Cost-effectiveness of ranibizumab and aflibercept to treat diabetic macular edema from a US perspective: analysis of 2-year Protocol T data

Abstract

Aims: Protocol T (NCT01627249) was a head-to-head study conducted by the Diabetic Retinopathy Clinical Research Network that compared intravitreal aflibercept, bevacizumab, and ranibizumab for the treatment of diabetic macular edema (DME). A cost-effectiveness analysis accompanying the 1-year data of Protocol T revealed that aflibercept was not cost-effective vs ranibizumab for all patients, but could have been cost-effective in certain patient sub-groups if the 1-year results were extrapolated out to 10 years. The present study evaluated the cost-effectiveness of US Food and Drug Administration-approved anti-vascular endothelial growth factor agents (ranibizumab, aflibercept) for treatment of DME using the 2-year data from Protocol T.

Methods: Costs of aflibercept 2.0 mg or ranibizumab 0.3 mg, visual acuity (VA)-related medical costs, and quality-adjusted life-years (QALYs) were simulated for eight VA health states. Treatment, adverse event management, and VA-related healthcare resource costs (2016 US dollars) were based on Medicare reimbursement and published literature. VA-related health utilities were determined using a published algorithm. Patients were stratified by baseline VA: 20/40 or better; 20/50 or worse.

Results: Total 2-year costs were higher, and QALYs similar, for aflibercept vs ranibizumab in the full cohort ($44,423 vs $34,529; 1.476 vs 1.466), 20/40 or better VA sub-group ($40,854 vs $31,897; 1.517 vs 1.519), and 20/50 or worse VA sub-group ($48,214 vs $37,246; 1.433 vs 1.412), respectively. Incremental cost-effectiveness ratios in the full cohort and 20/50 or worse VA sub-group were $986,159/QALY and $523,377/QALY, respectively. These decreased to $711,301 and $246,978 when analyses were extrapolated to 10 years.

Limitations: Key potential limitations include the fact that VA was the only QALY parameter analyzed and the uncertainty surrounding the role of better- and worse-seeing eye VA in overall functional impairment.

Conclusions: This analysis suggests that aflibercept is not cost-effective vs ranibizumab for patients with DME, regardless of baseline vision.

Keywords:

• Anti-vascular endothelial growth factor
• aflibercept
• cost-effectiveness analysis
• diabetic macular edema
• health-related quality of life
• ranibizumab

JEL classification codes:

• I11
• I19

Introduction

Diabetic retinopathy (DR) is a leading cause of vision impairment in adults with diabetes, affecting nearly 8 million people in the US alone¹. Diabetic macular edema (DME) is an advanced vision-threatening complication of DR². As the diabetes population increases, vision loss from DME has become an increasing socioeconomic concern³.

Vascular endothelial growth factor (VEGF) plays a critical role in DME⁴, and VEGF inhibition can slow disease progression and improve vision. Two anti-VEGF therapies (Lucentis [ranibizumab], Genentech, Inc., South San Francisco, CA, and Eylea [aflibercept], Regeneron Pharmaceuticals, Inc., Tarrytown, NY) are approved by the US Food and Drug Administration (FDA) to treat DME. Although bevacizumab has been used off label for DME, it is not FDA approved for any ophthalmic condition⁵. Significant visual acuity (VA) improvements have been observed with all three anti-VEGF agents^6–9.

The Diabetic Retinopathy Clinical Research Network (DRCR.net) Protocol T study (NCT01627249) compared intravitreal aflibercept, bevacizumab, and ranibizumab in eyes with DME^10,11. After 1 year, vision improved with all three agents, but the relative effect was dependent on baseline VA. In patients with mild vision loss (20/40 or better), vision gains were comparable across treatments. For patients with baseline VA of 20/50 or worse, vision gains were greatest with aflibercept¹⁰. At year 2, aflibercept superiority over bevacizumab was retained, but aflibercept superiority vs ranibizumab was no longer statistically significant in 20/50 or worse-seeing eyes¹¹.

Generally, for conditions that affect vision, efficacy is measured by improvements in VA, with less focus on how changes in VA affect health-related quality of life (HRQoL). However, cost-effectiveness of treatments and their impact on HRQoL are also of critical importance when considering an appropriate treatment for DME.

It is important to understand the most cost-effective use of FDA-approved anti-VEGF agents for DME, considering the high injection frequency in the first few years of treatment. For example, the median number of injections in Protocol T was nine for aflibercept and 10 for ranibizumab in the first year¹⁰ and 15 for both drugs over 2 years¹¹. A cost-effectiveness analysis (CEA) accompanying the 1-year data of Protocol T revealed that aflibercept was not cost-effective vs ranibizumab for all patients using a model with a 1-year time horizon, but was cost-effective in patients with baseline VA of 20/50 or worse using a 10-year time horizon¹².

We conducted a CEA using 2-year VA data from DRCR.net Protocol T¹¹. The focus of this analysis was a US payer perspective, and comparators were limited to the FDA-approved anti-VEGF agents ranibizumab and aflibercept.

Methods

Protocol T study design and treatment algorithms

The Protocol T (NCT01627249) study design has been described previously^10,11 and is summarized briefly below. The protocol adhered to the tenets of the Declaration of Helsinki and was approved by local institutional review boards or a central institutional review board if the site did not have a local board.

Adults ≥18 years of age with diabetes, center-involved DME, and best-corrected VA (BCVA) letter score of 78–24 (approximate Snellen equivalent, 20/32–20/320) were randomized to aflibercept 2.0 mg, ranibizumab 0.3 mg, or bevacizumab 1.25 mg¹⁰. Follow-up was every 4 weeks during year 1, and every 4–16 weeks during year 2. Study drugs were administered at baseline and as frequently as every 4 weeks based on BCVA and optical coherence tomography criteria¹¹. Starting at 24 weeks, an injection could be withheld if specific criteria were met. At 24 weeks, patients could be considered for laser photocoagulation therapy based on protocol-defined criteria¹⁰.

CEA design

A decision analytic model (Figure 1) was developed using 2-year VA (Snellen) data from the DRCR.net Protocol T study^10,11 to assess economic and HRQoL impacts of DME-related vision impairment. The model, in Microsoft Excel (Microsoft Corporation, Redmond, WA), included eight mutually exclusive health states defined by VA for treated and fellow eyes separately (Figure 1). Distribution of patients in each health state varied monthly based on treatment efficacy (in study and treated fellow eyes) or natural history (untreated fellow eyes). Health state- and treatment-specific costs and health utilities were assigned for each month.

Figure 1. Model schematic: eight VA health states with the five most common 1-year transitions in Protocol T for the (a) study eye (solid arrows) and (b) untreated fellow eye (dashed arrows). Abbreviation. VA, visual acuity.

Final treatment algorithms and anti-VEGF medical resource use were based on published Protocol T results^10,11, or assumptions when published estimates were unavailable. The model simulated VA outcomes for two baseline VA sub-groups: (a) 20/40 or better and (b) 20/50 or worse. Model VA outcomes were based on patient-level data at baseline, year 1 and year 2 for study and fellow eyes.

Costs

The US payer perspective was the focus of the base case analysis, so we have primarily modeled direct medical costs. However, we did conduct sensitivity analyses that took a societal perspective in order to capture some indirect costs, which included caregiver and nursing home costs. Non-monetary or intangible costs (e.g. pain and suffering associated with a disease) were included only to the extent that they may be captured in the VA health state utility estimates.

Costs (2016 US dollars) for anti-VEGF treatment, adverse event (AE) management, and VA-related resources were based on Medicare reimbursement and the literature (Table 1).

Table 1. Cost and health utility model parameters.

Parameter	Base case estimate	Range^a	References
Treatment-related costs
Ranibizumab product (0.3 mg)^b	$1,170/vial	$936–1,404	• Administration and laser costs based on Medicare reimbursement; product costs based on WAC
Aflibercept product (2.0 mg)^c	$1,850/vial	$1,480–2,220
Intravitreal administration (CPT 67028)	$103	—
Laser treatment (CPT 67210)	$525	$420–630
AE probabilities (2 years) and 1-time costs
Ranibizumab-related AEs			• Probabilities of AEs based on rates of pre-specified AEs in a recently published study^11 • Medical resources used to treat/manage AEs based on clinical expert opinion • Resource costs based on Medicare reimbursement
Traction retinal detachment	<1%/$2,093	—
Vitreous hemorrhage	5%/$3,377	—
Increased intraocular pressure	16%/$1,130	—
Cataract	0%/$3,125	—
Ocular inflammation	2%/$63	—
Non-fatal myocardial infarction	3%/$57,213 (year 1); $20,873 (year 2)	—
Non-fatal stroke	5%/$47,972 (year 1); $24,411 (year 2)	—
Vascular death	4%/$5,825	—
Total expected ranibizumab AE costs	$1,986 (year 1); $3,246 (year 2)	$1,589–2,383 (year 1); $2,597–3,895 (year 2)
Aflibercept-related AEs
Traction retinal detachment	1%/$2,093	—
Vitreous hemorrhage	7%/$3,377	—
Increased intraocular pressure	17%/$1,130	—
Cataract	1%/$3,125	—
Ocular inflammation	3%/$63	—
Non-fatal myocardial infarction	3%/$57,213 (year 1); $20,873 (year 2)	—
Non-fatal stroke	<1%/$47,972 (year 1); $24,411 (year 2)	—
Vascular death	1%/$5,825	—
Total expected aflibercept AE costs	$1,326 (year 1); $1,784 (year 2)	$1,061–1,591 (year 1); $1,427–2,141 (year 2)
Annual incremental expected medical costs relative to normal vision
Depression			• Estimates from published literature^13^,^14 were inflated to 2016 USD using the Medical Care component of the Consumer Price Index^15
Health states 1–3 (20/40 or better)	$0	—
Health states 4–6 (>20/40–20/80)	$864	$681–1,037
Health state 7 (>20/80–20/200)	$1,204	$963–1,445
Health state 8 (worse than 20/200)	$1,366	$1,093–1,639
Fractures
Health states 1–3 (20/40 or better)	$0	—
Health states 4–6 (>20/40–20/80)	$921	$737–1,105
Health state 7 (>20/80–20/200)	$1,022	$818–1,226
Health state 8 (worse than 20/200)	$1,253	$1,002–1,504
Skilled nursing facility
Health states 1–3 (20/40 or better)	$0	—
Health states 4–6 (>20/40–20/80)	$824	$659–989
Health state 7 (>20/80–20/200)	$1,074	$859–1,289
Health state 8 (20/200 or worse)	$1,552	$1,242–1,862
Nursing home
Health states 1–3 (20/40 or better)	$0	—
Health states 4–6 (>20/40–20/80)	$6,121	$4,897–7,345
Health state 7 (>20/80–20/200)	$7,472	$5,978–8,966
Health state 8 (worse than 20/200)	$11,061	$8,849–13,273
Annual expected caregiver costs associated with VA
Health states 1–2 (better than 20/32)	$372	$298–446	• Caregiver time estimates from published literature^18 were monetized in 2016 USD using hourly wage data^19
Health states 3–4 (20/32 to >20/50)	$2,241	$1,793–2,689
Health states 5–6 (20/50 to >20/80)	$5,882	$4,706–7,058
Health state 7 (20/80 to >20/200)	$23,325	$18,660–27,990
Health state 8 (20/200 or worse)	$77,370	$61,896–92,844
VA health state utilities
Health state 1 (20/25 or better)	0.799	0.719–0.879	• Czoski-Murray et al.^16
Health state 2 (>20/25–20/32)	0.744	0.670–0.818
Health state 3 (>20/32–20/40)	0.707	0.636–0.778
Health state 4 (>20/40–20/50)	0.670	0.603–0.737
Health state 5 (>20/50–20/63)	0.633	0.570–0.696
Health state 6 (>20/63–20/80)	0.597	0.537–0.657
Health state 7 (>20/80–20/200)	0.505	0.455–0.556
Health state 8 (worse than 20/200)	0.321	0.289–0.353

Costs are expressed in 2016 USD.

^a Minimum and maximum of range used in one-way sensitivity analyses.

^b Lucentis, Genentech, Inc., South San Francisco, CA.

^c Eylea, Regeneron Pharmaceuticals, Inc., Tarrytown, NY.

Abbreviations. AE, adverse event; CPT, Current Procedural Terminology; USD, US dollars; VA, visual acuity; WAC, wholesale acquisition costs.

Wholesale acquisition costs (WAC) for medications and 2016 Medicare reimbursement rates were used to calculate treatment costs (combined drug, fees for administration, procedure costs) and costs associated with AEs (procedures, physician services, inpatient admissions). WAC price was used instead of average sales price because the latter fluctuates by quarter and is similar to WAC price. WAC for both aflibercept and ranibizumab has remained unchanged since 2016. Costs for ongoing clinical management were assumed to be similar across both treatments and were not included.

Ranibizumab and aflibercept AE rates included in the model were estimated based on rates of pre-specified AEs from Protocol T¹¹ (Table 1). AEs associated with adjunctive laser treatment were infrequent and not considered in the analysis. AE management algorithms were based on consultation with a retina specialist. Healthcare resources, including diagnostic tests, therapeutic procedures, or medications needed to manage ocular AEs, were quantified. For non-ocular AEs (myocardial infarction [MI], stroke, vascular death), management included inpatient admission.

Visual impairment costs were limited to those associated with depression, fractures, and skilled nursing facility use to reflect a US payer perspective in the base case analyses (Table 1)^13,14. These costs were inflated to 2016 US dollars using the Medical Care component of the Consumer Price Index¹⁵.

Health utilities

The quality-adjusted life-year (QALY) is used to evaluate HRQoL. A published algorithm¹⁶ was used to translate patient VA data into utilities, which approximate HRQoL. Utility scores and duration of life are used to calculate QALYs. Utilities were based on those determined by the time tradeoff technique in patients with simulated vision impairment¹⁶. The utility equation developed by Czoski-Murray et al.¹⁶ includes a constant term as well as parameters for VA (defined by logarithm of the minimum angle of resolution) and age, allowing for more granular utility calculation compared with approaches using more rigid VA strata or not considering the impact of age on utility. A mean age of 61 years (from Protocol T) and VA in both better- and worse-seeing eyes were used to estimate overall utility for each patient.

Impact of better- and worse-seeing eye VA

Overall visual function and HRQoL in patients with eye diseases likely depend on VA in both better- and worse-seeing eyes. The degree to which each eye impacts daily activities, HRQoL, and risk of events such as falls is under debate¹⁷. However, ignoring the effect of poor VA in the worse-seeing eye or under-estimating the benefits of very good VA in the better-seeing eye likely results in inaccurate assessments of overall functionality and HRQoL.

A recent literature review concluded that the impact of VA in the worse-seeing eye is more important than previously thought¹⁷. However, the magnitude of the effect may vary depending on the eye disease and whether central vs peripheral vision is affected. Therefore, our model incorporated the effect of the better- and worse-seeing eye through a weighting function (based on clinical expert opinion). For example, VA-related costs and utilities were weighted 75%:25% (better:worse) to calculate an overall VA-related cost and utility for each patient. The impact of this approach was assessed in sensitivity analyses by varying the weighting function.

Analyses

Base case CEAs

For each patient, a monthly VA health state was estimated for study and fellow eyes based on patient-specific baseline and 1-year/2-year VA as described above. Vision-related costs were assigned to each VA health state (Table 1) and summed over the 2-year time horizon for each patient after applying the weighting function^{11,13–16,18,19}. Similarly, QALYs for each patient were calculated from VA health state-specific utilities after incorporating better- and worse-seeing eye VA with weighting function. Treatment costs were assigned based on the number of injections and laser treatments for each treatment. Bilateral treatment initiated at baseline was assumed to require twice the usual injections per patient, unless bilateral treatment was initiated at some time after baseline, in which case it was assumed to require 1.5 times the number of injections. We were not able to discern the precise time when bilateral treatment began (if it did not begin at baseline), so we estimated that the average time was near mid-study. Consequently, we estimated that patients receiving bilateral treatments received 1.5 times the average number of injections of those treated unilaterally throughout the study. Total costs and QALYs for each comparator were obtained by averaging each of these outcomes for all patients in each treatment sub-group. Costs and QALYs in future years (years 2+ in base case and extrapolated analyses) were discounted to net present value using a 3% annual rate²⁰.

While different willingness-to-pay thresholds exist for CEAs, a drug is often considered cost-effective if the incremental cost-effectiveness ratio (ICER) is <$150,000/QALY²¹. ICER for cost-effectiveness of aflibercept vs ranibizumab was calculated by dividing the difference in total cost for each drug by the difference in QALYs over the 2-year analysis period.

Sensitivity analyses

To assess robustness of the results, parameter estimates were varied in one-way and multi-way sensitivity analyses. In addition, extrapolation of results to 10 years and probabilistic sensitivity analyses (PSAs) were performed. Extrapolation and PSAs were conducted according to an approach described by Ross et al.¹². To reflect a societal perspective, a sensitivity analysis including nursing home^13,14 and caregiver costs¹⁸ was performed. Finally, a multi-way analysis was conducted, varying the better-seeing eye:worse-seeing eye weighting function. This analysis affects both VA-related costs as well as QALYs.

Extrapolation analysis

To extrapolate our cost-effectiveness results beyond 2 years, we adapted the approach used by Ross et al.¹². As was done by Ross et al., we accounted for the likely decrease in the number of injections beyond year 2 by using the 5-year data from the Protocol I study²². This may have resulted in an under-estimation of the number of injections required over the longer term because 44% of eyes in the ranibizumab plus deferred laser group in Protocol I (there was no ranibizumab monotherapy group) received at least one session of focal/grid laser treatment. It is unclear how laser treatment may have affected the ability of an eye to meet re-treatment criteria for ranibizumab. Other parameter estimates were updated from the Ross et al. model, if necessary, as described below.

Survival was projected based on 2011 US life tables published by the Centers for Disease Control and Prevention²³, and further adjusted to account for higher mortality in patients with diabetes²⁴. Patients could suffer from MI and/or stroke based on event incidence in Protocol T¹¹; one-time and ongoing annual disutilities²⁵ were applied to patients with these events. Long-term MI and stroke costs were estimated using published results of a large medical claims analysis²⁶ and inflated to 2016 US dollars using the Medical Care component of the Consumer Price Index¹⁵. We calculated an annual nursing home cost as the product of the percentage of patients with stroke discharged to long-term care (16.4%)²⁷, and the average daily cost of nursing home care ($235)²⁸. The annual result ($14,067) was used in extrapolation analysis scenarios. Costs, life-years, and QALYs were quantified up to a time horizon of 10 years.

Bootstrap probabilistic analysis

To evaluate uncertainty in parameter estimates, a PSA was conducted. Utility and cost parameters were included as distributions, and sampled treatment groups of equal size to Protocol T were created via random draw with replacement. Parameter values were drawn from their respective distributions for 5,000 simulations, with total costs, QALYs, and ICERs quantified for each simulation. The probability of either treatment alternative being deemed to be cost-effective at specific willingness-to-pay thresholds was determined.

Results

Protocol T patient population and VA outcomes

Of the 414 patients who completed year 1 in Protocol T, 191 ranibizumab- and 200 aflibercept-treated patients had baseline, 1-year, and 2-year VA data and were included in the current analysis. Of these, 97 and 103 patients treated with ranibizumab or aflibercept, respectively, had baseline VA of 20/32–20/40, and 94 and 97 patients treated with ranibizumab or aflibercept, respectively, had baseline VA of 20/50 or worse. For patients who completed year 2, the median number of intravitreal injections over 2 years was 15 in both the aflibercept and ranibizumab groups (interquartile range [IQR] = 11–17 and 11–19, respectively; global p = 0.08) and 5 (IQR = 2–7) and 6 (IQR = 2–9) injections during year 2, respectively¹¹.

Base case cost and QALYs

Full cohort

Total 2-year costs for aflibercept 2.0 mg in the full cohort exceeded ranibizumab 0.3 mg treatment costs by nearly $10,000 ($44,423 vs $34,529, respectively; Table 2). Approximately 86% and 74% of total costs were from drug costs for aflibercept and ranibizumab, respectively. Nominal cost differences between treatments were observed for product administration and adjunctive laser therapy. Aflibercept yielded a marginally greater number of QALYs than ranibizumab 0.3 mg (1.476 vs 1.466, respectively).

Table 2. Base case and extrapolated 10-year cost, QALY, and ICER results.

	Base case			Extrapolated 10-year costs
	AFL 2.0 mg	RBZ 0.3 mg	Difference (AFL – RBZ)	AFL 2.0 mg	RBZ 0.3 mg	Difference (AFL – RBZ)
Full cohort
Treatment costs	$43,693	$33,645	$10,049	$95,458	$74,675	$20,784
Product	$38,069	$25,661	$12,407	$67,275	$44,132	$23,143
Administration	$2,120	$2,259	–$140	$3,746	$3,885	–$140
Laser procedure	$394	$493	–$98	$541	$640	–$98
AEs	$3,111	$5,232	–$2,121	$23,897^b	$26,018^b	–$2,121^b
Visual impairment/blindness costs^a	$730	$885	–$155	$908	$1,084	–$176
Total costs	$44,423	$34,529	$9,894	$96,367	$75,759	$20,608
Total QALYs	1.476	1.466	0.010	5.978	5.949	0.029
ICER ($/QALY gained)	$986,159/QALY gained for AFL vs RBZ			$711,301/QALY gained for AFL vs RBZ
VA sub-group 1 (20/40 or better at baseline)
Treatment costs	$40,519	$31,577	$8,942	$92,284	$72,607	$19,678
Product	$35,062	$23,761	$11,301	$64,268	$42,232	$22,036
Administration	$1,952	$2,092	–$140	$3,578	$3,718	–$140
Laser procedure	$394	$493	–$98	$541	$640	–$98
AEs	$3,111	$5,232	–$2,121	$23,897^b	$26,018^b	–$2,121^b
Visual impairment/blindness costs^a	$335	$320	$15	$426	$383	$43
Total costs	$40,854	$31,897	$8,957	$92,710	$72,989	$19,721
Total QALYs	1.517	1.519	–0.002	6.131	6.164	–0.032
ICER ($/QALY gained)	RBZ dominates AFL			RBZ dominates AFL
VA sub-group 2 (20/50 or worse at baseline)
Treatment costs	$47,064	$35,779	$11,285	$98,829	$76,809	$22,020
Product	$41,262	$27,623	$13,639	$70,468	$46,094	$24,374
Administration	$2,297	$2,432	–$134	$3,923	$4,058	–$134
Laser procedure	$394	$493	–$98	$541	$640	–$98
AEs	$3,111	$5,232	–$2,121	$23,897^b	$26,018^b	–$2,121^b
Visual impairment/blindness costs^a	$1,150	$1,467	–$318	$1,420	$1,807	–$387
Total costs	$48,214	$37,246	$10,967	$100,249	$78,616	$21,633
Total QALYs	1.433	1.412	0.021	5.816	5.728	0.088
ICER ($/QALY gained)	$523,377/QALY gained for AFL vs RBZ			$246,978/QALY gained for AFL vs RBZ

^a Depression, fractures, skilled nursing facility.

^b Two-year AEs and 10-year myocardial infarction/stroke.

Abbreviations. AE, adverse event; AFL, aflibercept; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; RBZ, ranibizumab; VA, visual acuity.

Two-year costs of managing important AEs were somewhat greater (Δ$2,121) in ranibizumab- than aflibercept-treated patients; however, AE costs made up only 15.1% of the total ranibizumab 0.3 mg treatment costs.

VA sub-group 1 (20/40 or better at baseline)

Total treatment costs for aflibercept 2.0 mg exceeded ranibizumab 0.3 mg treatment costs by nearly $9,000 ($40,854 vs $31,897, respectively; Table 2). Nominal cost differences between treatments were observed for product administration and adjunctive laser therapy. Visual impairment costs were extremely low given the relatively good VA of this sub-group. Consistent with VA outcomes of Protocol T, QALYs accrued in each treatment arm were nearly equivalent (1.517 and 1.519 QALYs for aflibercept and ranibizumab, respectively).

VA sub-group 2 (20/50 or worse at baseline)

Total costs for patients in the 20/50 or worse VA sub-group were ∼17–18% higher than costs for the 20/40 or better VA sub-group, primarily because of the additional injections received, but also because of higher visual impairment costs. Two-year aflibercept 2.0 mg treatment cost was ∼$11,000 more than ranibizumab 0.3 mg treatment ($48,214 vs $37,246, respectively; Table 2); the vast majority of treatment cost was due to drug cost. As expected, given the Protocol T 1-year cost-effectiveness results¹², aflibercept 2.0 mg yielded a slightly greater number of QALYs than did ranibizumab (1.433 vs 1.412, respectively).

Base case ICER analysis

Full cohort

Because the 2-year cost difference between aflibercept 2.0 mg and ranibizumab 0.3 mg was substantial ($9,894) and the incremental benefit quite small (Δ0.010 QALYs), the resulting ICER of $986,159/QALY gained (Table 2) far exceeded the ICER threshold of $150,000/QALY gained. Hence, aflibercept is not cost-effective vs ranibizumab.

VA sub-group 1 (20/40 or better at baseline)

Calculating an ICER comparing the two treatment alternatives was not possible in the 20/40 or better VA sub-group (Table 2). When a treatment is both less costly and equally (or more) effective than an alternative, it is said to ‘dominate’ the other treatment. In this case, ranibizumab 0.3 mg is less costly and yields an equal number of QALYs, and, thus, dominates aflibercept 2.0 mg in the sub-group of patients with good VA at baseline.

VA sub-group 2 (20/50 or worse at baseline)

Because 2-year cost was much higher for aflibercept and the incremental benefit quite small for aflibercept vs ranibizumab, ICER was $523,377/QALY (Table 2). This exceeds the ICER threshold of $150,000/QALY gained, and, thus, aflibercept is not cost-effective vs ranibizumab.

Sensitivity analyses

One-way sensitivity analysis

Product cost is the key driver of overall 2-year costs; therefore, modification of drug cost or number of injections had the largest impact on the results (Figure 2; Supplementary Table S1). Utility estimates for each VA health state also had a substantial impact on the absolute number of QALYs accrued. Alternative assumptions regarding the number of injections administered to the fellow eye in patients treated bilaterally decreased costs by ∼12–13%. Changes in all other parameter estimates resulted in negligible changes in total costs or QALYs.

Figure 2. One-way sensitivity analysis results for the full cohort^a. ^aExcept where otherwise noted, parameters varied by ±20%. ^bIncrease in parameter estimate to 75th percentile value; decrease in parameter estimate to 25th percentile value. ^cThe base case analysis excluded nursing home and caregiver costs; therefore, these costs were added in the sensitivity analysis. ^dBase case injection frequency assumptions for patients treated in the fellow eye (a) from baseline through 1 year or more (2× median number of injections) or (b) any time before 2 years but not at baseline (1.5×) were reduced to 1.5× and 1.25×, respectively. Abbreviations. ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; VA, visual acuity.

In only a few scenarios did the parameter estimate modification change the conclusions. For the full cohort, aflibercept only became cost-effective (ICER = $19,930) when the median number of aflibercept injections was reduced from 15 over 2 years to 11 (lower 25th percentile), while no change was made to ranibizumab injections (also 15 over 2 years). In the 20/40 or better VA sub-group, ranibizumab remained dominant in all scenarios. In the 20/50 or worse VA sub-group, aflibercept was cost-effective, with a 20% decrease in aflibercept costs (ICER = $129,570), or when only considering the lower 25th percentile of injections (aflibercept dominant).

Better-seeing eye:worse-seeing eye sensitivity analysis

Because the results were most sensitive to the better-seeing eye weighting applied to VA-related costs and utilities, this parameter was varied over a range of 100–0% (with the base case weighting the better-seeing eye by 75%; Table 3). From a payer perspective, ICER for aflibercept compared with ranibizumab for the full cohort was >$150,000/QALY when only the worse-seeing eye was considered. In the 20/40 or better VA sub-group, ICER was >$2.5 million/QALY for aflibercept vs ranibizumab. In the 20/50 or worse VA sub-group, ICER was >$150,000/QALY for aflibercept vs ranibizumab.

Table 3. Multi-way sensitivity analysis of the better-seeing eye:worse-seeing eye weighting function.

Analysis	AFL 2.0 mg total costs	RBZ 0.3 mg total costs	AFL 2.0 mg QALYs	RBZ 0.3 mg QALYs	ICER ($/QALY gained)
Full cohort
Base case results, 75%:25% (better:worse)	$44,423	$34,529	1.476	1.466	$986,159
100%:0% (better:worse)	$44,078	$34,112	1.509	1.507	$3,565,898
50%:50% (better:worse)	$44,769	$34,947	1.443	1.426	$568,689
25%:75% (better:worse)	$45,115	$35,365	1.409	1.385	$397,794
0%:100% (better:worse)	$45,460	$35,783	1.376	1.344	$304,824
0%:100% (better:worse) with NH and CG costs	$63,358	$59,110	1.376	1.344	$133,814
VA sub-group 1 (20/40 or better at baseline)
Base case results, 75%:25% (better:worse)	$40,854	$31,897	1.517	1.519	RBZ dominates AFL
100%:0% (better:worse)	$40,609	$31,664	1.543	1.547	RBZ dominates AFL
50%:50% (better:worse)	$41,098	$32,130	1.490	1.490	RBZ dominates AFL
25%:75% (better:worse)	$43,343	$32,363	1.464	1.462	$5,049,321
0%:100% (better:worse)	$41,587	$32,596	1.437	1.434	$2,543,876
0%:100% (better:worse) with NH and CG costs	$52,170	$44,464	1.437	1.434	$2,180,333
VA sub-group 2 (20/50 or worse at baseline)
Base case results, 75%:25% (better:worse)	$48,214	$37,246	1.433	1.412	$523,377
100%:0% (better:worse)	$47,761	$36,638	1.473	1.465	$1,345,712
50%:50% (better:worse)	$48,667	$37,855	1.392	1.359	$321,353
25%:75% (better:worse)	$49,120	$38,463	1.352	1.305	$229,986
0%:100% (better:worse)	$49,573	$39,072	1.311	1.252	$177,905
0%:100% (better:worse) with NH and CG costs	$75,238	$74,224	1.311	1.252	$17,187

Abbreviations. AFL, aflibercept; CG, caregiver; ICER, incremental cost-effectiveness ratio; NH, nursing home; QALY, quality-adjusted life-year; RBZ, ranibizumab; VA, visual acuity.

From a payer perspective, ICER for aflibercept was not cost-effective in this sensitivity analysis. Only when the societal perspective was considered (nursing home and caregiver costs included) were any ICERs <$150,000/QALY.

Extrapolation analysis

The base case analysis time horizon (2 years) was chosen to reflect the availability of patient-level data from Protocol T. Analyses extrapolated to 10 years were conducted using a method adapted from Ross et al.¹² (Table 2). When outcomes and costs were extrapolated to 10 years, ICERs for aflibercept vs ranibizumab decreased to $711,301 for the full cohort and $246,978 for the 20/50 or worse VA sub-group (Table 2), but still exceeded the accepted ICER threshold. Moreover, total costs were ∼2.0–2.3 times higher than the base case, driven largely by anti-VEGF drug costs and inclusion of long-term management costs associated with MI and stroke, while total QALYs were 4.0 times higher (Table 2). ICERs remained well in excess of $150,000/QALY for both the full cohort and 20/50 or worse VA sub-group, and ranibizumab remained dominant over aflibercept for patients in the 20/40 or better VA sub-group.

Probabilistic analysis

To evaluate the impact of cost and utility parameter uncertainty on the base case results, PSAs were performed. Supplementary Figure S1 presents the results of the full cohort PSA as a scatterplot, with the base case result and willingness-to-pay ICER threshold of $150,000/QALY indicated by a red circle and dashed line, respectively. At a $150,000/QALY ICER threshold, the probability of aflibercept being considered cost-effective was 0.1%, 0.0%, and 2.5% for the full cohort, 20/40 or better VA sub-group, and 20/50 or worse VA sub-group, respectively. Therefore, there was a 97.5–100% probability, depending on the cohort, that aflibercept would not be considered cost-effective either because it was dominated (ranibizumab more effective but less costly – upper left quadrant of the scatterplot) or because the ICER exceeded the willingness-to-pay threshold.

Discussion

This study represents the first CEA of the 2-year Protocol T study results, demonstrating that treatment with aflibercept 2.0 mg is not cost-effective when compared with ranibizumab 0.3 mg for treatment of DME in all patients, regardless of baseline vision.

In the current study, QALYs were similar between both treatments regardless of baseline vision. In both VA sub-groups, 2-year total costs for treatment were greater for aflibercept than for ranibizumab. A cost-effectiveness threshold for product adoption of <$150,000/QALY has been recommended based on a review of >1,500 US cost-utility analyses²¹. The results from this analysis indicate that in the 20/50 or worse VA sub-group comparing aflibercept with ranibizumab, the calculated ICER ($523,377/QALY) far exceeded conventional ICER thresholds to be cost-effective. An ICER was not calculated in the 20/40 or better VA group because ranibizumab 0.3 mg was less costly and equally effective, and is therefore the dominant therapy. Although ICERs were sensitive to assumptions about relative weighting of the better- and worse-seeing eye on patient outcomes, taking into account differences in VA improvements between the two agents, aflibercept 2.0 mg is not a cost-effective option.

Impact of parameter uncertainty on the outcomes of interest was assessed by a PSA, which specifies distributions for parameters rather than point estimates and allows for assessments of variability in total cost and QALY results. Our analysis demonstrated a 97.5–100.0% probability (depending on the cohort) that aflibercept would not be considered cost-effective either because it was dominated by ranibizumab or because the ICER exceeded the willingness-to-pay threshold.

The DRCR.net analysis of the cost-effectiveness of aflibercept 2.0 mg, ranibizumab 0.5 mg, and bevacizumab 1.25 mg based on 1-year data from Protocol T found that neither aflibercept nor ranibizumab were cost-effective compared with bevacizumab regardless of baseline VA for treatment of DME (2016 WAC, $1,850, $1,170, and ∼$60/dose, respectively)¹². Our study, using longer-term follow-up, provides different conclusions than the 1-year analysis. This is because, when comparing aflibercept with ranibizumab, the DRCR.net analysis based on 1-year follow-up found aflibercept may be cost-effective compared with ranibizumab in some circumstances in both the base case and sensitivity analyses. For example, in their 10-year extrapolation, aflibercept was found to be cost-effective compared with ranibizumab for patients with a baseline VA of 20/50 or worse (ICER $135,000/QALY). However, our analysis, using 2-year follow-up, found that aflibercept is not cost-effective for this patient sub-group (ICER $246,978/QALY).

The Protocol T 1-year CEA also used a different model than used in the present study. In the Protocol T analysis, utilities were assigned based on conversion tables from Brown et al.²⁹. In the current study, utilities were based on those determined by the time tradeoff technique in patients whose visual impairment was simulated¹⁶. The utility equation includes a constant term in addition to parameters for VA (defined by logarithm of the minimum angle of resolution) and age, which allows for a more granular utility calculation compared with approaches using more rigid VA strata or not considering the impact of age on utility. Consistent with our findings, the DRCR.net analysis demonstrated that when bevacizumab is not available for treatment of DME, aflibercept is not cost-effective vs ranibizumab¹².

Model and analysis limitations

A potential limitation was that VA data represented the only QALY parameter analyzed. Furthermore, as in the main Protocol T clinical papers^10,11, the CEA reported herein explored two VA sub-groups and, because patients were not randomized according to sub-group, there may be subtle differences unaccounted for in our analysis. To evaluate potential differences that could impact the CEA, we explored numerous parameters in sensitivity analyses. With the exception of extreme or unrealistic scenarios, these parameters had very little impact on the results, and did not change the conclusions.

This analysis specifically excluded bevacizumab, which is not approved for ophthalmic use but is commonly used off label to treat DME. In the 1-year CEA of Protocol T, bevacizumab was the most cost-effective option despite having the least improvement in VA¹². Because differences in VA improvement between all agents only narrowed at year 2, bevacizumab would have remained the most cost-effective option. However, because differences in VA outcomes between aflibercept and ranibizumab were no longer statistically significant at year 2, a new CEA comparing these two agents was warranted.

Another limitation is the uncertainty surrounding the role of better- and worse-seeing eye VA in overall functional impairment¹⁷, specifically risk of events such as falls and overall HRQoL. For example, it is unclear to what extent functional levels and HRQoL would differ in two patients with same better-seeing eye VA but different worse-seeing eye VA. A study assessing impact of VA loss on utilities showed that taking only the better-seeing eye into account may under-estimate the impact of VA loss, particularly in cases in which the better-seeing eye has no vision loss³⁰. Our analyses addressed this issue using a novel approach, and we varied our assumptions related to this issue in the sensitivity analyses. More research on this issue is needed to ensure more informed assumptions can be used in future CEAs.

Generalizability of the results could be affected by differences in the patient population or treatment frequency between the Protocol T study and typical clinical practice. Protocol T only enrolled patients with center-involved DME who had not received any anti-VEGF therapy within the previous 12 months. The typical clinical practice is likely to see patients with different types of DME and treatment histories. Moreover, a recent retrospective analysis of a large US claims database found that the adjusted number of injections per year was 3.3 and 3.9 for ranibizumab and aflibercept, respectively, in treatment-naive patients with DME, and 3.8 and 4.2 in previously treated patients with DME³¹. This is considerably less than the 15 injections over 2 years for both aflibercept and ranibizumab in Protocol T. It should also be noted that the results of this study cannot be generalized to other countries. Protocol T was conducted in the US and only US prices and costs were used in the model.

In addition, differences in the number of letters of VA gained between ranibizumab and aflibercept are sometimes minimal and may not be clinically meaningful. It is generally accepted that a difference of ≥5 letters (1 line) of BCVA on the Early Treatment Diabetic Retinopathy Study eye chart is required to be clinically meaningful.

Conclusions

By providing a better understanding of relative differences in costs and QALYs of the two FDA-approved drugs for DME, the findings of this analysis may help payers, patients, and retina specialists when considering appropriate DME treatment.

Results from our CEA of the 2-year data from Protocol T indicate that treatment with aflibercept is more costly than ranibizumab and yields similar or nominal increases in QALYs, and is therefore not cost-effective vs ranibizumab. These results are consistent with the 1-year DRCR.net analysis of Protocol T¹².

Transparency

Declaration of funding

Genentech, Inc., South San Francisco, CA, provided support for the study and participated in the design and conduct of the study, and data collection, management, and interpretation. Third-party writing assistance was provided by Amy Lindsay, PhD, of Envision Pharma Group and funded by Genentech, Inc.

Declaration of financial/other interests

NH is a consultant for Alimera, Allegro, Allergan, BioTime, Gemini, Genentech, Inc., Katalyst, Novartis, Regeneron, and Roche. SBD is a consultant for Genentech, Inc. YR and VG are employees of Genentech, Inc., and hold stock/stock options in Roche. The peer reviewers on this manuscript have received an honorarium from JME for their review work. In addition, a reviewer on this manuscript discloses their role as a consultant for Genentech, Inc., and Novartis, while another reviewer discloses their role as a consultant for Bayer and Novartis. The reviewers have no other relevant financial relationships or otherwise to disclose.

Author contributions

All authors were involved in the conception and design of the study, the analysis and interpretation of the data, and critical review and revision of the paper for intellectual content. All authors provided final approval of the version to be published and agree to be accountable for all aspects of the work.

Previous presentations

Portions of these data were presented at the Association for Research in Vision and Ophthalmology Annual Meeting; May 7–11, 2017; Baltimore, MD.

Acknowledgements

Funding was provided by Genentech, Inc., a member of the Roche Group, for the study and third-party writing assistance, which was provided by Amy Lindsay, PhD, of Envision Pharma Group.