3 The company's submission | Pembrolizumab for advanced melanoma not previously treated with ipilimumab | Guidance

Clinical effectiveness

3.1 The company presented clinical‑effectiveness evidence for pembrolizumab from 2 clinical trials: KEYNOTE‑006 and KEYNOTE‑001. KEYNOTE‑006 was a randomised, international, multicentre, phase III trial of pembrolizumab 10 mg/kg every 2 weeks (n=279) or every 3 weeks (n=277), continued until disease progression or unacceptable toxicity, compared with ipilimumab 3 mg/kg every 3 weeks, continued for 4 doses (n=278). Results were analysed at 2 planned interim analyses, after 6 months of follow‑up (September 2014) and after 9–12 months of follow‑up (March 2015). After the second interim analysis the study was stopped, because the primary endpoint had been met. KEYNOTE‑001 was a combined phase I and II study. Evidence was presented from a sub‑study of this trial, referred to as KEYNOTE‑001 part D: this was a randomised, open‑label study comparing pembrolizumab 2 mg/kg every 3 weeks (n=51; the licensed dose) with 10 mg/kg every 3 weeks (n=52). Both KEYNOTE‑006 and KEYNOTE‑001 part D included people with advanced melanoma, with or without BRAF mutations, who had not had ipilimumab before (previous treatment with 1 or 2 other therapies was permitted).

3.2 In KEYNOTE‑006, pembrolizumab was associated with statistically significant increases in both progression‑free survival (first interim analysis) and overall survival (second interim analysis), compared with ipilimumab (table 1). Pembrolizumab was also associated with statistically significantly higher overall response rates compared with ipilimumab (table 1; p<0.001). Pre‑specified subgroup analyses based on demographic and clinical characteristics suggested that the treatment effect associated with pembrolizumab was generally consistent across subgroups, although some variations in effect based on the line of therapy and the expression of ligands for the 'programmed death 1' protein (termed 'PD‑L1 status') were seen. No significant differences in clinical effectiveness between pembrolizumab doses (that is, 10 mg/kg every 3 weeks, 10 mg/kg every 2 weeks or 2 mg/kg every 3 weeks) were seen in either KEYNOTE‑006 or KEYNOTE‑001 part D.

Table 1 Clinical‑effectiveness outcomes in KEYNOTE‑006

	Pembrolizumab		Ipilimumab n=278
	10 mg/kg every 3 weeks n=277	10 mg/kg every 2 weeks n=279	Ipilimumab n=278
Progression‑free survival (interim analysis 1)
Median: months (95% CI)	4.1 (2.9–6.9)	5.5 (3.4–6.9)	2.8 (2.8–2.9)
Hazard ratio versus ipilimumab (95% CI)	0.58 (0.47–0.72) p<0.00001	0.58 (0.46–0.72) p<0.00001
Overall survival (interim analysis 2)
Overall survival at 6 months¹: % (95% CI)	87.3 (82.7–90.7)	84.8 (80.0–88.5)	74.5 (68.7–79.4)
Hazard ratio versus ipilimumab (95% CI)	0.69 (0.52–0.90) p=0.00358	0.63 (0.47–0.83) p=0.00052
Overall response (interim analysis 1)
Overall response rate: % (95% CI)	32.9% (27.4–38.7)	33.7% (28.2–39.6)	11.9% (8.3–16.3)
Abbreviations: CI, confidence interval; n, number of patients. ¹Median overall survival not reached.

3.3 The company presented adverse event data from KEYNOTE‑006, and stated that pembrolizumab was generally well tolerated. It was associated with a similar number of adverse events, but fewer drug‑related grade 3–5 adverse events, serious adverse events, drug‑related serious adverse events and adverse events leading to withdrawal from the trial, compared with ipilimumab (table 2). Pembrolizumab was also associated with fewer high‑grade, immune‑related adverse events than ipilimumab, and fewer people in the pembrolizumab groups withdrew from the trial because of immune‑related adverse events (table 2). The most common treatment‑related adverse events with both pembrolizumab and ipilimumab were fatigue, diarrhoea, rash and itching. The most common grade 3–5 immune‑related adverse events associated with pembrolizumab were colitis and hepatitis.

Table 2 Summary of adverse events in KEYNOTE‑006

	Ipilimumab		Pembrolizumab (both arms combined)
	n	%	n	%
n	256		555
Patients with 1 or more AE	239	93%	539	97%
Drug‑related grade 3–5 AE	51	20%	65	12%
Serious AE	77	30%	140	25%
Drug‑related serious AE	45	18%	49	9%
Stopped because of an AE	34	13%	50	9%
Immune‑related AEs
Patients with 1 or more AE	47	18%	109	20%
Grade 3–5 AE	30	12%	30	5%
Stopped because of an AE	14	5%	15	3%
Abbreviations: AE, adverse event; n, number of patients.

3.4 The company compared the clinical effectiveness of pembrolizumab with ipilimumab, dabrafenib, vemurafenib and dacarbazine in a series of network meta‑analyses. The analyses were performed in a Bayesian framework using a fixed‑effects model, and were based on data from KEYNOTE‑006 and 5 other trials identified in the systematic review. The company presented results from a series of analyses, including 4 alternative network scenarios, 4 time points, and separate analyses of pembrolizumab as either a first‑ or second‑line treatment. It stated that, for the outcomes of progression‑free survival and overall survival, pembrolizumab appeared to have a similar efficacy to vemurafenib and dabrafenib, in people who have had no previous treatment. For example, in network scenario '3b' at the 6‑month time point, pembrolizumab was associated with a hazard ratio for progression‑free survival of 0.80 (95% credible interval [CrI] 0.32–1.92) compared with dabrafenib, and 0.67 (95% CrI 0.37–1.14) compared with vemurafenib. The corresponding hazard ratios for overall survival were 0.96 (95% CrI 0.46–1.93) and 0.75 (95% CrI 0.40–1.34), compared with dabrafenib and vemurafenib respectively. The company highlighted that pembrolizumab was associated with greater progression‑free survival and overall survival than both ipilimumab (scenario 3b, 6‑month timepoint: hazard ratio for progression‑free survival 0.45, 95% CrI 0.33–0.62; hazard ratio for overall survival 0.59, 95% CrI 0.41–0.84) and dacarbazine (scenario 3b, 6‑month timepoint: hazard ratio for progression‑free survival 0.36, 95% CrI 0.21–0.59; hazard ratio for overall survival 0.54, 95% CrI 0.30–0.91) for previously untreated disease, and was at least as effective as ipilimumab for people who have had 1 previous line of treatment (scenario 3b, 6‑month timepoint: hazard ratio for progression‑free survival 0.74, 95% CrI 0.48–1.12; hazard ratio for overall survival 0.80, 95% CrI 0.52–1.22).

Cost effectiveness

3.5 The company presented an economic model comparing pembrolizumab (2 mg/kg every 3 weeks) with ipilimumab, dabrafenib and vemurafenib. The model comprised 3 states: pre‑progression, post‑progression and death. The model used a cycle length of 1 week and a time horizon of 30 years (lifetime), taking the perspective of the NHS and personal social services, with costs and benefits discounted at a rate of 3.5% per year. The model included costs associated with melanoma treatment, costs in each health state (based on a study of resource use for melanoma treatment in the UK), management of adverse events and complications, and care at the end of life.

3.6 The proportion of people in the each health state was based on estimates of progression‑free survival and overall survival:

Progression‑free survival was estimated using Kaplan–Meier curves from the KEYNOTE‑006, BRIM‑3 and BREAK‑3 clinical trials, extrapolated to 30 years based on a Gompertz model (for pembrolizumab and ipilimumab) or a monthly risk of progression taken from NICE's technology appraisal guidance on ipilimumab for previously untreated advanced (unresectable or metastatic) melanoma (for dabrafenib and vemurafenib).
Overall survival was estimated by initially using Kaplan–Meier data from clinical trials (for the first 50–60 weeks of the model), followed by published mortality risks based on data from a pooled analysis of long‑term survival data for people with melanoma treated with ipilimumab (Schadendorf et al. [2015]; applied to pembrolizumab and ipilimumab) and NICE's technology appraisal guidance on ipilimumab for previously untreated advanced (unresectable or metastatic) melanoma (applied to dabrafenib and vemurafenib). Long‑term survival was based on mortality rates in a published registry study (Balch et al. [2001]).

3.7 Utility values were estimated using EuroQol EQ‑5D data from KEYNOTE‑006, by assuming that quality of life decreases as people approach the last months of life. The utility scores decreased from 0.82, for people who were more than 360 days before death, to 0.33 for people in the 30 days before death.

3.8 The company's base‑case results are summarised in table 3. These results were based on the discount in the patient access scheme for pembrolizumab and the list prices for all other drugs, and therefore were not used for decision‑making; they are included here for illustration only. Results from the company's model including the patient access schemes for pembrolizumab and all 3 comparators were presented by the ERG; these results are commercial in confidence and cannot be reported here.

Table 3 Results of the company's base‑case analysis (including pembrolizumab patient access scheme, list price for all comparators; not used for decision‑making)

BRAF mutation‑positive disease
	Total cost	Total LYG	Total QALYs	Incr cost	Incr LYG	Incr QALYs	ICER (£/QALY)
Dabrafenib	£71,029	3.41	2.17	‑	‑	‑	‑
Pembrolizumab	£76,689	5.08	3.14	£5660	1.67	0.97	£5852
Vemurafenib	£83,384	2.74	1.73	£6695	−2.34	−1.40	Dominated
Ipilimumab	£97,873	4.37	2.69	£21,185	−0.71	−0.44	Dominated
BRAF mutation‑negative disease
	Total cost	Total LYG	Total QALYs	Incr cost	Incr LYG	Incr QALYs	ICER (£/QALY)
Pembrolizumab	£76,689	5.08	3.14	‑	‑	‑	‑
Ipilimumab	£97,873	4.37	2.69	£21,185	−0.71	−0.44	Dominated
Abbreviations: ICER, incremental cost‑effectiveness ratio; Incr, incremental; LYG, life years gained; QALYs, quality‑adjusted life years. Dominated: provides fewer QALYs at greater cost than the comparator.

3.9 In a deterministic sensitivity analysis, the model results for all comparisons were most sensitive to the extrapolation of progression‑free survival for pembrolizumab (shape and treatment effect in the Gompertz model). In the probabilistic sensitivity analysis, the total costs associated with pembrolizumab increased by £10,996 compared with the deterministic results, and the total QALYs decreased by 0.02, whereas the results for ipilimumab, dabrafenib and vemurafenib did not change substantially. The company stated that the change in the results for pembrolizumab was caused by uncertainty in the extrapolation of progression‑free survival from the KEYNOTE‑006 trial, which led to a small number of iterations with high treatment costs. The company also presented 33 scenario analyses, to explore the effects of key assumptions on the model results. These analyses explored the effects of varying the progression‑free survival and overall survival estimates, time horizon, utility estimates, treatment and terminal care costs, treatment duration for pembrolizumab, and the discounting rate. The company stated that these analyses showed that the cost effectiveness of pembrolizumab was robust to most sources of uncertainty.

ERG comments

3.10 The ERG stated that KEYNOTE‑006 was well designed and well conducted. It considered that the population was representative of patients seen in the UK NHS, and patient characteristics were well balanced across treatment groups. However, it noted 3 key concerns about this trial:

The dosage of pembrolizumab (10 mg/kg every 3 weeks) did not match the licensed dose (2 mg/kg every 3 weeks). The ERG noted that the European Public Assessment Report (EPAR) states that no differences between the licensed dose and the studied dose are to be expected. Although this was largely based on data from patients who had previous therapy with ipilimumab, the ERG cautiously accepted this conclusion.
The trial was stopped early, so the overall survival data were immature. The ERG was unclear whether the true impact of pembrolizumab on survival will be identified.
The trial specified a maximum treatment duration of 24 months. The ERG considered that the effect of this rule on clinical outcomes is unknown.

3.11 The ERG considered that the clinical assumptions used in the company's network meta‑analysis were reasonable, but the methods of the analysis were flawed. It stated that the populations in the control arms were not comparable, the analysis did not correctly reflect changing hazard ratios over time, and the methods used to adjust for treatment switching in the key trial for vemurafenib may not have been adequate. The ERG considered that the network meta‑analysis did not provide valid treatment effect estimates, particularly for pembrolizumab compared with dabrafenib and vemurafenib.

3.12 The ERG's critique of the company's economic model suggested that the model was generally consistent with the NICE reference case. However, it highlighted that the structure of the model led to counterintuitive results – specifically, that pembrolizumab would become less cost effective if its effectiveness at delaying disease progression increased. The ERG stated that this was because delaying disease progression was associated with additional treatment costs but no increase in quality of life. In addition, the ERG expressed concerns about the modelling of overall survival, progression‑free survival, treatment costs and quality of life. It considered that there were limitations in the methods of extrapolation for both progression‑free survival and overall survival. In particular, for overall survival, the ERG stated that there was a risk of selection bias in the data taken from the study by Schadendorf et al. (2015), and there were limitations in the algorithm used to adjust for patient characteristics and the long‑term survival data (from Balch et al. [2001]) used to project long‑term survival. The ERG highlighted that the company's estimates for mortality risk in people treated with pembrolizumab changed erratically during the course of the model and were not clinically plausible. For the analysis of progression‑free survival, the ERG noted limitations including the use of centrally assessed progression, the company's choice of censoring rule, inappropriate use of the proportional hazards assumption and incomplete adjustment for differences in patient characteristics between the dabrafenib, vemurafenib and pembrolizumab trials. The ERG described concerns about the duration of treatment, the weight distribution of the population, and the administration costs of ipilimumab, which meant that the costs of treatment were not accurately estimated. It considered that there were important limitations in the estimation of utility, because of the use of EQ‑5D data based on patients from all regions (rather than UK or European patients only) and the assumption that utility did not change when disease progressed. The ERG highlighted that 75–90% of the cost differences between treatments in the company's model could be attributed to direct treatment costs, and that 87.5% of the health gain with pembrolizumab occurred after 12 months. It therefore considered that the key factors affecting results of the model were drug costs, duration of treatment and overall survival.

3.13 The ERG presented a series of exploratory analyses to address their principal concerns about the company's model (see section 3.12). In particular, it changed the modelling of overall survival (using methods developed for NICE's technology appraisal guidance on ipilimumab for previously treated advanced (unresectable or metastatic) melanoma), progression‑free survival and treatment duration, and amended the treatment costs and utility scores. It also presented 2 scenario analyses, in which the duration of progression‑free survival was extended by 3 or 6 years for people whose disease had not progressed after 2 years; the ERG presented results for all analyses using the patient access scheme price for pembrolizumab and the list prices for ipilimumab, dabrafenib and vemurafenib (summarised below) and also using patient access schemes for all drugs (commercial in confidence; cannot be reported here). The ERG's amendments, combined, increased the costs associated with pembrolizumab by £6593 and reduced the total QALYs by 0.18. The amendments also reduced the costs and QALYs associated with ipilimumab (by £2558 and 0.17 respectively), increased the costs and QALYs for vemurafenib (by £7027 and 0.5 respectively), and increased the costs but reduced the QALYs for dabrafenib (by £3,238 and 0.02 respectively). Both of the scenario analyses substantially increased the costs associated with pembrolizumab, but had no effect on the other treatments.

3.14 Full details of all the evidence are available.