frequentist method: OR 1.29, 95% CI 0.91 to 1.83; Bayesian method: OR 1.28, 95% CI 0.45 to 3.68). No statistically significant differences in mean change in best‑corrected visual acuity at 12 months were shown between aflibercept 2 mg every 8 weeks and ranibizumab 0.5 mg treatment as needed (frequentist method: mean difference 0.83, 95% CI −1.57 to 3.23; Bayesian method: mean difference −2.87, 95% CI −10.02 to 4.30). When the manufacturer repeated the analysis after excluding the DETAIL study, the results for the outcome of mean change in best‑corrected visual acuity at 12 months were similar (frequentist method: mean difference 1.35, 95% CI −1.08 to 3.77; Bayesian method: mean difference 1.15, 95% CI −3.92 to 6.09).

3.10 The manufacturer did not present a network meta‑analysis of outcomes at 24 months because VIEW 1 and 2 both allowed treatment switching after 12 months from a fixed dosing regimen of aflibercept 2 mg every 8 weeks to a treatment as needed regimen (aflibercept fixed 2 mg every 8 weeks/treatment as needed). Therefore, two‑step indirect comparisons, based on the Bucher method, were used to compare aflibercept fixed 2 mg every 8 weeks/treatment as needed with ranibizumab 0.5 mg treatment as needed. The indirect comparisons included data from 3 studies: VIEW 1 and 2, and CATT, a 2‑year study that compared ranibizumab 0.5 mg with bevacizumab for treating patients with wet age‑related macular degeneration (CATT Research Group, 2012). The CATT study presented data for ranibizumab as an identical switch trial and as fixed dose or treatment as needed only. Both sets of data from the CATT study were analysed for the indirect comparison.

3.11 The results of the manufacturer's indirect comparison for the outcomes at 24 months also showed no statistically significant differences between aflibercept fixed 2 mg every 8 weeks/treatment as needed and ranibizumab 0.5 mg treatment as needed in the proportion of patients who maintained vision (relative risk 0.99, 95% CI 0.93 to 1.07) or gained vision (relative risk 0.88, 95% CI 0.61 to 1.28). No statistically significant differences in mean change in best‑corrected visual acuity at 24 months were shown between aflibercept 2 mg every 8 weeks and ranibizumab 0.5 mg treatment as needed (mean difference 0.31, 95% CI −4.33 to 3.71).

3.12 The manufacturer highlighted concerns about the validity of the network meta‑analyses and indirect comparisons because of the heterogeneity between the included studies. On the basis of a quality assessment checklist, the manufacturer found that 3 of the included studies had a high risk of bias. The manufacturer also noted that several of the studies had different baseline characteristics in terms of ETDRS letter score, treatment as needed re‑treatment criteria, proportion of men, central retinal thickness and numbers of injections. The manufacturer commented that sensitivity analyses were performed with regard to the heterogeneity but that the results were unchanged.

3.13 The manufacturer stated that the safety and tolerability of aflibercept compared with ranibizumab for up to 96 weeks was included as a secondary objective in the VIEW 1 and 2 studies. On the basis of the safety analysis dataset, no clinically meaningful differences were reported between aflibercept and ranibizumab for treatment‑emergent adverse events, with similar incidences of reported events between treatment arms. The most common treatment‑emergent adverse events (reported in at least 5% of patients treated with aflibercept in VIEW 1 and 2) were: conjunctival haemorrhage (26.7%), eye pain (10.3%), vitreous detachment (8.4%), cataract (7.9%), vitreous floaters (7.6%), and increased intraocular pressure (7.2%). The incidence of arterial thromboembolic events (including non‑fatal myocardial infarction, non‑fatal stroke or vascular stroke), which are potentially related to anti‑VEGF treatment, was also similar between the aflibercept (3.3%) and ranibizumab (3.2%) treatment groups in VIEW 1 and 2.

Cost effectiveness

3.14 The economic evidence provided by the manufacturer included a literature review, which identified one published cost‑effectiveness analysis of aflibercept in US patients with wet age‑related macular degeneration, and a de novo cost–utility analysis. The manufacturer developed a Markov state‑transition cohort model simulating cohorts of people with wet age‑related macular degeneration receiving aflibercept 2 mg every 8 weeks or ranibizumab 0.5 mg treatment as needed. The model assumed a cycle length of 1 month based on the level of monitoring associated with ranibizumab treatment, and used a lifetime horizon (25 years based on a starting age of 74 years). An NHS and personal social services perspective was taken and costs and benefits were discounted at 3.5%.

3.15 The economic model included a total of 30 health states defined by a combination of different levels of visual acuity in both eyes (the treated eye and the second eye) in addition to the absorbing health state of death. For each health state, visual acuity in the treated eye or second eye was defined according to 5 possible levels on the ETDRS scale, ranging from no visual impairment (ETDRS more than 80 letters) to blindness (ETDRS fewer than 36 letters) with 3 intermediate levels (ETDRS 66–80 letters; 51–65 letters and 36–50 letters). In each model cycle, people were assumed to have the median visual acuity of each ETDRS range and moved to the median value of either the adjacent state or the state 2 levels higher or lower, based on the number of letters gained or lost. For each health state, the patient could either be on or off active treatment.

3.16 The economic model included a 5‑year treatment period on the basis of clinical opinion which suggested that patients are likely to continue treatment beyond 24 months. For the first 2 years, clinical-effectiveness data at baseline, 52 and 96 weeks from the last observation carried forward population in the VIEW 2 study were used to estimate the visual acuity of people receiving aflibercept. The probabilities of gaining and losing visual acuity in year 1 were applied to the VIEW 2 patient distribution at baseline and the probabilities of gaining and losing visual acuity in year 2 were applied to the modelled year 1 distribution. The visual acuity of people receiving ranibizumab for the first 2 years of the model was estimated from the relative risks of improving and maintaining vision for aflibercept compared with ranibizumab taken from the manufacturer's network meta‑analysis and indirect comparison of aflibercept 2 mg every 8 weeks with ranibizumab 0.5 mg treatment as needed. During this period, people who were defined as being in the blind health state received treatment with ranibizumab or aflibercept. However, this did not continue in years 3 to 5 on the basis of clinical opinion which suggested that the blind eye is unlikely to benefit from treatment. Simple linear interpolation was used to populate the monthly model cycles for year 1 (cycles 1–12) and year 2 (cycles 13–24). The annual rates of treatment discontinuation in year 1 (2.7%) and year 2 (3.5%) were assumed to be identical between both treatment groups and were based on an average of the discontinuation rates reported in the VIEW 2 and CATT studies.

3.17 For years 3 to 5 in the model, it was assumed that people on active treatment would remain in the same health state that they were in after 2 years. Because no statistically significant differences in clinical effectiveness were identified in the indirect comparison of aflibercept with ranibizumab, identical assumptions were made for both treatment groups during this period. In the absence of available trial data, clinical opinion was used to estimate the annual probability of treatment discontinuation in years 3 to 5 (18.7%), which were also assumed to be identical between treatment groups. From year 6 it was assumed that all people in both treatment groups discontinued active treatment and started best supportive care.

3.18 The manufacturer assumed that clinical effectiveness in the treated eye was independent of effectiveness in the second eye. Clinical‑effectiveness data for the second eye while on treatment was calculated using the same methodology applied to the treated eye. The manufacturer assumed that wet age‑related macular degeneration involvement in both eyes was 0% at the start of the model and that people developed wet age‑related macular degeneration in the second eye from year 3. The manufacturer also assumed that all people in the model who developed wet age‑related macular degeneration in the second eye from year 3 were treated. On the basis of a meta‑analysis by Wong et al. (2008) of patients with wet age‑related macular degeneration receiving no active treatment, the manufacturer estimated a 0.65% monthly probability of developing wet age‑related macular degeneration in the second eye. For people who were not receiving active treatment, clinical‑effectiveness data from Wong et al. were used to estimate the monthly probability of losing either 15 letters (0.56%) or 30 letters (1.56%) with the remaining people maintaining stable visual acuity.

3.19 The manufacturer stated that there is limited evidence of a relationship between wet age‑related macular degeneration and an increased risk of mortality and that, on the basis of data from the VIEW 1 and 2 studies, it is unlikely that there is any difference in mortality between aflibercept and ranibizumab. Therefore, age‑specific all‑cause mortality from UK life tables was used for both treatment groups. For people who were blind in both eyes, an excess risk of mortality was taken from a UK study of older patients with visual impairment (Thiagarajan et al. 2005).

3.20 To estimate the health‑related quality of life associated with each health state corresponding to visual acuity in both eyes, EQ‑5D data from VIEW 2 were transformed to utility values using the UK population tariff. A pooled dataset of all trial arms at baseline, 52 weeks and 96 weeks was used by the manufacturer. The manufacturer adjusted the utility values for 4 of the health states in the model to maintain the assumption that utility values decrease consistently with worsening visual acuity. This was achieved by taking the average of the utility values above and below the anomalous value. Utility values were not adjusted for age in the model. The resulting utility values applied in the model are academic in confidence and therefore not reported here. The manufacturer stated that, because of the low rates of adverse events observed in the VIEW 1 and 2 trials and the small differences observed between the aflibercept and ranibizumab treatment groups, the impact of adverse events on health‑related quality of life was not included in the base‑case analysis. However, in a scenario analysis, the manufacturer included the loss in utility associated with adverse ocular events taken from 2 separate studies identified in a systematic literature review (Brown et al. 2007; Gower et al. 2010). These utility decrements were subtracted from the utility values for the health states defined by visual acuity and included retinal haemorrhage (−0.300), vitreous haemorrhage (−0.305), endophthalmitis (−0.300), cataract (−0.142) and retinal detachment (−0.27).

3.21 The manufacturer included the costs of drug treatment, including drug acquisition, administration and monitoring costs. The drug acquisition costs incorporated the confidential discount applied to the list price of aflibercept approved as part of the patient access scheme. The manufacturer of ranibizumab has also previously agreed a revised patient access scheme with the Department of Health in 2013 (as revised in the context of NICE technology appraisal guidance 274), in which it applied a revised discount to ranibizumab for all indications. At the time of submission for this appraisal, the manufacturer of aflibercept was unaware of the size of the confidential discount and therefore presented a range of scenario analyses, which applied discounts to the list price of ranibizumab ranging from 10% to 50%, in increments of 5%.

3.22 The resource use and unit costs associated with treatment and monitoring visits were based on Hospital Episode Statistics (HES 2010/11) and NHS reference costs (2011/12). The manufacturer assumed that in year 1 people treated with aflibercept had their treatment administration and monitoring at the same visit (one‑stop model), and that 50% of people treated with ranibizumab followed a one‑stop model and 50% had separate visits for treatment and monitoring (two‑stop model). In years 2 to 5, the manufacturer assumed that 50% of people in both treatment groups followed a one‑stop model and 50% followed a two‑stop model. The manufacturer assumed that treatment with both aflibercept and ranibizumab occurred as a weighted average of a day‑case visit (55%) and outpatient visit (45%), resulting in a total cost of £257.45 per treatment visit. It was assumed that people in both treatment groups would need one fluorescein angiography (£117) before starting treatment.

3.23 The manufacturer assumed that people receiving aflibercept had 7 injections in the first year and 4 injections in the second year based on the treatment frequency recommended in the summary of product characteristics and the VIEW 2 study. It was assumed that people receiving ranibizumab had 8 injections in the first year and 6 injections in the second year based on NICE technology appraisal guidance 155 and the European Medicines Agency assessment report for the 2011 revision of the summary of product characteristics. Based on clinical specialist opinion, the manufacturer assumed that people in both treatment groups had 4 injections in years 3 to 5.

3.24 The manufacturer assumed that separate monitoring visits included the cost of an ophthalmologist outpatient visit (£80) and an optical coherence tomography (£117), resulting in a total cost of £197 per monitoring visit. The frequency of monitoring visits in the first 2 years of the model was also based on the summary of product characteristics for both treatments. People receiving aflibercept had 7 monitoring visits in year 1 and 6 monitoring visits in year 2 and people receiving ranibizumab had 12 monitoring visits in years 1 and 2. People receiving aflibercept in a one‑stop model had their treatment and monitoring at the same visit and therefore needed no separate monitoring visits in the first year and 2 separate visits in the second year. People receiving aflibercept in a two‑stop model in the second year had their treatment and monitoring at separate visits and therefore needed 6 separate monitoring visits in the second year. People receiving ranibizumab had 4 separate monitoring visits in the first year and 6 separate visits in the second year in a one‑stop model and 12 separate monitoring visits in the first 2 years in a two‑stop model. On the basis of clinical specialist opinion, people in years 3 to 5 in both treatment groups had 3 separate monitoring visits in the one‑stop model and 7 separate monitoring visits in a two‑stop model.

3.25 The manufacturer estimated the costs associated with blindness for people who were defined as being blind in both eyes (ETDRS score under 36 letters). The manufacturer applied cost data taken from a published UK costing study of blindness in people with age‑related macular degeneration (Meads and Hyde 2003). This study estimated the costs associated with a range of items including low‑vision aids, rehabilitation, residential care, district nursing, community care and the cost of treating complications including depression and falls. After adjusting for inflation, the total estimated annual cost of blindness was £585. Because of the low incidence of adverse events reported in the VIEW 1 and 2 studies, the manufacturer did not apply the costs of adverse events in the base‑case analysis.

3.26 The manufacturer's base‑case deterministic cost‑effectiveness results (including the patient access scheme for aflibercept but not for ranibizumab) showed that aflibercept dominated ranibizumab because it resulted in lower costs and higher quality‑adjusted life years (QALYs; 7.77 compared with 7.76). When the manufacturer applied a discount to the list price of ranibizumab, ranging from 10 to 50%, aflibercept continued to dominate ranibizumab.

3.27 The manufacturer performed one‑way sensitivity analysis using a net monetary benefit approach because aflibercept dominated ranibizumab in the base‑case analysis^[1].The deterministic and probabilistic sensitivity analyses used the discounted price for aflibercept agreed under the patient access scheme and the list price for ranibizumab. The results of the one‑way sensitivity analyses indicated that the cost effectiveness of aflibercept was most sensitive to the drug acquisition costs, frequency of injections and monitoring visits, proportion of people in one‑stop and two‑stop models, discount rates and the relative risk of gaining or losing visual acuity with ranibizumab treatment. The manufacturer stated that, in all sensitivity analyses, aflibercept continued to dominate ranibizumab. Results of the manufacturer's probabilistic sensitivity analysis showed that aflibercept had a 100% probability of being cost effective compared with ranibizumab if the maximum acceptable incremental cost‑effectiveness ratio (ICER) was £20,000 per QALY gained.

3.28 The manufacturer also conducted a number of scenario analyses, which included the discounted price for aflibercept but not for ranibizumab. Two scenarios involved varying the frequency of injections and monitoring: applying the average number of injections reported in years 1 and 2 of the VIEW 2 and CATT trials for aflibercept and ranibizumab respectively, and applying monthly monitoring visits for ranibizumab and bi‑monthly monitoring visits for aflibercept in years 3 to 5. One scenario involved applying the same clinical‑effectiveness data for both treatments so that the same proportions of people gaining or losing visual acuity were applied in both treatment groups. One scenario applied alternative utility values from a study by Czoski‑Murray et al. (2009) in which members of the general public valued levels of visual impairment that were simulated by custom‑made contact lenses, using the time trade‑off method. One scenario modelled the impact of adverse ocular events in the ranibizumab treatment group, which included retinal haemorrhage, vitreous haemorrhage, endophthalmitis, cataract and retinal detachment taken from a separate trial of ranibizumab in patients with wet age‑related macular degeneration (Boyer et al. 2009). Another scenario applied clinical‑effectiveness estimates equivalent to best‑supportive care, taken from Wong et al., in years 3–5 for both treatment groups. For all scenario analyses, aflibercept either continued to dominate ranibizumab or resulted in net cost savings (when the same proportions of people gaining or losing visual acuity were applied in both treatment groups).

ERG critique of manufacturer's submission

3.29 The ERG considered that the clinical‑effectiveness evidence from the VIEW 1 and 2 studies was of good quality without any obvious sources of bias. The ERG noted that the manufacturer used the last observation carried forward approach to calculate missing data for the primary outcome of the proportion of people who maintained vision at week 52 in VIEW 1 and 2. The ERG considered that this approach may have introduced bias because it can artificially stabilise disease, which may be inappropriate for a progressive disease such as wet age‑related macular degeneration. After clarification, the manufacturer provided the observed results at week 52 for the outcome of maintained vision from the per protocol and full analysis datasets, which were similar to the original results based on the last observation carried forward approach. The ERG also ran the network meta‑analysis for the outcome of maintained vision at 12 months using observed data from VIEW 1 and 2 and found that the results were similar to the original results obtained using the last observation carried forward approach. Therefore, the ERG was satisfied that the use of last observation carried forward did not substantially impact the results for the primary outcome at week 52 in VIEW 1 and 2.

3.30 The ERG agreed with the manufacturer that there were concerns about the validity of the network meta‑analyses and indirect comparison because of heterogeneity between the included studies. The ERG noted that the manufacturer had conducted sensitivity analyses with regard to heterogeneity, but commented it was not clear what these sensitivity analyses were. The ERG also noted that the network meta‑analysis for the outcome of mean change from baseline in best‑corrected visual acuity at 12 months excluded a treatment arm from one of the studies included in the analysis (ranibizumab fixed dose 0.3 mg arm in the DETAIL study). The ERG was concerned about the validity of the results as a result of this omission and therefore repeated the analysis including this treatment arm. The ERG found that, although this did not significantly alter the results in terms of the mean difference in change in best‑corrected visual acuity between treatment arms, the results of the network analyses and indirect comparison should be interpreted with caution.

3.31 In its critique of the manufacturer's economic model, the ERG noted that clinical‑effectiveness data for the aflibercept treatment group were derived from the VIEW 2 study rather than pooled data from VIEW 1 and 2. The ERG also noted that the proportion of people treated with aflibercept who gained or lost visual acuity at 52 weeks and 96 weeks was broadly similar between VIEW 1 and 2 and that there was no suggestion of bias arising from the choice of VIEW 2 data rather than pooled clinical‑effectiveness data. However, the ERG highlighted that there were discrepancies between the clinical‑effectiveness data from VIEW 2 and the modelled population in terms of the proportion of people who gained or maintained visual acuity at week 52. The ERG commented that it was unclear what clinical‑effectiveness data were used for the aflibercept group in the model.

3.32 The ERG noted that people treated with aflibercept in VIEW 1 and 2 received an average of 7.5 and 7.7 injections in the first year of both studies. The ERG also noted that the dosing schedule suggests that people who remain on treatment would need 8 injections in year 1. Therefore, the ERG considered that it may have been more reasonable for the manufacturer to model 8 injections of aflibercept in year 1. The ERG also noted that the average number of 4 aflibercept injections in year 2 of the model, which were taken from the VIEW 2 study, had been annualised from 44 weeks to 52 weeks to account for the study duration (96 weeks) which was slightly shorter than 2 years (104 weeks).The ERG considered that, on the basis of the weighted average number of injections of ranibizumab (7.4) in a treatment as needed dosing regimen in studies included in the manufacturer's systematic review, the number of ranibizumab injections in year 1 of the model should probably have been 7 rather than 8. The ERG commented that the network meta‑analysis for visual acuity outcomes at 24 months relied largely upon data from the CATT study, in which patients treated with ranibizumab had an average of 5.7 injections in a 'treatment as needed' regimen in year 2. The ERG therefore considered that the manufacturer's assumption of 6 ranibizumab injections in the second year of the model was reasonable.

3.33 The ERG noted that the manufacturer reported relative risks of maintaining and gaining visual acuity from its network meta‑analysis and indirect comparison between baseline and 12 months and between 12 months and 24 months. However, the ERG considered that the results of the manufacturer's systematic review and indirect comparison at 24 months were the relative risks of gaining or maintaining visual acuity between baseline and 24 months. The ERG also noted that applying the 24‑month relative risks to the probability of gaining or maintaining visual acuity between 12 months and 24 months resulted in more people in the aflibercept treatment group gaining or maintaining visual acuity compared with ranibizumab at 24 months. However, the ERG considered that, because the estimated relative risks of gaining and maintaining visual acuity for aflibercept compared with ranibizumab from baseline to 24 months were less than 1, fewer people in the aflibercept treatment group should have gained or maintained visual acuity compared with ranibizumab at 24 months.

3.34 The ERG considered that the manufacturer's approach to modelling second‑eye involvement was incorrect. The ERG noted that the probabilities of gaining or maintaining visual acuity with aflibercept or ranibizumab during the first 2 years of treatment were not applied to the second eye and that there was no incidence of second‑eye involvement in years 1 and 2 of the model. The ERG also noted that, although the baseline prevalence of wet age‑related macular degeneration in the second eye was 19% in the pooled VIEW 1 and 2 population, the manufacturer had assumed that people in both treatment groups had no visual impairment or wet age‑related macular degeneration in their second eye at the start of the model. Furthermore, the ERG considered that the model did not allow for sensible consideration of the timing of second‑eye involvement because the effect of treatment on visual acuity in the second eye and the costs of treating any second‑eye involvement were limited to years 3 to 5. Therefore, the ERG concluded that the manufacturer's economic model in its current form is a 'one‑eye model' that should be limited to considering the cost effectiveness of aflibercept as unilateral treatment for wet age‑related macular degeneration.

3.35 Because the ERG concluded that the manufacturer's model may be limited to being a one‑eye model, it suggested that further consideration should be given as to whether people received treatment in their better‑seeing eye or their worse‑seeing eye and the resulting impact on health‑related quality of life. The ERG considered that the manufacturer's assumption of no second‑eye involvement in years 1 and 2 resulted in the model being a worse‑seeing eye model, with the additional assumption of the second eye having no visual impairment. On the basis of the manufacturer's EQ‑5D utility values from VIEW 2, the ERG suggested a narrower range of utility values for the 5 health states defined by visual acuity in a worse‑seeing eye model. For a better‑seeing eye model, the ERG suggested that utility values should be taken from a study by Brown (1999) that measured vision‑related utility values using the time trade‑off method in 325 people from the USA with impaired vision (Snellen scale 20/40) in at least 1 eye. The ERG noted from the Brown study that, among people who had good vision in their better‑seeing eye, the worse‑seeing eye contributed little to health‑related quality of life. The utility values taken from the Brown study ranged from 0.920 to 0.621 for the 5 health states defined by visual acuity in the manufacturer's model, a range that the ERG noted was similar to the range of utility values from the VIEW 1 study under the assumption of the worse‑seeing eye being blind.

3.36 The ERG considered that it was unclear why all patients in the aflibercept group followed a one‑stop monitoring model and 50% of patients in the ranibizumab group followed one‑stop and 50% followed a two‑stop model in the first year of the economic model. If patients in the aflibercept group followed a two‑stop model, they would therefore have 7 separate monitoring visits in year 1. The ERG also considered that the manufacturer's estimated cost per treatment visit of £257, which was based on a weighted average of outpatient and day‑case visits from 2010/11 HES data, may have been too high. On the basis of 2011/12 HES data, the ERG estimated a lower weighted average cost of £129.46 per treatment visit. The ERG also noted that, in the appraisal of ranibizumab for treating diabetic macular oedema (NICE technology appraisal guidance 274), the manufacturer estimated a total cost of £143 per treatment visit. The ERG also considered that the manufacturer's estimated cost of £117.26 for an optical coherence tomography (based on a fluorescein angiography) may have been too high and that a lower cost of £51.27 (based on a 20‑minute ultrasound scan) may have been more appropriate. The ERG also noted that the manufacturer's estimate of the annual costs of blindness was implemented as a monthly cost in the model.

3.37 The ERG conducted exploratory analyses, which involved the following modifications to the manufacturer's model:

3.38 The ERG applied the changes outlined in section 3.37 in 2 scenario analyses for the worse‑seeing eye model and 2 scenario analyses for the better‑seeing eye model. The first scenario for each model adopted the manufacturer's interpretation that its indirect comparison of aflibercept with ranibizumab at 24 months provided relative risks of maintaining and gaining visual acuity from 12 to 24 months. In this first scenario, the ERG retained the proportions of people maintaining and gaining visual acuity in the manufacturer's original model. The second scenario for each model adopted the ERG's interpretation that the manufacturer's indirect comparison at 24 months provided relative risks of gaining and maintaining visual acuity from baseline to 24 months. In this second scenario, the ERG retained the baseline distribution of visual acuity from the manufacturer's original model.

3.39 The ERG incorporated the confidential discount applied to the list price of aflibercept and a range of discounts (from 0 to 50% in increments of 5%) to the list price of ranibizumab in its scenario analyses as outlined in section 3.38. In the ERG's first scenario analysis for the worse‑seeing eye model, aflibercept either dominated ranibizumab (discount 0–45%) or resulted in an ICER of £60,153 per QALY gained (discount 50%). In the ERG's first scenario analysis for the better‑seeing eye model, aflibercept either dominated ranibizumab (discount 0–45%) or resulted in an ICER of £9002 per QALY gained (discount 50%). In the ERG's second scenario analysis for the worse‑seeing eye model, aflibercept resulted in lower costs and lower QALYs compared with ranibizumab when a discount range of 0–45% was applied to the list price of ranibizumab, with ICERs ranging from £1,692,511 to £108,180 saved per QALY lost. In the ERG's second scenario analysis for the better‑seeing eye model, the ICERs for aflibercept compared with ranibizumab ranged from £261,432 to £16,710 saved per QALY lost when a discount range of 0–45% was applied to the list price ranibizumab. When the ERG applied a 50% discount to the list price of ranibizumab, aflibercept was dominated by ranibizumab for both the worse‑seeing eye and better‑seeing eye models.

3.40 Full details of all the evidence are in the manufacturer's submission and the ERG report.