Nivolumab with fluoropyrimidine- and platinum-based chemotherapy for untreated unresectable advanced, recurrent, or metastatic oesophageal squamous cell carcinoma

3.1 The patient experts explained that there are no screening tools to identify oesophageal squamous cell carcinoma and often the condition is diagnosed at an advanced stage. They explained that people with the condition can struggle to maintain their fitness and manage their condition. The patient experts described that the condition and treatment side effects massively affect patients' quality of life, social experience and relationships with family and carers. They highlighted that advanced oesophageal squamous cell carcinoma is less survivable than other cancers.

3.2 NICE's technology appraisal guidance recommends pembrolizumab with platinum- and fluoropyrimidine-based chemotherapy for untreated advanced oesophageal and gastro-oesophageal junction cancer (TA737) in adults whose tumours express PD‑L1 with a combined positive score (CPS) of 10 or more. The clinical experts explained that pembrolizumab plus chemotherapy is widely used when it is suitable. However, patient experts were aware that some people do not have pembrolizumab plus chemotherapy despite its suitability. The clinical and patient experts agreed there is an unmet need in people for whom treatment with pembrolizumab plus chemotherapy is not suitable. The clinical experts explained that clinicians would value an additional treatment option because there may be circumstances in which nivolumab is preferred over pembrolizumab. The committee concluded that patients and clinicians would welcome a new treatment for untreated unresectable advanced, recurrent, or metastatic oesophageal squamous cell carcinoma.

3.3 Suitability of nivolumab and pembrolizumab is assessed by measuring PD‑L1 status. The way that PD‑L1 positivity was measured and defined differed in the marketing authorisations for nivolumab and pembrolizumab. In the CheckMate-648 trial around half of participants in the intention-to-treat (ITT) population had tumour cell PD‑L1 expression with a level of 1% or more in line with nivolumab's marketing authorisation (measure of PD‑L1 expression on tumour cells as a ratio percentage). Pembrolizumab was also a suitable option for more than half of these participants because their tumours expressed PD‑L1 with a CPS of 10 or more, in line with its marketing authorisation (measure of PD‑L1 expression on tumour cells and immune cells as a ratio score). This means that both nivolumab and pembrolizumab were suitable for some people. In the pembrolizumab clinical trial (KEYNOTE-590), 51% of people had tumours that expressed PD‑L1 with a CPS of 10 or more. PD‑L1 status defined by tumour cell expression was not reported. The clinical experts highlighted that there is no clear clinical justification for attributing each specific test to either nivolumab or pembrolizumab. The committee acknowledged the overlap and differences between the 2 tests and the complexity it introduces. The company explained that, in clinical practice, treatment decisions for oesophageal squamous cell carcinoma might vary at different hospitals. It also explained they will not be based on tumour cell expression and CPS collectively because the 2 tests for assessing PD‑L1 are independent of each other. The committee was aware that testing is time and resource intensive. The clinical experts explained that it was preferrable that the 2 tests (tumour cell expression and CPS) would be done concurrently, rather than sequentially, to determine whether nivolumab or pembrolizumab is suitable and permissible. However, they highlighted that treatment decisions are often guided by local health service protocols which may use different approaches. The clinical experts did not agree on whether sequential testing would occur in practice. Clinical experts noted that testing tumour cell PD‑L1 expression may be more accessible to clinicians than testing PD‑L1 expression through CPS. This is because most hospitals can do the tumour cell test but few hospitals have facilities to do the CPS test. This means CPS test results can take 2 weeks or more when external testing is needed. Clinical experts suggested that because tumour cell testing is more accessible than CPS testing, clinicians are more likely to request it first when testing sequentially. The clinical and patient experts were concerned that the waiting time for CPS results and requirement for concurrent testing for nivolumab and pembrolizumab suitability could delay starting a suitable immunotherapy. They highlighted that a delay could disadvantage people, because it is better to start treatment as soon as possible so that the immune response associated with immunotherapies is generated as early as possible. There was uncertainty on whether both tests would be done sequentially or concurrently. The committee heard that in clinical practice there were implementation issues for pembrolizumab (TA737; see section 3.4). However, the committee's remit is to compare the clinical and cost effectiveness of new interventions with currently available treatments for the same indication. This is to ensure patients and the NHS access the most effective, best value treatments. So, the committee's evaluation of nivolumab should include a NICE-recommended cost-effective comparator treatment which was available and used in NHS practice such as pembrolizumab. However, the committee noted the implementation issues for pembrolizumab and strongly concluded that tests for pembrolizumab and nivolumab suitability should be done concurrently. This is to minimise unnecessary delays in accessing treatment.

3.4 The company maintained that pembrolizumab plus chemotherapy was recommended too recently to be considered standard of care. It considered chemotherapy to be the main comparator for nivolumab plus chemotherapy. The clinical experts mentioned that in their centres they had observed slower uptake of pembrolizumab than anticipated. They attributed this to testing capacity issues associated with the COVID‑19 pandemic and differences in how diagnostic services do CPS tests. The Cancer Drugs Fund (CDF) lead stated that use of pembrolizumab plus chemotherapy had been increasing slowly, which is not unusual when there is a test needed for treatment. Despite differing opinions on the uptake of pembrolizumab, the clinical experts and CDF lead agreed that pembrolizumab should be considered as a comparator in this appraisal. The committee concluded that:

3.5 CheckMate 648 is a randomised, three-arm, open-label, placebo-controlled trial (n=970). It compared cisplatin and fluorouracil, with or without nivolumab, and nivolumab plus ipilimumab, as first-line treatment for unresectable advanced, recurrent or metastatic oesophageal cancer. The nivolumab plus ipilimumab and chemotherapy trial arm was not included in the decision problem. The primary outcomes were progression-free survival and overall survival in the ITT population (n=645 for the nivolumab plus chemotherapy and chemotherapy trial arms). Progression-free survival and overall survival were reported for people with squamous cell oesophageal carcinoma whose tumour cells expressed PD‑L1 at 1% or more (n=315). The committee concluded that the data from this subgroup is appropriate for decision making when only nivolumab is suitable (see section 3.4).

3.6 Data on progression-free survival and overall survival outcomes from the CheckMate 648 subgroup (see section 3.5) are academic in confidence and cannot be presented here. The committee concluded that nivolumab plus chemotherapy improves both progression-free survival and overall survival compared with chemotherapy alone.

3.7 Nivolumab plus chemotherapy was indirectly compared with pembrolizumab plus chemotherapy in the absence of a direct trial comparison. Data for pembrolizumab came from the KEYNOTE‑590 trial. There was no efficacy data from the KEYNOTE‑590 population in whom nivolumab plus chemotherapy was suitable (tumour cell PD‑L1 expression of 1% or more). The indirect treatment comparison (ITC) used outcomes from the population for whom pembrolizumab plus chemotherapy was suitable (PD‑L1 expression with a CPS of 10 or more). Progression-free survival and overall survival outcome data from CheckMate 648 and KEYNOTE‑590 were indirectly compared. In KEYNOTE‑590, progression-free survival data was restricted to oesophageal cancer with squamous cell carcinoma or adenocarcinoma histologies, and overall survival data was restricted to squamous cell histology. A comparability assessment between the 2 trials showed that CheckMate 648 included more people with an Asian family background and fewer people had metastatic disease. The company compared some baseline characteristics from people in both trials whose tumours had PD‑L1 expression with a CPS of 10 or more. The ERG agreed that this subgroup appeared comparable to the ITT populations in the trials. However, it was noted that conclusions around trial comparability are limited because the only characteristics presented were age, Asian family background, Eastern Cooperative Oncology Group status and metastatic disease. Also, these characteristics were only available as an average of both trial arms in KEYNOTE‑590 (pembrolizumab plus chemotherapy and chemotherapy alone).

One clinical expert highlighted that people in the 2 trials are likely to be comparable, based on the overall survival that was observed in the control arms of both trials. They also said there is little clinical understanding of how outcomes may differ by how PD‑L1 expression was measured (tumour cell expression or CPS). Furthermore, they stated there is also uncertainty in how outcomes may differ in the broader oesophageal cancer population and in tumours which express PD‑L1. Another clinical expert commented that while CheckMate 648 included only people with oesophageal squamous cell carcinoma, KEYNOTE‑590 included people with squamous cell carcinoma and adenocarcinoma, located in either the oesophagus or the gastroesophageal junction. This made it difficult to reach any conclusions about trial comparability. The committee noted there was uncertainty about the comparability of the 2 trials used in the ITC.

3.8 The company explored various ITC analyses to estimate the relative treatment effect of nivolumab plus chemotherapy versus pembrolizumab plus chemotherapy in terms of progression-free survival and overall survival. The company's analyses included people from CheckMate 648 for whom both nivolumab and pembrolizumab were suitable (see section 3.4). This generated time-varying progression-free survival hazard ratios (HRs) which favoured pembrolizumab plus chemotherapy from 3 months. The overall survival HRs favoured nivolumab plus chemotherapy after 6 months. The ERG's base-case analyses included people from CheckMate 648 for whom pembrolizumab was suitable (see section 3.4). These generated time-varying progression-free survival and overall survival HRs which favoured pembrolizumab plus chemotherapy over all time points. The ERG favoured including people for whom pembrolizumab was suitable, to maintain maximum trial comparability because the information on suitability of nivolumab in KEYNOTE‑590 was not available (see section 3.3 and section 3.7). The company and ERG base-case analyses also differed in the distributions used to extrapolate survival and which treatments were used as the baseline to scale pembrolizumab survival estimates. In both analyses, the HR credible intervals were wide and crossed 1. The ERG commented that the methodologies used in the ITC were robust and there were some arguments for using the company's base-case analysis. However, the ERG concluded that it was difficult to determine whether 1 treatment extended progression-free survival or overall survival more than the other.

The clinical experts maintained that the effectiveness of nivolumab and pembrolizumab (in terms of response or survival) is almost the same in other cancers, highlighting an immunotherapy 'class effect'. They expect this effect to be consistent in treating oesophageal squamous cell carcinoma tumours. They further explained that comparing nivolumab and pembrolizumab across different definitions of PD‑L1 positivity and trial datasets was 'risky' in terms of validity. The CDF lead explained that pembrolizumab is administered less frequently than nivolumab, which may provide added benefit to people and also reduce the burden on NHS cancer services because fewer hospital visits would be needed.

The committee acknowledged the complexity of calculating a reliable relative treatment effect in the comparison of nivolumab and pembrolizumab. The clinical experts, company, ERG and committee all agreed that no definitive evidence of superiority of one treatment over the other had been presented.

3.9 The company presented a 3‑state partitioned survival model to estimate the cost effectiveness of nivolumab plus chemotherapy compared with chemotherapy and pembrolizumab plus chemotherapy. The 3 health states were progression-free survival, progressive disease, and death. The ERG agreed that the company's model structure captured all relevant health states and that partitioned survival models are widely used in cancer modelling. The ERG's concerns on the model structure related to the modelling of effectiveness associated with subsequent therapy. The company included a 2‑year stopping rule for nivolumab. Although this is not included in the marketing authorisation, it reflects the use in the trial and expected use in clinical practice, in line with nivolumab use in other cancers. The committee concluded that the company's model structure was acceptable for decision making.

3.10 The company used Kaplan–Meier data from CheckMate 648 to model overall survival in both treatment arms, with a log-normal extrapolation from 6.9 months. The company justified the semi-parametric approach stating it better reflects the changing hazard observed after 20 months in both trial arms and noted a clear inflection point for the chemotherapy arm smoothed hazard plot at around 6 months. The ERG did not agree with the company's approach and used a parametric log-logistic extrapolation in their base case. It justified the parametric extrapolation by noting no clear inflection point for the nivolumab plus chemotherapy smoothed hazard plot. It also found a reasonable correspondence of the landmark overall survival observed in CheckMate 648 with the parametric overall survival analysis. The committee heard that a semi-parametric extrapolation of overall survival was broadly accepted by the committee in TA737. The committee acknowledged there was uncertainty in the long-term overall survival and there were arguments for each approach.

3.11 There was evidence of treatment switching in CheckMate 648, with a proportion of both arms who had subsequent systemic and anti-PD‑L1 therapies (including nivolumab and pembrolizumab). The company considered that a decreasing hazard rate (mortality rate) observed in the chemotherapy arm of CheckMate 648 was implausible. The ERG argued that it might be partly because of in-trial switching to an anti-PD‑L1 therapy. The ERG also noted that, although only a fraction of people in the trial switched to an anti-PD‑L1 therapy, the company assumed in its economic model that all people who had subsequent treatment in clinical practice would have an anti-PD‑L1 therapy. The ERG suggested that this might lead to a bias where the effectiveness of chemotherapy had been underestimated in the economic model. To overcome the bias in results, it preferred an adjustment of overall survival in the model using methods outlined in NICE's Decision Support Unit technical support document 16. The company elected against including a switching adjustment in their base-case modelling of overall survival, commenting that few people switched to an anti-PD‑L1 therapy in the trial. It also explained that the switching adjustment methods in the technical support document place large demands on limited data which can create more uncertainty. The company presented 2 switching adjustment scenario analyses: a rank-preserving structural failure time model and a model-based approach. As the incremental cost-effectiveness ratio (ICER) of nivolumab plus chemotherapy versus chemotherapy increased markedly in these scenarios, the ERG maintained that the ICER is underestimated without a switching adjustment. The committee heard that switching was not included in TA737. While the ERG acknowledged this, it again referred to the clinical evidence that it said demonstrated switching and the difference between the proportion switching in the trial and that which would occur in clinical practice. A clinical expert explained that switching to an anti-PD‑L1 therapy may introduce some bias, but it was not expected to be impactful on overall survival. The committee concluded that the switching may have an impact on the overall survival in both trial arms, but that its overall impact was uncertain.

3.12 In its base case, the company assumed there was no waning of treatment effect after stopping nivolumab. It commented that there was evidence of a robust and durable treatment effect lasting beyond discontinuation for immunotherapies. It stated that if any treatment waning was to occur it would be 5 years after starting therapy. It referenced NICE's technology appraisal of nivolumab for untreated HER2-negative advanced gastric, gastro-oesophageal junction or oesophageal adenocarcinoma in which the committee considered a 5‑year assumption of treatment waning to be plausible. The ERG disagreed with this assumption and included the waning of treatment effect on overall survival for nivolumab from 2.5 years to 4.0 years after starting therapy and 6 months after stopping therapy. The ERG justified the assumptions, explaining that for some overall survival parametric functions the nivolumab plus chemotherapy treatment effect versus chemotherapy alone increased over time, which was considered implausible. Additionally, the ERG highlighted that treatment waning assumptions were not disregarded by the committee in TA737. However, it noted that no conclusions about the appropriateness of waning were made by the committee. The company commented that there was no clear justification for the 2.5 years to 4.0 years treatment waning assumptions preferred by the ERG.

The clinical experts commented that some treatment waning is possible and it is more likely when nivolumab plus chemotherapy is discontinued early. The committee considered that based on the maturity of the data in CheckMate 648, the treatment effect may be captured within the observed results making an adjustment of the extrapolation of overall survival for treatment waning unnecessary.

The committee acknowledged that the treatment waning effect of pembrolizumab was discussed in TA737 and that no conclusions on the appropriateness of waning were stated in the final guidance. It concluded that assuming a lifelong treatment effect of nivolumab may be over optimistic and some treatment waning might be expected. However, there was not enough evidence to be precise about when this would occur. The committee concluded that it would consider scenarios with and without treatment waning in its decision making, only if they impact the final decision on cost effectiveness.

3.13 The reduction in the risk of death in CheckMate 648 was similar to that of background mortality. The company and the ERG agreed that mortality should not be any lower than background (all-cause) mortality. To enforce a plausible minimum rate of mortality, a background mortality rate was included in the model, additional to the predicted overall survival. The company maintained that the approach resulted in very little difference in the overall survival. However, the ERG considered this approach to double-count mortality in the model and preferred the use of alternative methods which prevent implausibly low mortality hazard with any overall survival extrapolation. The committee acknowledged the justifications for each modelling approach but did not conclude which was preferable because each assumption had a minimal impact on the cost-effectiveness estimates.

3.14 Utility values in the model were calculated using EQ‑5D data from CheckMate 648. The company attributed utilities to the progression-based health states in its base case, using the average utility reported in the 2 trial arms. The ERG explained that because the EQ‑5D scores reported were consistently higher for chemotherapy compared with nivolumab plus chemotherapy, treatment-specific utilities were more appropriate. The clinical experts suggested toxicity associated with nivolumab therapy may explain lower utility values compared with chemotherapy alone. The clinical experts stated that they would expect the opposite in practice, with people who had nivolumab plus chemotherapy likely to have higher utility because of better disease control. The CDF lead and clinical experts did not agree with using the treatment-specific utilities. This is based on their experience that health-related quality of life disutilities associated with anti-PD‑(L)‑1 immunotherapies toxicity would be outweighed by the health benefit derived from response to therapy. Because treatment-specific utilities are clinically plausible, the committee concluded that treatment-specific utilities were not appropriate.

3.15 A proportion of people in CheckMate 648 delayed or missed doses of their trial treatment. In its base-case analysis, the company adjusted treatment costs based on the mean relative dose intensity observed in the trial. These treatment costs were then weighted by the time each patient spent on treatment. The ERG explained that time on therapy was not relevant for capturing the relative dose intensity and omitted this from its base case. The company stated that the treatment discontinuation biased the treatment cost calculations. The committee noted uncertainty in both approaches taken.

3.16 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. It considered whether nivolumab plus chemotherapy met the end of life criteria compared with chemotherapy for people in whom pembrolizumab is unsuitable. The company and ERG both agreed that data suggests that life expectancy in this population is less than 24 months. The observed median overall survival benefit with nivolumab plus chemotherapy in CheckMate 648 was larger than the additional 3‑month extension to life needed by the criteria (data are academic in confidence and cannot be presented here). The committee considered whether nivolumab plus chemotherapy met the end of life criteria compared with pembrolizumab plus chemotherapy for people in whom pembrolizumab is suitable. The company and ERG both agreed, based on the KEYNOTE‑590 analyses, that life expectancy in this population is less than 24 months. Based on the ITC analyses (see section 3.8), the median overall survival benefit of nivolumab plus chemotherapy compared with pembrolizumab plus chemotherapy did not exceed the additional 3‑month extension to life needed by the criteria (the data cannot be reported here because the company submitted it as academic in confidence). The committee concluded that nivolumab met the end of life criteria compared with chemotherapy but that criteria were not met compared with pembrolizumab plus chemotherapy.

3.17 The company's base-case comparison of nivolumab with pembrolizumab included the following assumptions:

3.18 The company's comparison of nivolumab plus chemotherapy with chemotherapy for people in whom pembrolizumab is not suitable used the same assumptions as for the comparison with pembrolizumab plus chemotherapy (see section 3.17). The ERG's base case included the same assumptions as the company in its exclusion of an adjustment on survival for switching (see section 3.11). The ERG's assumptions differed from the company in the approach to modelling progression-free survival and overall survival, and parametric extrapolation was modelled with a waning of nivolumab treatment effect between 2.5 years and 4 years (see section 3.12). Additionally, treatment-specific progression utilities were included in the model (see section 3.14), and the relative dose intensity was not weighted by time on treatment (see section 3.15). The ICERs cannot be reported here because of confidential commercial arrangements for nivolumab and subsequent treatments. Including the ERG's assumptions increased the ICER compared with the company's base case, however the company's cost-effectiveness estimates are below the end of life threshold of £50,000 per quality-adjusted life year (QALY) gained. While the ERG's cost-effectiveness estimates are above £50,000 per QALY gained, the committee did not agree with the assumption of treatment-specific utilities. Removing this assumption provided cost-effectiveness estimates which were below £50,000 per QALY gained. Because end of life criteria have been met (see section 3.16), the committee concluded that nivolumab plus chemotherapy is likely a cost-effective use of NHS resources when pembrolizumab is unsuitable.

3.19 Patient experts explained that people living in the most deprived areas are more likely to be diagnosed with oesophageal squamous cell carcinoma later than others. The committee noted that the issues surrounding the delays in diagnosis are unable to be addressed in technology appraisal guidance.

3.21 The committee noted that the company's and ERG's base-case conclusions aligned in that nivolumab plus chemotherapy was dominated by pembrolizumab plus chemotherapy, that is higher cost and lower efficacy. The committee acknowledged the uncertainty around relative treatment effects for the 2 treatments, but concluded that nivolumab plus chemotherapy was unlikely to be cost effective compared with pembrolizumab plus chemotherapy.

The committee noted that the company's base case suggested that nivolumab plus chemotherapy is likely to be cost effective compared with chemotherapy, for people for whom pembrolizumab is not suitable. This is when considering nivolumab as a life-extending treatment for people with short life expectancy (see section 3.17). The committee also noted that the ERG's base case suggested that it was unlikely to be cost effective compared with chemotherapy alone, even in the context of a life-extending treatment for people with a short life expectancy (see section 3.16). However, nivolumab was cost effective when treatment-specific utilities, which the committee deemed as inappropriate, were removed from the ERG's analyses and when considering end of life criteria (see section 3.17).

Despite the remaining areas of uncertainty, it was agreed that the cost-effectiveness estimates for nivolumab when pembrolizumab is not suitable are likely to be within the range usually considered a cost-effective use of NHS resources. This is for a life-extending treatment for people with short life expectancy. Therefore, nivolumab with platinum- and fluoropyrimidine-based chemotherapy is recommended for use in the NHS as an option for untreated unresectable advanced, recurrent, or metastatic oesophageal squamous cell carcinoma, in adults whose tumour cells express PD‑L1 at 1% or more. It is recommended only if pembrolizumab plus chemotherapy is unsuitable. The committee recalled the current implementation issues for testing (see section 3.3). It therefore also concluded that testing for suitability of nivolumab and pembrolizumab should be done at the same time.