Pembrolizumab with axitinib for untreated advanced renal cell carcinoma

3.1 In England, kidney cancer is expected to cause about 3,783 deaths every year, with 10,759 new cases per year. Of people with kidney cancer, 80% have renal cell carcinoma. A patient expert explained that treatment with pembrolizumab with axitinib had been positive. This was because their tumour had reduced and there were no notable side effects with the treatment, unlike their experience with other treatment options. Patient experts confirmed that people with untreated renal cell carcinoma felt that the side effects of treatment could substantially affect quality of life. The committee recognised that for advanced renal cell carcinoma there is a high unmet need for both patients and healthcare professionals. Also, there is an unmet need for treating non-clear cell renal cell carcinoma specifically. Overall, an option that improved survival and reduced side effects would be welcomed by patients and clinicians to allow more choice of treatment and individualised care plans.

3.2 The committee considered the current treatment pathway for renal cell carcinoma. First-line options for treating metastatic renal cell carcinoma include tivozanib, sunitinib and pazopanib. Pazopanib is most likely to be used out of these. Avelumab with axitinib is recommended through the Cancer Drugs Fund (CDF). Cabozantinib is only recommended for patients with intermediate or poor risk renal cell carcinoma. Nivolumab with ipilimumab is also only recommended for patients with intermediate or poor risk renal cell carcinoma, through the CDF. Treatment options, in particular cabozantinib, can be difficult to tolerate because of the side effects. Clinical experts expected that patients who are less frail would be offered combination therapy instead of single agents. This is because of enhanced tolerability and a longer duration of disease control when using 2 effective treatments together (noting that the IMDC criteria corresponds to prognosis, rather than a score of frailty). During technical engagement, clinical experts estimated that over 50% of people who had first-line treatment would have subsequent treatment. The CDF clinical lead and the clinical experts explained that if patients have first-line treatment with pembrolizumab (a checkpoint inhibitor) plus axitinib (a tyrosine kinase inhibitor [TKI]), then they would be unable to have nivolumab (another checkpoint inhibitor) or axitinib monotherapy later in the treatment pathway. It is likely that subsequent treatment options would then be considered from a combination of current first-line and second-line options. The committee concluded that pembrolizumab with axitinib was likely to have a substantial effect on the care pathway. But, this effect would be similar to that of other immunotherapy combinations for first-line renal cell cancer.

3.3 The clinical evidence came from KEYNOTE‑426, an open-label, randomised phase 3 trial that compared pembrolizumab plus axitinib with sunitinib (median follow up of 12.8 months). The primary outcome measures in KEYNOTE‑426 were overall survival (hazard ratio 0.53; 95% confidence interval 0.38 to 0.74, p=0.00005) and progression-free survival (hazard ratio 0.69; 95% confidence interval 0.57 to 0.84, p=0.00014). Median survival was not reached in either arm. There was no evidence presented comparing pembrolizumab plus axitinib with tivozanib or pazopanib. However, tivozanib and pazopanib were assumed to have equal efficacy and safety to sunitinib. This was in line with previous NICE technology appraisal guidance on:

3.4 There was no direct evidence comparing pembrolizumab plus axitinib with cabozantinib for the IMDC intermediate and poor-risk subgroup. The company did a network meta-analysis using data from KEYNOTE‑426 and CABOSUN (a randomised phase 2 trial of cabozantinib [n=79] compared with sunitinib [n=78]). The committee noted the small sample size of CABOSUN. Also, the network meta-analysis did not find a significant difference in progression-free survival or overall survival for the indirect comparison of pembrolizumab plus axitinib with cabozantinib. Overall, the committee considered that the evidence base for the intermediate and poor-risk subgroup was weak.

3.5 Clinical experts expected that pembrolizumab with axitinib would offer a durable response, but they were not certain about the size of the response. They suggested that a different survival trajectory between pembrolizumab with axitinib and sunitinib could be expected. This was because of the differences in the biological mode of action between an immunotherapy and a TKI. The clinical experts explained that immunotherapy was expected to not only attack and kill the cancer cells, but also re-programme the immune system to recognise and adapt to attack and kill future cancer cells. This mode of action differed from a single TKI. The clinical experts expected a durable sustained response after treatment that was not expected with treatment from a single TKI. However, the NICE Decision Support Unit technical support document 14 advises that both arms should have the same extrapolation distribution unless there is substantial justification. There was theoretical justification to use different distributions for each of the trial arms. However, there was no robust evidence to support the argument that the different mode of action of the drugs would result in different survival trajectories. The committee acknowledged that the overall survival data were immature. Therefore, it was appropriate to consider various scenarios presented, including analyses when different distributions were applied. However, the committee concluded that there was no robust evidence to justify using different distributions to extrapolate survival for each of the trial arms.

3.6 The committee considered which distribution was the most appropriate to model the overall survival for pembrolizumab with axitinib. The log-logistic distribution used by the company had optimistic survival estimates compared with clinical estimates. The committee also examined the progression-free survival data and survival curves from KEYNOTE‑426, noting that there were data for about 20 months of follow up. It also noted that disease had progressed in most people before 20 months, regardless of treatment. This led the committee to question both the size and length of response, and given this, whether it was valid to assume different survival trajectories for the different treatments. At consultation, the company submitted long-term follow-up data from KEYNOTE‑035. This was a phase 1 study of pembrolizumab with axitinib in advanced untreated renal cell cancer. It suggested treatment effect continued past 5 years (the exact results are confidential and cannot be reported here), supporting using the log-logistic distribution. Given the short follow up in KEYNOTE‑426, clinical experts stated that extrapolation of mature data from other sources (such as KEYNOTE‑035) was important to inform long-term survival estimates. However, the ERG noted that the KEYNOTE‑035 study was not designed to assess overall survival, which was one of a number of secondary outcomes. This, along with the small number of patients in the trial, meant that the results of KEYNOTE-035 should be interpreted with caution. The committee concluded that the survival estimates from the log-logistic distribution in the company base case were optimistic.

3.7 To account for age-related increase in mortality, overall survival was capped at the general population mortality rates in the company model. For pembrolizumab with axitinib, this change happened at around 20 years and suggested that about 17% of people were 'cured'. The committee asked the company whether it had examined cure fractions or if it had considered a 'cure' model to estimate survival. The company confirmed that cure fractions had not been considered in the economic modelling and did not intend to do a 'mixture' cure model. The committee concluded that overall survival for pembrolizumab and axitinib may have been overestimated. This was because switching to the same mortality as the general population at about 20 years was likely to be overly optimistic.

3.8 The Weibull curve was the ERG and technical team's preferred distribution for extrapolating overall survival for both pembrolizumab with axitinib and sunitinib. The committee heard from clinical experts that a rising hazard rate, which was a characteristic of the chosen Weibull distribution, was not expected for people who had pembrolizumab with axitinib. Therefore, the committee agreed that the chosen Weibull distribution was likely to give pessimistic survival estimates.

3.9 The committee concluded that the most plausible survival estimates were likely to fall within the range created by the log-logistic and Weibull distribution used in the company base case and the ERG and technical team base cases, respectively. At consultation, the company submitted new analyses using the exponential distribution to estimate overall survival for both pembrolizumab with axitinib and sunitinib. The company considered this approach unfavourable to pembrolizumab with axitinib. In this distribution, long-term survival estimates fell within the range considered plausible by the committee and a smaller proportion of patients were predicted to be 'cured'. The ERG maintained its initial preference for the Weibull distribution but noted that the exponential was also plausible. The committee agreed to take both the log-logistic and exponential distributions into account in its decision making. However, it noted that considerable uncertainty remained because of the immaturity of the evidence.

3.10 The committee acknowledged that assumptions about treatment effect duration would affect expected survival. Clinical experts explained that there could be a long-term response with continued use of a TKI, but it would not be a durable response and would stop when treatment was stopped. Immunotherapy was expected to provide a durable response after stopping the treatment because of its mode of action but this has not yet been confirmed with clinical evidence. One clinical expert estimated that this could happen in around 15% of people having pembrolizumab with axitinib. Expert opinion on long-term survival also varied, with between 35% and 50% of people estimated to be alive 5 years after starting treatment. There was a large amount of uncertainty surrounding the estimates at 10 years and 20 years. Although the committee thought a durable response was possible, immaturity of the data meant that this was based on clinical opinion, scientific reasoning and anecdotal evidence. The committee noted that in previous NICE appraisals of checkpoint inhibitors when length of treatment was capped at 2 years, the committee:

3.11 There was a 2-year stopping rule in the economic model for pembrolizumab. Treatment with axitinib continued until second-line treatment was needed, for example, because of disease progression. Based on long-term follow-up data from other checkpoint inhibitors, the committee found it reasonable to assume some duration of response. However, it agreed that it could not generalise the size of this effect from one cancer to another. It also recalled that there were only data for about 20 months of follow up from KEYNOTE-426. It considered scenarios when the treatment effect stopped after 3 years, 5 years and 10 years (that is, treatment effect continued to 1 year, 3 years and 8 years after stopping pembrolizumab). The committee noted that there could be uncertainty in the economic model if treatment waning effects were applied in a scenario with continued axitinib treatment, or in scenarios when there was an implicit assumption of cure in the model. Therefore, the scenario analyses were interpreted with caution. The committee concluded that the immaturity of the data made any estimation of treatment waning effect highly uncertain. But, it accepted scenarios when a waning effect was applied after 5 years.

3.12 At consultation, the company provided a scenario to model treatment effect waning after 5 years, based on individual responses to pembrolizumab with axitinib in the KEYNOTE-426 trial. People whose disease responded to treatment were modelled so that they were assumed to have the base-case hazard rate. A waning effect was only applied to people whose disease did not respond to treatment (16.2% of the full population). For these people, there was a gradual decrease in hazard rate for progression-free and overall survival between years 5 and 10. After this hazard rates equalled that of sunitinib. The ERG was unclear why the waning effect had only been applied to people whose disease did not respond, given that for most people disease had progressed after 5 years. It stated that applying a treatment waning effect to the entire population was appropriate and provided a scenario using this method. The committee concluded that a treatment waning effect should be applied to all patients having pembrolizumab with axitinib, regardless of treatment response.

3.13 In KEYNOTE-426, the maximum pembrolizumab treatment duration was 2 years from the first dose, when treatment must be stopped. This was not reflected in the summary of product characteristics, which states that treatment should continue until disease progression or unacceptable toxicity. For pembrolizumab for other indications, a 2‑year stopping rule was applied. The committee noted that the 2‑year stopping rule was included in company's economic model, and concluded that it was appropriate. KEYNOTE-426 allowed treatment to stop and restart within the 35 cycles. It also allowed for another 17 cycles of retreatment because of relapse if the patient had stopped at 35 cycles or stopped because of complete remission. Clinical experts noted that these patients were those most likely to have an ongoing treatment effect, so retreatment rates were expected to be low. The committee noted that the follow up of 20 months was shorter than the 2‑year stopping rule. So, KEYNOTE‑426 did not give any information about the likely impact of the 2‑year stopping rule on effectiveness, the proportion of patients who would restart treatment with pembrolizumab after having had 35 cycles, or the effectiveness of retreatment. The committee concluded that a 2‑year treatment stopping rule in line with the clinical- and cost-effectiveness evidence was appropriate. However, it was uncertain how this might work in the NHS if the technology were recommended.

3.14 At consultation, the company presented retreatment data from a later data cut of KEYNOTE‑426 (results are confidential and cannot be presented here). The company discouraged modelling of retreatment because of insufficient evidence and a lack of robust statistical methods to account for bias and confounders. However, it presented a scenario to show the effect on the incremental cost-effectiveness ratio (ICER). A cost was applied to 4.8% of the full population. This was based on pembrolizumab retreatment rates seen in the phase 3 KEYNOTE‑006 and KEYNOTE‑010 trials for melanoma and non-small-cell lung cancer, respectively. Clinical experts noted that the immaturity of the data meant that retreatment had little impact on the effectiveness estimates. However, the committee shared the ERG's concerns that these data were collected in different cancer types and used pembrolizumab as a monotherapy. It concluded that the company's retreatment rate was not generalisable to renal cell cancer and the scenario presented was not appropriate for decision making.

3.15 Clinical experts confirmed that markers of disease progression, such as tumour size, may not have a strong correlation with quality of life. This suggests that a time-to-death approach to estimate health-related quality of life could be reasonable. The committee compared the utility values used for the progression-free and progressed states with those using the time-to-death approach in the company base case. The company also provided a scenario when utilities were calculated by progression status and differentiated by treatment. They were higher for pembrolizumab with axitinib than those calculated for sunitinib for each respective health state. The committee noted the decrement in quality of life between the progression-free and progressed states. It considered how the utility data were collected in KEYNOTE‑426. Findings from all the methods to analyse utility data may be biased and give overly optimistic estimates. This is because data collection on health-related quality of life stopped shortly after progression. This may have led to informative censoring bias and uncertainty in estimates for health-related quality of life at the end stages of disease. Clinical experts commented that they would expect post-progression quality of life to be influenced by subsequent-line treatments and this may be higher than estimated using the study data. Patient experts confirmed that patients might feel the need to complete the questionnaire with more positive responses to be able to continue treatment. The committee concluded that, without further data, post-progression utility values from both the published literature and KEYNOTE-426 were acceptable for decision making.

3.16 The committee did not comment further on the appropriateness of including or excluding an age-related decrement to the model. This was because overall survival estimates were highly uncertain. However, findings from both scenarios (with and without age-related decrements) were considered in the committee's decision making.

3.17 The cost-effectiveness results are commercial in confidence and cannot be reported here. The committee considered all scenarios from the company, ERG and technical team to establish when pembrolizumab with axitinib could be considered cost effective. It agreed that the company's original base-case ICER was likely to be optimistic because of using the log-logistic distribution for extrapolation and applying a lifetime treatment effect. Also, when using either the log-logistic or exponential distribution, every scenario presented was above the normally acceptable range, taking into account all commercial arrangements. This applied to both the overall renal cell carcinoma population and the intermediate and poor-risk subgroup when using either the company's or ERG's assumptions of treatment waning effect. However, the technical team and ERG base-case ICERs were likely to be pessimistic, because they used the Weibull distribution in the extrapolation of survival (see section 3.8). ICERs of alternative scenarios provided by the technical team and the ERG also did not fall below £30,000 per quality-adjusted life year (QALY) gained. These scenarios included all commercial arrangements added to the analyses for either the overall renal cell carcinoma population or for the intermediate and poor-risk subgroup. The committee concluded that the most plausible ICER was within the range presented by the company base case and the technical team base case. So, the most plausible ICER was above the range normally considered cost effective.

3.18 The committee considered the advice about life-extending treatments for people with a short life expectancy in NICE's guide to the methods of technology appraisal. The committee, ERG and company agreed that pembrolizumab with axitinib does not meet end-of-life criteria for the overall renal cell carcinoma population. The first end-of-life criterion (that treatment is indicated for patients with a short life expectancy, normally less than 24 months) in the intermediate and poor-risk group was not met. This was because the median overall survival in the sunitinib arm of CHECKMATE‑214 was 26 months, with the mean value likely to be higher. Overall survival for the poor-risk group could not be estimated from the economic model because it was not considered as a distinct subgroup. The committee noted that the CABOSUN trial included few patients who had poor-risk disease (15 for cabozantinib and 15 for sunitinib). This meant an overall survival estimate for cabozantinib, the standard of care in this population, would be highly uncertain. The committee concluded there was no evidence to support that the first end-of-life criterion was met in any of the IMDC risk groups. Therefore, pembrolizumab with axitinib did not meet the criteria to be considered as a life-extending treatment at the end of life.

3.19 The committee discussed the arrangements for the CDF agreed by NICE and NHS England in 2016, noting NICE's Cancer Drugs Fund methods guide (addendum):

3.20 The company and clinical experts considered that pembrolizumab with axitinib was innovative. They noted pembrolizumab with axitinib had a notable survival benefit and expected that the treatment would have a durable response. A clinical expert commented that, observationally, the technology seemed to have an improved adverse event profile when compared with other combination treatments. The committee agreed that these were important potential benefits of pembrolizumab with axitinib. However, it had not been presented with evidence of any additional benefits that could not be captured in the QALY measurements.

3.21 No equality or social value judgement issues were identified.

3.22 The committee concluded that the most plausible ICER, when commercial discounts were taken into account, was above the range that NICE normally considers to be a cost-effective use of NHS resources. It therefore concluded that pembrolizumab with axitinib is not recommended for untreated advanced renal cell carcinoma.