Aflibercept for treating visual impairment caused by macular oedema secondary to central retinal vein occlusion

The manufacturer submitted evidence of clinical effectiveness for aflibercept compared with ranibizumab in people with macular oedema secondary to central retinal vein occlusion (CRVO). The main sources of evidence presented in the manufacturer's submission came from 2 randomised controlled trials that compared aflibercept with sham injection in people with macular oedema secondary to CRVO (COPERNICUS and GALILEO). In both trials the included patients had been diagnosed less than 9 months before the start of the trial and they had not received previous treatment for CRVO.

COPERNICUS was a randomised, double-blind, multicentre trial conducted in 6 non-European countries. From week 0 to week 24, patients in the intervention group (n=114) received aflibercept every 4 weeks and patients in the comparator group (n=73) received a sham injection every 4 weeks. From week 24 to week 52 patients in both groups received aflibercept if they met protocol-specified retreatment criteria, and received a sham injection if retreatment was not indicated. After the first year, patients continued in a 1‑year extension phase (up to 100 weeks) with aflibercept as needed (no sham injection). Patients were retreated with aflibercept if any of the following conditions were met: increase of 50 micrometres or more in central retinal thickness on optical coherence tomography (OCT) compared with lower previous measurement, new or persistent cystic retinal damages of sub-retinal fluid on OCT or persistent diffuse oedema of 250 micrometres or more in the central subfield on OCT, or a loss or gain of 5 letters or more between the current and most recent visit. All patients were eligible to receive pan retinal photocoagulation at any time if they developed neovascularisation. The average age of the patients was 66.3 years and most patients were male (57%). The mean best corrected visual acuity (BCVA) at baseline was 50.0 letters and 68% of patients had perfused retinal occlusion, which was defined as fewer than 10 disc areas of capillary non-perfusion on fluorescein angiography. The manufacturer did not report any statistically significant differences in the baseline characteristics between the 2 groups.

GALILEO was a randomised, double-blind, multicentre trial conducted in 10 European and Asian-Pacific countries. None of the study centres was located in the UK. From week 0 to week 24, patients in the intervention group (n=103) received aflibercept every 4 weeks and patients in the comparator group (n=68) received a sham injection every 4 weeks. From week 24 to week 52, patients in the intervention group received aflibercept if they met protocol-specified retreatment criteria, or sham injection. Patients were assessed monthly using retreatment criteria as in COPERNICUS. Patients in the comparator group continued to receive sham injection from week 24 to week 52. From week 52 to week 76, all patients received aflibercept if they met the retreatment criteria, or sham injection, every 8 weeks. All patients were eligible to receive pan retinal photocoagulation at any time if they developed neovascularisation. The average age of the patients was 61.5 years and most patients were male (56%). The mean BCVA at baseline was 52.2 letters. Perfused retinal occlusion, defined as fewer than 10 disc areas of capillary non-perfusion on fluorescein angiography, was present in 86% of patients in the intervention group and 79% of patients in the comparator group. The manufacturer stated that there was a slight imbalance in mean central retinal thickness between the 2 groups (683.20 micrometres in aflibercept compared with 638.66 micrometres in sham). The manufacturer stated that these values are considered to be close to the baseline central retinal thickness values from other CRVO trials, including COPERNICUS. The manufacturer did not report any statistically significant differences in the other baseline characteristics between the 2 groups.

The manufacturer used the same statistical analysis for the results from COPERNICUS and GALILEO and the intention-to-treat protocol was not used. In the primary efficacy analyses, data from all randomised patients who received any trial medication and had a baseline assessment and at least 1 efficacy assessment after baseline were included (full analysis set). Data were analysed according to the group to which patients were randomised. In the sensitivity analyses, data from all patients in the full analysis set who received at least 5 injections of trial medication and did not have any major protocol violations or deviations were included (per protocol population). Data were analysed according to which treatment patients received. In the safety analyses, data from all randomised patients who received any trial medication were included (safety population). Data were analysed according to which treatment patients received.

The primary outcome of both COPERNICUS and GALILEO was the proportion of eyes with a gain of 15 or more letters in BCVA from baseline to week 24. Statistically significantly more patients gained 15 or more letters at 24 weeks with aflibercept than with sham in COPERNICUS (aflibercept 64/114 [56%] and sham 9/73 [12%], p<0.001 adjusted by region [North America compared with rest of the world] and baseline BCVA [greater than 20/200 compared with 20/200 or lower]). There were also statistically significantly more patients in GALILEO who gained 15 or more letters at 24 weeks with aflibercept than with sham (aflibercept 62/103 [60%] and sham 15/68 [22%], p<0.0001 adjusted by region [Europe compared with Asia/Pacific] and baseline BCVA [greater than 20/200 compared with 20/200 or lower]). The manufacturer performed a meta-analysis of the data from COPERNICUS and GALILEO at week 24 for the number of patients gaining 15 or more letters and the mean change in BCVA from baseline. The odds and relative risk of gaining 15 or more letters at 24 weeks was statistically significantly higher in the group receiving aflibercept compared with the group receiving sham injection (odds ratio 6.85 [95% confidence interval 4.08 to 11.51]; relative risk 3.28 [95% confidence interval 2.25 to 4.79]). Subgroup analysis showed that baseline perfusion status (presence or absence of ischaemia) did not appear to have any significant effect on response rates.

From week 24 to week 52, patients in both groups in COPERNICUS received aflibercept treatment as needed and patients in GALILEO continued to receive treatment according to their group allocation at the start of the trial. At 52 weeks, there were statistically significantly more patients who had gained 15 or more letters in the group initially allocated to receive aflibercept than in the group initially allocated to receive sham injection in COPERNICUS (aflibercept 63/114 [55%] and sham 22/73 [30%], p<0.001) and in GALILEO (aflibercept 62/103 [60%] and sham 22/68 [32%], p=0.0004).

From week 52 onwards, patients in both groups in COPERNICUS and GALILEO received aflibercept treatment as needed. At 76 weeks there were statistically significantly more patients who had gained 15 or more letters in the group initially allocated to receive aflibercept than in the group initially allocated to receive sham in GALILEO (aflibercept 59/103 [57%] and sham 20/68 [29%], p<0.0004). At 100 weeks there were statistically significantly more patients who had gained 15 or more letters in the group initially allocated to receive aflibercept than in the group initially allocated to receive sham in COPERNICUS (aflibercept 56/114 [49%] and sham 17/73 [23%], p=0.0003).

Secondary outcomes in both trials included mean change at 24 weeks from baseline in BCVA, central retinal thickness, and the proportions of patients progressing to ocular neovascularisation. The mean change in BCVA and the mean change in central retinal thickness from baseline to 24 weeks were statistically significantly greater in the aflibercept group compared with the sham group in both COPERNICUS and GALILEO. The percentage of patients progressing to neovascularisation at week 24 was statistically significantly greater in the sham group compared with the aflibercept group in COPERNICUS, but there was no statistically significant difference between the groups in GALILEO.

The manufacturer performed a network meta-analysis to compare aflibercept with ranibizumab and dexamethasone, because head-to-head comparison data were not available from randomised controlled trials. Six high-quality trials were included in the network meta-analysis. Two trials compared ranibizumab with sham injection (CRUISE and ROCC), 2 trials compared aflibercept with sham injection (COPERNICUS and GALILEO), and 2 trials compared dexamethasone intravitreal implant with sham injection (GENEVA 008 and GENEVA 009). Data for bevacizumab were not included in the analysis because the manufacturer did not consider treatment with bevacizumab to be the standard of care for people with CRVO and only 2 small studies of moderate-to-low quality were available (Epstein et al. 2012 and Wittstrom et al. 2012). Epstein compared bevacizumab with sham in 60 patients and Wittstrom compared bevacizumab with pan-retinal photocoagulation in 19 patients.

The network meta-analysis was performed only on 6-month (24 weeks) trial data because of the switching of patients between treatment groups in some of the trials. For the comparison of ranibizumab with aflibercept, there was no statistically significant difference in the odds or relative risk of gaining 15 or more letters or of losing 15 or more letters. There was also no statistically significant difference in the mean change in BCVA from baseline to 24 weeks. For the comparison of dexamethasone with aflibercept, the odds and relative risk of gaining 15 or more letters from baseline to 24 weeks were statistically significantly smaller with dexamethasone and the mean change in BCVA was statistically significantly larger with aflibercept. The odds and relative risk of losing 15 or more letters were not reported. The results of the network meta-analysis were submitted as commercial in confidence and cannot be presented.

Impact on health-related quality of life was measured by NEI VFQ-25 in COPERNICUS and by NEI VFQ-25 and EQ-5D in GALILEO. From baseline to 24 weeks, there was a statistically significant greater mean change in NEI VFQ-25 score in the aflibercept group of both COPERNICUS (aflibercept 7.2, standard deviation [SD] 12.1, and sham 0.8, SD 9.8; p=0.001) and GALILEO (aflibercept 7.5, SD not reported, and sham 3.5, SD not reported; p=0.001). From baseline to 52 weeks, there was a statistically significant greater mean change in NEI VFQ-25 score in the aflibercept group in GALILEO (aflibercept 7.8 and sham 4.5, p=0.0049) but there was no statistically significant difference in mean change in NEI VFQ-25 total score in COPERNICUS (aflibercept 7.5 and sham 5.1, p=0.216). The mean changes in NEI VFQ-25 total score at week 76 in GALILEO and week 100 in COPERNICUS were not statistically significantly different. EQ-5D values were reported for the European subset of the population in GALILEO and there were no significant differences in EQ-5D scores. The EQ-5D data were submitted as academic in confidence and therefore cannot be presented.

The most common ocular treatment emergent adverse events in both trials and both groups were conjunctival haemorrhage, reduced visual acuity, eye pain, retinal haemorrhage, and increased intraocular pressure. There were deaths from arrhythmia, acute myocardial infarction, oesophageal adenocarcinoma, and pneumonia in the sham group of COPERNICUS. The exact number of deaths in this group was submitted as academic in confidence and cannot be presented. No deaths occurred in the aflibercept group of COPERNICUS or in either group in GALILEO. The manufacturer did not report whether the number of adverse events was statistically significantly different between the groups.

The manufacturer stated that the number of patients who had adverse events in the trials included in the network meta-analysis was too low to conduct a robust network meta-analysis on safety end points.

The manufacturer developed a cost–utility Markov model that evaluated the cost effectiveness of aflibercept compared with ranibizumab in people with macular oedema secondary to CRVO. There were 25 health states in the model, defined by the BCVA in both the eye receiving treatment (the study eye) and the non-treated eye (the second eye), in addition to death. The health states were defined by a 15-letter range in BCVA. The model had 4-weekly cycles and a time horizon of 30 years, which was effectively a lifetime horizon given that the baseline age of the cohort was 64 years. Patients could move into an improved health state, remain in the same health state, or move into a worse health state. The BCVA for the patients' second eye was assumed to remain constant over time and so second eye involvement was not included in the model.

The baseline distribution of the patient population between the 25 health states of the model was inferred from the pooled COPERNICUS and GALILEO baseline distributions of the study eye and second eye. The model assumes that the distributions are independent, resulting in an inferred 2-eyed patient distribution with the largest proportion of patients having BCVA of 64 to 50 in their study eye and BCVA of 80 or more in their second eye.

The transition probabilities for aflibercept for the first 6 cycles (0 to 24 weeks) of the model were based on pooled data from COPERNICUS and GALILEO. To determine the transition probabilities for improvement in BCVA with ranibizumab for the first 6 cycles (0 to 24 weeks), the relative risk of gaining 15 or more letters with ranibizumab that was calculated in the network meta-analysis (see section 3.9) was applied to the aflibercept probabilities. The transition probabilities for moving to a worse BCVA with ranibizumab were assumed to be the same as with aflibercept. The transition probabilities for remaining in the same health state with ranibizumab were calculated by subtracting the transition probability for moving to a better health state and the transition probability for moving to a worse health state from 1. From cycle 6 to 13 (week 24 to week 52) it was not possible to use pooled data from COPERNICUS and GALILEO because patients in COPERNICUS were able to change treatments after week 24. Patients receiving ranibizumab in CRUISE were also able to change treatments after week 24. The manufacturer therefore chose to assume that, for both aflibercept and ranibizumab, patients' vision was maintained and patients remained in the same health state for cycles 6 to 13. From cycle 13 (week 52) onwards, it was assumed that for both aflibercept and ranibizumab, patients' BCVA deteriorated, following a natural disease history progression that remained constant over time (Klein et al. 1991).

Because the second eye was assumed to have a constant BCVA, only the BCVA of the study eye was modelled. The manufacturer assumed an indefinite duration of treatment benefit, based on the treatment benefit seen at 24 weeks.

The EQ-5D data collected in GALILEO were used for health-related quality-of-life data in the economic model. The utility values used in the base-case analysis were based on the EQ-5D value averaged across all 4 time points (0, 24, 52, and 76 weeks) for the European population in both the aflibercept and sham injection groups of GALILEO. The utility values obtained from these scores were then analysed based on the 'worse-seeing eye' of the patients, to reflect that patients enrolled to the 2 aflibercept trials were predominantly tested in their 'worse-seeing eye'. A total of 121 patients were included in the analyses, with 440 observations across all time points and across the 2 treatment groups. Each observation was assigned to 1 of the 5 health state BCVA ranges based on the BCVA achieved in the patient's 'worse-seeing eye'. The assignment was irrespective of whether the 'worse-seeing eye' was the study eye or the second eye. For each BCVA range, the average utility values were then estimated across these observations. The 'worse-seeing eye' utility values from the EQ-5D trial data for health states 1 to 5 were submitted as academic in confidence and cannot be presented. The 'worse-seeing eye' utility values were attributed to the 25 health states in the model based on the lower of the 2 BCVA scores represented in a health state. The manufacturer did not use the NEI VFQ-25 data collected in COPERNICUS and GALILEO because they stated that EQ-5D is the preferred measure in the NICE reference case. The manufacturer identified a relevant study by Czoski-Murray et al. (2009) that was used to obtain utility values used in the scenario analysis.

Adverse events were not modelled in the base-case analysis because the manufacturer stated that antivascular endothelial growth factors (anti-VEGFs; including ranibizumab and aflibercept) have similar safety profiles to each other. Raised intraocular pressure, cataracts and retinal tears were modelled in the scenario analysis.

Total costs for treatment were calculated from the unit costs for aflibercept or ranibizumab, administration, and a monitoring visit, multiplied by the number of treatment and monitoring visits needed. The direct drug costs in the model incorporated the confidential discount applied to the list price of aflibercept approved as part of the patient access scheme. The manufacturer assumed that 52.38% of administration visits for aflibercept and ranibizumab would take place in an outpatient setting and the remaining in a day-case setting. A weighted average was used to derive an administration cost of £257 for each drug. Monitoring visit costs were £197 for each treatment.

A one-stop model was applied, which assumed that administration visits can double as monitoring visits. It was assumed that aflibercept would need 5.75 administration and monitoring visits, and ranibizumab would need 5.50 administration and monitoring visits, from week 0 to week 24. From week 24 to week 52 it was assumed that aflibercept would need 2.55 administration visits and 3.50 monitoring visits, and that ranibizumab would need 3.30 administration visits and 4.40 monitoring visits. A cost associated with blindness was applied each month for the first and second year when the 'better-seeing eye' was declared blind (BCVA of 35 letters or fewer). The costs associated with blindness for year 1 and year 2 were submitted as commercial in confidence and cannot be presented.

The manufacturer's base-case cost-effectiveness results (incorporating the patient access scheme for aflibercept but not for ranibizumab) showed that aflibercept dominated (that is, was less expensive and more effective than) ranibizumab because it resulted in more quality-adjusted life years (incremental quality adjusted life years [QALYs] 0.054) and lower costs (incremental costs −£2,937).

The manufacturer did not know the level of discount in the patient access scheme for ranibizumab because it is confidential so it applied a range of discounts to the list price of ranibizumab in its sensitivity analysis. When the manufacturer applied a discount to the list price of ranibizumab ranging from 0 to 50% in increments of 5%, aflibercept continued to dominate ranibizumab until the price of ranibizumab was discounted by 50%. When the price of ranibizumab was reduced by 50%, aflibercept had higher costs and more QALYs than ranibizumab, resulting in an incremental cost-effectiveness ratio (ICER) of £5,871 per QALY gained. The ICERs resulting from the sensitivity analyses for the other key drivers of the model were not reported by the manufacturer.

The manufacturer conducted one-way sensitivity analyses with and without the confidential discount applied to the list price of aflibercept. When using the discounted price of aflibercept, a net monetary benefit approach (calculated by multiplying the incremental QALYs by £20,000 and then subtracting the incremental costs) was used because aflibercept dominated ranibizumab in the base-case analysis. The results of the sensitivity analyses indicated that the model was sensitive to changes in the number of ranibizumab injections from 0 to 24 weeks and from 25 to 52 weeks, the relative risk of gaining 15 or more letters when comparing aflibercept with ranibizumab, the number of aflibercept injections from 25 to 52 weeks, and the number of monitoring treatments for ranibizumab from 0 to 52 weeks.

The manufacturer conducted 3 scenario analyses. The first scenario analysis used treatment durations of 2 years and 4 years rather than 1 year. This showed that aflibercept continued to dominate ranibizumab with incremental QALYs of 0.054 with 2 or 4 years of treatment (incremental costs with discounted aflibercept price were not reported). The second scenario used utility values for the 'better-seeing eye' from Czoski-Murray et al. (2009), irrespective of whether the 'better-seeing eye' was the study eye or the second eye. In Czoski-Murray, a value of 0.828 is applied if either eye is in health state 1. If the highest BCVA in either eye was health state 2, a value of 0.735 was applied. If the highest BCVA in either eye was health state 3, a value of 0.627 was applied. If the highest BCVA in either eye was health state 4, a value of 0.519 was applied. If both eyes were in health state 5, a value of 0.469 was applied. Using Czoski-Murray utility values, aflibercept continued to dominate ranibizumab with incremental QALYs of 0.028 (incremental costs with discounted aflibercept price not reported). The third scenario modelled the inclusion of the costs of adverse events (cataracts, intraocular pressure, and retinal tear) that were not included in the base-case analysis. Aflibercept continued to dominate ranibizumab with incremental QALYs of 0.054 (incremental costs with discounted aflibercept price not reported).

The manufacturer presented a fourth scenario analysis in which aflibercept was compared with dexamethasone. In the deterministic analysis, aflibercept was associated with more QALYs (incremental QALYs 0.189) and higher costs (incremental costs £612) which resulted in an ICER of £3,236 per QALY gained.

No subgroups were identified by the manufacturer for analysis.

The ERG commented that the manufacturer did not include bevacizumab or clinical observation as comparators even though they were listed as comparators in the final scope issued by NICE. The ERG noted that the manufacturer stated that it did not include bevacizumab or clinical observation as comparators because they are no longer considered routine or best practice since the publication of positive NICE technology appraisal guidance on ranibizumab for treating visual impairment caused by macular oedema secondary to retinal vein occlusion and NICE technology appraisal guidance on dexamethasone intravitreal implant for the treatment of macular oedema secondary to retinal vein occlusion. The ERG noted that bevacizumab has been widely used in the NHS and that patients may be kept under observation if there are contraindications to anti-VEGF treatment (such as allergy and eye infections) or if they refuse intravitreal treatment. The ERG questioned whether bevacizumab and clinical observation should have been included as comparators.

The ERG noted that it was not clear what proportion of patients in COPERNICUS and GALILEO had ischaemia or severe ischaemia, because different definitions of ischaemia and severe ischaemia exist. It suggested that the proportion of patients with ischaemia or severe ischaemia in the COPERNICUS and GALILEO trials may be lower than the proportion of patients with ischaemia or severe ischaemia and CRVO in England and Wales. The ERG questioned whether the results of COPERNICUS and GALILEO are applicable to patients with ischaemia or severe ischaemia.

The ERG highlighted that stopping rules were not used in the manufacturer's model. The manufacturer highlighted that no additional stopping rules were recommended in NICE's technology appraisal guidance on ranibizumab for treating visual impairment caused by macular oedema secondary to retinal vein occlusion. The ERG noted from the summary of product characteristics that stopping rules should be used for deterioration in visual acuity after 3 injections and if there is no fall in oedema fluid or central retinal thickness. The ERG was aware that the summary of product characteristics also states that continued treatment is not recommended if there is no improvement in visual and anatomic outcomes over the course of the first 3 injections. The ERG questioned whether a stopping rule should have been implemented in the model.

The ERG considered that the manufacturer's model does not incorporate the relative risk of losing 15 or more letters. It highlighted that the relative risk of losing 15 or more letters is in favour of ranibizumab and detrimental to dexamethasone, and not including the relative risk of losing 15 or more letters could affect the model results for the efficacy of aflibercept. The ERG questioned whether the relative risk of losing 15 or more letters should have been included in the model.

The ERG noted that, in the manufacturer's model, any net gain in visual acuity at 6 months broadly persists through the model lifetime. It highlighted that treatment was only received for 1 year. The ERG questioned whether it was reasonable to assume that the net gain at week 24 will remain for the lifetime of the patient.

The ERG suggested that the implementation of the health-related quality-of-life data in the manufacturer's model may be conservative. It highlighted that, in the base-case analysis, it is assumed that the EQ-5D data are the health-related quality-of-life data for the 'worse-seeing eye'. The ERG noted that, as a consequence, this is only applied when the study eye is being modelled as the 'worse-seeing eye'. It also understood that it is assumed that the Czoski-Murray utility values are the health-related quality-of-life data for the 'better-seeing eye'. The ERG noted that, as a consequence, this is only applied when the study eye is being modelled as the 'better-seeing eye'. It questioned the appropriateness of this approach.

The ERG considered that the utility values in the manufacturer's base-case analysis were obtained from the EQ-5D data from GALILEO and the utility values used in one of the manufacturer's scenario analyses were obtained from Czoski-Murray et al. (2009). It noted that utility values from Brown (1999) were used in NICE's technology appraisal guidance on dexamethasone intravitreal implant for the treatment of macular oedema secondary to retinal vein occlusion and NICE's technology appraisal guidance on ranibizumab for treating visual impairment caused by macular oedema secondary to retinal vein occlusion. The ERG questioned which utility values are the most appropriate to use in the cost-effectiveness analysis.

The ERG argued that the costs of administration and monitoring for aflibercept and ranibizumab appeared to be overestimates. It stated that the outpatient administration costs could be reduced by costing them as 'Outpatient procedures: BZ23Z: Minor vitreous retinal procedures' and that the dedicated monitoring visit costs could be reduced by costing them as 'RA23Z: Ultrasound Scan, less than 20 minutes'.

The ERG noted that the calculation used for the costs of raised intraocular pressure appears to be inappropriate. It stated that the costs of medication were unnecessarily divided by 13 and the costs of inpatient therapies were divided by 6.

The ERG considered the application of the costs of blindness in the aflibercept and ranibizumab groups in the model appears to be inappropriate. It stated that the manufacturer's model only considers the incidence events of blindness for the aflibercept and ranibizumab groups, rather than considering the incident events and the prevalence of blindness as it does in the dexamethasone group.

The ERG argued that the costs of blindness appear to have been underestimated. It stated that the manufacturer used Meads and Hyde (2003) as their source of the costs of depression. The ERG stated that McCrone et al. (2008) provides a more recent and more accurate estimate of the costs of depression.

The ERG noted that the manufacturer assumed that 52.38% of administration visits for anti-VEGF therapy (ranibizumab and aflibercept) would take place in an outpatient setting and the remaining in a day-case setting. This results in an average weighted administration cost of £257. The ERG considered that all administration visits would take place in an outpatient setting, resulting in an administration cost of £181. In NICE's technology appraisal guidance on dexamethasone intravitreal implant for the treatment of macular oedema secondary to retinal vein occlusion, the ranibizumab administration visit was costed as an office-based outpatient procedure. The ERG queried which administration cost was the most appropriate.

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

• The number of dexamethasone administrations in year 1 was changed from 4.00 to 1.86.

• The cost per aflibercept or ranibizumab administration was changed from £257.45 to £180.73.

• The cost per dedicated monitoring visit for aflibercept or ranibizumab was changed from £197.00 to £130.01.

• The number of dedicated monitoring visits for aflibercept was changed from 2.43 to 0.95.

• The number of dedicated monitoring visits for ranibizumab was changed from 2.03 to 1.10.

• The cataract rate in the ranibizumab group was changed from 3.3% to 1.6%.

• The duration of quality of life impact of raised intraocular pressure was changed from 1 day per cycle to 1 cycle.

• The duration of quality of life impact of cataract was changed from 1 cycle to 3 cycles.

• The duration of quality of life impact of retinal tears was changed from 1 cycle to 4 cycles.

• The cost of raised intraocular pressure was changed from £33 to £4.

• The blindness mortality multipliers were changed from 1.54 to 0.00.

• The costs of blindness for aflibercept and ranibizumab were applied to both newly incident and prevalent cases of blindness instead of only the incident cases of blindness.

• The costs of blindness were increased. The costs were submitted as commercial in confidence and therefore cannot be presented.
  
  The ERG also corrected an error in the manufacturer's calculation of adverse event rates for aflibercept and dexamethasone from month 7 to 12.

The ERG incorporated the confidential discount applied to the list price of aflibercept, but not the confidential discount for ranibizumab, in its exploratory analysis outlined in section 3.40. It showed that aflibercept dominated ranibizumab because it resulted in lower costs and higher QALYs (incremental costs −£3,049, incremental QALYs 0.053).

The ERG was aware that the discounts agreed in the patient access schemes for aflibercept and ranibizumab are confidential. Therefore, it applied a discount ranging from 0 to 50% to the list price of ranibizumab in its exploratory analysis as well as applying the discount to the list price of aflibercept. With the discounted price of aflibercept and a 0 to 45% reduction in the list price of ranibizumab, aflibercept dominated ranibizumab because it is less costly (incremental costs ranged from −£3,049 to −£122) and has a greater QALY gain (incremental QALYs 0.053) than ranibizumab. With the discounted price of aflibercept and a 50% reduction in the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £3,820 per QALY gained (incremental cost £203, incremental QALY 0.053).

The ERG carried out sensitivity analyses around the relative risk of losing 15 or more letters at 6 months, the administration cost, duration of treatment, adverse events, and the source of utility values. The ERG used the discount to the list price of aflibercept and reduced the list price of ranibizumab by a value of either 0% or 10 to 50% in increments of 5%.

• When the relative risk of losing letters was included in the model and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£3,005 to −£78, incremental QALYs 0.003). When a 50% discount was applied to the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £86,789 per QALY gained (incremental cost £247, incremental QALY 0.003).

• When the administration cost was reduced in the model and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£3,083 to −£156, incremental QALYs 0.053). When a 50% discount was applied to the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £3,176 per QALY gained (incremental cost £169, incremental QALY 0.053).

• When the duration of treatment was extended to 2 years and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£4,422 to −£285, incremental QALYs 0.053). When a 50% discount was applied to the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £3,274 per QALY gained (incremental cost £175, incremental QALY 0.053).

• When the duration of treatment was extended to 5 years and a discount of 0 to 50% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£6,838 to −£31, incremental QALYs 0.053).

• When costs of adverse events were included in the model and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£3,066 to −£139, incremental QALYs 0.053). When a 50% discount was applied to the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £3,489 per QALY gained (incremental cost £187, incremental QALY 0.053).

• When Czoski-Murray utility values with the 'worse-seeing eye' were used in the model and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£3,049 to −£122, incremental QALYs 0.053). When a 50% discount was applied to the list price of ranibizumab the ICER for aflibercept compared with ranibizumab was £3,851 per QALY gained (incremental cost £203, incremental QALY 0.053).

• When Brown utility values with the 'worse-seeing eye' were used in the model and a discount of 0 to 45% was applied to the list price of ranibizumab, aflibercept dominated ranibizumab (incremental costs ranged from −£3,049 to −£122, incremental QALYs 0.040). When a 50% discount was applied to the list price of ranibizumab, the ICER for aflibercept compared with ranibizumab was £5,076 per QALY gained (incremental cost £203, incremental QALY 0.040).

The ERG also presented a deterministic ICER for aflibercept compared with dexamethasone. Incorporating the patient access scheme for aflibercept, the ICER for aflibercept compared with dexamethasone was £12,265 per QALY gained (incremental cost £2,285, incremental QALY 0.186).

The ERG also carried out exploratory analyses on the comparison of aflibercept with dexamethasone, incorporating the patient access scheme for aflibercept. Increasing the duration of treatment to 2 years resulted in an ICER for the comparison of aflibercept with dexamethasone of £14,034 per QALY gained and increasing it to 5 years resulted in an ICER of £18,699 per QALY gained. For the comparison of aflibercept with dexamethasone, using Czoski-Murray utility values with the 'worse-seeing eye' resulted in an ICER of £12,868 per QALY gained, and with the 'better-seeing eye' resulted in an ICER of £18,740 per QALY gained. Using Brown utility values with the 'worse-seeing eye' resulted in an ICER of £16,833 per QALY gained, and with the 'better-seeing eye' resulted in an ICER of £28,523 per QALY gained.

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