3 The manufacturer's submission

The Appraisal Committee (section 9) considered evidence submitted by the manufacturer of ranibizumab and a review of this submission by the Evidence Review Group (ERG; section 10).

3.1 The manufacturer submitted evidence of clinical and cost effectiveness for ranibizumab compared with grid laser photocoagulation in people with visual impairment caused by macular oedema secondary to branch retinal vein occlusion (BRVO) and for ranibizumab compared with best supportive care in people with visual impairment caused by macular oedema secondary to central retinal vein occlusion (CRVO). The manufacturer stated that there was no direct or indirect evidence comparing the clinical effectiveness of ranibizumab with bevacizumab or dexamethasone intravitreal implant (which were defined as comparators in the scope for the appraisal). However, a comparison of the cost effectiveness of ranibizumab with dexamethasone intravitreal implant was included in the manufacturer's submission; this was not the case for bevacizumab (see sections 3.9 and 3.10).

3.2 The main sources of evidence presented in the manufacturer's submission came from the BRAVO and CRUISE randomised controlled trials (RCTs). These evaluated the efficacy of ranibizumab, compared with a sham procedure, for treating visual impairment caused by macular oedema secondary to BRVO and to CRVO respectively. The BRAVO (n=397) and CRUISE (n=392) trials were both 3‑armed RCTs carried out at multiple centres in the USA. Patients were randomised equally to sham injection, monthly intraocular ranibizumab 0.3 mg or monthly intraocular ranibizumab 0.5 mg. Both trials included people with visual impairment caused by macular oedema who had been diagnosed in the 12 months before study initiation. Patients entered a 6‑month treatment phase during which monthly injections were given, beginning on day 0. In the treatment phase of BRAVO, patients in both the sham injection and ranibizumab groups could receive grid laser photocoagulation for rescue treatment from 3 months. In both BRAVO and CRUISE, the treatment phase was followed by a 6‑month observation phase during which all groups (that is, the sham group and the 2 ranibizumab groups) could receive ranibizumab as needed. Patients in the observation phase of BRAVO (but not CRUISE) could receive grid laser photocoagulation for rescue treatment from 3 months (that is, at month 9 of the study). The final treatment in both BRAVO and CRUISE was given at month 11, with a final study visit at month 12. Patients who completed the 12‑month BRAVO and CRUISE trials could enter an open‑label extension study (HORIZON).

3.3 The primary outcome in both BRAVO and CRUISE was the mean change from baseline in best corrected visual acuity (BCVA) score in the study eye at 6 months. BCVA score was measured using the Early Treatment of Diabetic Retinopathy Study (ETDRS) eye chart, in which a score of 85 letters corresponds to normal visual acuity ('20/20 vision'). Secondary outcomes reported in both BRAVO and CRUISE included mean change from baseline in BCVA score over time up to 6 and 12 months, and the proportion of patients gaining or losing more than 15 letters in BCVA score at 6 and 12 months compared with baseline. The trials also reported results for several exploratory outcomes, including the mean change from baseline in the National Eye Institute Visual Function Questionnaire‑25 (NEI VFQ‑25) composite score up to 6 months. The NEI VFQ‑25 has 25 questions that are designed to measure the effect of visual impairment on daily functioning and quality of life.

3.4 This appraisal considered the 0.5 mg dose of ranibizumab, which is the only dose with a UK marketing authorisation. In BRAVO, 91.7% of patients in the sham group and 95.4% in the ranibizumab group were treated in the 'worse‑seeing eye' (that is, the eye affected by RVO). In CRUISE, 90.0% of patients in the sham group and 92.3% in the ranibizumab group were treated in their 'worse‑seeing eye'. The mean number of ranibizumab injections in the treatment phase was 5.7 (BRAVO) and 5.6 (CRUISE). The average number of ranibizumab injections in the observation phase was 2.7 (BRAVO) and 3.3 (CRUISE). More than 80% of patients from the sham injection group in both BRAVO and CRUISE received ranibizumab as needed during the observation phase. During the first 6 months of the BRAVO trial, grid laser photocoagulation was used in 57.6% of patients in the sham injection group and in 21.4% of the patients in the ranibizumab group. Over the 12‑month study period in BRAVO, 61.4% of patients in the sham (plus ranibizumab) group and 34.4% of patients in the ranibizumab group received rescue treatment with grid laser photocoagulation.

3.5 In BRAVO, at month 6, patients in the ranibizumab group had gained an average of 18.3 letters (95% confidence interval [CI] 16.0 to 20.6) from baseline BCVA score. This gain was statistically significant compared with the gain of 7.3 letters (95% CI 5.1 to 9.5) in the group receiving sham injection (p<0.0001). At month 12 of the BRAVO trial (that is, at the end of the 6‑month observation period, during which all patients could receive ranibizumab as needed), the 0.5 mg ranibizumab group reported an average gain in BCVA baseline score of 18.3 letters (95% CI 15.8 to 20.9) compared with the sham (plus ranibizumab) group, which had gained 12.1 letters (95% CI 9.6 to 14.6, p<0.01). The observed improvement at month 6 from baseline in the NEI VFQ‑25 composite score was statistically significantly greater in patients receiving ranibizumab (10.4 points, 95% CI 8.3 to 12.4) than in patients receiving sham injection (5.4 points, 95% CI 3.6 to 7.3; p<0.005). The manufacturer reported that overall the BRAVO trial showed a clinically meaningful and statistically significant effect of ranibizumab on visual acuity and patient‑reported outcomes based on the NEI VFQ‑25 at 6 months. The manufacturer carried out a post‑hoc analysis stratified by rescue treatment with grid laser photocoagulation to investigate the effects of adding this treatment to ranibizumab. The manufacturer concluded that treating patients with grid laser photocoagulation as well as ranibizumab did not lead to the efficacy of ranibizumab being overestimated.

3.6 At month 6 in the CRUISE trial, patients in the ranibizumab group achieved a statistically significant mean gain in BCVA score from baseline of 14.9 letters (95% CI 12.6 to 17.2) compared with the sham group, who gained 0.8 letters (95% CI −2.0 to 3.6, p<0.0001). The manufacturer reported that the improvements in BCVA in the ranibizumab group at month 6 were generally maintained through to month 12 with treatment as needed (13.9 letters [95% CI 11.5 to 16.4] for ranibizumab; 7.3 letters [95% CI 4.5 to 10.0] for sham [plus ranibizumab] group; p<0.001). Patients receiving ranibizumab 0.5 mg showed statistically significantly greater improvements in patient‑reported outcomes as measured by the NEI VFQ‑25 (6.2 points, 95% CI 4.3 to 8.0) than patients receiving sham injection (2.8 points, 95% CI 0.8 to 4.7; p<0.05).

3.7 For patients who entered the open‑label extension study (HORIZON), ranibizumab 0.5 mg was given at intervals of at least 30 days. Sixty‑seven per cent of patients from BRAVO and 60% of patients from CRUISE completed month 12 of HORIZON. The primary outcome for the HORIZON extension study was mean change from HORIZON baseline in BCVA score up to 24 months. The manufacturer presented results from the first 12 months. From the BRAVO trial baseline, patients receiving sham (plus ranibizumab) and those receiving 0.5 mg ranibizumab had mean gains in BCVA score of 15.6 letters and 17.5 letters respectively. From the CRUISE trial baseline, patients receiving sham (plus ranibizumab) and those receiving 0.5 mg ranibizumab had mean gains in BCVA score of 7.6 and 12.0 letters respectively (no confidence intervals reported).

3.8 Adverse events were reported at 6 months and 12 months in both BRAVO and CRUISE trials, and for a further 12 months' follow‑up in the HORIZON extension study. In BRAVO, at 6 months there were 7 ocular adverse events (5.4%) in the ranibizumab group compared with 17 (13%) in the sham group, excluding occurrences of raised intraocular pressure. Non‑ocular serious adverse events (potentially related to vascular endothelial growth factor [VEGF] inhibition) at 6 months were higher in the ranibizumab group (5 events [3.8%]) than in the sham group (1 event [0.8%]). In CRUISE, at 6 months there were 13 ocular adverse events (10.1%) in the ranibizumab group compared with 25 (19.4%) in the sham group, excluding occurrences of raised intraocular pressure. In CRUISE, non‑ocular serious adverse events (potentially related to VEGF inhibition) were similar in both the ranibizumab and sham groups (3 [2.3%] and 2 [1.6%] respectively). The most common adverse event reported in BRAVO and CRUISE at 12 months was cataract, with 8 (6.2%) and 9 (7%) instances associated with ranibizumab treatment respectively; in the sham (plus ranibizumab) group, 3 (2.6%) and 2 (1.8%) instances of cataract were reported for the treatment period of 6 to 12 months. Instances of raised intraocular pressure were reported in both BRAVO and CRUISE at 6 months but were academic in confidence, and therefore not reported here. In the HORIZON extension study, the incidence of any adverse event in the sham (plus ranibizumab) and ranibizumab groups was 2.2% and 5.8% respectively for the patients (with BRVO) recruited from BRAVO, 5.2% and 3% respectively for the patients (with CRVO) recruited from CRUISE.

3.9 A systematic review was undertaken to identify RCTs involving potential comparators for ranibizumab in the treatment of visual impairment caused by macular oedema secondary to RVO. The manufacturer discussed the feasibility of conducting a formal indirect comparison of ranibizumab with dexamethasone intravitreal implant or bevacizumab in CRVO, and an indirect comparison of ranibizumab with dexamethasone intravitreal implant, bevacizumab or grid laser photocoagulation in BRVO. For a comparison of ranibizumab and bevacizumab in CRVO, the manufacturer identified a study by Faghihi et al. (2008) but stated that there was not enough information about the baseline characteristics of patients in the study. For a comparison with bevacizumab in BRVO, studies by Moradian et al. (2011) and Russo et al. (2009) were identified. The manufacturer stated that an indirect comparison could not be conducted without bias because the length of time since diagnosis of macular oedema differed in Moradian et al. (2011) and BRAVO, and because the trial duration was different in all 3 studies. The manufacturer also considered that bevacizumab was not an appropriate comparator because it did not consider that its use in the NHS was routine or best practice. The manufacturer stated that an indirect comparison of ranibizumab and dexamethasone intravitreal implant could not be undertaken for CRVO or BRVO because of the population differences in trials involving these 2 treatments: patients had different lengths of time since diagnosis of macular oedema, different baseline ranges of BCVA and different retinal thickness in the BRAVO and CRUISE trials compared with the GENEVA studies (which compared dexamethasone with sham injection). The manufacturer also stated that ranibizumab could not be compared indirectly with grid laser photocoagulation because of fundamental differences in trial design between BRAVO, which was sham‑injection‑controlled, and the laser studies BVOS (1984) and Battaglia et al. (1999), which were not.

3.10 Although no formal indirect comparison of ranibizumab with other drug treatments was performed, the relative systemic safety profiles of ranibizumab and bevacizumab were discussed in the manufacturer's submission. The manufacturer stated that ranibizumab was associated with a better safety profile than bevacizumab. The manufacturer provided data from 3 large retrospective studies by Carneiro et al. (2011), Curtis et al. (2010) and Gower et al. (2011) in support of this statement, but these studies compared bevacizumab with ranibizumab for the treatment of age‑related macular degeneration (AMD) rather than RVO. The manufacturer acknowledged that AMD manifests later in life than RVO, and so the average age of patients in the BRAVO and CRUISE trials was lower than in the studies of AMD.

3.11 For evidence of cost effectiveness, the manufacturer submitted a Markov state transition model comparing treatment with ranibizumab with grid laser photocoagulation (standard care) for visual impairment caused by macular oedema secondary to BRVO and with best supportive care for CRVO. Treatment was modelled over a 15‑year time horizon for a hypothetical cohort of 1000 patients with visual impairment because of macular oedema secondary to RVO, with a starting age of around 66 years. Eight BCVA health states and death are included in the model structure, with each health state having an associated utility and mortality risk depending on whether the 'better‑seeing eye' or 'worse‑seeing eye' is treated. In the manufacturer's base‑case analysis, it was assumed that all patients are treated in their 'better‑seeing eye'. People move through the model in monthly cycles, accumulating the utility associated with each health state they enter, together with the costs of treatment and subsequent monitoring. Additional costs and disutility associated with blindness were applied for people with a visual acuity equal to or less than 35 letters in the 'better‑seeing eye'. The model assumed that a person's risk of mortality would increase with worsening visual acuity in the 'better‑seeing eye'. A published study by Christ et al. (2008) was used to provide the risk levels by ETDRS bands. The manufacturer asserted that mortality associated with RVO would not be expected to result in any additional risk of mortality over and above that of the general population and as a consequence, the model did not include an assumption of excess mortality associated with RVO.

3.12 Transition probabilities were determined monthly and subsequently used to calculate overall monthly transition probabilities for months 0 to 1, months 2 to 6 and months 7 to 12. For CRVO, the probabilities derived from the sham group of the CRUISE trial for months 2 to 6 were applied to months 2 to 6, 7 to 12 and 13 to 24 in the best supportive care arm of the model. The manufacturer stated that this was because there were no comparative data beyond month 6; from this point in the trial treatment with ranibizumab could be given to people in either arm as needed. Similarly for BRVO, no comparative data existed beyond month 6. However, this was further complicated by the use of grid laser photocoagulation as a rescue treatment in both arms of the trial beyond month 3. Therefore the probabilities for months 7 to 12 were pooled from the sham and ranibizumab groups of BRAVO and applied to months 7 to 12 and months 13 to 24 in both arms of the model. Dexamethasone intravitreal implant was incorporated into the model by a combination of applying relative risks from 2 trials (GENEVA studies) and assigning probabilities observed in the control groups of the BRAVO and CRUISE trials.

3.13 The manufacturer conducted a systematic review of the literature to identify utility values for populations with visual impairment because of RVO, with priority given to populations with macular oedema secondary to BRVO or CRVO. From the results of this review, the manufacturer chose a study by Brown (1999) as the source of utilities for the model, stating that this was the only relevant study that reported utility values related to visual acuity. This is a US study assessing preferences for different levels of visual acuity in patients with vision loss from various causes, 7% of whom had RVO. The manufacturer's model applies different utility values to each BCVA health state, depending on whether the 'better‑seeing eye' or 'worse‑seeing eye' is treated. Although separate utility values for visual acuity in the 'better‑seeing eye' and 'worse‑seeing eye' were available from the study by Brown, the manufacturer stated that there was little difference between the worst and best health states for people treated in their 'worse‑seeing eye' and therefore assumed a value of 0.85 for all BCVA health states for people treated in the 'worse‑seeing eye' (that is, no gain from treatment). In addition, the base‑case model assumed all people would be treated in their best‑seeing eye and therefore this issue of utility gain for people treated in their 'worse‑seeing eye' was not relevant to the base case. For people treated in their 'best‑seeing eye' the maximum utility gain from treatment was 0.41 (that is, the difference between the best visual health state of 0.92 and the worst visual health state of 0.51). Utilities were not adjusted for age.

3.14 Costs included intervention and comparator costs, administration costs and follow‑up visits. Grid laser photocoagulation (BRVO only) was assumed to incur no cost but an administration cost as an outpatient procedure was applied. The ranibizumab injection administration visit was costed as an office‑based outpatient procedure, whereas the dexamethasone intravitreal implant injection was based on a weighted average of the cost of an outpatient procedure (25%) and day case procedure (75%) to account for its greater complexity. For ranibizumab, the frequency of injections was taken from the BRAVO and CRUISE trials. In addition, patients experiencing adverse events had an associated cost applied, and patients considered to be blind had the additional costs associated with blindness.

3.15 All of the manufacturer's base‑case incremental cost‑effectiveness ratios (ICERs) that were presented included an approved patient access scheme, which was offered by the manufacturer at the time of submission (and which was subsequently superseded). In the base case for BRVO, the ICER for ranibizumab compared with grid laser photocoagulation was £20,494 per quality‑adjusted life year (QALY) gained. In the base case for CRVO, the ICER for ranibizumab compared with best supportive care was £8643 per QALY gained. The base‑case ICERs for ranibizumab compared with dexamethasone for BRVO and CRVO were £5486 and £7174 per QALY gained respectively. Incremental costs and QALYs for the base‑case results were commercial in confidence and therefore cannot be presented here.

3.16 The manufacturer performed a deterministic sensitivity analysis and found the model to be sensitive to the frequency of injections and follow‑up visits. The manufacturer performed scenario analyses to assess the impact of varying the proportion of people treated in their 'worse‑seeing eye' and commented that this was a key driver of cost effectiveness. The manufacturer also presented probabilistic sensitivity analyses and concluded that the probability that ranibizumab was cost effective when compared with grid laser photocoagulation in BRVO was 45.5% and 57.2% at thresholds of £20,000 and £30,000 per QALY gained respectively. For ranibizumab compared with best supportive care in CRVO, the probability of cost effectiveness was estimated by the manufacturer to be 74.5% and 83.3% at thresholds of £20,000 and £30,000 per QALY gained respectively.

Evidence Review Group comments on the manufacturer's submission

3.17 The ERG noted that most patients with retinal ischaemia were excluded from the BRAVO and CRUISE trials, because one of the exclusion criteria was brisk afferent pupillary defect, which, as the manufacturer stated, equates to severe retinal ischaemia. Therefore, the ERG considered that the results of any analyses could only be applied to people without retinal ischaemia.

3.18 The ERG considered that the concomitant use of grid laser photocoagulation from month 3 confounded the results of the BRAVO study and that definite conclusions could not be drawn about the effects of ranibizumab compared with sham injection or grid laser photocoagulation alone. The ERG noted that there was not enough evidence to conclude that grid laser photocoagulation had no effect in the ranibizumab group. It also noted that the treatment period of the BRAVO trial was not long enough to capture any benefits of grid laser photocoagulation on patient outcomes, which may last longer than 3 years. Furthermore, clinical advice to the ERG suggested that ranibizumab and grid laser photocoagulation would not be used together to treat patients in clinical practice.

3.19 The ERG noted that from month 6 onwards people were allowed to have ranibizumab as needed and therefore considered the data up to month 6 to be the most relevant data for determining the comparative effectiveness of ranibizumab in treating visual impairment caused by macular oedema secondary to RVO. However, the ERG noted that this period may not be long enough to determine the long‑term effects of ranibizumab.

3.20 The ERG questioned the manufacturer's view that an indirect estimate of the efficacy of ranibizumab compared with bevacizumab, dexamethasone and grid laser photocoagulation (for BRVO only) was not possible. Although the ERG accepted that there were differences in baseline characteristics between patients in the CRUISE, BRAVO and GENEVA trials (see section 3.9), it stated that this would not prevent an indirect comparison between ranibizumab and dexamethasone, and would likely favour ranibizumab. This was because at baseline the mean duration of macular oedema secondary to CRVO was 3 months in CRUISE and 5 months in GENEVA, and a greater mean duration of RVO tends to result in a poorer response to treatment. The ERG suggested that the impact of any bias could have been explored through critical assessment. The ERG agreed with the manufacturer that it was not possible to incorporate bevacizumab for people with CRVO into an adjusted indirect comparison because only 1 study of bevacizumab for CRVO had been identified in the manufacturer's submission (Faghihi et al. 2008), and this did not adequately report baseline characteristics. For BRVO, the ERG considered that studies of bevacizumab reported in the manufacturer's submission (see section 3.9) were suitable for inclusion in an indirect comparison with the first 3 months of data from BRAVO (that is, before rescue treatment with grid laser photocoagulation was permitted). Again, the ERG accepted that there would be some bias in this comparison, but overall the biases would likely favour ranibizumab because the duration of macular oedema in BRAVO was longer than in the study by Moradian et al. and because the Moradian study included more patients with ischaemia than BRAVO. The ERG highlighted that this could be explored in critical assessment.

3.21 From the trials reported in the manufacturer's submission, the ERG was able to construct a linear network of trials using BRAVO (ranibizumab compared with sham), Moradian et al. 2011 (bevacizumab compared with sham) and Russo et al. 2009 (bevacizumab compared with grid laser photocoagulation). The ERG commented that although the results should be treated with caution because they are exploratory, they estimated an approximately 3‑letter improvement in visual acuity with ranibizumab over bevacizumab and an 8‑letter improvement with ranibizumab over grid laser photocoagulation at month 3. However, the ERG did not consider the difference between bevacizumab and ranibizumab to be clinically meaningful from this analysis.

3.22 The ERG noted that in the base‑case analysis the model assumed all patients were treated in the 'better‑seeing eye', despite the fact that over 90% of patients in the BRAVO and CRUISE trials were treated in their 'worse‑seeing eye'. The ERG considered that it was not reasonable to assume equivalent gains in utility and reductions in costs when treating a patient in their 'worse‑seeing eye'. The ERG considered the manufacturer's use of a 'better‑seeing eye' model to be inappropriate because RVO is predominantly a unilateral condition, and therefore most patients would receive treatment in their 'worse‑seeing eye' only.

3.23 The ERG considered the pooled transition probabilities for ranibizumab, which the manufacturer stated had been necessary to account for the effect of grid laser photocoagulation in people with BRVO. The ERG commented that pooling would lead to an overestimate of the efficacy of ranibizumab because the benefit seen in patients in the sham group who received ranibizumab after the first 6 months would be added to the continued effect of ranibizumab in those patients initially randomised to receive ranibizumab. The ERG conducted sensitivity analyses using unpooled transition probabilities. It noted that the ICER for ranibizumab compared with grid laser photocoagulation (for BRVO) with this change alone increased the manufacturer's original base‑case from £20,494 to £52,004 per QALY gained for months 7 to 12, and ranibizumab was dominated (was less clinically effective and more expensive) for months 13 to 24, and months 7 to 12 plus months 13 to 24 together. Incremental costs and QALYs for the base‑case results were commercial in confidence and therefore cannot be presented here.

3.24 The ERG considered the manufacturer's exploratory economic analysis that incorporated dexamethasone intravitreal implant. The ERG commented that there was a potential bias towards ranibizumab in the manufacturer's approach (see section 3.20). The ERG conducted an exploratory indirect comparison of ranibizumab with dexamethasone intravitreal implant, which provided relative risks of an improvement in visual acuity of 10 letters (2 lines) or more for patients with macular oedema secondary to BRVO and CRVO (a relative risk of less than 1 favours ranibizumab). The relative risks increased from 0.55 to 0.79 for ranibizumab compared with dexamethasone in BRVO. For CRVO, the corresponding figures were 0.30 to 0.40. The ERG commented that the relative risks calculated from the manufacturer's model were more favourable to ranibizumab in both BRVO and CRVO. Moreover, because the ERG's indirect comparison was known to be biased towards ranibizumab, the manufacturer's approach to modelling dexamethasone was largely biased towards ranibizumab. However, the ERG commented that these results were exploratory and should therefore be interpreted with caution.

3.25 The ERG noted the manufacturer's assumption in the economic model that there is no mortality risk attributable to RVO. The ERG identified a UK‑based study (Tsaloumas et al. 2000) that concluded that patients with RVO were at a significantly greater risk of death from myocardial infarction than the general population. The ERG was of the opinion that it would have been appropriate to include the relative risk of 1.6 reported by Tsaloumas in the manufacturer's base‑case analysis.

3.26 The ERG noted the manufacturer had applied a mortality risk associated with the visual acuity level in the patient's 'better‑seeing eye' using data from Christ et al. (2008). The ERG noted that this mortality risk would only apply to patients being treated in their 'better‑seeing eye'. The ERG commented that if the model was amended to treat those with visual impairment in their 'worse‑seeing eye' it would be appropriate to use a mortality risk associated with 'some' visual impairment in these patients.

3.27 The ERG noted that the utility values for visual acuity in the 'better‑seeing eye' were taken from Brown (1999) rather than Czoski‑Murray et al. (2009), an earlier version of which had been used (referred to as Brazier et al. 2006) in Ranibizumab and pegaptanib for the treatment of age\u2011\related macular degeneration (NICE technology appraisal guidance 155). The ERG noted that the manufacturer's model assumes utilities are independent of age, although age adjustment is expected to have minimal impact on the ICERs. However, the ERG commented that age adjustment of the utilities presented by Czoski‑Murray et al. (2009) was not necessary because age had already been adjusted for. The ERG commented that the study by Czoski‑Murray et al. (2009) should therefore be used as the source for utility associated with visual acuity in the 'better‑seeing eye' in this assessment. For the 'worse‑seeing eye' the ERG was of the opinion that the available evidence from the Brown publication (which reported utilities by worse‑ and better‑seeing eyes) suggested the maximum gain from treating a person's 'worse‑seeing eye' would be 0.1.

3.28 The ERG carried out exploratory analyses varying several parameters. Assuming that only 10% (and not 100%) of people are treated in their 'better‑seeing eye', applying utilities derived from Czoski‑Murray et al. (2009), a 0.1 overall benefit associated with treating the 'worse‑seeing eye' and an increased risk of mortality associated with RVO, the ICERs for ranibizumab compared with best supportive care and with dexamethasone in CRVO were £43,760 and £37,443 per QALY gained respectively. This formed the ERG's base‑case estimate for CRVO. The ERG did not present further economic evaluation of ranibizumab compared with grid laser photocoagulation because it considered that the confounded data from BRAVO (in which grid laser photocoagulation was permitted in the ranibizumab arm after 3 months of treatment) was insufficient to inform an indirect comparison and be used in the economic model. When the ERG applied their preferred assumptions as above for ranibizumab compared with dexamethasone intravitreal implant in BRVO, this increased the manufacturer's base‑case ICER from £5486 to £31,122 per QALY gained. The ERG's exploratory analyses highlighted that the key drivers that increased the manufacturer's base‑case ICERs were amending the proportion of patients treated in their 'better‑seeing eye' (10% instead of 100%) and the assumption of some benefit associated with treating the 'worse‑seeing eye'.

3.29 The ERG commented that its comparisons of ranibizumab with dexamethasone intravitreal implant for both BRVO and CRVO used relative risks derived from the manufacturer's model (0.55 for BRVO and 0.30 for CRVO) rather than those derived from the ERG's indirect comparison (0.79 and 0.40 respectively). The ERG commented that this would bias the results in favour of ranibizumab and if the ERG's suggested relative risks were applied, the ICER would increase further. The ERG also commented that the efficacy of dexamethasone was potentially underestimated because of differing patient characteristics in the trials that informed the comparison (patients had differing durations of macular oedema from diagnosis in the GENEVA trials compared with BRAVO and CRUISE). Therefore, the manufacturer's base‑case ICERs may be underestimates. The ERG also highlighted that the ICER generated for ranibizumab compared with dexamethasone from the BRAVO trial is derived using the pooled transition probabilities in the original submission. Using the unpooled transition probabilities would increase the ICER further.

3.30 The ERG conducted an exploratory cost‑minimisation analysis for ranibizumab compared with bevacizumab, assuming equivalent efficacy for the 2 treatments (in BRVO and CRVO) and an acquisition cost of £50 per month for bevacizumab. The ERG presented data on the incremental costs of ranibizumab compared with bevacizumab that included commercial‑in‑confidence information and so cannot be presented here. The ERG's analysis using the manufacturer's model suggested that ranibizumab would need to generate 1.5 times more QALYs than bevacizumab (each month between months 2 and 6) in macular oedema secondary to BRVO (without considering the revised patient access scheme implemented in NICE technology appraisal guidance 274) to give an ICER at the top end of the range usually considered cost effective. Ranibizumab would need to generate 1.7 times more QALYs than bevacizumab for macular oedema secondary to CRVO (without considering the revised patient access scheme implemented in NICE technology appraisal guidance 274) to give an ICER at the top end of the range usually considered cost effective.

Revised economic model submitted by the manufacturer during consultation

3.31 In response to the consultation on the appraisal consultation document the manufacturer submitted a revised cost–utility analysis addressing the Committee's concerns about the original model. The revised economic model included the ERG's preferred assumptions relating to 10% of patients being treated in their 'better‑seeing eye', utilities from Czoski‑Murray et al. (2009). The manufacturer also applied some alternative assumptions (see sections 3.33 to 3.36) to present revised base‑case cost‑effectiveness estimates for ranibizumab compared with dexamethasone in both BRVO and CRVO.

3.32 The manufacturer considered that the ERG's approach to deriving utilities from Czoski‑Murray et al. (2009) underestimated the utility gains associated with improving visual acuity. The manufacturer therefore provided an alternative to the ERG's derivation of utilities by applying a regression equation from the Czoski‑Murray et al. (2009) publication to derive utilities for each of the 8 BCVA health states in a similar way to that employed in the guidance on Ranibizumab and pegaptanib in age\u2011\related macular degeneration (NICE technology appraisal guidance 155). The manufacturer also applied a 0.3 overall benefit of treating the 'worse‑seeing eye' because it felt that a 0.1 benefit, as applied by the ERG, did not capture the difference in utility for people with blindness in their 'worse‑seeing eye'.

3.33 The manufacturer did not apply an excess mortality risk specifically associated with RVO in its revised model because the available evidence was conflicting, and it considered the evidence base to be inconclusive. In addition, the manufacturer highlighted that an excess mortality risk had not been included in the evidence submitted for the published guidance on Dexamethasone intravitreal implant in macular oedema secondary to retinal vein occlusion (NICE technology appraisal guidance 229). However, the manufacturer did amend the model to include mortality risk associated with visual impairment in the 'worse‑seeing eye' as originally suggested by the ERG (see section 3.26).

3.34 The manufacturer acknowledged the Committee's concerns about the use of pooled transition probabilities during months 7–24 of the BRAVO trial, which were originally used to account for the confounding effect of patients being treated with grid laser photocoagulation at the same time as treatment with ranibizumab. In its revised model, the manufacturer applied data for months 7–24 from only the ranibizumab arm of BRAVO to inform the transitions of all BRAVO patients.

3.35 The manufacturer's revised model included updated adverse event rates for year 2 for ranibizumab as well as updated adverse event rates for dexamethasone for year 1, based on 12‑month outcomes from the GENEVA studies published since the manufacturer's original submission. In addition, the manufacturer considered that the dexamethasone re‑treatment frequency included in the original model (every 6 months) was conservative and therefore applied a re‑treatment frequency of 4 months to the revised model.

3.36 Finally, the manufacturer applied a lifetime time horizon instead of the 15‑year time horizon included in the original submission in order to derive its base‑case cost‑effectiveness estimates. The manufacturer stated that this was consistent with the approach taken in the guidance on Dexamethasone intravitreal implant in macular oedema secondary to retinal vein occlusion (NICE technology appraisal guidance 229).

3.37 The manufacturer highlighted in its consultation response that although patients with brisk afferent pupillary defect were excluded from the BRAVO and CRUISE studies, this only represents those at the more severe end of the ischaemic spectrum and therefore did not exclude those with milder ischaemia.

3.38 After consultation on the appraisal consultation document, the manufacturer responded to the Committee's concerns regarding the extent of bias generated by the differences in the duration of macular oedema between GENEVA and BRAVO/CRUISE. The manufacturer commented that the extent of bias was not known and that the implications for the ICER may be minimal. They also re‑examined the reported mean durations of macular oedema in the GENEVA (dexamethasone) and BRAVO and CRUISE (ranibizumab) studies noting that in the GENEVA studies, the mean duration of macular oedema was assessed at a screening visit which occurred at least 1 month earlier than in the BRAVO and CRUISE studies. The manufacturer further noted that its cost‑effectiveness analysis of ranibizumab compared with dexamethasone was already favourable to dexamethasone because the relative effectiveness of ranibizumab and dexamethasone was assessed at 3 months, and conservative assumptions related to the number of dexamethasone injections and the rate of adverse events associated with dexamethasone were used.

3.39 In their response to the appraisal consultation document consultation the manufacturer reiterated their view that bevacizumab is not a valid comparator because it does not satisfy the definition of a comparator as set out in the NICE Guide to the methods of technology appraisal because, in their view, the use of bevacizumab is not routine or best practice, and because bevacizumab is not licensed for RVO. The manufacturer further highlighted that the studies included in the ERG's indirect comparison of ranibizumab with bevacizumab had methodological shortcomings and that the inclusion of the study by Russo et al. (2009) was inappropriate because standard deviations were not reported in this study for the values of change from baseline. The manufacturer considered the ERG's conclusion about the direction of bias in the indirect comparison to be overly speculative and not evidence‑based. In addition the manufacturer reiterated its concerns over comparisons with an unlicensed drug that might compromise patient safety, further noting that the absence of a full pharmacovigilance programme (normally funded by the drug sponsor), would mean that the cost of safety surveillance would be a significant burden to the NHS and was not included in the ERG's cost‑minimisation analysis. The manufacturer provided an estimate of the per‑patient cost of a basic pharmacovigilance programme, which was submitted as commercial in confidence and cannot therefore be reported. Finally, the manufacturer highlighted that the ERG's use of a cost‑minimisation analysis is fundamentally flawed and that it was not appropriate to assume equivalent safety and efficacy of ranibizumab and bevacizumab, an assumption required for cost‑minimisation methodology, when this has not been established.

Evidence Review Group's comments on the manufacturer's revised model

3.40 The ERG noted the manufacturer's approach to applying the 'better‑seeing eye' utilities from Czoski‑Murray et al. (2009) to the model and agreed that this was appropriate because it provided utilities for each of the health states in the model rather than the ERG's smaller set of utility values (see section 3.32), and was consistent with the approach taken by the Assessment Group for the guidance on Ranibizumab and pegaptanib in age\u2011\related macular degeneration (NICE technology appraisal guidance 155). The ERG considered the manufacturer's approach of applying a maximum utility gain of 0.3 from treatment of the 'worse‑seeing eye'. The ERG noted that the manufacturer extrapolated the 0.1 utility loss estimated by Brown et al. (2001) to apply further loss in the 'worse‑seeing eye'. However, the ERG considered that the evidence presented by Brown et al. (2001) suggested that further deterioration in visual acuity in the 'worse‑seeing eye' did not affect utility and therefore applying a 0.3 utility gain was not evidence‑based. They also considered that it lacked face validity.

3.41 The ERG noted the manufacturer's response suggesting that the extent of bias towards ranibizumab in comparison with dexamethasone was overestimated (in the original ERG report). The ERG acknowledged that there is uncertainty surrounding the extent of bias but that this is difficult to quantify. With respect to duration of macular oedema at baseline, the ERG did not have access to data from BRAVO and CRUISE to adjust for the difference in the timing of measurement of duration of macular oedema compared with GENEVA. The ERG also noted that although the manufacturer's revised model uses data from months 7–12 of the ranibizumab arm of BRAVO to inform all transition probabilities from month 7 to 24, this assumes equivalent efficacy between dexamethasone and ranibizumab when given as needed. The ERG noted that it is not clear whether this assumption is conservative, and that evidence provided by the manufacturer from the BRAVO trial showed that there is a decline in ranibizumab's efficacy when given as needed rather than as monthly treatment.

3.42 The ERG considered the manufacturer's consultation response to dexamethasone re‑treatment frequency and adverse events associated with dexamethasone. The ERG noted that the manufacturer's approach should have also modelled the effectiveness of dexamethasone re‑treatment given less frequently than every 4 months (for example, every 5 months) and that the manufacturer did not take into account that a higher re‑treatment frequency would result in a more stable efficacy for dexamethasone. In addition, the ERG agreed that a higher re‑treatment frequency for dexamethasone would have the impact of increasing the number and severity of adverse events.

3.43 The ERG noted the manufacturer's response to the issue of ischaemic disease. The ERG agreed that the exclusion criterion in BRAVO and CRUISE of brisk afferent pupillary defect would not exclude patients with minor ischaemic disease. However, the ERG noted that none of the key trials used in the ERG's exploratory analysis, including BRAVO and CRUISE, reported data on the number of patients with baseline macular ischaemia. Therefore, the ERG highlighted that it could not be assumed that patients with ischaemia could have been included in BRAVO and CRUISE and stated that the effects of ranibizumab in the subgroup of patients with macular oedema secondary to ischaemic RVO were unknown.

Patient access scheme as revised in the context of NICE technology appraisal guidance 274

3.44 The manufacturer submitted a revised patient access scheme in 2013 for consideration in this appraisal (as revised in the context of NICE technology appraisal guidance 274), in which it applied a revised discount to ranibizumab for all indications (see section 2.4). The manufacturer did not submit any additional clinical‑effectiveness data but submitted an economic model that incorporated the revised patient access scheme discount and employed all of the revised assumptions from the manufacturer's response to consultation (outlined in more detail in sections 3.31 to 3.39). In summary these were as follows:

  • deriving utilities from Czoski‑Murray et al. (2009) for each of the 8 BCVA health states in a similar way to that employed in the guidance on Ranibizumab and pegaptanib in age\u2011\related macular degeneration (NICE technology appraisal guidance 155) by applying a regression equation from Czoski‑Murray et al. (2009)

  • applying a 0.3 overall utility benefit for treating the 'worse‑seeing eye'

  • including mortality risk associated with visual impairment as suggested by the ERG

  • applying data for months 7–24 from only the ranibizumab arm of BRAVO to inform the transition probabilities of all BRAVO patients

  • including updated adverse event rates for year 2 for ranibizumab as well as updated adverse event rates for dexamethasone for year 1

  • applying a lifetime time horizon to be consistent with the published guidance for Dexamethasone intravitreal implant for macular oedema secondary to retinal vein occlusion (NICE technology appraisal guidance 229).

3.45 By applying all of their revised assumptions (see section 3.44), the manufacturer's revised base‑case ICERs (with the revised patient access scheme as implemented in NICE technology appraisal guidance 274) for ranibizumab compared with dexamethasone were £2370 and £6995 per QALY gained for BRVO and CRVO respectively. For ranibizumab compared with best supportive care in CRVO the ICER was £13,851 per QALY gained, whereas for ranibizumab compared with grid laser photocoagulation in BRVO, the ICER was £23,073 per QALY gained.

3.46 The manufacturer performed a deterministic sensitivity analysis and found the model to be sensitive to the frequency of injections and follow‑up visits; for example in BRVO, increasing the number of injections in year 2 (from 2.5 to 6) increased the ICER for ranibizumab compared with dexamethasone from £2370 to £9892 per QALY gained.

3.47 The manufacturer provided scenario analyses (one‑way sensitivity analysis) which explored the effect of changing some of the parameters in the model individually. The manufacturer explored the effect of applying a 0.2 overall utility gain for treating the 'worse‑seeing eye'. This increased the base‑case ICERs of ranibizumab compared with dexamethasone from £2370 to £3029 per QALY gained for BRVO, and from £6995 to £9005 per QALY gained for CRVO. For ranibizumab compared with best supportive care in CRVO, the ICER increased from £13,851 to £18,332 per QALY gained. For ranibizumab compared with grid laser photocoagulation, the manufacturer's ICER increased from £23,073 to £30,778 per QALY gained.

3.48 Other scenario analyses included reducing the mean number of ranibizumab injections in year 2 (the revised number of mean injections was submitted as academic in confidence and cannot be presented). This reduced the ICERs of ranibizumab compared with dexamethasone by 54% in BRVO and by 58% in CRVO. A scenario which included longer follow‑up data based on year 2 of the HORIZON extension study reduced the ICERs for ranibizumab compared with dexamethasone from £2370 to £1599 per QALY gained and from £23,073 to £20,911 per QALY gained for ranibizumab compared with grid laser photocoagulation in BRVO. The corresponding scenario analysis for CRVO was not presented.

3.49 The manufacturer also presented probabilistic sensitivity analyses and concluded that from the base‑case results the probability that ranibizumab was cost effective when compared with grid laser photocoagulation in BRVO was 44.2% at a threshold of £20,000 per QALY gained and 58.6% at a threshold of £30,000 per QALY gained. For ranibizumab compared with best supportive care in CRVO, the probability of cost effectiveness was estimated by the manufacturer to be 67.9% and 82.0% at thresholds of £20,000 and £30,000 per QALY gained respectively.

ERG critique of manufacturer's revised base case with the patient access scheme as revised in the context of NICE technology appraisal guidance 274

3.50 The ERG noted that the revised model used to inform the current patient access scheme submission is the same as that submitted as part of the manufacturer's response to the original consultation, but that not all issues raised by the Committee have been addressed. Therefore, the ERG's views outlined in sections 3.40 to 3.43 (relating to deriving utilities, transition probabilities, re‑treatment frequency and adverse events associated with dexamethasone and ischaemic disease) still apply. Overall, the ERG accepted the manufacturer's approach to:

  • modelling 90% of people in the model as being treated in their 'worse‑seeing eye'

  • the use and implementation of the 'better‑seeing eye' utilities derived from Czoski‑Murray et al. (2009)

  • excess mortality associated with visual impairment in the 'worse‑seeing eye'

  • updated adverse events.

    However, the ERG maintained that the available evidence relating to utility gain from treating the 'worse‑seeing eye' suggests a utility decrement of 0.1 (rather than the manufacturer's assumption of 0.3) between the best and worst 'worse‑seeing eye' BCVA health states. Therefore, the ERG provided amended exploratory cost‑effectiveness estimates to include the assumption of a 0.1 utility decrement. For BRVO, this increased the manufacturer's base‑case ICER for ranibizumab compared with grid laser photocoagulation from £23,073 to £44,713 per QALY gained. In the ERG's incremental analysis in BRVO, dexamethasone was extendedly dominated by ranibizumab (that is, the ICER for dexamethasone compared with grid laser photocoagulation was higher than for ranibizumab compared with grid laser photocoagulation) and the ICER compared with best supportive care was £44,713 per QALY gained. The ERG highlighted that there is considerable uncertainty in the comparisons of ranibizumab with grid laser photocoagulation and with dexamethasone as a result of the confounded data from BRAVO (grid laser photocoagulation was permitted in the ranibizumab arm after 3 months of treatment) used to inform the comparison with grid laser photocoagulation and the absence of a direct comparison with dexamethasone.

3.51 For ranibizumab compared with dexamethasone in BRVO, the ERG's amendment of including a 0.1 utility decrement increased the manufacturer's base‑case ICER from £2370 to £4092 per QALY gained. The ERG performed the same amendment for CRVO, and the ICER for ranibizumab compared with best supportive care increased from £13,851 to £26,263 per QALY gained, and for ranibizumab compared with dexamethasone the ICER increased from £6995 to £12,306 per QALY gained. In the ERG's incremental analysis for CRVO, dexamethasone was extendedly dominated by ranibizumab (that is, the ICER for dexamethasone compared with best supportive care was higher than for ranibizumab compared with grid laser photocoagulation) and the ICER for ranibizumab compared with best supportive care was £26,263 per QALY gained. The ERG highlighted that the absence of a direct comparison of ranibizumab with dexamethasone generates considerable uncertainty in these results. In particular, the manufacturer assumed that from month 7 onwards, the efficacy of dexamethasone is equivalent to ranibizumab when given as needed (rather than monthly). The ERG note that it remains unclear whether this assumption would lead to bias towards or against ranibizumab.

NICE Decision Support Unit report

3.52 Following the Committee's consideration of comparators in this appraisal, and in line with NICE processes (specifically section 3.5.49 of the Guide to the single technology appraisals process), the NICE Board asked for additional work to be commissioned from the NICE Decision Support Unit (DSU) related specifically to the consideration of intravitreal bevacizumab as a comparator.

3.53 The DSU report considered 4 questions:

  • What evidence is there relating to the pharmaceutical quality of reformulated bevacizumab as used in eye conditions in general?

  • How widespread is intravitreal bevacizumab use in the UK?

  • What is the evidence for efficacy of intravitreal bevacizumab in adults with RVO (and diabetic macular oedema) specifically?

  • What evidence is there regarding adverse events for intravitreal bevacizumab in eye conditions in general?

3.54 The DSU report noted that the process of diluting and aliquoting bevacizumab into the smaller doses required for intravitreal injections requires a 'specials' licence issued by the Medicines and Healthcare products Regulatory Agency (MHRA) and can be performed by hospital pharmacists or on a larger scale by specialist units under tightly controlled conditions. The DSU identified Moorfields Pharmaceuticals (a manufacturing arm of Moorfields Eye Hospital NHS Foundation Trust) and Liverpool and Broadgreen University Hospitals pharmacy as the 2 major suppliers of intravitreal bevacizumab in the UK, both of which hold 'specials' licences. The DSU report highlighted that the greatest risk from reformulation of bevacizumab is infection such as endophthalmitis, which can lead to loss of vision or even the eye itself, and that there has been a warning issued about this by the US Food and Drug Administration (FDA). Reports of sterile endophthalmitis or uveitis by Moorfields to the MHRA have resulted in the recall of 27 batches of bevacizumab.

3.55 The DSU report investigated the extent of use of intravitreal bevacizumab in the UK by reviewing commissioning policy documents, data from the 2 major suppliers of intravitreal bevacizumab, and a survey of consultant ophthalmologists. The findings suggested that there is substantial use of intravitreal bevacizumab across the UK in eye conditions in general and that its use is even more widespread in private practice.

3.56 The report also reviewed the evidence relating to efficacy of intravitreal bevacizumab specifically in RVO. The DSU identified 5 RCTs that examined the effectiveness of bevacizumab on BCVA in people with RVO, 3 of which were in populations with CRVO; the remaining 2 studies were in BRVO. The studies all suggested that intravitreal bevacizumab appeared to confer some improvement in BCVA in both BRVO and CRVO compared with sham injection. However, because 3 of the studies were only available as conference abstracts, detailed data were not available. In addition, the DSU highlighted that interpretations of the findings should be made with caution because of the small number of studies with relatively small sample sizes and differences in participants' age, gender distribution and type of RVO. The studies also had relatively short follow‑up durations (the maximum was 24 weeks).

3.57 The report assessed the evidence relating to adverse events associated with intravitreal bevacizumab in eye conditions in general. A total of 22 RCTs were identified, which evaluated the safety of bevacizumab compared with laser therapy, sham injection, triamcinolone, ranibizumab, pegaptanib and observational control. In addition, 67 observational non‑RCT studies were included in the safety review of intravitreal bevacizumab. Overall, the DSU report commented that adverse event rates following intravitreal bevacizumab treatment were low when compared with other intravitreal treatments, sham injection and laser therapy and most of these studies were in people with AMD, diabetic macular oedema or RVO. Most outcomes were not significantly different between groups. The DSU report noted that higher rates of adverse events have been reported in the bevacizumab group in the head‑to‑head studies of intravitreal bevacizumab and ranibizumab (CATT and IVAN trials in AMD) and although this was not significant in the IVAN trial, when added to the meta‑analysis with the CATT trial, the overall finding was statistically significant. Overall the DSU considered that the 22 RCTs offer the most robust assessment of adverse events. The DSU commented that the evidence on safety of intravitreal bevacizumab from observational studies was inconclusive. However, with respect to larger studies, observational data from Curtis et al. suggested no difference in the risk of adverse events between bevacizumab and ranibizumab, and another population‑based case–control study reported no relationship between the risk of systemic events such as myocardial infarction, venous thromboembolism, stroke or congestive heart failure and the administration of intravitreal bevacizumab or ranibizumab.

3.58 Comments on the DSU report consultation highlighted that the quality of reformulated bevacizumab might vary between studies and in clinical practice, and there are concerns about the reports of endophthalmitis. However it is unclear how this compares with ranibizumab. Consultees commented that there was insufficient evidence to evaluate the safety of intravitreal bevacizumab, while other consultees noted that the pooled analysis of IVAN and CATT trials, which compared ranibizumab and bevacizumab directly (in people with AMD), showed significantly higher serious systemic adverse events in the bevacizumab group. Some consultees noted that although there is some favourable evidence for efficacy of intravitreal bevacizumab for RVO in comparison with sham injection, the evidence is limited. However, some consultees noted that the use of intravitreal bevacizumab could be substantial but may have declined since the publication of Dexamethasone intravitreal implant for the treatment of macular oedema secondary to retinal vein occlusion (NICE technology appraisal guidance 229).

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

  • National Institute for Health and Care Excellence (NICE)