Dexamethasone intravitreal implant for the treatment of macular oedema secondary to retinal vein occlusion

The manufacturer originally submitted evidence of clinical and cost effectiveness for dexamethasone versus best supportive care (observation). Following the consultation after the first Appraisal Committee meeting, the manufacturer submitted a comparison versus bevacizumab. No comparison was made with triamcinolone (Kenalog formulation), which was defined in the scope as a comparator for the treatment of macular oedema following both BRVO and CRVO. Similarly, dexamethasone was not compared with grid laser photocoagulation for non-ischaemic BRVO.

In the manufacturer's original submission, evidence of clinical effectiveness versus observation was based on two identical randomised, sham-controlled, three-arm parallel-group studies of dexamethasone intravitreal implant in people with macular oedema secondary to BRVO or CRVO. Both studies (GENEVA 008 and GENEVA 009) had an initial 6-month treatment period followed by a 6-month open-label extension in which all patients in both arms of the study who met the re-treatment criteria received a dexamethasone implant. Patients were re-treated if best corrected visual acuity (BCVA) was less than 84 letters or the retinal thickness by optical coherence tomography was more than 250 µm in the central 1 mm macular subfield and, in the investigators' opinion, the procedure would not put the patient at significant risk. All participants had macular oedema for 6 weeks to 12 months before study entry. Participants were allocated in a 1:1:1 ratio to receive a 700-microgram dexamethasone intravitreal implant (n=427), a sham implant (n=426) or a 350-microgram dexamethasone implant (n=414). This appraisal considered the 700-microgram dose which is the only dose which has a UK marketing authorisation. The sham group had a needleless applicator pressed against the conjunctiva actuated with a click. Investigators were masked to study treatment. The results were presented separately for people with retinal vein occlusion (RVO) and the subgroups of people with macular oedema secondary to CRVO, BRVO, BRVO with macular haemorrhage and BRVO with previous laser treatment. People with BRVO who are eligible for laser therapy were not included as a subgroup. The results from the two studies (GENEVA 008 and GENEVA 009) were pooled and this formed the basis of the evidence considered by the Committee, although the data were also available separately for each study in the manufacturer's submission.

The results of the pooled analysis showed that for the total RVO population 21.3% of the 427 patients in the intention-to-treat population receiving dexamethasone had an improvement in BCVA from baseline of at least 15 letters at day 30 compared with 7.5% of 426 patients in the sham group. This rose to 29.3% at day 60 (compared with 11.3% in the sham group) but returned to 21.8% and 21.5% at day 90 and day 180 respectively (compared with 13.1% and 17.6% in the sham group). The differences were statistically significant at day 30 (p<0.001), 60 (p<0.001) and 90 (p=0.008) but not at day 180 (p>0.05). The results for patients who were re-treated at day 180 were presented as academic-in-confidence information and are therefore not presented here.

For the CRVO subgroup, 21.3% of patients in the dexamethasone group had an improvement in BCVA from baseline of at least 15 letters compared with 6.8% in the sham group at day 30 (p<0.001). At day 60, 28.7% in the dexamethasone group had an improvement in BCVA of at least 15 letters compared with 8.8% in the sham group (p<0.001). There was no statistically significant difference between the groups at days 90 and 180.

In the subgroup with BRVO, 21.3% of patients receiving dexamethasone had an improvement in BCVA from baseline of at least 15 letters at day 30 compared with 7.9% in the sham group (p0.001). The corresponding figures for the subgroup with BRVO with macular haemorrhage were 22.0 % and 8.8 % respectively (p≤0.001). Both subgroups had statistically significant differences between patients treated with dexamethasone and the sham group at days 60 and 90, but not at day 180. In the subgroup with BRVO and previous laser therapy 22.2% had an improvement in BCVA from baseline of at least 15 letters at day 30 compared with 2.8% in the sham group (p=0.028). Differences between the dexamethasone and sham groups were also statistically significant at days 60 (p<0.001), 90 (p=0.011) and 180 (p=0.022).

The cumulative response rate for time to achieve an improvement in BCVA of at least 15 letters from baseline in the study eye was statistically significant for dexamethasone versus sham. The difference in mean change from baseline BCVA, the categorical change from baseline BCVA and proportion of patients with an improvement in BCVA of at least 10 letters from baseline in the study eye were statistically significantly higher for dexamethasone versus sham at days 30, 60, 90 and 180 in the pooled analysis. For all RVO at 180 days, the most common adverse events were raised intraocular pressure, eye pain and ocular hypertension. Intraocular pressure was raised in 25.2 % of patients treated with dexamethasone compared with 1.2 % in the sham group. Of patients treated with dexamethasone, 7.4% had eye pain compared with 3.8% in the sham group. Ocular hypertension was experienced by 4% of patients in the treated group compared with 0.7% in the sham group. Presence of anterior chamber cells and retinal neovascularisation were also reported. Other reported adverse events were retinal detachment, retinal tears and cataract. Safety data for the re-treated population (receiving a second implant by day 180) were presented as academic-in-confidence information and are therefore not presented here.

Following the consultation after the first Appraisal Committee meeting, the manufacturer submitted a mixed-treatment comparison of dexamethasone versus bevacizumab. The network of evidence from a systematic review was for BRVO only and included the BRVO data from the GENEVA trials, a non-randomised study by the Branch Vein Occlusion Study group comparing laser with sham treatment (n=78) and a randomised study by Russo et al. comparing laser with bevacizumab (n=30). The outcome was improvement in BCVA using standard effect sizes and a fixed-effects model. Dexamethasone was less effective than bevacizumab with a difference in BCVA of 1.74 letters (95% confidence interval [CI] 9.57 to 6.19) when assessed at day 180. However dexamethasone was more effective than bevacizumab when assessed at day 60, with a gain of 2.55 letters (95% CI 5.28 to 10.48).

For evidence of cost effectiveness, the manufacturer submitted a de novo Markov model that compared treatment with dexamethasone with sham injection in people with macular oedema and vision loss following CRVO or BRVO. Treatment was modelled over a lifetime horizon based on the transition of people between five health states based on the Early Treatment Diabetic Retinopathy Study (EDTRS) measurement of BCVA in the affected eye and death. The worst health state represented visual acuity less than or equal to 38 letters, which equated to severe visual impairment. The best health state was visual acuity of 69 letters or over. The mean BCVA of people in the model was 54 letters, which equates to the second best health state. The patient population was based on data from the GENEVA trials. The model assumed that 90% of people would present with macular oedema in the 'worse-seeing' eye. The model had a cycle length of 1 month for the first 3 months following presentation with RVO, followed by a 3-month cycle in months 4 to 6 and 6-monthly cycles thereafter. Patients entering the model received dexamethasone or observation. Up to 12 months, transition probabilities were based on pooled patient-level data from the GENEVA studies, including the open-label extension. Beyond 12 months, data were extrapolated from 6- to 12-month data for treatment and re‑treatment and 3- to 6-month data for sham. Treatment duration was assumed to last for 2.5 years in people with BRVO and
3 years in people with CRVO; thereafter visual acuity was assumed to be stable.

The data inputs for the manufacturer's model included utility values estimated using the Visual Function Questionnaire Utility Index (VFQ-UI) and mapped onto the health states using an algorithm from a study eliciting preferences from the general population. Resource use was identified from a systematic review of the literature and input from clinical specialists. Costs included drug cost and medical resource use (hospital visits, monitoring, costs associated with blindness and the cost of treating adverse events, including raised intraocular pressure, cataracts, retinal tears/detachment). Costs associated with treating adverse events were assumed to increase with the third and fourth treatment.

Key assumptions of the economic model included:

• 90% of people treated would have macular oedema in the 'worse-seeing' eye

• the stabilisation of visual acuity for 2.5 years in people with BRVO and 3 years in people with CRVO

• re-treatment at 6-monthly intervals with a maximum of five injections for BRVO and six injections for CRVO (with assumptions over the number of treatments received)

• a risk of involvement of the other eye of 6.5% in the first year (for those with initial RVO in their 'worse-seeing' eye)

• blindness and an excess mortality hazard of 1.54 associated with a BCVA in the 'better-seeing' eye of 38 or fewer letters (measured by the EDTRS).
  
  Sensitivity analyses included varying utility estimates, costs, stabilisation of visual acuity at day 360, extrapolation assumptions, mortality, involvement of the other eye, discounting, re-treatment, and people with a worse BCVA on entering the model. Results were presented for the entire RVO population and the subgroups of CRVO, BRVO with macular haemorrhage, BRVO with previous laser therapy, BRVO with a diagnosis of 90 days or less at the time of treatment, and BRVO with a diagnosis of more than 90 days at the time of treatment. In the original sensitivity analyses (manufacturer original submission), the factors having the largest impact on estimates of cost effectiveness for the total population were costs associated with vision loss (costs of residential care and the uptake of residential care), affected eye (proportion of people treated for macular oedema in the 'worse-seeing' eye) and rates of discount.

Following the consultation after the first Appraisal Committee meeting, the model was amended by the manufacturer, resulting in a revised base-case ICER. The revised base case used a ratio of 25% outpatient versus 75% day case procedures for administration and an adjustment to the way that the costs of vision loss were applied to the 'better-seeing' eye in patients whose BCVA in the affected eye falls below < 20/200. An adjustment of 25% plus a further 10% uplift applied every 6 months was made to the average annual costs associated with severe visual impairment. The model continued to use only the last 3 months of observation data from the GENEVA studies.

In the revised base case for all RVO, the total incremental cost was £3,698 for dexamethasone compared with observation and the incremental QALYs were 0.21. The incremental cost-effectiveness ratio (ICER) was £17,558 per QALY gained for dexamethasone intravitreal implant compared with observation in all RVO. In the revised base case for CRVO, the total incremental cost was £4,732 and the incremental QALYs were 0.29. The cost per QALY gained was £16,522 for dexamethasone compared with observation. In the revised base case for BRVO with macular haemorrhage, the total incremental cost was £3,119 and the incremental QALYs were 0.18. The incremental cost per QALY gained was £17,741. In the revised base case for BRVO with previous laser therapy, the total incremental cost was £1,857 and the incremental QALYs were 0.29. The incremental cost per QALY was £6,361 for dexamethasone compared with observation for BRVO with previous laser treatment.

In addition, the manufacturer also provided additional scenario analyses in response to requests in the appraisal consultation document. Alternative scenario analyses for the re-treatment rate included scenarios in which proportions re-treated were as at day 180 for the five injections after the first injection in people with CRVO, proportions re-treated were as at day 180 for the four injections after the first injection in people with BRVO and proportions re-treated were varied between the two extremes of the base case and the GENEVA studies. When proportions re-treated were as at day 180, the ICERs for CRVO, BRVO with macular haemorrhage and BRVO with previous laser treatment increased from £16,522, £17,741 and £6,361 to £22,083, £45,878 and £16,548 per QALY gained respectively. When the mid-point was used, the ICERs for CRVO, BRVO with macular haemorrhage and BRVO with previous laser treatment increased from £16,522, £17,741 and £6,361 to £20,257, £31,123 and £10,876 per QALY gained respectively.

Following the consultation after the first Appraisal Committee meeting, the manufacturer submitted an additional cost–utility analysis versus bevacizumab for BRVO. This used the mixed-treatment analysis to obtain treatment effects for bevacizumab compared with dexamethasone (a non-significant mean gain of 1.74 letters for bevacizumab compared with dexamethasone at 6 months). Compared with bevacizumab, in the base case for all RVO, the net marginal benefit for dexamethasone (where net marginal benefit is larger than zero and treatment with dexamethasone considered cost effective) was £1,927 at £30,000 per QALY gained. The manufacturer submitted a cost-minimisation analysis for dexamethasone versus bevacizumab in CRVO assuming equivalent efficacy for bevacizumab and dexamethasone in CRVO. This was associated with a cost saving of £4,463 with dexamethasone which was mainly a result of a lower frequency of injections with dexamethasone compared with bevacizumab and fewer subsequent follow-up visits.

The manufacturer also submitted a second exploratory cost-minimisation analysis for BRVO and a comparison using ranibizumab as a substitute for bevacizumab. This analysis was submitted as commercial in confidence because of its exploratory nature and so is not presented here.

The manufacturer also provided more information about the location and extent of macular haemorrhage in the subgroup of patients for whom laser treatment was not considered appropriate because of macular haemorrhage. In the GENEVA trials, the location and extent was assessed by standardised fundus photographs evaluated by trained graders using a standardised grading protocol (the ETDRS macular oedema grading protocol) and masked to patient group. Retinal haemorrhage in the macula was defined when the grader was at least 90% certain that the retinal haemorrhage was present in the grid (macular area).

The ERG considered the GENEVA trials to be of high quality. Although there was a statistically significant increase in the BCVA based on the mean letter score with the dexamethasone implant, the ERG did not consider this to be clinically significant because most patients did not achieve a 15-letter improvement from baseline. However, a higher proportion had an improvement of at least 10 letters. The effectiveness of the dexamethasone implant appeared to peak at around 60 days. The ERG highlighted that the trial protocol did not allow for early re-treatment and during the trial and open-label follow-on patients received only two injections of dexamethasone. The ERG noted that the main benefit from re-treatment was in patients whose condition had responded during the initial 180-day trial period. The ERG also commented that the number of treatments needed in practice is not known and that clinical opinion estimated a maximum of six.

The ERG also highlighted that because the trial included a maximum of two dexamethasone treatments, the impact of up to six treatments on the incidence of adverse events was not known. The ERG also expressed concern over the size of the needle which is larger for dexamethasone intravitreal implant than for other treatments. The ERG stated that the main weaknesses in the evidence were lack of long-term follow-up data and data on earlier re-treatment before 180 days.

The ERG considered the robustness of the manufacturer's original model. It highlighted cost inputs (particularly the cost of dexamethasone administration and the cost of severe visual impairment), structural assumptions in the model (such as the duration of trial data on which to base extrapolation of health states in the treatment and observation arms, assumptions related to the stability of visual acuity in 'resolved' patients, the proportion of people who will present with RVO in their 'worse-seeing' eye and the modelling of fellow eye involvement).

The ERG considered that a number of the unit costs applied in the model had been overestimated. For example, the cost of administering the dexamethasone intravitreal implant might have been overestimated because the implant could be given on an outpatient basis (£150) but costs were based on day-case (£648) care in the manufacturer's submission. The ERG also estimated that the cost of residential care was £16,999 instead of £23,972 as used in the base case, and the cost of cataract extraction was £789 rather than £965 as in the base-case model.

The ERG conducted a sensitivity analysis on the unit costs applied in the manufacturer's model and other assumptions related to the extrapolation of effectiveness data beyond the trial. According to the ERG, key uncertainties related to the extrapolation of data remain in the evaluation of cost effectiveness. The likely maximum number of dexamethasone administrations and frequency of re-treatment, the likelihood of resolution, the likelihood of cataract development and extraction, the likelihood of involvement of the other eye and the likelihood of the RVO leading to macular oedema are all important aspects of this uncertainty.

The ERG highlighted factors in the manufacturer's original sensitivity analyses that appeared to have a particular impact on cost effectiveness. When visual acuity was assumed to be stable after a year with no further dexamethasone treatments (rather than at 2.5 years with BRVO and 3 years with CRVO) the base-case ICER increased from £7,368 to £10,764 per QALY gained for the RVO population. When those not treated were all assumed to have the observation transition probabilities applied up to 2.5 years for BRVO and 3 years for CRVO (rather than transition probabilities weighted by proportion of people who were not treated who resolved at day 180 and those who discontinued treatment for other reasons), the base-case ICER increased from £7,368 to £24,924 per QALY gained for the RVO population. When the proportions re‑treated were based on the re-treatment rate in the trial (day 180) for the five injections after the first in CRVO (85.7%) and the four injections after the first in BRVO (78.8%), this increased the base-case ICER from £7,368 to £19,100 per QALY gained for the RVO population. When there was a decline in vision in 1.5% of patients in each health state, worsening by one health state every 6 months was assumed (compared with visual stability from 2.5 years for BRVO and 3 years for CRVO in the base case). This had a small effect on the base-case ICER, which increased from £7,368 to £7,685 per QALY gained for the RVO population.

The ERG also conducted its own additional exploratory analyses on the manufacturer's original economic model which updated the modelling of fellow eye involvement. The ERG compared the manufacturer's model, which included the Weibull function for fellow eye involvement, with the same model with no fellow eye involvement. At the same time, the ERG also varied the proportion of people entering the model with macular oedema in the 'worse-seeing' eye from 90% (as in the manufacturer's model) to 97% (based on the proportion treated in the 'worse-seeing' eye in the trials). The alternative assumption of no fellow eye involvement changed the base-case ICERs for CRVO, BRVO with macular haemorrhage and BRVO with previous laser from £6,041, £7,987 and dominant to £17,279, £34,277 and £11,905 per QALY gained respectively. The alternative assumption of 97% with macular oedema in the 'worse-seeing' eye changed the base-case ICER for CRVO, BRVO with macular haemorrhage and BRVO with previous laser from £6,041, £7,987 and dominant to £15,800, £10,206 and dominant per QALY gained respectively.

In addition, the ERG questioned the way in which 6-month data from the open-label phase were used for the extrapolation of results with dexamethasone treatment and the use of 3- to 6-month data from the trial phase for extrapolation in the observation arm of the original model. After the first Committee meeting the Committee had requested that all data be included in a revised model. When extrapolation was based on 6- to 12-month data from the open-label phase of the trial for dexamethasone treatment and 0- to 6-month data for the observation arm (with 90% of people being treated in the 'worse-seeing' eye) the base-case ICER increased from £6,041 to £15,395 per QALY gained for the RVO population.

The manufacturer submitted two additional original base-case models: one involving structural changes in the modelling of fellow eye involvement, and another which also included lower costs of dexamethasone administration. The revised modelling of fellow eye involvement took account of age with differential survival by collapsing the model into two health states. Without inclusion of lower costs for dexamethasone administration, the revised model was associated with higher ICERs for dexamethasone versus observation of £10,271 per QALY gained for all RVO, £8,165 for CRVO, £11,403 for BRVO with macular haemorrhage, and dominance for dexamethasone over observation for BRVO with previous laser treatment. With revised cost assumptions (outpatient appointments rather than day-case costs, reduced costs of residential care and reduced costs of cataract removal), ICERs were reduced to £7,616 per QALY gained for all RVO, £6,221 for CRVO, £8,848 for all BRVO, £8,313 for BRVO with macular haemorrhage, and dominance for dexamethasone over observation for BRVO with previous laser treatment.

The ERG noted that the manufacturer's model applied the cost associated with severe visual impairment to people with visual acuity less than or equal to 38 letters in the 'worse-seeing' eye. The ERG noted that the cost of severe visual impairment should only be applied when both eyes have visual acuity less than 38 letters. The ERG conducted a sensitivity analysis in which the cost of severe visual impairment was applied when both eyes entered the worse state (visual acuity less than 38 letters) and assumed perfect correlation in the BCVA of both eyes and no correlation between the eyes. This increased the base-case ICER for CRVO from £6,221 to £15,956 per QALY gained when there was perfect correlation and to £18,091 per QALY gained when there was no correlation between eyes. The corresponding base-case ICERs for BRVO with macular haemorrhage increased from £8,313 to £9,674 and £21,443 per QALY gained respectively.

Following the consultation after the first Appraisal Committee meeting, the ERG noted that the manufacturer's revised analysis of cost effectiveness included several deviations from the analysis requested by the Committee at its first meeting. The manufacturer included administration costs for 75:25 day-case to outpatient procedures whereas the Committee had requested day-case costs for all patients. No bevacizumab vial sharing was included in the base case. Extrapolation from data for the last 3 months of observation in the GENEVA studies was used although the Committee requested that all 6 months of data be used. In addition, severe visual impairment was not modelled as requiring bilateral involvement. Cost of reduction of severe visual impairment was reduced by 25% as a proxy with 10% applied every 6 months. The ERG noted that the 10% was not applied annually and it was unclear from where the 10% figure was derived. When the ERG removed this 10% uplift per cycle, the manufacturer's estimated ICERs for CRVO, BRVO with macular haemorrhage and BRVO with previous laser treatment increased from £16,522, £17,741 and £6,361 to £22,831, £23,847 and £12,857 per QALY gained respectively. When the ERG included all 6 months of observational data from the GENEVA studies, the manufacturer's estimated ICERs for CRVO, BRVO with macular haemorrhage and BRVO with previous laser treatment increased from £16,522, £17,741 and £6,361 to £25,336, £38,489 and £11,650 per QALY gained respectively.

The ERG considered that a number of the unit costs applied in the revised model had been overestimated for bevacizumab and the costs used favoured dexamethasone. For example, administration costs were used for 25% outpatient and 75% day-case procedures, there was uncertainty over the calculation of the administration costs for outpatients, a high bevacizumab cost was used and a large number of bevacizumab administrations included. Also, the ERG stated that the frequency of cataracts with steroids was underestimated. When the ERG made adjustments for these costs (a cost of bevacizumab of £50, less frequent bevacizumab administration and follow-up based on the Pan American Collaborative Retina Study group data, and 100% outpatient procedures), the base-case cost minimisation for CRVO, which estimated a cost saving of £4,463 with dexamethasone versus bevacizumab, became a cost saving for bevacizumab of £2,127 versus dexamethasone.

The ERG highlighted that there was a considerable amount of data on bevacizumab treatment from trials including bevacizumab, laser therapy and triamcinolone in RVO (992 observed patient-years) and agreed that this type of observational and uncontrolled data generates uncertainty. The ERG concluded that more data could have been included in the network model by using data on triamcinolone treatment.

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