Ranibizumab for treating choroidal neovascularisation associated with pathological myopia

The manufacturer submitted evidence of clinical and cost effectiveness for ranibizumab compared with verteporfin photodynamic therapy (vPDT) in people with choroidal neovascularisation associated with pathological myopia. Pathological myopia is a chronic condition characterised by excessive lengthening of the eye and degenerative changes at the back of the eye. These changes to the eye can cause blood vessels to leak or bleed into the retina in a process known as choroidal neovascularisation. This can result in visual impairment, in particular a loss of central vision. The manufacturer did not provide a comparison with bevacizumab, which is listed as a comparator in the scope for this appraisal. It did not consider bevacizumab to be a valid comparator because it is unlicensed for this condition and not routinely used.

The main sources of evidence presented in the manufacturer's submission came from a Novartis phase 3 trial (RADIANCE) and 2 other randomised trials (Gharbiya 2010; Iacono 2012). Gharbiya (2010) and Iacono (2012) compared ranibizumab with bevacizumab. However, the manufacturer did not present data from the bevacizumab arm of these trials.

RADIANCE compared ranibizumab with vPDT in people with visual impairment caused by choroidal neovascularisation secondary to pathological myopia. The trial was a randomised, double‑blind, multicentre study conducted in 20 countries, which compared 2 groups of patients using ranibizumab (n=222) with 1 group using vPDT (n=55). On day 1 of treatment, patients in the ranibizumab groups received 0.5 mg of ranibizumab and patients in the vPDT group were given 6 mg/m² of verteporfin intravenously, followed by a light dose of 50 J/cm² at an intensity of 600 mW/cm² for 83 seconds. In the ranibizumab disease activity group (n=116) and the vPDT group, patients were re‑treated if visual impairment caused by intra or subretinal fluid, or active leakage secondary to pathological myopia, was seen. Treatment was continued until these effects were no longer seen. In the ranibizumab disease stabilisation group (n=106), patients were re‑treated if there was a loss of best corrected visual acuity (BCVA) because of disease activity. Treatment was continued until BCVA was stable for 3 consecutive monthly assessments.

The primary end point of RADIANCE was the mean average change in BCVA between baseline and months 1 to 3, measured using the Early Treatment of Diabetic Retinopathy Study (ETDRS) eye chart, in which a score of 85 letters corresponds to normal visual acuity. Gains in BCVA (reported as mean±standard deviation [SD]) were statistically significantly greater in both ranibizumab groups (disease activity group; 10.6±7.3 letters, p<0.0001 compared with vPDT, disease stabilisation group; 10.5±8.2 letters, p<0.0001 compared with vPDT) than in the vPDT group (2.2±9.5 letters). The secondary end points included the proportion of patients gaining 10 or more or 15 or more letters, mean change in BCVA, and changes in central retinal thickness from baseline. Both of the ranibizumab groups had statistically significantly more patients gaining 10 or more letters or 15 or more letters than the vPDT group. There was no statistically significant difference between either of the ranibizumab groups compared with the vPDT group in mean change in BCVA or in mean change in central retinal thickness. The length of follow‑up was 12 months for the 2 ranibizumab groups. After 3 months, 72% of the patients in the vPDT group received ranibizumab. Therefore, the manufacturer did not compare the results of the vPDT group with the results of the ranibizumab groups after the initial 3‑month period.

The 2 other randomised trials (Gharbiya 2010; Iacono 2012) were single‑centre trials conducted in Italy comparing ranibizumab with bevacizumab. The manufacturer did not present the data for the bevacizumab arm for either trial. The Iacono (2012) study was a double‑blind clinical trial in people with subfoveal choroidal neovascularisation secondary to pathological myopia (55 eyes; ranibizumab=27, bevacizumab=28) with a follow‑up period of 18 months. Gharbiya (2010) was an interventional study in people with subfoveal or juxtafoveal choroidal neovascularisation secondary to pathological myopia and evidence of leakage from the choroidal neovascularisation lesion (32 eyes; ranibizumab=16, bevacizumab=16) with a follow‑up period of 6 months. The mean (±SD) change in BCVA was 9±NR (not reported) letters in the ranibizumab arm of the Iacono (2012) study and 17.3±11.1 letters in the ranibizumab arm of the Gharbiya (2010) study. It was not reported how many patients gained 10 or more letters in the Iacono study, although 7 (30%) gained 15 or more letters in the ranibizumab arm. In the ranibizumab arm of the Gharbiya (2010) study, 12 (75%) patients gained 10 letters or more and 9 (56%) gained 15 letters or more. The mean change in retinal thickness was not reported in the Iacono (2012) study. In the ranibizumab arm of the Gharbiya (2010) study, the mean change in retinal thickness was -45±NR micrometres.

The manufacturer identified 6 non‑randomised studies relevant to the decision problem. All 6 studies investigated the use of ranibizumab in patients with choroidal neovascularisation secondary to pathological myopia, with follow‑up times ranging from a mean of 8 months to a median of 17 months. One study was a multicentre phase 2 study (the REPAIR study) and the other 5 studies were prospective case‑series (Calvo‑Gonzalez 2011; Lalloum 2010; Ouhadj 2010; Silva 2010; Vadala 2011). A statistically significant change in BCVA from baseline to time of assessment was shown in 4 of the 6 studies. The number of patients who gained 15 or more letters at follow‑up ranged from 24% to 47%.

Adverse effects of ranibizumab were reported in RADIANCE. Ocular adverse events in the ranibizumab disease activity group were 16 (14%, 0 severe), 31 (26%, 1 severe), and 44 (37%, 1 severe) and in the ranibizumab disease stabilisation group were 29 (27%, 0 severe), 38 (36%, 0 severe), and 46 (43%, 1 severe) by 3, 6, and 12 months respectively. There were 5 (9%) ocular adverse events in the vPDT group by 3 months, of which none were severe. Non‑ocular adverse events in the ranibizumab disease activity group were 30 (25%, 1 severe), 42 (36%, 3 severe), and 51 (43%, 6 severe) and in the ranibizumab disease stabilisation group were 27 (26%, 0 severe), 38 (36%, 1 severe), and 48 (45%, 3 severe) by 3, 6, and 12 months respectively. There were 6 (11%) non‑ocular adverse events in the vPDT group by 3 months, of which none were severe. There were no systemic or significant ocular adverse events in the Iacono (2012) or Gharbiya (2010) trials. REPAIR reported adverse events that occurred in 2 or more patients. Ocular adverse events occurred in 29 (45%) patients and non‑ocular adverse events in 39 (60%) patients over 12 months. Calvo‑Gonzalez (2011) reported that 2 eyes developed anterior uveitis over a mean follow‑up of 16 months. The other 4 non‑randomised studies (Lalloum 2010; Ouhadj 2010; Silva 2010; Vadala 2011) reported that no systemic or ocular adverse events were observed, with the mean follow‑up ranging from 8 to 17 months.

Impact on health‑related quality of life was measured in RADIANCE. The change in National Eye Institute Visual Functioning Questionnaire 25 item (NEI VFQ‑25) composite score from baseline to 3 months (reported as mean±SD) was statistically significantly higher for the 2 ranibizumab groups (disease activity group; 4.3±10.1, p<0.05 compared with vPDT, disease stabilisation group; 5.3±14.0, p<0.05 compared with vPDT) compared with the vPDT group (0.3±12.6). The mean (±SD) change in the EQ‑5D questionnaire from baseline to 3 months was 2.3±55.0, 4.2±NR, and 2.1±NR for the ranibizumab disease activity, ranibizumab disease stabilisation, and vPDT groups respectively. The mean (±SD) reduction in Work Productivity and Activity Impairment Questionnaire: General Health (WPAI‑GH) score from baseline to 3 months was 22.0±55.0, 21.9±75.2, and 10.2±59.9 for the ranibizumab disease activity, ranibizumab disease stabilisation, and vPDT groups respectively. The statistical significance of the differences between the groups for the EQ‑5D and WPAI‑GH scores were not reported.

The manufacturer developed a cost–utility Markov model that evaluated the cost effectiveness of ranibizumab compared with vPDT in people with choroidal neovascularisation associated with pathological myopia. There were 8 health states in the model, defined by the BCVA in the treated eye in addition to the absorbing health state of death. The health states were defined by a 10‑letter range in BCVA. The model had 3‑monthly cycles and a lifetime time horizon.

The transition probabilities for the first cycle of the model (baseline to month 3) for both ranibizumab and vPDT were based on RADIANCE. For the next 3 cycles (months 4 to 12), the transition probabilities between health states were derived from RADIANCE for ranibizumab and from the Verteporfin in Photodynamic Therapy (VIP) trial for vPDT. VIP compared vPDT with photodynamic therapy in 120 patients with subfoveal choroidal neovascularisation secondary to pathological myopia. For cycles 5 onwards (1 year onwards), a slow worsening of visual acuity was assumed, based on natural disease progression reported in Yoshida (2002) for the base case and an additional 6 natural history studies for the other transition probabilities. The model included crossover from the better‑seeing eye to the worse‑seeing eye and vice versa as patients changed health states.

A baseline rate of bilateral involvement (that is, both eyes affected by choroidal neovascularisation) of 15% was derived from 2 published studies (Cohen 1996; Hampton 1983) and the model assumed no incidence of choroidal neovascularisation secondary to pathological myopia after baseline measurement. Based on expert opinion, the manufacturer estimated a recurrence of choroidal neovascularisation in 6% of patients each year after the first 2 years of modelling. The manufacturer assumed an indefinite duration of treatment benefit, based on the treatment benefit seen at year 1.

Base‑case utility values for the better‑seeing eye were taken from a published study of the UK general population in which BCVA health states were simulated with contact lenses that created the effects of age‑related macular degeneration (Czoski‑Murray et al. 2009). They ranged from 0.850 in patients with a BCVA of 86 to 100 letters to 0.353 for those with a BCVA of less than 25 letters. Base‑case utility values for the worse‑seeing eye were calculated from the values for the better‑seeing eye, with the assumption that the maximum utility gain in the worse‑seeing eye was 0.1. These utilities therefore ranged from 0.850 for a BCVA of 86 to 100 letters, to 0.750 for a BCVA of less than 25 letters.

Disutilities were defined as adverse events that occurred in more than 5 patients and were suspected to be related to the study drug or ocular injection in RADIANCE (for ranibizumab) or VIP (for vPDT). Disutilities were conjunctival haemorrhage (ranibizumab; 8.5%, vPDT; 0%), increased intraocular pressure (ranibizumab; 4.2%, vPDT; 0%), visual disturbance (ranibizumab; 0%, vPDT; 14.8%), and injection site adverse events (ranibizumab; 0%, vPDT; 9.9%).

Total costs for treatment were calculated from the unit costs, administration costs, and the cost of a monitoring visit multiplied by the total number of treatment visits and monitoring visits needed. The cost of blindness was calculated as £17,326 in the first year and £17,245 in each year after.

The manufacturer's base‑case deterministic cost‑effectiveness analysis results showed that ranibizumab dominated vPDT (that is, it was more effective and less costly), resulting in more QALYs (13.18 compared with 12.75) and lower costs (£9694 compared with £12,455). The manufacturer similarly presented base‑case probabilistic results which showed that ranibizumab dominated vPDT.

The manufacturer conducted one‑way sensitivity analyses using a net monetary benefit approach (calculated by multiplying the incremental QALYs by £20,000 and then subtracting the incremental costs) because ranibizumab dominated vPDT in the base‑case analysis. The sensitivity analysis showed that the model was sensitive to changes in the unit cost of ranibizumab and vPDT, the number of ranibizumab injections in the first and second year, the starting age of the patient group, the discount rate for benefits and the maximum utility gain in the worse‑seeing eye. The results of the manufacturer's sensitivity analysis showed that ranibizumab remained dominant up to a unit cost of £783 (range £0 to £3750) and when up to 12 injections were needed in either year 1 or year 2 (range 0 to 12, with vPDT given 3.4 times per year). Scenario analyses showed that ranibizumab remained dominant when other methods for calculating transition probabilities, such as keeping transition probabilities constant across all visual acuity levels, and other sources of natural history data (Bottoni et al. 2001; Hampton et al. 1983; Hotchkiss et al. 1981; Kojima et al. 2006; Secretan et al. 1997; Tabandeh et al. 1999; Yoshida et al. 2002), were used, and when the maximum gain in utility for the worse‑seeing eye is 0.2 or 0.3. The sensitivity analysis showed that there was a 100% probability of ranibizumab being cost effective if the maximum acceptable ICER was £20,000 or £30,000 per QALY gained.

The manufacturer conducted 3 scenario analyses. The first scenario analysis involved calculating the transition probabilities from patient‑level data using 3 different methods. The base‑case method used probabilities that were dependent on the current BCVA level and assumed the patient could move from any health state to any other health state in each cycle. The second method used probabilities that were dependent on the patients' current BCVA level for the top 2 health states only, so that a patient could only gain or lose up to 2 health states in each cycle. The third method used a constant probability across all BCVA levels, regardless of the patient's current BCVA level, and assumed that patients could only gain or lose up to 2 health states each cycle. The second scenario analysis involved using different sources for calculating transition probabilities beyond year 1. The third scenario analysis involved using different values for the maximum utility gains for the worse‑seeing eye. Ranibizumab continued to dominate vPDT in all of the scenario analyses.

The ERG commented that the manufacturer did not include bevacizumab as a comparator even though it was included in the NICE appraisal scope. The ERG noted that the manufacturer stated that bevacizumab is unlicensed for use in the UK for choroidal neovascularisation associated with pathological myopia and that use of bevacizumab is not established practice in the UK for this indication. The ERG stated that although vPDT has a UK marketing authorisation for treatment of choroidal neovascularisation, it is rarely used in clinical practice.

The ERG found 2 head‑to‑head trials of bevacizumab and ranibizumab. The ERG noted that the manufacturer had included these 2 trials in their submission, but had only presented data from the ranibizumab arms. The ERG stated that neither of these studies showed statistically significant differences between the ranibizumab and bevacizumab arms in mean change from baseline in BCVA, mean change in central retinal thickness, or in the number of patients gaining 10 or more or 15 or more letters.

The ERG noted that in RADIANCE the primary end point was at 3 months, and the ERG's clinical specialist thought that 12 months should be the minimum to assess longer-term efficacy of treatment. The ERG stated that in VIP, the statistically significant difference between the vPDT and photodynamic therapy groups in the primary end point at 3 months was no longer seen at 24 months, and that this could also be true for ranibizumab. The ERG believed that it was unlikely that a 3‑month follow‑up period would provide adequate information about potential adverse effects of the anti‑VEGF treatment.

The ERG noted that geographic atrophy, which is an advanced form of dry age‑related macular degeneration in which the rods and cones of the retina degenerate, is a common feature in patients with pathological myopia. It stated that the development of geographic atrophy or extension of pre‑existing geographic atrophy has been recognised as a potential side effect in patients with age‑related macular degeneration having anti‑VEGF treatment. The ERG was concerned that geographic atrophy was not assessed in RADIANCE because it can affect long‑term visual outcomes.

The ERG noted that there was a difference between the patient populations in the RADIANCE and VIP trials. It was concerned that RADIANCE included a greater proportion of patients with non‑subfoveal involvement. The ERG stated that this may affect the comparability of the trials, because patients with subfoveal involvement tend to have a worse prognosis. The ERG suggested that the difference in the number of patients with subfoveal involvement in the 2 trials may overestimate the benefit of ranibizumab.

The ERG noted that the model accounted for the possibility of the better‑seeing eye becoming the worse‑seeing eye, and vice versa, as patients change health states. The ERG stated that the method used by the manufacturer may underestimate the net quality‑adjusted life year (QALY) gains and costs of blindness that may arise from the more effective treatment.

The ERG questioned whether an appropriate source had been used for the health‑related quality-of-life data in the model. The ERG identified the Brown et al. (1999) study, which measured health‑related quality of life directly from patients with impaired vision in at least 1 eye, producing a narrower range of utility values than the study by Czoski‑Murray et al. (2009).

The ERG noted that the number of ranibizumab injections needed in years 2 and 3 may have been underestimated. It described a study by Franqueira et al. (2012) that reported results of a 3‑year retrospective study of 40 eyes with choroidal neovascularisation associated with pathological myopia. The mean number of injections in the study was 2.4 in year 2. The ERG suggested that 1.7 injections in year 2 would be a more reasonable assumption than the 1 injection in year 2 assumed in the manufacturer's model.

The ERG commented that the costs of blindness may have been overestimated. This was driven by the different costs of residential care calculated by the ERG and the manufacturer. The ERG suggested that a cost of blindness of £7510 in the first year and £7429 in each subsequent year, based on 2011 Personal and Social Services Research Unit costs and 30% of people being privately funded, was a more reasonable assumption.

The ERG noted that there were health state probabilities included in the manufacturer's model that were populated by relatively few patients. It was unsure whether the trials provided sufficient patient‑level data to be able to sensibly populate a model with 8 health states and a 64 cell transition probability matrix. Therefore the ERG was concerned about the reliability of the manufacturer's probability modelling.

The ERG noted some uncertainty about the use of mortality multipliers in the manufacturer's model. It stated that the definition of visual impairment in Christ et al. (2008), which was used by the manufacturer as a source of the multipliers, was ambiguous.

The ERG noted that EQ‑5D data were collected in RADIANCE but were not included in the manufacturer's submission. The ERG requested the EQ‑5D data during the clarification process and these were provided by the manufacturer. The ERG commented that the data indicated that changes in the BCVA of the worse‑seeing eye had no impact on patients' health‑related quality of life.

The ERG stated that the manufacturer's assumption that treatment benefit would continue indefinitely was optimistic. The ERG performed exploratory analyses that incorporated alternative durations of treatment benefit of 1, 5, 10 and 20 years. This caused the net savings, QALYs and health benefits to decrease compared with those in the manufacturer's model. Ranibizumab remained dominant compared with vPDT even for a 1‑year duration of treatment benefit.

The ERG highlighted minor errors in the manufacturer's model, in the calculation of the quarterly proportion of patients worsening, derived from natural history data. These errors were acknowledged in the manufacturer's clarification responses. The ERG corrected the errors in their exploratory analysis of the manufacturer's model.

The ERG conducted an exploratory analysis which involved the following modifications to the manufacturer's model:

• Brown et al. (1999) as a source of utility values in addition to Czoski‑Murray et al. (2009)

• changed the dose of ranibizumab in year 2 from 1 to 1.7

• changed the costs of blindness from £17,326 in year 1 and £17,245 in each subsequent year to £7510 and £7429 respectively

• changed the mortality multiplier for blindness (BCVA of 35 letters or less) from 1.54 to 1.48

• corrected the calculation of the quarterly proportion of patients worsening.

In the ERG's exploratory analysis, ranibizumab dominated vPDT. The total cost of ranibizumab was £10,055 and of vPDT was £12,529 (incremental cost -£2474). Using utility values from Brown et al. (1999), the total QALYs were 14.514 for ranibizumab and 14.170 for vPDT (incremental QALYs 0.344). Using utility values from Czoski‑Murray et al. (2009), the total QALYs were 13.105 for ranibizumab and 12.838 for vPDT (incremental QALYs 0.266).

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