Ranibizumab for treating diabetic macular oedema

The manufacturer performed a systematic review of the evidence on the clinical effectiveness of ranibizumab. The review identified 4 randomised controlled trials (RCTs) that included ranibizumab in people with diabetic macular oedema – RESTORE, Diabetic Retinopathy Clinical Research Network Protocol I (DRCR.net), RESOLVE and READ‑2. The manufacturer focused its submission on RESTORE and DRCR.net. The 2 other RCTs did not receive detailed attention, because the manufacturer judged them to be of less direct relevance to the decision problem. It stated that RESOLVE had limited application to the appraisal because it did not present a comparison with laser photocoagulation, which the manufacturer believed to be the most relevant comparator for an analysis concentrating on practice in the UK. The manufacturer stated that READ‑2 did not provide high-quality evidence because follow-up was of shorter duration than in other included studies, the schedule for treatment differed from that in the summary of product characteristics for ranibizumab and the manufacturer believed the trial may have had methodological shortcomings.

RESTORE was an industry-sponsored, multicentre (73 centres in 13 countries), sham-controlled randomised trial that compared ranibizumab plus sham laser photocoagulation (n=116) with ranibizumab plus laser photocoagulation (n=118) and laser photocoagulation plus sham injections (n=111). The trial lasted for 1 year, and participants were followed beyond 1 year, but did not necessarily remain on the treatment to which they had been randomised. RESTORE included people aged over 18 years with type 1 or type 2 diabetes and haemoglobin A1c (HbA1c) lower than 10% (86 mmol/mol). The trial protocol stated that, for each participant, only 1 eye should be treated, even if both eyes had disease. The eye with the worse vision was treated unless the investigator deemed it appropriate to treat the eye with the better vision. According to the trial protocol, 20% of participants had their better-seeing eye treated. Best corrected visual acuity (hereafter, 'visual acuity') was measured using 'ETDRS (Early Treatment Diabetic Retinopathy Study)-like' charts, in which a score of 85 letters corresponds to normal visual acuity ('20/20 vision'). An eye was eligible for randomisation if visual acuity was between 78 and 39 letters. Participants who had previous laser photocoagulation were included in the trial. Ranibizumab or sham injections were administered monthly in months 1 to 3; after this, they continued on a monthly basis until vision was stabilised for 2 visits or visual acuity reached 85 letters or more. Treatment with monthly injections was restarted if there was a decrease in visual acuity caused by progression of diabetic macular oedema and continued until the same criteria were fulfilled. Laser photocoagulation or sham laser photocoagulation was administered on day 1 and repeated at intervals of at least 13 weeks, if deemed necessary by the treating clinician. RESTORE was judged to satisfy all methodological quality criteria assessed by the manufacturer. At the time of the original submission, full details of RESTORE had not been published.

DRCR.net, an RCT funded by the US National Institutes of Health, was conducted at 52 clinical sites in the United States. The trial protocol stipulated that the trial would last for 3 years; however, at the time of submission for NICE technology appraisal 237, only 12-month follow-up data were available. Participants were aged over 18 years and had type 1 or type 2 diabetes. An eye was eligible for randomisation if it had centre-involving macular oedema and a visual acuity of between 78 and 24 letters using the Electronic-Early Treatment Diabetic Retinopathy Study (E-ETDRS) visual acuity test (again, a score of 85 letters corresponds to normal visual acuity). People who had previous laser photocoagulation were included in the trial. Randomisation was by eye (rather than by participant) and a participant could have both eyes involved in the study. Each eligible eye was randomised to receive either a sham injection plus laser photocoagulation (n=293 eyes), ranibizumab plus prompt laser photocoagulation (within 3 to 10 days of first ranibizumab injection; n=187 eyes) or ranibizumab with the possibility of subsequent (deferred) laser photocoagulation (at least 24 weeks after the first ranibizumab injection; n=188 eyes). In practice, only 28% of the ranibizumab plus deferred laser arm received laser treatment at any time during the first study year. A fourth group in DRCR.net received triamcinolone; this group is not included in this appraisal because triamcinolone is not currently used in clinical practice in the UK for diabetic macular oedema and was not in the scope for this appraisal. The randomisation protocol specified that, in participants with 2 eligible eyes, 1 eye would receive pharmacological treatment (plus prompt or deferred laser photocoagulation) and the other eye would receive prompt laser photocoagulation alone. Investigators administered ranibizumab or sham injections every 4 weeks until the fourth study visit (that is, after 12 weeks of treatment). At subsequent 4‑weekly visits, the decision to give another injection depended on visual acuity and retinal thickness of the treated eye. Investigators repeated laser photocoagulation or sham laser photocoagulation, if needed, at intervals of at least 13 weeks (3‑monthly). The manufacturer judged that DRCR.net satisfied all the methodological quality criteria it assessed (although it noted that participants randomised to ranibizumab plus deferred laser were aware of their allocated treatment).

The primary outcome measure of both RESTORE and DRCR.net was mean change in visual acuity in the treated eye after 12 monthly follow-up visits. The RESTORE analysis was based on the average of changes in visual acuity from baseline, measured monthly over the period from month 1 to month 12 ('mean average change'), whereas DRCR.net compared the visual acuity measured at baseline with that measured at 12 months ('mean change'). In RESTORE, the visual acuity of eyes randomised to ranibizumab monotherapy rose by a mean average of 6.1 letters, and eyes randomised to ranibizumab plus laser photocoagulation gained a mean average of 5.9 letters. In RESTORE, eyes randomised to laser photocoagulation alone gained fewer letters (0.8) than eyes randomised to either ranibizumab-containing arm (p<0.001). In DRCR.net, visual acuity rose by an average of 9 letters after 12 months in eyes randomised to ranibizumab plus either prompt or deferred laser photocoagulation compared with an average of 3 letters in eyes randomised to laser photocoagulation alone (p<0.001 for either ranibizumab-containing arm compared with laser photocoagulation alone). The manufacturer provided a meta-analysis of mean changes from baseline visual acuity at 12 months combining results from RESTORE and DRCR.net. This suggested that the visual acuity of eyes treated with ranibizumab plus laser photocoagulation gained an average of 5.83 more letters than the visual acuity of eyes treated with laser photocoagulation alone (95% confidence interval [CI] 4.07 to 7.59, p<0.001; fixed-effects and random-effects models estimate identical results).

In both RESTORE and DRCR.net, a series of subgroup analyses examined the primary outcome measure in participants categorised according to baseline characteristics. In all subgroups analysed in both trials, the visual acuity of participants randomised to ranibizumab-containing treatment improved more than the visual acuity of those randomised to laser photocoagulation. The manufacturer's submission noted that, in RESTORE, gains in visual acuity associated with ranibizumab were greatest in participants with baseline central retinal thickness of 300 micrometres or more and participants with a baseline visual acuity of fewer than 74 letters; the manufacturer presented no evidence on the statistical significance of these differences.

An alternative approach to presenting the results of visual acuity testing reports the proportion of participants in whom the treated eye improved or worsened by an amount reflecting a clinically significant change in vision, usually a gain or loss of 10 letters. The manufacturer's submission reported that, in RESTORE, the proportions of participants gaining 10 letters in visual acuity in their treated eye after 12 months of treatment were 37% in those randomised to ranibizumab monotherapy, 43% in those randomised to ranibizumab plus laser photocoagulation and 15% in those randomised to laser photocoagulation alone (p<0.001 for either ranibizumab-containing arm compared with laser photocoagulation alone; p-values taken from European Medicines Agency Assessment Report). The proportions of participants who lost 10 letters of visual acuity in their treated eye after 12 months were 3%, 4% and 13% respectively (p<0.05 for either ranibizumab-containing arm compared with laser photocoagulation alone; p values calculated by the NICE technical team). In DRCR.net, after 12 months of treatment a gain of 10 letters in visual acuity was reported in 47% of eyes treated with ranibizumab plus deferred laser photocoagulation, 51% of eyes treated with ranibizumab plus prompt laser photocoagulation and 28% of eyes treated with laser photocoagulation alone (p<0.001 for either ranibizumab-containing arm compared with laser monotherapy). After 12 months a loss of 10 letters in the treated eye was reported in 3%, 3% and 13% of eyes respectively (p≤0.001 for either ranibizumab-containing arm compared with laser photocoagulation alone). The manufacturer provided a meta-analysis of categorical visual acuity data from RESTORE and DRCR.net after 12 months of treatment. This suggested that the visual acuity of eyes randomised to ranibizumab plus laser photocoagulation was approximately twice as likely to improve by 10 letters than the visual acuity of eyes randomised to laser photocoagulation alone (relative risk=2.15, 95% CI 1.43 to 3.22, p<0.001; random-effects model; fixed-effects model produced similar results). Eyes treated with ranibizumab plus laser photocoagulation were over 3 times less likely to lose 10 letters in visual acuity (relative risk=0.28, 95% CI 0.15 to 0.53, p<0.001; random-effects model; fixed-effects model produced similar results).

RESTORE measured vision-related quality of life using the National Eye Institute Visual Function Questionnaire-25 (NEI VFQ‑25), which has 25 questions designed to measure the effect of visual impairment on daily functioning and quality of life. The mean changes from baseline in the composite score and in subscales related directly to vision were in favour of ranibizumab compared with laser photocoagulation alone (p<0.05), but benefit was not demonstrated in subscales addressing dependency and driving (p>0.05). The RESTORE investigators also administered assessments of health-related quality of life, including EuroQol‑5D (EQ‑5D). The manufacturer did not report the results directly, although its economic model relied on an analysis incorporating the EQ‑5D data (see section 3.12).

The manufacturer stated that ranibizumab has a favourable safety profile, emphasising that extensive evidence is available from the use of ranibizumab in the treatment of wet age-related macular degeneration. Although none of the RCTs of ranibizumab in diabetic macular oedema were designed primarily to assess safety outcomes, no significant differences were observed between arms in the frequency of ocular and non-ocular adverse events. None of the studies reported death rates.

The economic evidence provided by the manufacturer in its original submission comprised a brief literature review (identifying no relevant published analyses) and a de novo cost–utility analysis. The cost–utility analysis used a Markov model simulating cohorts of people with diabetic macular oedema receiving ranibizumab monotherapy, ranibizumab plus laser photocoagulation, or laser photocoagulation alone. The model had 3‑monthly cycles and a base-case time horizon of 15 years. It assumed (simulated) a starting population with diabetic macular oedema with a mean age of 63 years and visual acuity scores of between 75 and 36 letters. Health states were defined by visual acuity in the treated eye, rather than both eyes, and used 10‑letter categories (with the exception of the best and worst states), resulting in 8 health states excluding death: 0 to 25 letters, 26 to 35 letters, 36 to 45 letters, 46 to 55 letters, 56 to 65 letters, 66 to 75 letters, 76 to 85 letters and 86 to 100 letters.

Based on the average number of injections received in RESTORE, the manufacturer's original model included 7 ranibizumab injections in the first year for both the ranibizumab monotherapy arm and the combination therapy arm. The manufacturer included an additional stopping rule, assuming that people with a visual acuity of 76 letters or more in the treated eye would not receive active treatment and would incur no treatment costs. In the second year of the model, based on data from DRCR.net, the manufacturer assumed 3 further ranibizumab injections in the ranibizumab monotherapy arm and 2 further ranibizumab injections in the combination therapy arm. For both the combination therapy arm and the laser photocoagulation alone arm, the model assumed 2 laser photocoagulation treatments in the first year and 1 additional laser treatment in the second year. The original model did not simulate any ranibizumab treatment or laser photocoagulation after the second year. The manufacturer accepted that, in practice, additional laser photocoagulation would take place after the second year, but assumed that there would be no difference between the modelled arms in this respect and therefore any costs and effects would cancel each other out.

To estimate the health-related quality of life associated with each health state corresponding to vision in the model, EQ‑5D data from RESTORE were transformed to utility values using standard social tariffs and then related to visual acuity in the treated eye using linear regression. In this way, a mathematical relationship was assumed between visual acuity in the treated eye and the health-related quality of life of people with diabetic macular oedema. Tests of interaction demonstrated that neither treatment allocation nor duration of treatment at the time of measurement had a significant impact on the relationship between utility and visual acuity category. Therefore, the utility value was derived for each visual acuity category using the whole data set, without any adjustment variables. The results of the regression suggested that, depending on visual acuity in the treated eye, the utility value ranged from 0.860 to 0.547. The utility value for the best health state (visual acuity 86 to 100 letters) resulting from the regression was 0.831; however, this result was considered anomalous, because it was based on relatively few data points and it was thought unrealistic that utility should be worse than in the next-best state, in which vision is inferior. Therefore, the manufacturer set the utility value for a visual acuity of 86 to 100 letters to be equal to the utility value for a visual acuity of 75 to 85 letters (that is, 0.860). The manufacturer did not apply any treatment-specific utilities and specifically did not apply a decrement in utility value to reflect adverse effects associated with the treatments.

Depending on the period of follow-up being simulated, the model used 3 different sets of data and assumptions to estimate the probability of changing between visual acuity states (transition probabilities). The first phase of the model reflected the first year of treatment, for which the manufacturer drew transition probabilities directly from changes in visual acuity observed among individual participants in RESTORE. In the original model, this data set included a proportion of treated eyes with a visual acuity of more than 75 letters at baseline, although eyes with equivalent vision were not simulated in the model. In the second and third phases of the model, the manufacturer based the transition probabilities on a 3‑way matrix that reflected the estimated likelihood of improvement, deterioration or no change in the visual acuity of a given eye. The manufacturer assumed that the same probabilities applied to all states (with the exception of assuming that an eye with the best possible visual acuity could not improve and that an eye with the worst possible visual acuity could not worsen). The second phase of treatment reflected a period during which visual acuity was assumed to remain constant. This was based on the DRCR.net study, which the manufacturer interpreted as showing that average visual acuity was maintained between 12 months and 24 months for all treatments. Therefore, the manufacturer assumed equal probabilities of improving or worsening in all model arms. In the original model, this phase comprised the second year of treatment only. In the third phase, the model assumed no treatment effect: based on an informal review of epidemiological literature, the manufacturer assumed quarterly probabilities of 0.025 for improvement and 0.035 for deterioration in all arms (implying that the probability of visual acuity staying the same was 0.94 in any quarter). In the original model, this final phase began at the start of year 3, and continued until the end of the model. According to this approach, the relative difference in treatment effect observed at the end of the initial treatment phase (year 1) was preserved for the remainder of the model. Thus, beyond 1 year, the visual acuities of participants previously treated with ranibizumab remained constant during the second year and after that declined at the same rate as the visual acuities of participants previously treated with laser photocoagulation. As a result, the relative difference in vision between treatments estimated at the end of the first year was preserved for the remainder of the model (that is, for the next 14 years).

In its model, the manufacturer assumed that the probability of dying is the same for all health states (so no additional risk of death was associated with worsening vision). Mortality rates were derived from life tables for England and Wales, modified by a relative risk representing the additional hazard of death for the modelled cohort when compared with the general population. In the original model, the manufacturer used a relative risk of 1.27, drawn from published literature comparing the risk of death for people with clinically significant diabetic macular oedema with the risk of death for people with diabetes, but without diabetic macular oedema. The manufacturer modelled adverse events on the basis of observed, treatment-specific rates of 4 events (cataract, endophthalmitis, retinal detachment and vitreous haemorrhage) in a pooled analysis of RESTORE and DRCR.net. The manufacturer included only the costs associated with these complications; it did not assume that the presence of these adverse events lowers quality of life (utility).

The cost of ranibizumab included in the original model was £761.20 per injection (this was before the manufacturer submitted a patient access scheme; the manufacturer submitted the patient access scheme after publication of the appraisal consultation document for NICE technology appraisal guidance 237, and revised it for the rapid review submission). Unit costs of a visit to an eye clinic for a check-up and/or treatment were based on NHS reference costs (2008/09). The manufacturer assumed that treatment with both ranibizumab and laser photocoagulation occurs on an outpatient basis, and costs £150.00 per visit. For combination therapy, the manufacturer assumed that ranibizumab injections and laser photocoagulation would occur at the same visit, and cost £184.00 per visit. The manufacturer assumed that the full cost of laser photocoagulation is encompassed within the NHS reference cost for the clinic visit and included no additional costs for buying and maintaining the equipment. The costs (£126.00 each) of visits to monitor people (for vision and recurrence of disease) were also included: in the original model, people receiving ranibizumab monotherapy had 12 visits in the first year and 10 visits in the second year; people receiving combination therapy had 12 visits in the first year and 8 visits in the second year; those receiving laser photocoagulation alone in the first and second years, and all people from the third year onwards, had 4 visits per year. For the ranibizumab-containing arms, a visit for treatment was assumed to include monitoring as well. The same assumption was not applied for people receiving laser photocoagulation alone; that is, people receiving laser photocoagulation needed separate visits for treatment and monitoring.

In the model, the manufacturer applied estimated costs associated with severe vision loss for people with the lowest visual acuity in the treated eye (0 to 25 or 26 to 35 letters), regardless of vision in the non-treated eye. These costs reflected the additional resource use associated with people who are eligible to register as severely sight impaired (blind). The costs accounted for a range of items including low-vision aids, rehabilitation, residential care, district nursing, community care and the cost of treating complications including depression and falls. The manufacturer drew cost data largely from a published costing study of blindness in the UK that focused on people with age-related macular degeneration (Meads and Hyde 2003), with costs updated or adjusted for inflation as appropriate. The total cost applied was £6067 in the first year and £5936 in subsequent years.

In its deterministic base case, the manufacturer's original model estimated an incremental cost-effectiveness ratio (ICER) of £19,075 per quality-adjusted life year (QALY) gained for ranibizumab monotherapy compared with laser photocoagulation alone. The model predicted that combining laser photocoagulation with ranibizumab would be more expensive and less effective than ranibizumab alone; that is, ranibizumab alone dominated combination therapy. The manufacturer also presented a series of deterministic sensitivity analyses in which single parameters (or related groups of parameters) were varied across plausible ranges. These analyses suggested that the model was most sensitive to the time horizon: when the time horizon was limited to 10 years, the estimated ICERs for ranibizumab monotherapy compared with laser photocoagulation alone rose by approximately 50%, to £30,367 per QALY gained. Most sensitivity analyses repeated the base-case finding that ranibizumab monotherapy dominates combination therapy with ranibizumab plus laser photocoagulation. Probabilistic sensitivity analysis (based on 10,000 Monte-Carlo simulations) suggested that the probability of ranibizumab monotherapy providing best cost–utility compared with laser alone was 49.3% and 76.8% at thresholds of £20,000 and £30,000 per QALY gained respectively. The probability of combination therapy providing best cost–utility compared with ranibizumab monotherapy was 19.4% and 17.8% at the same thresholds (these estimates have been corrected from the manufacturer's submission for NICE technology appraisal guidance 237 by the NICE technical team and verified by the ERG).

The manufacturer also presented a series of deterministic subgroup analyses in which first-year transition probabilities were derived from analyses of the RESTORE trial limited to participants with the characteristic(s) in question. The manufacturer changed no other parameters in the model. The manufacturer noted that several of the analyses were based on small numbers of participants. There were large variations between the results of some subgroups:

• Limiting the analysis to people with good glycaemic control (HbA1c less than 8%) produced an ICER of £13,196 per QALY gained for ranibizumab monotherapy compared with laser photocoagulation. In people with poorer glycaemic control (HbA1c 8% or more) ranibizumab monotherapy had an ICER of £36,383 per QALY gained when compared with laser photocoagulation.

• In people who previously had laser photocoagulation, ranibizumab monotherapy was associated with an ICER of £29,660 per QALY gained compared with laser photocoagulation. The equivalent figure for the subgroup who had not previously had laser photocoagulation was £12,675 per QALY gained.

• The ICERs for ranibizumab monotherapy compared with laser photocoagulation in people with baseline visual acuity of 36 to 45 letters, 46 to 55 letters, 56 to 65 letters and 66 to 75 letters were £52,704, £7645, £42,477 and £12,198 per QALY gained respectively.
  
  However, the manufacturer considered that in light of the limitations related to the small number of trial participants in some of the subgroups, the relative cost effectiveness of ranibizumab monotherapy in these subgroups is uncertain and should be interpreted with caution.

In response to consultation on the original appraisal consultation document, the manufacturer submitted a revised cost–utility analysis, addressing reservations the Committee had expressed about the original model and submitted a first patient access scheme. Several consultees and commentators, including patient and professional groups, agreed with the Committee that the manufacturer's original economic model had given an unrealistic representation of likely clinical practice in some respects.

The manufacturer stated that the revised model should be 'considered a better-seeing eye model' – that is, it should be thought of as simulating a treatment strategy in which people with diabetic macular oedema received treatment in their better-seeing eye only. This was a change from the manufacturer's original economic analysis, in which the modelled treatment strategy assumed that people would largely receive treatment in their worse-seeing eye. The manufacturer observed that cost–utility models produce lower ICERs when simulating treatment in the better-seeing eye of people with diabetic macular oedema than when simulating treatment in the worse-seeing eye. This is partly because the health-related quality of life of people with visual impairment is associated primarily with vision in the better-seeing eye and partly because the costs of severe visual impairment depend on vision in the better-seeing eye. Thus, in general, treatments that maintain or improve vision in the better-seeing eye will be favoured in economic analyses. The manufacturer cited the precedent of NICE's technology appraisal guidance on ranibizumab and pegaptanib for the treatment of age-related macular degeneration in which the Assessment Group's model had explicitly assumed that only the better-seeing eye was treated, but the Committee had made recommendations on the assumption that treatment would be given to the first eye to present clinically, be it the better- or worse-seeing eye. For these reasons, the manufacturer felt it would be helpful to present the Committee with a model that explicitly assumed treatment of the better-seeing eye, although the data on which the model was based were drawn predominantly from people whose worse-seeing eye had been treated, and it did not otherwise change the way the model associated vision in the worse-seeing eye with quality of life.

The manufacturer accepted the view the Committee had expressed in the appraisal consultation document that its original model had underestimated the hazard of death associated with diabetic macular oedema by not including the hazard associated with diabetes itself. Its revised model used a higher relative risk of death of 2.45 for people with diabetic macular oedema compared with the general population. The manufacturer derived this value by combining an estimate of the additional hazard of death associated with diabetes (1.93 compared with the general population, from an English epidemiological study by Mulnier et al. 2006) with an estimate of the additional hazard of death independently associated with macular oedema among people with diabetes in Wisconsin, USA (1.27, as reported by Hirai et al. 2008). The manufacturer stated that the revised relative risk of 2.45 was more plausible, although it might overestimate the true hazard associated with diabetic macular oedema. It noted that the revised model predicted that 43% of the cohort would remain alive after 15 years (at age 78), whereas the original model had suggested that 65% of people would be alive at that time.

In NICE technology appraisal guidance 237 the Committee considered at its first meeting that, by assuming people whose visual acuity rose to 76 letters or higher would stop receiving ranibizumab, the manufacturer's original model had not reflected likely clinical practice. Acknowledging this view, the manufacturer removed the stopping rule from the base case of its revised model. For related reasons, the manufacturer also revised the effectiveness evidence used to simulate the first year of treatment. The original model drew transitions between visual acuity states from those observed in the whole population in RESTORE, including participants whose visual acuity had been higher than 75 letters at the start of treatment. In its revised model, the manufacturer calculated visual acuity state transitions in the first year of treatment using only participants whose visual acuities matched those of the assumed starting population in the model (that is, only people with baseline visual acuities of 36 to 75 letters).

The manufacturer accepted that there were uncertainties around the assumption in its original model that people would need only 2 years of treatment with ranibizumab. Its revised model assumed that people receiving ranibizumab would receive an average of 2 injections in a third year of treatment and 1 injection in a fourth year of treatment. In the laser photocoagulation arm, the model assumed once-yearly treatments for years 3 and 4. To reflect the benefit that would accrue from these additional treatments, the manufacturer extended its original assumption that visual acuity would be maintained in all arms in year 2 of the model to encompass years 3 and 4 – that is, vision remained stable from years 2 to 4 and then declined equally in the group treated with ranibizumab and the group treated with laser photocoagulation.

When considering the manufacturer's original model, the Committee had expressed concern that it did not account for the need to treat both eyes in a significant proportion of people. The manufacturer did not alter its base case to address this issue. However, it provided a scenario analysis, using its revised model, which simulated treatment in both eyes for 35% of people. This analysis assumed that, in people with bilateral disease, both eyes would be treated and monitored at the same visit, with ranibizumab drug and treatment costs doubled. The analysis applied reduced costs associated with severe visual impairment because fewer people would go blind in both eyes. The analysis assumed that treating the second eye would result in utility gains a quarter the magnitude of those achieved by treating the first eye; this is because the health-related quality of life of people who can see well with both eyes is only a little better than the health-related quality of life of people who can see well with 1 eye. The model calculated this figure by applying a 25% uplift to the QALYs generated by ranibizumab.

In NICE technology appraisal guidance 237, the Committee had noted that the range of utility values used in the manufacturer's original model was broader than would be expected according to the assumptions of the model. The Committee had suggested that 1 possible explanation for this was that the regression model used to define the relationship between visual acuity and utility (see section 3.12) had not accounted for confounding variables reflecting the effect of diabetic comorbidities on health-related quality of life. To address this point, the manufacturer's consultation comments included an 'extended' analysis, which re-estimated the relationship between visual acuity and utility using a model containing additional covariates: age, sex, duration of diabetes, blood pressure control at baseline (categorised dichotomously), baseline HbA1c and a variable indicating whether each treated eye was the participant's better- or worse-seeing one. This analysis suggested that, apart from visual acuity, only sex was a significant predictor of utility (p<0.05). The manufacturer concluded that the utility function used in the original model had not been confounded by factors relating to diabetic comorbidities and retained the same utility values in its revised base case. However, the manufacturer also presented a scenario analysis that used the utility values estimated in the extended regression analysis, including all non-significant covariates.

When considering the manufacturer's original model, the Committee had questioned the validity of assuming that the relative improvement in vision achieved in the first year would persist for the duration of the model (see section 3.13), that is, that vision would deteriorate equally in the groups. The manufacturer responded to this point in its consultation comments, arguing that there was no evidence to suggest that gains in visual acuity would diminish at different rates depending on the treatment. It also cited the precedent of NICE's technology appraisal guidance on ranibizumab and pegaptanib for the treatment of age-related macular degeneration, in which the Committee had accepted an analogous approach, as a reasonable basis for decision-making.

In NICE technology appraisal guidance 237, the Committee had expressed concern that the unit cost of injection procedures used in the manufacturer's original model (£150.00; see section 3.15) might underestimate the true costs of administering ranibizumab. In its response to consultation, the manufacturer provided a 'bottom-up' estimate of the cost of an intravitreal injection visit, compiling separate estimates of the costs associated with consulting individual members of an ophthalmology clinic team and including overheads. This resulted in an estimated unit cost of £142.91, which the manufacturer took as evidence that the value used in the original model had accurately reflected the administration costs of ranibizumab. The manufacturer emphasised that charges applied to visits for ranibizumab injections are subject to local agreements between commissioners and providers, and may not always reflect the true costs of service delivery.

In NICE technology appraisal guidance 237, the Committee had initially concluded that the manufacturer's original model overestimated the cost savings of ranibizumab-based therapy that would be achieved by avoiding or delaying severe vision loss. The model drew the costs of severe vision loss from visual acuities that fell below 35 letters in treated eyes, but the data sources from which the manufacturer drew these costs used visual acuity in the better-seeing eye. By stating that its revised model should be considered to simulate treatment in the better-seeing eye, the manufacturer suggested it had removed this problem.

The results of the manufacturer's revised model in NICE technology appraisal guidance 237 included a subgroup analysis estimating the cost–utility of ranibizumab in people with the greatest degree of macular oedema (central foveal thickness greater than 400 micrometres). The manufacturer provided this analysis in response to comments from clinical specialists, reported in the original appraisal consultation document, suggesting that laser photocoagulation may be less effective in a thicker, more oedematous retina. For this reason, the manufacturer stated that such people represented a 'clinically plausible' subgroup in which ranibizumab could be expected to have a greater relative effect when compared with laser photocoagulation. The manufacturer confirmed that the trial protocol for RESTORE had pre-specified subgroup analyses according to 3 categories of central foveal thickness: less than 300 micrometres, 300 to 400 micrometres and greater than 400 micrometres. In its response to consultation, the manufacturer stated that it had carried out tests of the statistical significance of differences in clinical outcome according to baseline central foveal thickness category. However, the results of these tests were not presented in the submitted documentation.

The manufacturer's revised model also included a patient access scheme reflecting its agreement with the Department of Health in 2011 that ranibizumab will be made available to the NHS at a discounted price (level of discount confidential; see section 2.4).

The manufacturer's revised base case focused solely on the comparison of ranibizumab monotherapy with laser photocoagulation. It estimated that ranibizumab is associated with an ICER of £30,277 per QALY gained (disaggregated cost and QALY estimates are unavailable because of the confidentiality of the patient access scheme). In the subgroup of people with central foveal thickness greater than 400 micrometres, the equivalent ICER was £21,418 per QALY gained.

The manufacturer's scenario analysis simulating treatment in both eyes for 35% of people resulted in an ICER of £44,355 per QALY gained for ranibizumab monotherapy compared with laser photocoagulation. In the subgroup of people with central foveal thickness greater than 400 micrometres, the equivalent ICER was £35,719 per QALY gained.

The manufacturer provided a scenario analysis adopting utilities re-estimated from RESTORE data using an extended model with additional covariates reflecting baseline characteristics of participants and factors relating to diabetic comorbidities (see section 3.26). This resulted in an increase in the ICER from its base case of £30,277 per QALY gained to £33,857 per QALY gained for ranibizumab monotherapy compared with laser photocoagulation.

A further series of scenario analyses adopted alternative estimates of utility drawn from various published sources. When utility values from the better-seeing eye study by Lloyd et al. were used, the ICER was £24,779 per QALY gained. When utility was estimated according to an equation published by Sharma et al., associating visual acuity in the better-seeing eye with health-related quality of life, the ICER was between £12,312 and £12,610 per QALY gained, depending on the version of the equation used. A final analysis adopted utility values estimated in a study by Czoski-Murray et al. (referred to as Brazier et al. in NICE technology appraisal guidance 237), in which members of the general public valued levels of visual impairment that were simulated by custom-made contact lenses, using the time trade-off method. Participants wore the same lenses in both eyes, so the resulting utility values reflected bilateral impairment of vision. This was the source of utility values the Committee had judged most accurately reflected the health-related quality of life associated with visual impairment in NICE's technology appraisal guidance on ranibizumab and pegaptanib for the treatment of age-related macular degeneration. When these values were used in the revised ranibizumab model, the ICER was £23,664 per QALY gained. This ICER has been amended from the ICER given in the manufacturer's consultation comments to correct an error in the utility values used (identified by the NICE technical team and confirmed by the manufacturer in correspondence).

In its response to consultation on the appraisal consultation document for NICE technology appraisal guidance 237, the manufacturer provided additional arguments on the suitability of bevacizumab as a comparator. The manufacturer stated that bevacizumab could not be considered routine or best practice, because NHS experience is 'limited to experimental or compassionate use' and 'there are no robust data to demonstrate the safety, effectiveness and quality of the product'. The manufacturer argued that the optimum dose of bevacizumab for intraocular use is not established. It summarised 'emerging safety signals for the use of unlicensed intravitreal bevacizumab', and emphasised that any cost–utility analysis including bevacizumab would have to include the costs of an NHS pharmacovigilance programme. The manufacturer also reviewed the evidence that might be used to perform an indirect comparison of ranibizumab with bevacizumab. It concluded that significant methodological and clinical differences between studies precluded a valid analysis.

NICE technology appraisal guidance 237 did not recommend ranibizumab for the treatment of visual impairment due to diabetic macular oedema. After publication of NICE technology appraisal guidance 237, the manufacturer submitted a revised patient access scheme in which it applied a revised discount to ranibizumab for all indications (see section 2.4) to be considered as a rapid review of the original guidance.

As this was a rapid review, the manufacturer did not submit any additional clinical effectiveness data. However, in addition to the revised patient access scheme, the manufacturer submitted an amended economic model that attempted to address 6 specific concerns that were raised by the Committee in section 4.29 of NICE technology appraisal guidance 237 as follows:

• By not accounting for the need to treat both eyes in a large proportion of people with diabetic macular oedema, the manufacturer's revised base-case model underestimated the benefits and – to a greater degree – the costs of treatments. The manufacturer's scenario analysis simulating treatment in both eyes for 35% of people provided a more realistic reflection of likely clinical practice.

• The range of utility values used in the manufacturer's revised base case was broader than would be expected according to the assumptions of the model. The Committee preferred the manufacturer's scenario analysis adjusting for factors that may influence the relationship between visual acuity and health-related quality of life.

• The model underestimated the amount of ranibizumab that people with diabetic macular oedema were likely to need over time. Basing the number of injections for year 2 of the model's ranibizumab monotherapy arm on experience in DRCR.net overlooks the fact that the participants in the trial also received laser photocoagulation, which clinicians believe may have a ranibizumab-sparing effect. The declining number of ranibizumab injections assumed in years 3 and 4 is not evidence-based, and is unlikely to lead to stable vision during that period, as assumed. It may also have been unrealistic to assume that ranibizumab treatment will not continue beyond 4 years.

• The model's assumption that the relative benefit achieved during the treatment phase lasts indefinitely was unrealistic. If NICE's technology appraisal guidance on ranibizumab and pegaptanib for the treatment of age-related macular degeneration is considered as a precedent, then it should be noted that the model in that appraisal had a shorter time horizon, which limited the Committee's uncertainty about extrapolating treatment effects into the future.

• The model applied unequal assumptions about treatment and monitoring visits for people treated with ranibizumab and those treated with laser photocoagulation.

• The manufacturer's model overestimated the degree of glycaemic control that would be expected in people treated in clinical practice in the UK.

In response to the Committee's concerns, the manufacturer amended its economic model with the following revisions:

• To approximate an ICER for treating both eyes, the manufacturer multiplied the ICER from the revised model, considered by the manufacturer to represent a better-seeing eye model, by a factor of 1.5. The manufacturer noted that this approach was consistent with that taken by the Committee in NICE's technology appraisal guidance on ranibizumab and pegaptanib for the treatment of age-related macular degeneration, when it observed that a policy of treating the first eye to come to clinical attention would result in substantially higher costs, but fewer savings and lower utility gains, than a policy of treating only the better-seeing affected eye. The manufacturer did not make any additional changes to the model to address this issue.

• To address the Committee's concerns about utility, the manufacturer considered that the utility values estimated in a study by Czoski-Murray et al. (2009) were those preferred by the Committee. In this study, the investigators developed a regression model that estimated the contribution of visual acuity (measured by the LogMAR [logarithm as the minimal angle of resolution] scale) to health-related quality of life, adjusting for age. The regression model was subsequently used by the manufacturer to estimate utility values for each of the 8 visual acuity health states (defined by the ETDRS scale) after converting the upper and lower limits of the ETDRS scale to its LogMAR equivalent. Based on a mean age of 65 years, the estimated utility values used in the manufacturer's new base-case analysis ranged from 0.869 for the best health state to 0.353 for the worst health state.

• To address the Committee's concerns about the number of ranibizumab injections people with diabetic macular oedema would need, the manufacturer assumed that people receiving ranibizumab alone would need a total of 14 ranibizumab injections over 4 years: 7 injections in the first year, 4 injections in the second year, and 3 injections in the third year. These assumptions were based on a 2-year extension of the RESTORE study, which showed that trial participants needed a decreasing number of ranibizumab injections from the first year to the third year. The manufacturer assumed that no injections were needed in the fourth year. The manufacturer also carried out a threshold analysis to estimate the maximum number of ranibizumab injections that could be administered over the time horizon of the model while maintaining an ICER below £30,000 per QALY gained.

• To address the Committee's concerns about extrapolating the relative benefit of treatment with ranibizumab beyond the initial treatment phase, the manufacturer reduced the time horizon of the model from 15 years to 10 years. The manufacturer also conducted a threshold analysis to explore the highest rate at which vision could worsen in people treated with ranibizumab and maintain an ICER below £30,000 per QALY gained. To do this, the manufacturer assumed from the fourth year onwards, quarterly probabilities of 0.025 for improvement and 0.055 for deterioration for people receiving ranibizumab monotherapy and quarterly probabilities of 0.025 for improvement and 0.035 for deterioration for people receiving laser photocoagulation alone.

• To address the Committee's concerns about care provided during visits to the eye clinic, the manufacturer assumed that for people receiving either ranibizumab monotherapy or laser alone, a visit for treatment also included monitoring. People receiving ranibizumab monotherapy visited the eye clinic 12 times in the first year, 8 times in the second year and 6 times in the third year; those receiving laser photocoagulation alone had 4 visits per year in the first, second and third years, and people on either treatment had 2 visits per year in the fourth year.

• Despite the Committee's concerns that the ICER for ranibizumab monotherapy compared with laser photocoagulation would be higher in people with poor glycaemic control, the manufacturer did not present further subgroup analyses according to the degree of glycaemic control. The manufacturer noted the Committee's considerations in NICE technology appraisal guidance 237 that analyses restricted to people with good glycaemic control (HbA1c less than 8%) and poor glycaemic control (HbA1c 8% or more) were 'exploratory'. The manufacturer also suggested that these subgroup analyses need careful interpretation because of the small sample sizes, which resulted in a very small number of people in extreme health states influencing the results.

The manufacturer's model for the rapid review submission, including the revised patient access scheme, compared ranibizumab monotherapy with laser photocoagulation. The manufacturer presented separate ICERs for treating the better-seeing eye and for treating both eyes. In the base case, the manufacturer estimated that treating the better-seeing eye with ranibizumab was associated with an ICER of £14,137 per QALY gained and that treating both eyes with ranibizumab was associated with an ICER of £21,205 per QALY gained. The manufacturer estimated that an additional 4 injections of ranibizumab can be given in years 4 to 9 (resulting in a total of 18 injections) for the ICER when treating both eyes to remain below £30,000 per QALY gained. The manufacturer also estimated that the rate of deterioration in vision (visual acuity) for people treated with ranibizumab in both eyes from year 4 onwards would need to be more than 1.5 times higher than that for people treated with laser photocoagulation for the ICER to increase to £30,000 per QALY gained. The manufacturer conducted one-way sensitivity analyses that suggested that the model was most sensitive to changes to the time horizon and to utility values. When the time horizon was limited to 5 years the ICER associated with treating both eyes was £41,568 per QALY gained. When the manufacturer instead used utility values from the study by Lloyd et al. (representing the contribution to utility from the better-seeing eye), the ICER associated with treating both eyes was £43,716 per QALY gained. The manufacturer provided no probabilistic sensitivity analyses in its rapid review submission.

To address the Committee's concerns about the validity of the manufacturer's previous subgroup analyses (submitted in response to the consultation on the original appraisal consultation document for NICE technology appraisal guidance 237) according to retinal thickness and the inconsistent relationship between retinal thickness and cost effectiveness, the manufacturer presented additional subgroup analyses according to the degree of retinal thickness. For the rapid review, the manufacturer presented subgroup analyses based on central retinal (rather than foveal) thickness, arguing that this more reliably measures retinal thickness than central foveal thickness. The manufacturer acknowledged that the pattern of cost-effectiveness estimates for the 3 subgroups defined by central foveal thickness had been erratic, and may have been influenced by small sample sizes (see section 3.42). Therefore, the manufacturer combined the 2 subgroups with lower values of central retinal thickness to create 2 subgroups (less than 400 micrometres and 400 micrometres or greater) of similar size. The manufacturer presented post hoc tests of the statistical significance of differences in clinical outcome according to baseline central retinal thickness, which suggested that laser photocoagulation was less effective in people with central retinal thickness of 400 micrometres or more (p<0.01) than in people with thicker retinas. In response to Committee comments in NICE technology appraisal guidance 237 that the manufacturer should explore subgroup-specific parameters for all model inputs, and not only effectiveness, the manufacturer also adjusted other model parameters according to the 2 subgroups, including distribution of visual acuity at baseline and treatment frequency in the first year. For people with central retinal thickness of 400 micrometres or more, the ICER associated with treating only the better-seeing eye was £8881 per QALY gained and the ICER associated with treating both eyes was £13,322 per QALY gained. For people with central retinal thickness of less than 400 micrometres, the ICER associated with treating the better-seeing eye was £28,861 per QALY gained and the ICER associated with treating both eyes was £43,292 per QALY gained.