Dexamethasone intravitreal implant for treating diabetic macular oedema

Clinical effectiveness

1.1 The company identified 6 randomised controlled trials (RCTs) that compared dexamethasone intravitreal implant with a relevant comparator in adults with diabetic macular oedema (DMO; MEAD‑010, MEAD‑011, study 024, PLACID, NCT00035906 and BEVORDEX).

1.2 MEAD‑010 and MEAD‑011 were identical in design and provided the key data for dexamethasone intravitreal implant in the company submission. The trials compared dexamethasone 700 micrograms and dexamethasone 350 micrograms with sham procedure in adults who had been treated before with medical or laser photocoagulation therapy or if laser photocoagulation therapy was not suitable. MEAD‑010 included 494 patients and took place at 59 study centres in 10 countries, including countries in Australasia, North America, Europe, Asia and Africa. MEAD‑011 included 554 patients and took place at 72 study centres in 14 countries, including countries in South America, Europe, Australasia, Asia and North America. Patients were included in the trials if they had a baseline best corrected visual acuity (BCVA) between 34 and 68 letters and a baseline central retinal thickness (CRT) of 300 micrometres or more. Both trials lasted between 36 and 39 months. In both trials, patients had the first treatment on the day of randomisation. They were evaluated for re‑treatment at 6 months and then every 3 months, although treatment was not given more often than every 6 months. Patients were eligible for re‑treatment if retinal thickness in the 1 mm central macular subfield was greater than 225 micrometres (until May 2010) or 175 micrometres (from May 2010), or if optical coherence tomography showed evidence of residual retinal oedema consisting of intraretinal cysts or any regions of increased retinal thickening. The primary outcome in both trials was mean BCVA average change from baseline which was performed using analysis of covariance (ANCOVA) with the treatment as a fixed effect and the baseline BCVA as a covariate. For patients with no post‑baseline BCVA assessment, the average change from baseline was 0.

1.3 Study 024 was an open‑label trial comparing dexamethasone 700 micrograms with ranibizumab 0.5 mg. Patients having dexamethasone intravitreal implant were treated at baseline, month 5 and month 10.

1.4 PLACID compared dexamethasone 700 micrograms plus laser photocoagulation with sham procedure plus laser photocoagulation in 253 patients. Patients had either dexamethasone intravitreal implant or sham procedure on the day of randomisation. At 1 month all patients had laser photocoagulation treatment. Patients could have up to 3 additional laser photocoagulation treatments (at months 4, 7 and 10) and 1 additional dexamethasone intravitreal implant treatment or sham procedure (at month 6 or 9).

1.5 NCT00035906 compared dexamethasone 700 micrograms and dexamethasone 350 micrograms with observation in 171 patients. The population included people with DMO and people with macular oedema associated with uveitis, retinal vein occlusion or Irvine–Gass syndrome, that persisted at least 90 days after laser photocoagulation or medical treatment. Patients had a single treatment at randomisation and were followed for 90 days.

1.6 BEVORDEX compared dexamethasone 700 micrograms with bevacizumab 1.25 mg in 88 eyes. Dexamethasone intravitreal implant was not given more than every 4 months.

1.7 For full details of the trials, please see the company's submission.

Cost effectiveness

1.27 The company submitted an economic evaluation that, in the base case, compared dexamethasone intravitreal implant with a watch‑and‑wait approach for patients with DMO that has not responded to non‑corticosteroid treatment or for whom such treatment is unsuitable, and compared dexamethasone intravitreal implant with ranibizumab for patients with DMO who have a pseudophakic lens. The company carried out additional analyses comparing dexamethasone intravitreal implant with fluocinolone acetonide intravitreal implant in people with disease that has not responded adequately to non‑corticosteroids, and with bevacizumab, watch‑and‑wait and laser photocoagulation in people with a pseudophakic lens.

1.28 For patients with DMO that is considered unsuitable for non‑corticosteroid therapy, dexamethasone intravitreal implant was considered as a first‑ or second‑line treatment option. For patients with DMO that has not responded adequately to non‑corticosteroid therapy (such as ranibizumab, bevacizumab and laser photocoagulation), dexamethasone intravitreal implant was considered as a second‑line treatment. The company considered watch‑and‑wait to be the most appropriate comparator for patients with disease that has not responded adequately to non‑corticosteroid therapy, and those for whom non‑corticosteroid therapy is not suitable. The company used data from the whole DMO population as a proxy for both populations because the available evidence did not suggest a differential efficacy between them and the general DMO population.

1.29 For patients with DMO who have a pseudophakic lens, dexamethasone intravitreal implant was considered as a first‑ or second‑line treatment. The company considered the most appropriate comparator to be ranibizumab, as this is the most common first‑line treatment for DMO. The analysis was based on the network meta‑analysis for patients with a pseudophakic lens. This included data from the subgroup of patients with a pseudophakic lens in the pooled MEAD trials and data for the subgroups of people with a pseudophakic lens in the other trials in the network if available. If data from people with a pseudophakic lens was not available in the trials, data for the whole DMO population was used instead to enable the network to be constructed.

1.30 The model had 3‑monthly cycles and a time horizon of 15 years. A half‑cycle correction and a discount rate of 3.5% for quality‑adjusted life‑years (QALYs) and costs were applied. An NHS and personal social services perspective was used.

1.31 There were 6 health states in the model defined by the BCVA changes in each eye, regardless of whether the eye was treated, in addition to the absorbing health state of death. Both eyes could transition independently between the 6 visual acuity states. The health states were defined by a 10‑letter range in BCVA:

1.32 The model allowed BCVA changes in both eyes to be modelled independently, with the 'better‑seeing eye' (BSE) and 'worse‑seeing eye' (WSE) defined at baseline and fixed throughout the time horizon. The baseline distribution of vision across visual acuity states was reported as confidential and cannot be presented here. Patients entering the model could be affected by DMO in either their BSE or WSE (unilateral DMO), or both eyes (bilateral DMO), with the proportions determined by the pooled number of patients in these groups in the dexamethasone intravitreal implant treatment arms of the MEAD trials. The proportions are assumed to vary by population.

1.33 Patients with DMO in 1 eye at baseline could develop DMO in their other eye ('fellow eye involvement') and move to bilateral treatment. In the model, this could occur only at the end of year 1 or year 2. Patients with bilateral DMO were assumed to have the same treatment in both eyes. Patients could discontinue treatment because of adverse events or loss of efficacy of treatment.

1.34 Patients were at risk of death at all times during the model. The risk of all‑cause mortality was applied to all patients, adjusted for the additional mortality from diabetes and from DMO. The model assumed that mortality occurred equally across all visual acuity states. There was no additional mortality from blindness in the base case (although this was tested in sensitivity analyses).

1.35 For the baseline effect, the 3‑monthly probabilities of eyes treated with dexamethasone intravitreal implant transitioning between visual acuity states were based on the dexamethasone arm of the pooled MEAD trials. For the relative effect, the transition probabilities for watch‑and‑wait were calculated by applying the relative risks for sham procedure from the network meta‑analysis to the 3‑month transition probabilities for dexamethasone intravitreal implant (baseline treatment). For the relative effects of ranibizumab (and bevacizumab, laser photocoagulation and fluocinolone acetonide intravitreal implant in the sensitivity and scenario analyses) the transition probabilities were calculated by applying the relative risks from the network meta‑analysis to the 3‑month transition probabilities for dexamethasone intravitreal implant (baseline treatment). If treatment was discontinued, visual acuity was assumed to follow the natural history of vision in eyes with DMO based on Mitchell et al. (2012). Eyes without DMO were assumed to maintain constant vision.

1.36 The model included data for 5 adverse events: cataracts, raised intraocular pressure (IOP), retinal detachment, endophthalmitis and vitreous haemorrhage. Data was taken from the clinical trials included in the network meta‑analysis. Data for watch‑and‑wait was taken from a natural history study (the Blue Mountains study). The risk of adverse events was assumed to be equal for the general DMO population and the population with a pseudophakic lens, except that there was no risk of cataract in the population with a pseudophakic lens. Adverse effects did not have any effect on health‑related quality of life in the model.

1.37 The company concluded that the published utility values used in NICE's technology appraisal guidance on ranibizumab for treating diabetic macular oedema and fluocinolone acetonide intravitreal implant for treating chronic diabetic macular oedema after an inadequate response to prior therapy were subject to a large number of limitations:

1.38 Health‑related quality of life in the model was dependent on the patient's visual acuity. The company conducted its analyses using Visual Function Questionnaire Utility Index (VFQ‑UI) data, which was calculated from the 25‑item National Eye Institute Visual Function Questionnaire (NEI VFQ‑25) data collected in the MEAD clinical trials. This data related specifically to the DMO population. A regression model was used to estimate utility values for each patient, based on the BCVA of a patient's BSE and WSE. It included vision in the BSE and in the WSE separately as exploratory variables, allowing both eyes to contribute independently to the utility equation used in the economic modelling. The BCVA of the BSE had a higher impact on the estimated utility than the BCVA of the WSE. VFQ‑UI data calculated from the NEI VFQ‑25 data collected in the MEAD trials was used directly to estimate utilities in the model. EQ‑5D values obtained from the MEAD clinical trials were used in the sensitivity analyses. The company performed a systematic review for publications with additional health‑related quality‑of‑life data, but did not find any relevant studies. The utility values associated with the different visual acuity states are reported as confidential and cannot be presented here.

1.39 The company used NHS reference costs and the Monthly Index of Medical Specialities to cost the resources associated with treatment of DMO, including: intervention costs, monitoring and test costs, health state costs and adverse event costs. Treatments were costed as follows: dexamethasone intravitreal implant, £870.00; ranibizumab, £742.17; and fluocinolone acetonide intravitreal implant, £5,500. Laser photocoagulation was assumed to have zero cost because all treatment centres were thought to have access to existing equipment. Watch‑and‑wait was also associated with zero cost. Bevacizumab was assumed to have an acquisition cost of £50.00 in line with the lower limit reported in the NICE Decision Support Unit report on bevacizumab in eye conditions: issues related to quality, use, efficacy and safety. Ranibizumab and fluocinolone acetonide intravitreal implant both have a confidential patient access scheme, and scenario analyses varying the discount to the list price were provided by the company. The cost of laser photocoagulation administration was assumed to be £116.68. All intravitreal injections were assumed to be given in an outpatient setting at a cost of £116.68. If a day‑case procedure was used in the sensitivity analyses, the cost was assumed to be £356.35.

1.40 The assumed total costs per round of treatment were different for unilateral and bilateral disease, except for laser photocoagulation which was assumed to have the same cost for both (£116.68). Dexamethasone intravitreal implant was assumed to have a total cost of £986.68 for treating unilateral disease and £1,944.19 for bilateral disease. For ranibizumab, the total cost based on its list price was £858.85 for unilateral disease and £1,659.36 for bilateral disease. The total cost of bevacizumab was £166.68 for unilateral disease and £275.02 for bilateral disease. Fluocinolone acetonide intravitreal implant was assumed to have a total cost of £5,616.68 for treating unilateral disease and £11,204.19 for bilateral disease (based on its list price).

1.41 The costs of monitoring and tests used were as follows (all sourced from NHS reference costs): routine monitoring visit, £80.04; optical coherence tomography, £18.06; fluorescein angiography, £116.68; and IOP check, £80.04. The costs of monitoring and treatment were assumed to be equal across all health states. In addition, if BCVA in the BSE fell below 35 letters (severe vision loss), there were a number of additional costs including community care, residential care, hip replacement and depression (total cost per patient per year for severe vision loss is £16,755.23).

1.42 The average number of monitoring visits used in the model was taken from the NICE technology appraisal guidance on fluocinolone acetonide intravitreal implant for the treatment of chronic diabetic macular oedema after an inadequate response to prior therapy for watch‑and‑wait and fluocinolone acetonide intravitreal implant, NICE technology appraisal guidance on ranibizumab for treating diabetic macular oedema for ranibizumab, and the summary of product characteristics and clinical opinion for dexamethasone intravitreal implant. In the model it was assumed there would be 4 monitoring visits each year for watch‑and‑wait, dexamethasone intravitreal implant and laser photocoagulation. It was assumed there would be 12 visits in year 1, 10 visits in year 2, and 4 visits in year 3 for ranibizumab and bevacizumab.

1.43 The average number of treatments per year used in the model was taken from the MEAD trials (dexamethasone intravitreal implant), FAME (fluocinolone acetonide intravitreal implant), RESTORE (ranibizumab), BOLT (bevacizumab, trial data to year 2 and then assumed to be equal to ranibizumab), and PROTOCOL I (laser photocoagulation, trial data to year 2 and then the last observation was carried forward). The model assumed a maximum treatment duration of 3 years. The number of treatment visits for dexamethasone intravitreal implant and laser photocoagulation are reported as confidential and cannot be presented here. It was assumed that there would be 1 visit in year 1 and 0.26 visits in years 2 and 3 for fluocinolone acetonide intravitreal implant. The model assumed 7 treatment visits in year 1, 3.9 visits in year 2 and 2.9 visits in year 3 for ranibizumab, and 9 visits in year 1, 4 visits in year 2 and 2.9 visits in year 3 for bevacizumab. It was assumed that there were no treatment visits with watch‑and‑wait. The model allowed the use of rescue therapy with laser photocoagulation for some interventions, although not in the comparison of dexamethasone intravitreal implant with watch‑and‑wait.

1.44 Adverse events were associated with costs in the model, all taken from NHS reference costs. The cost of a cataract extraction procedure was assumed to be £865.56. The total average cost of treating raised IOP per patient was £262.40 for medical management and £1,222.93 for surgical management (costs stated here are those used in the company model). The cost of re‑attachment of the retina following retinal detachment was £1,685.00. The cost of vitreous biopsy following endophthalmitis was £1,393.00. The cost of a vitrectomy procedure following vitreous haemorrhage was £1,685.00.

1.45 In the company's base case, dexamethasone intravitreal implant dominated watch‑and‑wait for patients with DMO that does not respond adequately to non‑corticosteroid therapy, or for whom such treatment is unsuitable (incremental costs: ‑£1,469; incremental QALYs 0.0656). In the full population of people with a pseudophakic lens, treatment with ranibizumab resulted in a deterministic incremental cost‑effectiveness ratio (ICER) of £50,905 per QALY gained (incremental costs £6,004, incremental QALYs 0.1179) compared with dexamethasone intravitreal implant when the list price of ranibizumab was used. The corresponding probabilistic ICER was £89,531 per QALY gained (incremental costs £6,710, incremental QALYs 0.0749). When a discount of 50% was applied to the list price of ranibizumab, ranibizumab dominated dexamethasone intravitreal implant in the deterministic analysis (incremental costs ‑£716, incremental QALYs 0 1179) and probabilistic analysis (incremental costs ‑£15, incremental QALYs 0.0749).

Company sensitivity analyses and scenarios

1.46 The company carried out 1‑way sensitivity analyses and scenario analyses to assess the impact on the deterministic results.

1.47 In the company's sensitivity analyses for patients with DMO that has not responded adequately to non‑corticosteroid therapy, or for whom such treatment is unsuitable, dexamethasone intravitreal implant remained cost‑effective compared with watch‑and‑wait. The ICERs were most sensitive to changes to the cost of residential care and the percentage of patients requiring residential care.

1.48 In the company's sensitivity analyses for people with a pseudophakic lens, dexamethasone intravitreal implant remained cost effective compared with ranibizumab at list price. The ICERs were most sensitive to changes to the relative risk of worsening vision from the network meta‑analysis and the proportion of outpatient procedures for ranibizumab.

1.49 The company performed 28 scenario analyses. The scenarios that had a significant impact on the ICER are reported in sections 1.50 to 1.54.

1.50 For patients with DMO that has not responded adequately to non‑corticosteroid therapy, or for whom such treatment is unsuitable, dexamethasone intravitreal implant continued to dominate in the majority of the scenarios. When 1‑year and 5‑year time horizons were used, dexamethasone intravitreal implant had an ICER of £1,822,946 and £127,034 per QALY gained compared with watch‑and‑wait. A 10‑year time horizon resulted in an ICER of £6,365 per QALY gained with dexamethasone intravitreal implant compared with watch‑and‑wait. Assuming that the person had DMO in their WSE with no fellow eye involvement resulted in an ICER of £131,276 per QALY gained for dexamethasone intravitreal implant compared with watch‑and‑wait.

1.51 For patients with DMO that has not responded adequately to non‑corticosteroid therapy, the ICER for fluocinolone acetonide intravitreal implant at list price compared with dexamethasone intravitreal implant was £24,591 per QALY gained (incremental costs £1,953; incremental QALYs 0.0794). When a discount of 10% was applied to the cost of fluocinolone acetonide intravitreal implant, the ICER decreased to £10,241 per QALY gained. When the discount was increased to 20% or more, fluocinolone acetonide intravitreal implant dominated dexamethasone intravitreal implant.

1.52 For people with a pseudophakic lens, dexamethasone intravitreal implant was dominated by both laser photocoagulation (incremental costs £7,359; incremental QALYs ‑0.0482) and bevacizumab (incremental costs £6,318; incremental QALYs ‑0.1491).

1.53 For people with a pseudophakic lens, the ICERs for ranibizumab compared with dexamethasone intravitreal implant with a 10% and 20% discount to the list price of ranibizumab were £39,510 and £28,116, respectively, per QALY gained. With a discount of 30% and 40% to the list price of ranibizumab, the ICERs were £16,721 and £5,327, respectively, per QALY gained. Ranibizumab dominated dexamethasone intravitreal implant when a discount of 50% was applied to the list price of ranibizumab.

1.54 For the other scenarios for patients with a pseudophakic lens, a discount of 50% to the list price of ranibizumab was used. Ranibizumab dominated dexamethasone intravitreal implant in most of the scenario analyses. When stable vision after discontinuing treatment was assumed, the ICER was £1,554 per QALY gained for ranibizumab. With a 1‑year, 5‑year and 10‑year time horizon the ICERs were £697,936, £47,729 and £7,564 respectively per QALY gained. When it was assumed that there was unilateral DMO in the WSE with no fellow eye involvement, the ICER was £57,384 per QALY gained with ranibizumab. When it was assumed that all injections were given as day cases, the ICER was £16,323 per QALY gained, and when it was assumed that 50% of injections were day cases, the ICER was £5,128 per QALY gained.

Company response to clarification

1.55 The company provided several additional analyses in response to clarification; the most important of these are described below.

1.56 In the first analysis, the baseline BCVA distribution in bilateral DMO was taken from the subgroup of patients with bilateral DMO, rather than from patients with unilateral DMO (as in the base case). In people with a pseudophakic lens, the cost effectiveness of dexamethasone intravitreal implant was improved, as it remained cost effective at higher discount to the price of ranibizumab (up to 39% of the list price). Ranibizumab at 50% discount price was not dominant anymore, although it was still cost effective compared with dexamethasone intravitreal implant at an ICER of £7,208 per QALY gained.

1.57 In the third and fourth analyses, the company used 3‑month transition probabilities for both watch‑and‑wait and dexamethasone intravitreal implant directly from the pooled data from the MEAD trials rather than from the network meta‑analysis. The ERG argued that the results of the economic analysis between dexamethasone intravitreal implant and watch‑and‑wait should be the same, whether the relative effect is taken from the MEAD trials (as in analyses 3 and 4) or from the network meta‑analysis (as in the company's base case). However, this is not the case. The ERG argued that this may be because the company used relative risks derived from the network meta‑analysis with the assumption that the 12‑month relative risks remained constant to year 3. The ERG believed that this assumption was incorrect because the pooled data from the MEAD trials showed that the relative effect of dexamethasone intravitreal implant versus sham procedure is not stable over 3 years. The ERG also argued that the differences in the economic analyses may be a result of the company's normalisation of the transition probabilities which were done so that the probabilities summed up to 1. This may have introduced bias in the company's analysis, although it is not clear how much and in which direction. The company argued that the results of the network meta‑analysis were more appropriate to use than the pooled MEAD data because the sham arm of the MEAD trials was likely to overestimate the true efficacy of a watch‑and‑wait strategy. The ERG agreed that the use of the MEAD sham data is likely to have overestimated the true efficacy of watch‑and‑wait. However, it highlighted that the MEAD trials were the only ones in the network meta‑analysis that provide relative effects for dexamethasone intravitreal implant compared with sham, and so these relative effects are also present in the results from the network meta‑analysis. In the third analysis, the company restricted movements between health states to a maximum of 1 state (as in the company's base case). This resulted in watch‑and‑wait dominating dexamethasone intravitreal implant. In the fourth analysis, there was unrestricted movement between health states. This resulted in an ICER of £1,411,676 for dexamethasone intravitreal implant compared with watch‑and‑wait. The ERG argued that the fourth analysis, which uses data directly from the MEAD trials, appears to be more reflective of relative clinical effects between dexamethasone intravitreal implant versus watch‑and‑wait for patients with DMO that is unsuitable for or insufficiently responsive to non‑corticosteroid therapy.

New company evidence submitted at ACD comments stage

1.58 The company received permission from NICE to submit new analyses in response to the appraisal consultation document (ACD) on patients who do not have a pseudophakic lens and with DMO that does not respond to non‑corticosteroid treatment, or for whom such treatment is unsuitable. The company's base‑case analyses including the new evidence used the head‑to‑head MEAD trial data, incorporated the corrections previously made by the ERG and included changes to 4 further areas:

1.59 In its new analyses, the company considered the true cost of residential care was unlikely to be wholly in the private sector or local authority. Instead, it used a weighted cost that was 95% of the cost of private sector residential care and 5% of the cost of local authority residential care, giving an annual residential care cost of £28,985. Implementing this change, together with the assumptions preferred by the committee in the ACD (see section 2.19), gave an ICER of £1,170,914 per QALY gained (incremental costs £6,753; incremental QALYs 0.0058) for dexamethasone intravitreal implant compared with watch‑and‑wait.

1.60 The company believed that the high discontinuation rates in the MEAD studies were a major source of uncertainty in its previous analyses. It considered that the natural history transition matrix from Mitchell et al. (2012) had likely overestimated BCVA in patients who discontinued because of a lack or loss of efficacy or who were censored from the study:

1.61 In its new analyses, the company estimated the outcomes for patients who were censored from or who discontinued the studies in both treatment arms using a last transition carried forward (LTCF) approach. For these patients, the last observed transition (that is, the change in BCVA between the last 2 visits before discontinuation or censoring) was applied in every cycle after discontinuing until the end of the initial 3‑year treatment period. Transition matrices were generated for each 3‑month cycle then cumulative LTCF matrices were combined with the observed transition matrices to give an estimated matrix for the total population, assuming no discontinuation from treatment. This methodology was applied to both treatment arms and re‑treatment rates were adjusted to reflect the lack of discontinuation. Adding this change to the assumptions in section 1.59 caused the ICER for dexamethasone intravitreal implant compared with watch‑and‑wait to drop from £1,170,914 per QALY gained to £148,403 per QALY gained (incremental costs £5,554; incremental QALYs 0.0374).

1.62 The company stated that there was no evidence to suggest that other model types would provide a better fit to the data derived from the MEAD trials than the linear regression approach. It noted that including an interaction term between BCVA in the BSE and BCVA in the WSE did not improve the model fit or provide a meaningful point estimate for the interaction. In its new base case, the company used published estimates of utility values from Czoski‑Murray et al. (2009) instead of those derived from the MEAD studies, which covered a narrower range. The company noted that in previous technology appraisals for DMO, the committee had preferred these published values. The company included scenario analyses using utility values from Brown (1999) and Brown et al. (2000) because these have also been discussed in other technology appraisals in DMO. Because these 3 studies reported only BSE utility values, the company estimated the utility values in the WSE by assuming that the change in the WSE was 30% of that in the BSE, which it said was consistent with assumptions in previous technology appraisals. Adding this change to the assumptions in section 1.61 caused the ICER for dexamethasone intravitreal implant compared with watch‑and‑wait to drop further from £148,403 per QALY gained to £50,280 per QALY gained (incremental costs £5,554; incremental QALYs 0.1105).

1.63 The company applied a clinical continuation rule for dexamethasone intravitreal implant in its economic model. It was assumed that treatment was not continued if patients did not gain at least 5 letters by month 6 after their first injection of dexamethasone intravitreal implant. Of the 338 patients remaining in the dexamethasone intravitreal implant arm at month 6 in the MEAD studies, 105 (31.1%) did not gain at least 5 letters by month 6. These patients were assigned transition probabilities associated with the natural history of vision (in line with Mitchell et al.). Applying only the continuation rule gave an ICER of £678,142 per QALY gained (incremental costs £5,347; incremental QALYs 0.0079). Adding this change to the assumptions described in section 1.62 resulted in the company's base‑case ICER using the new evidence submitted in response to the ACD (see below).

1.64 In its new analyses of patients who do not have a pseudophakic lens and with DMO that does not respond to non‑corticosteroid treatment, or for whom such treatment is unsuitable, the company's base‑case ICER for dexamethasone intravitreal implant was £14,978 per QALY gained (incremental costs £2,523; incremental QALYs 0.1685). According to the company's probabilistic analyses, the probability of dexamethasone being cost effective compared with watch‑and‑wait was 54% at a maximum acceptable ICER of £20,000 and 58% at a maximum acceptable ICER of £30,000 per QALY gained.

ERG comments on the company's main submission

ERG comments on new company evidence submitted at ACD comments stage

1.98 The ERG provided comments on the company's new evidence:

1.99 The ERG reviewed how the company had modelled the transition probabilities for patients who had discontinued treatment or had been censored. It noted that using the company's new LTCF approach instead of the original transition matrices considerably reduced the ICER for dexamethasone intravitreal implant versus watch‑and‑wait (from £1,170,914 per QALY gained to £148,403 per QALY gained; see sections 1.60 and 1.61). In its critique, the ERG said that it was clinically implausible for patients who discontinued treatment to have stable disease (that is, to remain in their health state at discontinuation). At the committee meeting, the company noted it appeared that the ERG had misinterpreted the company's approach. The company confirmed that patients did not remain in the same health state after discontinuing. Instead, the company assumed that no discontinuation or censoring occurred and used patient‑level data to anticipate what would happen in future cycles. The company confirmed it had modelled this by applying the last transition before discontinuing to the overall calculations of future movement between any 2 health states. The ERG agreed that this was an error in its report on the company's new evidence. Nevertheless, it considered the company's previous assumption (that patients who discontinued reverted to a natural history of vision decline) to be less biased.