Draft guidance consultation

No direct comparison data between pembrolizumab plus doublet chemotherapy and nivolumab plus doublet chemotherapy was available. So the company did a network meta-analysis (NMA) using data from KEYNOTE‑859 for pembrolizumab plus doublet chemotherapy and CHECKMATE‑649 for nivolumab plus doublet chemotherapy. CHECKMATE‑649 was an international, randomised open-label placebo-controlled trial, that compared nivolumab plus doublet chemotherapy with placebo plus doublet chemotherapy. The trial population was similar to KEYNOTE‑859 but also included people with unknown HER2 status and oesophageal adenocarcinoma. At the time of the original company submission no subgroup data was available for people who had pembrolizumab plus doublet chemotherapy and whose tumours express PD‑L1 with a CPS of 5 or more (from now, the CPS of 5 or more subgroup). This is the population for whom nivolumab plus chemotherapy is a treatment option. The company provided an indirect comparison in people whose tumours express PD‑L1 with a CPS of 10 or more (from now, the CPS of 10 or more subgroup), making the case that it expected the treatment effect in this group to be the same as the CPS of 5 or more subgroup. At the clarification stage, the company provided a post-hoc analysis for the CPS of 5 or more subgroup. But it did not consider the results to be statistically valid because this was not a pre-specified subgroup in KEYNOTE‑859. It explained that the trial was stratified by having a positive PD‑L1 CPS score (the CPS of 1 or more subgroup) or a CPS score of less than 1. It added that the trial was powered to detect any statistically significant differences between treatment arms in the CPS of 10 or more subgroup, which was a pre-specified subgroup. The committee noted that the CPS of 5 or more subgroup included more people than the CPS of 10 or more subgroup. So, it would expect the CPS of 5 or more subgroup to also be powered to detect any statistically significant differences between trial arms. The committee noted the EAG's concerns that published data on baseline characteristics was not available for CHECKMATE‑649. So it was not possible to determine whether there were any differences in the baseline characteristics of people in each subgroup between the 2 clinical trials, which may have biased the results. The expert also highlighted that comparing pembrolizumab with nivolumab across CPS subgroups can be difficult because of different PD‑L1 testing methods used in clinical practice. The test used to determine if a person can be treated with pembrolizumab is different to the test used for nivolumab and CPS scores are not equivalent across tests. The company used constant hazard ratios for the indirect comparison which assumes the proportional hazards assumption is met. The EAG commented that this approach was inconsistent with the company's proportional hazards test for the CPS of 1 or more subgroup and CPS of 5 or more subgroup. The company agreed with the EAG that using time-varying hazard ratios would have been more appropriate. The committee agreed that using time-varying hazard ratios would be its preferred method but that it was possible to reach a conclusion on the analyses using the company's approach. There were no statistically significant differences in overall survival or progression-free survival between pembrolizumab plus doublet chemotherapy and nivolumab plus doublet chemotherapy for the CPS of 1 or more subgroup or the CPS of 5 or more subgroup. There was no statistically significant difference in overall survival for the CPS of 10 or more subgroup. Progression-free survival was not reported for the CPS of 10 or more subgroup in CHECKMATE‑649, so no comparison could be made. The company considers the exact results of the NMA to be confidential, so they cannot be reported here. Overall, the committee agreed with the company and EAG that the results were consistent across the subgroups. Taking into account any potential differences between the trials, the committee concluded that pembrolizumab plus doublet chemotherapy and nivolumab plus doublet chemotherapy are similarly effective at treating HER2‑negative advanced gastric or GOJ adenocarcinoma in people whose tumours express PD‑L1 with a CPS of 5 or more.

The company suggested that any observed differences in the adverse event profiles for pembrolizumab and nivolumab from the trials are explained by the difference in their concomitant doublet chemotherapy. The clinical expert confirmed this assumption, noting that adverse events related to immunotherapy are the same between pembrolizumab and nivolumab and are manageable in clinical practice. The company also included a scenario that assumed that pembrolizumab and nivolumab adverse event profiles were equivalent, which had little impact on the incremental cost-effectiveness ratio (ICER). The committee concluded that pembrolizumab and nivolumab were similarly tolerable.

The company's economic model included a 35‑cycle maximum treatment duration for pembrolizumab based on KEYNOTE‑859. The company assumed that there was no treatment effect waning for pembrolizumab plus doublet chemotherapy in its base-case analysis. That is, that the treatment effect of pembrolizumab plus doublet chemotherapy would be maintained after stopping treatment rather than reducing over time (waning). The company noted that there was no clear evidence of treatment effect waning based on the independent estimation of survival curves for the intervention and comparator arms of the clinical trials. The company provided a scenario that applied gradual treatment effect waning 7 years after starting pembrolizumab plus doublet chemotherapy, that reduced to the same as the comparator arm over the next 2 years. The EAG commented that it was not reasonable to assume a lifetime treatment effect after pembrolizumab plus doublet chemotherapy has stopped. The EAG preferred to apply gradual treatment effect waning 5 years after starting treatment that reduced to the same as the comparator arm over the next 2 years. The company's and EAG's chosen timepoints for when treatment effect waning begins were influenced by treatment effect waning assumptions from previous technology appraisals. This included NICE's technology appraisal guidance on nivolumab with platinum- and fluoropyrimidine-based chemotherapy for untreated HER2-negative advanced gastric, gastro-oesophageal junction or oesophageal adenocarcinoma. In the previous appraisal, the company preferred a treatment waning effect starting 6.5 years after starting treatment with nivolumab plus doublet chemotherapy, whereas the EAG preferred a treatment waning effect starting 5 years after starting treatment. For both assumptions, the hazard of dying became the same as the comparator arm at the point of treatment waning. The committee heard from the clinical expert that the available follow-up data is for less than 5 years. So there is no data to demonstrate whether the treatment effect of pembrolizumab plus doublet chemotherapy is maintained in the longer term or not. They added that, for the 10 to 15% of people who have a complete response, treatment effect waning would not be expected. But, for people who have not had a response within 3 to 6 months, treatment effect waning would be expected as they would have moved on to less clinically effective follow-on treatments. The committee concluded that it was appropriate to apply treatment effect waning for pembrolizumab. It agreed that treatment effect waning starting either 5 years or 7 years after starting treatment and reducing to the same as the comparator after 2 years, were both plausible.

In the company model, the costs for people having doublet chemotherapy were capped at 6 cycles in line with what the company reported as NHS clinical practice. In KEYNOTE‑859, some people had doublet chemotherapy treatments for longer than 6 cycles. The EAG noted that capping doublet chemotherapy costs at 6 cycles does not account for the fact that overall survival and progression-free survival in both treatment arms were based on some people having some doublet chemotherapies for more than 6 cycles. The EAG noted that overall survival and progression-free survival in KEYNOTE‑859 may have been higher than what would be observed in clinical practice. Clinical experts explained that there isn't a cap on doublet chemotherapy in clinical practice although treatment beyond 6 cycles would be rare. This is because the treatment effect of doublet chemotherapy past 6 cycles is modest. So clinicians will aim to prescribe doublet chemotherapy for the shortest course possible to give a response without toxicity. The clinical experts also noted that the number of cycles could be influenced by the adverse event profiles of different doublet chemotherapy combinations and the use of concomitant immunotherapies. The company stated that a scenario in which the cap on chemotherapy cycles was not applied had a minimal effect on the cost-effectiveness results. The committee concluded that applying a cap of 6 cycles on the costs of doublet chemotherapy in the model was appropriate. So the company's method for modelling survival estimates reflected what would be expected in NHS practice.

Utility values were estimated using EQ‑5D data from the KEYNOTE‑859 trial using 2 different approaches. In its base case the company used a time-to-death approach and presented a health state approach in scenario analyses. The time-to-death approach estimates utilities using time intervals that describe life expectancy rather than utility values associated with non-progressed and progressed disease. The company explained that because of the limited collection of assessments with cancer progression in KEYNOTE‑859, health state utilities from the trial data may only reflect quality of life close to the time of cancer progression rather than the entirety of living with progressed cancer. The company considers the exact estimated utility values to be confidential, so they cannot be reported here. The EAG agreed that the time-to-death approach may be more appropriate to capture the quality of life for people with progressed cancer in this evaluation. The EAG also noted that using either a time-to-death or health state approach had a minimal effect on both the cost-effectiveness results and the company's preferred base case. The committee concluded that using a time-to-death approach to estimate utilities based on KEYNOTE‑859 was appropriate for decision making.

The committee considered the severity of the condition (the future health lost by people living with the condition and having standard care in the NHS). The committee may apply a severity modifier (a greater weight to quality-adjusted life years [QALYs]) if technologies are indicated for conditions with a high degree of severity. The company provided absolute and proportional QALY shortfall estimates in line with NICE's health technology evaluations manual. For the CPS of 1 or more subgroup, both the company's and EAG's shortfall analyses suggested that a severity weight of 1.2 should be applied to the QALYs. So the EAG applied a severity weight of 1.2 to the QALYs in its base-case analysis. But the company commented that the current evaluation would have met the end-of-life criteria used in NICE's previous health technology evaluation methods. It added that at the time those methods were applicable, they would have allowed for up to a 1.7 severity weight to be applied to the QALYs. So the company applied a severity weight of 1.7 to the QALYs in its base-case analysis, despite its shortfall analysis suggesting a severity weight of 1.2. The committee was aware that the methods for assessing end-of-life criteria had been replaced with the methods for applying a severity modifier. This was to better reflect society's preferences on the value of treatments for severe conditions. For the CPS of 5 or more subgroup, both the company's and EAG's shortfall analyses suggested a severity weight was not applicable. The committee agreed that current NICE methods for health technology evaluation should be followed. So, the committee concluded that, for the CPS of 1 or more subgroup, a severity weight of 1.2 should be applied to the QALYs. The committee also concluded that, for the CPS of 5 or more subgroup, no severity weight should be applied to the QALYs.