A United Kingdom-based economic evaluation of ranibizumab for patients with retinal vein occlusion (RVO)

Abstract

Objective:

This study compares the cost-effectiveness of intravitreal ranibizumab vs observation and/or laser photocoagulation for treatment of macular edema secondary to retinal vein occlusion in a UK-based model.

Methods:

A Markov model was constructed using transition probabilities and frequency of adverse events derived using data from the BRAVO, CRUISE, and HORIZON trials. Outcomes associated with treatments and health states were combined to predict overall health costs and outcomes for cohorts treated with each option.

Results:

In branch retinal vein occlusion, ranibizumab produced a gain of 0.518 quality-adjusted life years at an incremental cost of £8141, compared with laser photocoagulation. The incremental cost-effectiveness ratio was £15,710 per quality-adjusted life year, and the incremental cost per month free from blindness was £658. In central retinal vein occlusion, ranibizumab produced a gain of 0.539 quality-adjusted life years at an incremental cost of £9216, compared with observation only. The incremental cost-effectiveness ratio was £17,103, and the incremental cost per month free from blindness was £423.

Conclusions:

These incremental cost-effectiveness ratios are below the £20,000–30,000 range typically accepted as a threshold for cost-effectiveness. This suggests that ranibizumab may be regarded as a cost-effective therapy for patients with macular edema secondary to retinal vein occlusion, relative to grid laser photocoagulation (for BRVO) and observation (for CRVO). Limitations include sparse data for utilities associated with the severity of visual impairment in the WSE in patients with RVO. A lack of direct comparative evidence between ranibizumab and the dexamethasone intravitreal implant for the treatment of BRVO and CRVO and the infeasibility of an indirect comparison due to significant heterogeneity in trial designs prevented the inclusion of this treatment as a comparator in the Markov model.

Keywords::

• Cost-effectiveness
• Retinal vein occlusion
• Macular edema
• Ranibizumab

Introduction

Retinal vein occlusion (RVO) is an important cause of vision loss among elderly adults1 and may involve the central or branch retinal veins2. Occlusion of the retinal veins causes an increase in retinal capillary pressure resulting in up-regulation of vascular endothelial growth factor (VEGF) expression and a consequent increase in vascular permeability and new vessel proliferation within the iris and anterior chamber. As a result, blood and plasma are discharged into the retina, often causing complications including macular edema (MO) and varying degrees of ischemia, potentially leading to severe vision loss.

MO secondary to RVO can lead to a significant decrease in vision-related quality-of-life3,4. The prognosis for branch RVO (BRVO) is frequently more favorable than that for central RVO (CRVO)2. The current standard of care for BRVO, according to National Institute for Health and Care Excellence (NICE) guidelines, is grid laser photocoagulation, which aims to preserve vision while minimizing complications5,6. However, recent RCT evidence indicates that anti-VEGF therapies are significantly more effective at improving BCVA than laser photocoagulation7. Anti-VEGF therapy is now the standard of care for CRVO, replacing an observation-only approach which used rescue pan-retinal laser photocoagulation in the case of rubeosis iridis development8–10. The dexamethasone intravitreal implant has received a UK marketing authorization for MO secondary to either BRVO or CRVO. Laser anastomosis and optic nerve sheathotomy are surgical procedures that have been studied in clinical trials, but these options are rarely undertaken in current clinical practice11.

This study aimed to compare the cost-effectiveness of using ranibizumab vs observation for CRVO and vs grid laser photocoagulation therapy for BRVO. The perspective of the evaluation was that of the National Health Service (NHS) in England and Wales, as cost and resource use data from England were used.

Ranibizumab (Lucentis®, Genentech Inc., San Fransisco, CA and Novartis Pharma AG, Basel, Switzerland) is an anti-VEGF drug licensed for the treatment of patients with MO secondary to RVO. It is administered as an intravitreal injection and has a recommended posology of 0.5 mg once per month for 3 months and as required thereafter, based on clinical outcomes12. Ranibizumab has been studied in three pivotal RVO studies: BRAVO, which evaluated the treatment with ranibizumab in BRVO13; CRUISE, which evaluated the treatment with ranibizumab in CRVO14; and HORIZON (Cohort 2, comprising patients with MO after RVO who had completed the BRAVO and CRUISE studies)15–17. BRAVO and CRUISE efficacy and safety data were used to populate year 1 of the model, and HORIZON data were used to populate year 2.

The primary end-point for BRAVO and CRUISE was the mean change in best-corrected visual acuity (BCVA) letter score from baseline to month 6. The two studies demonstrated that ranibizumab 0.5 mg led to a statistically greater increase in BCVA compared with the control (in BRAVO, the change was +18.46 compared with +7.17 for ranibizumab and grid laser therapy14, respectively, while CRUISE reported changes of +14.81 for ranibizumab compared with +0.92 for the sham injection)13. Improvement in BCVA reported at 6 months was sustained when patients were treated for up to 12 months15,16.

Materials and methods

Model structure

A Markov model for MO secondary to RVO was developed in Microsoft Excel. Patients moved between nine different health states at monthly intervals (cycles). The health states were based on eight different BCVA intervals, using the Early Treatment Diabetic Retinopathy Study (ETDRS) letter scale18 and a ninth state, ‘death’ (Figure 1).

Figure 1. Model structure. Although different scales (e.g., logMAR, Snellen) exist to measure the severity of a patient’s visual impairment, the Early Treatment of Diabetic Retinopathy Study letter scale was chosen for this model because it is commonly used in clinical trials and is the scale used in the BRAVO and CRUISE clinical studies.

This model aimed to predict changes in BCVA, utility, resource use and costs in each patient with visual impairment due to MO secondary to RVO. Using monthly cycles in the first year, the model predicted each patient’s health state based on outcomes from BRAVO for ranibizumab vs grid laser photocoagulation14 and from CRUISE for ranibizumab vs sham injection/observation13.

The base-case analysis assumed a lifetime time horizon (i.e. follow-up of all patients until death) and applied a discount rate of 3.5% to costs and benefits, in line with UK guidelines for economic evaluation and health technology assessment19.

Treatment effectiveness

The model’s clinical inputs in year 1 were driven by data from the BRAVO and CRUISE trials (Table 1). Patient-level data were used to calculate monthly transition probabilities for each treatment arm separately. For ranibizumab, data from the 0.5 mg arm in BRAVO and CRUISE for up to 12 months were used. Transition probabilities for grid laser photocoagulation in BRVO were estimated from the active control arm of BRAVO for 6 months, because laser treatment in that trial was applied as per standard of care; thereafter, laser treatment was assumed to follow the same transition probabilities as ranibizumab. For observation in CRVO, transition probabilities were estimated from the control arm of CRUISE for 6 months and, for months 7–12, the same probabilities were assumed as for months 2–6 (the first month is omitted in order to avoid bias introduced by spontaneous resolution during this early period). Effectiveness data for ranibizumab in year 2 were drawn from the HORIZON (Cohort 2) study. In the absence of clinical trial data beyond year 2, it was assumed that patients achieved stability by the end of year 2 except for the natural aging effect of a progressive decrease in BCVA based on advice received from experts.

Table 1. Model inputs for effectiveness data (and adverse event rates for ranibizumab only).

	BRVO		CRVO
	Ranibizumab	Grid laser photocoagulation	Ranibizumab	Observation
Month 1	BRAVO	BRAVO	CRUISE	CRUISE
Months 2–6	BRAVO	BRAVO	CRUISE	CRUISE
Months 7–12	BRAVO	Same as ranibizumab	CRUISE	Same as months 2–6
Year 2	HORIZON (Cohort 2)	Same as ranibizumab	HORIZON (Cohort 2)	Natural history
Year 3+	Natural history	Natural history	Natural history	Natural history

BRVO, branch retinal vein occlusion; CRVO, central retinal vein occlusion.

Deterioration in BCVA was integrated into the model from year 3 onwards using data sourced from a population-based observational study20. This study reported that the number of patients with ‘mild’ symptoms increased from 2% in patients aged 43–54 to 9.4% between the ages of 65–74; the increase from ‘no problems’ to ‘mild’ is equivalent to around 2 lines of letters. Therefore, it was assumed that 7.4% of patients would worsen by this magnitude of reduced letter score over a period of 20 years, based on the assumption that no injections are administered from year 3 onwards, as described below. This progression rate was converted into a monthly probability for inclusion in the model, which translated to a probability of losing 2–4 lines of 0.031% per monthly cycle. The following formula was used for the above-mentioned conversion:

The economic model follows each patient’s BCVA in the treated eye, which in some cases was the better-seeing eye (BSE), due to ocular comorbidities in the fellow eye; in other cases it was the worse-seeing eye (WSE). Based on the proportions of patients in BRAVO and CRUISE whose study eye was the BSE or WSE at baseline, it was assumed that 10% of treatments would be in the patient’s BSE, rising to 21.5% at 12 months (due to two factors: first, the expectation that treatment would improve the BCVA in the study eye to the extent that, in some cases, it would become the patient’s BSE, and second, that the fellow eye would also become affected in some patients). The proportion of patients treated in their BSE impacts on the calculations for quality-of-life and mortality, because both use a weighted approach based on the number of patients with BCVA that is representative of their BSE and WSE, respectively.

Mortality

All-cause mortality was included in the model, using annual rates based on life tables for England and Wales. Studies have demonstrated a relationship between BCVA and risk of mortality21–23, with one study presenting results specific to patients with RVO23. It is, therefore, appropriate to assume that the worsening of BCVA in RVO is associated with increased mortality. When the BSE was affected, the risk ratio (RR) applied to each BCVA band was 1.00 for BCVA letter scores above 56, 1.23 for letter scores between 36–55, and 1.54 for letter scores below 3621. When the WSE was affected, a risk ratio was used only for BCVA levels below 35 letters score, and a ratio of 1.23 was then applied21.

Adverse events

For grid laser photocoagulation and observation, it was assumed that no ocular adverse events would be observed due to the limited data available on the occurrence of these adverse events. This conservative assumption is supported by 12-month results from SCORE6. Ocular adverse event frequencies for ranibizumab were included in the model based on combined rates observed within BRAVO and CRUISE. The rationale for including each adverse event was based on the relative prevalence and severity of each event (i.e., infrequent and non-severe events and events not associated with increased health resource utilization were not included). The rates of adverse events differed in the first and second years of treatment (Table 2).

Table 2. Adverse event rates in years 1 and 2.

	Year 1		Year 2
	Ranibizumab	Grid laser photocoagulation/observation	Ranibizumab	Grid laser photocoagulation/observation
Cataract	6.60%	0.00%	5.42%	0.00%
Increased IOP (treated with drug)	10.00%	0.00%	5.42%	0.00%
Increased IOP (treated with surgery)	0.00%	0.00%	0.49%	0.00%
Iris neovascularization	2.35%	0.00%	0.98%	0.00%
Endophthalmitis	0.40%	0.00%	0.40%	0.00%

IOP, intraocular pressure.

Fellow eye involvement

Fellow eye involvement (FEI) was assumed not to impact on the effectiveness of treatment in the model, as there are no data to indicate the relationship between the outcomes in each individual eye. The model was, therefore, built to capture only the cost implications of FEI (i.e. the occurrence of RVO in both of the patient’s eyes), which is a conservative assumption. When RVO occurred in the fellow eye, the cost of treatment was effectively doubled for those patients (because two eyes would be treated), although the cost of administration and follow-up was assumed to remain the same, since both eyes could be assessed at the same visit; however, bilateral involvement is often asynchronous.

The rate of FEI was assumed to be 6.5% at 12 months and 10% at 6 years. After 6 years, the rate for FEI was assumed to remain stable at 10%. These assumptions were previously used in a health technology assessment submission to NICE and were deemed by NICE to be plausible24.

Utilities

It is generally understood that quality-of-life in people with visual impairment is associated primarily with BCVA in the BSE. Utility values are usually considered to be lower in patients whose affected eye is the BSE than in those whose affected eye is the WSE, while utility gains from improving BCVA would generally be higher among patients treated in the BSE than in those treated in the WSE, although this has not been thoroughly investigated. It was, therefore, important to choose utilities that were specific to BSE or WSE25. Utilities by BCVA letter score were built into the model based on the algorithm developed by Czoski-Murray et al.26 and were used as the base-case for the BSE in the model, as these utilities adjust for age and reflect the societal value of visual impairment. In the absence of appropriate data for the WSE, it was assumed that patients could experience a maximum gain in utility score of 0.30 between the best and worst possible health states (this compares with 0.52 in the Czoski-Murray et al. algorithm for the BSE). These utility scores are shown in Table 3. Disutilities were applied to each patient experiencing an adverse event in the model, and were calculated as one-off events per patient, weighted by the expected duration of the quality-of-life impact.

Table 3. Utility scores by health state.

Health state	Utility score26	Utility score, assuming 0.30 overall gain
BCVA letter score 86–100	0.87	0.92
BCVA letter score 76–85	0.76	0.88
BCVA letter score 66–75	0.65	0.83
BCVA letter score 56–65	0.61	0.79
BCVA letter score 46–55	0.54	0.75
BCVA letter score 36–45	0.46	0.71
BCVA letter score 26–35	0.39	0.66
BCVA letter score <25	0.35	0.62
Death	0.00	0.00
Adverse events	Disutility	Duration
Cataract	−0.14	6 months
Increased IOP (treated with drug)	−0.01	1 day
Increased IOP (treated with surgery)	−0.01	6 months
Iris neovascularization	0.00	N/A
Endophthalmitis	−0.30	3 months

BCVA, best-corrected visual acuity; IOP, intraocular pressure; N/A, not applicable.

Resource use

The frequency of ranibizumab treatment during year 1 was taken from BRAVO and CRUISE directly (Table 4). This may be a conservative assumption given that patients were treated with a fixed 6-month regimen at the start of the trial, followed by as-needed dosing (this is different from the posology in the Committee for Medicinal Products for Human Use label). Year 2 rates for ranibizumab treatment were drawn from the HORIZON study, after excluding patients who did not enter the extension study. From year 3 onwards it was assumed that no injections would be required because it was expected that patients’ BCVA would have stabilized by then based on advice received from experts. It was assumed that patients would be followed-up monthly during year 1, and that most patients would attend bi-monthly monitoring during year 2.

Table 4. Frequency and cost of treatment and follow-up.

	BRVO		CRVO
	Ranibizumab	Grid laser photocoagulation	Ranibizumab	Observation
Year 1
Treatment visits	8.015	1.5^a	9.016	0.0^b
Follow-up visits	4.0^c	2.5^d	3.0^c	6.0^e
Year 2
Treatment visits	1.917	1.0^a	3.017	0.0^b
Follow-up visits	4.1^e	3.0^d	7.0^f	4.0^d
Year 3+
Treatment visits	0.0^b	0.0^b	0.0^b	0.0^b
Follow-up visits	4.0^d	4.0^d	4.0^d	4.0^d
Cost per treatment	£742.1733	£0^g	£742.1733	£0
Cost of administration	£192.00^h	£192.00^h	£192.00^h	£0
Adverse events
Cataract		£800.002
Increased IOP (treated with drug)		£31.6724
Increased IOP (treated with surgery)		£872.6324
Stroke		£10,28134
Iris neovascularization		£576 Three sessions of pan-retinal photocoagulation27
Endophthalmitis		£207027

^aAssumption6.

^bAssumption.

^cAssumption (based on a total of 12 visits of any type per year).

^dAssumption (based on a total of four visits of any type per year).

^eAssumption (based on a total of six visits of any type per year).

^fAssumption (based on a total of 10 visits of any type per year).

^gNo marginal depreciation cost was assumed for each time the equipment was used—only an administration cost was applied.

^hAccumulated cost of optical coherence tomography (ultrasound scan less than 20 min)27 and cost of injection visit (BZ23Z: outpatient procedures [office based])29.

BRVO, branch retinal vein occlusion; CRVO, central retinal vein occlusion; IOP, intraocular pressure.

For grid laser photocoagulation in patients with BRVO, treatment frequency was taken from the SCORE trial for years 1 and 26, because patients in BRAVO were switched to ranibizumab treatment after 6 months and this would have reduced their use of laser treatment15. No laser treatment was assumed from year 3 onwards. In addition, patients were assumed to have quarterly visits from year 1 onwards. A similar pattern of follow-up was assumed for observation in patients with CRVO, although in some countries CRVO patients may be followed more closely to assess the development of rubeosis2.

Costs

The level of resource use previously described was multiplied by unit costs in order to derive total costs. Unit costs were obtained from published UK sources in order to calculate the cost per intervention, for each of the treatments (Table 4)27. Patients experiencing adverse events were assumed to incur the cost of managing those events. These NHS reference costs included £800.00 for the treatment of cataract, £31.67 for drug-treated intra-ocular pressure (IOP), and £872.63 for surgery (Table 4)27.

A systematic review of published literature found no evidence of a link between disease severity and resource use in RVO. However, patients considered to be blind (BCVA letter score below 6/60 [ETDRS letters score lower than 35] in the BSE) have been demonstrated to incur significant costs. The costs associated with blindness were drawn from a previously published study28 and were inflated to 2011 prices29. The above-mentioned costs for blindness were applied when the patient is affected in the BSE and the BCVA letter score falls below 35.

Sensitivity analysis

One-way sensitivity analyses were performed when there was uncertainty about key parameter values. This enabled identification of the impact of different parameters on the costs and effects of treatment alternatives. These parameters are described in Table 5. In addition, a probabilistic sensitivity analysis (PSA) was performed. To do this, distributions were fitted to key parameters within the model (Table 5). For probabilities, beta-distributions were used, while cost parameters were fitted with gamma-distributions (beta-distributions are bound between the values of 0 and 1, whereas gamma-distributions produce only non-negative values). Normal distributions were assumed for parameters such as age, for which non-negative values were very unlikely, and log-normal distributions were used for risk ratios. For parameters such as the transition probability rates, the beta-distribution was determined by the trial results. Specifically, the α-input for each probability was represented by the number of patients who moved health state, while the β-value was determined by the number who did not move to that state.

Table 5. Parameter values for probabilistic sensitivity analyses.

Parameter	Mean	Variation	Type	α	β	Notes
General
Starting age, years	66.8	5	Normal	N/A	N/A	BRAVO and CRUISE
Treatment in the BSE at baseline, %	10	N/A	β	52	470	BRAVO and CRUISE
Treatment in the BSE at 12 months, %	21.5	N/A	β	112	410	BRAVO and CRUISE
Effectiveness multipliers
Treatment effectiveness probabilities: month 1	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Treatment effectiveness probabilities: months 2–6	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Treatment effectiveness probabilities: months 7–12	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Comparator effectiveness probabilities: month 1	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Comparator effectiveness probabilities: months 2–6	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Comparator effectiveness probabilities: months 7–12	1	0.1	Log-normal	N/A	N/A	Multiplier/assumption
Natural deterioration per monthly cycle	0.031%	N/A	β	3.1	9996.9	Assumes n = 10,000
Mortality RR (BCVA 36–55 letters in the BSE, <36 letters in the WSE)	1.23	0.1	Log-normal	N/A	N/A	Assumption
Mortality RR (BCVA <36 letters in the BSE)	1.54	0.1	Log-normal	N/A	N/A	Assumption
Quality-of-life
Utility multiplier (all)	1	0.05	Normal	N/A	N/A	Assumption
Disutility for cataract	−0.14	0.0284	Normal	N/A	N/A	Assumption (±20%)
Disutility for increased IOP (treated with drug)	−0.01	0.0026	Normal	N/A	N/A	Assumption (±20%)
Disutility for increased IOP (treated with surgery)	−0.01	0.002	Normal	N/A	N/A	Assumption (±20%)
Disutility for stroke	−0.27	0.052	Normal	N/A	N/A	Assumption (±20%)
Costs
Administration costs (multiplier)	1.000	0.200	γ	25	0.04	Assumption (±20%)
Follow-up costs (all treatments) (multiplier)	1.000	0.200	γ	25	0.04	Assumption (±20%)
Treatment visits year 1 (ranibizumab) BRVO	8.0	0.8	γ	100	0.08	Assumption (±10%)
Treatment visits year 2 (ranibizumab) BRVO	1.9	0.19	γ	100	0.025	Assumption (±10%)
Treatment visits year 1 (laser)	1.5	0.15	γ	100	0.02	Assumption (±10%)
Treatment visits year 2 (laser)	1.0	0.1	γ	100	0.01	Assumption (±10%)
Treatment visits year 1 (ranibizumab) CRVO	9.0	0.9	γ	100	0.09	Assumption (±10%)
Treatment visits year 2 (ranibizumab) CRVO	3.0	0.30	γ	100	0.038	Assumption (±10%)
Cost of cataract	£800.00	160	γ	25	32	Assumption (±20%)
Cost of increased IOP (treated with drug)	£31.67	6.33	γ	25	1.267	Assumption (±20%)
Cost of increased IOP (treated with surgery)	£872.63	174.5	γ	25	34.905	Assumption (±20%)
Cost of blindness	£6067.93	1213.59	γ	25	0.04	Assumption (±20%)
Adverse event rates
Cataracts (ranibizumab; year 1/year 2)	6.60%/5.42%	N/A	β	34	488	BRAVO and CRUISE
Iris neovascularization (ranibizumab; year 1/year 2)	2.35%/0.98%	N/A	β	52	470	BRAVO and CRUISE

BSE, best-seeing eye; BRVO, branch retinal vein occlusion; CRVO, central retinal vein occlusion; IOP, intraocular pressure; N/A, not applicable; RR, risk ratio.

Results

In BRVO, ranibizumab was shown to produce a gain of 0.518 quality-adjusted life years (QALYs) at an incremental cost (including drug acquisition, administration, monitoring and the cost of blindness) of £8141 when compared with grid laser photocoagulation over a lifetime horizon (Table 6). The incremental cost-effectiveness ratio (ICER) was, therefore, £15,710 per QALY. This result is below the £20,000–30,000 range, which has typically been regarded as a threshold for cost-effectiveness in the UK. In addition, the incremental cost per month free of blindness was £658.

Table 6. Base case cost-effectiveness results.

	Ranibizumab	Observation	Grid laser photocoagulation	Incremental
BRVO
Total costs	£17,908	£9767		£8141
QALYs	8.516	7.998		0.518
Months without blindness	194	181		12
Incremental cost per QALY				£15,710
Incremental cost per month free of blindness				£658
CRVO
Total costs	£20,646		£11,430	£9216
QALYs	7.383		6.844	0.539
Months without blindness	171		150	22
Incremental cost per QALY				£17,103
Incremental cost per month free from blindness				£423

BRVO, branch retinal vein occlusion; CRVO, central retinal vein occlusion; QALY, quality-adjusted life-year.

In CRVO, ranibizumab was shown to lead to a gain of 0.539 QALYs at an incremental cost of £9216 when compared with best supportive care (observation). The ICER was, therefore, £17,103 per QALY. The incremental cost per month free from blindness was £423 (Table 6). The results of the one-way sensitivity analysis are shown in Figure 2, whilst the probabilistic scatter plots and cost-effectiveness acceptability curves (CEACs) are shown in Figure 3.

Figure 2. Sensitivity analysis tornado diagrams for (a) BRVO and (b) CRVO.

Figure 3. Probabilistic scatter planes and cost-effectiveness acceptability curves for (a) BRVO and (b) CRVO.

As shown in the one-way sensitivity analysis, when the rate of natural deterioration of BCVA was increased, or when the overall mortality rate increased, the treatment became less cost-effective. This is because the ‘capacity to benefit’ (through duration of benefit) is reduced and any benefits of treatment are lost sooner. The model is relatively sensitive to the cost associated with blindness (i.e. when the patient’s BCVA in the BSE falls below 35 letters). Again, this is to be expected, because one of the key benefits of effective treatment is to reduce the burden of patients whose sight becomes significantly affected by the disease.

The PSA allows an overall assessment of the uncertainty within the model; it generates a ‘scatter plot’, which has a ‘downward-sloping’ pattern. This represents the likely inverse relationship between incremental cost and incremental effectiveness. For example, if the PSA distributions randomly generate a ‘favorable’ iteration (i.e., when treatment is much more effective than the comparator), this is likely to have two effects. First, the patient will gain QALYs through improved quality-of-life and, to a much lesser extent, through increased survival. Second, the patient will be likely to avoid costly complications such as blindness and will, therefore, see their incremental cost reduced. Similarly, a ‘less favorable’ iteration will see reduced QALYs and increased long-term costs. The PSA found that, when there is a willingness-to-pay threshold of £20,000 for each QALY gained, the probability of ranibizumab being cost-effective is 58.8% and 57.6% in BRVO and CRVO, respectively (Figure 3). At a willingness-to-pay threshold of £30,000 per QALY gained, the probability of ranibizumab being cost-effective is 69.0% for BRVO and 68.3% for CRVO.

Discussion

A key strength of economic models is that they allow the decision-maker to combine evidence from a variety of sources in order to derive a single assessment of the impact of an intervention upon utilities, survival, and overall treatment costs.

The model used in this analysis was based mainly on data derived from the BRAVO, CRUISE, and HORIZON (Cohort 2) trials. This allows for a robust comparison of ranibizumab vs grid laser photocoagulation (for BRVO) and observation (for CRVO). The results confirm that ranibizumab is a cost-effective treatment for patients with MO secondary to RVO relative to the current standard of care for BRVO and to observation for CRVO. There remains some uncertainty over the long-term comparative impact of treatment over the comparator, because all patients in BRAVO and CRUISE were able to receive ranibizumab after 6 months; future clinical trials are needed to evaluate this.

In this analysis, conservative assumptions were made for months 7–12 and in the second year of the model, assuming that transition probabilities were identical for ranibizumab and the standard of care arm for BRVO. For CRVO, the probabilities for observation were assumed to be the same as for months 2–6 during the rest of the first year, and stability was assumed for the second year. Given the significantly better outcomes observed for ranibizumab than control in the initial 6 months of the trials, this would seem to be a conservative approach, although it is also acknowledged that the potential delayed onset of effectiveness associated with grid laser photocoagulation could partially offset this. The model also assumes that patients achieve stability in visual acuity by the end of year 2 and, therefore, that no injections are required from year 3 onwards. The HORIZON trial reported a slight decline in visual acuity in patients with CRVO associated with fewer injections in the second year17; however, additional long-term clinical data will be needed to determine the long-term trajectory for visual acuity in patients treated with ranibizumab for MO secondary to RVO. When these become available, they may be incorporated into a future model.

A key limitation of the analysis is the lack of accurate data for utilities associated with the severity of visual impairment in the WSE in patients with RVO. While several studies have estimated utilities in patients based on the BCVA letter score in their BSE, there is very little evidence related to utilities based on BCVA in the WSE of patients. It is worth noting that the absolute value of utility does not drive the results of the model, but rather the incremental QALYs gained are largely driven by the slope of the utility curve. It is difficult to quantify the benefit of improving the BCVA in a patient’s WSE, and, therefore, it seems appropriate that future studies aim to distinguish the impact on utility of BCVA in the BSE and WSE.

Ideally, both of the patients’ eyes would be modeled simultaneously using a patient-level simulation to establish the BCVA in each eye. Although the computational burden of this approach is manageable, the evidence available to populate such a model is sparse, especially in RVO. Such a model would require detailed data on the complex relationship between BCVA in each eye, which was beyond the scope of this paper.

Dexamethasone intravitreal implants are also approved for the treatment of patients with MO secondary to RVO30,31. While laser photocoagulation (for BRVO) and observation (for CRVO) are the most appropriate comparators for ranibizumab, it would be desirable to also compare the cost-effectiveness of the dexamethasone implant with ranibizumab. However, there is a lack of direct evidence comparing ranibizumab and the dexamethasone implant for the treatment of BRVO and CRVO, preventing comparison using standard methods for the outcomes needed for this Markov model. Moreover, the studies and enrolled populations are too heterogeneous to permit an unbiased indirect comparison32. Further studies with direct comparisons between ranibizumab and the dexamethasone implant, or sophisticated and robust indirect treatment comparisons, would be needed before the cost-effectiveness of these two therapies can be evaluated robustly and compared.

Conclusion

In summary, the existing clinical and economic evidence suggest that ranibizumab is a cost-effective treatment for patients with MO secondary to RVO relative to grid laser photocoagulation (for BRVO) and to observation (for CRVO). The present study demonstrates that the significant gains in BCVA with ranibizumab lead to significant benefits for patients, while treatment costs are to a large extent compensated by the avoidance of blindness and visual impairment in a significant proportion of patients.

Transparency

Declaration of funding

This study was funded by Novartis Pharma AG.

Declaration of financial/other relationships

Matthew Taylor, Elçin Serbetci, and Lily Lewis are employees of York Health Economics Consortium Ltd. Alberto Ferreira is an employee of Novartis and owns shares in the company. Kerry Gairy was an employee of Novartis Pharmaceuticals UK Limited, Surrey, UK at the time the study was conducted. Julie Blouin was an employee of Novartis Pharmaceuticals Canada Inc. Dorval, Quebec, Canada at the time the study was conducted. Paul Mitchell is an academic and has received honoraria and travel support from Novartis and Bayer as a consultant. JME Peer Reviewers on this manuscript have no relevant financial relationships to disclose.

Acknowledgments

This manuscript has been prepared with editorial assistance from Dr Martin Bell of Oxford PharmaGenesis™ Ltd, Oxford, UK, funded by Novartis.