Pembrolizumab with carboplatin and paclitaxel for untreated metastatic squamous non-small-cell lung cancer

3.1 People with squamous NSCLC often have a poor quality of life, and a potential extension to life is important to them. Outcomes tend to be worse with squamous NSCLC than with non-squamous NSCLC because people have more smoking-related comorbidities. For people with squamous NSCLC whose tumours express PD‑L1 with a tumour proportion score of 0% to 49%, outcomes are particularly poor. This is because the only first-line treatment is platinum-based combination chemotherapy, if it is tolerated. In the original appraisal, the clinical experts explained that for people whose tumours express PD‑L1 at 50% or more, most clinicians would use pembrolizumab monotherapy to avoid chemotherapy toxicity. But they added that some people who need urgent clinical intervention (for example, people with impending major airway obstruction) may benefit from initial combination therapy with pembrolizumab and chemotherapy. The committee concluded that pembrolizumab combination therapy would be a welcome additional treatment option for untreated metastatic squamous NSCLC.

3.2 Treatment for lung cancer is defined by histology (non-squamous or squamous NSCLC) and PD‑L1 tumour proportion score. This is in line with NICE's guideline on lung cancer: diagnosis and management. First-line management of metastatic squamous NSCLC in clinical practice is platinum-based combination chemotherapy (that is, cisplatin or carboplatin and either gemcitabine, paclitaxel or vinorelbine) for people whose tumours express PD‑L1 at less than 50%. In NICE's technology appraisal guidance on pembrolizumab for untreated PD-L1-positive metastatic NSCLC, pembrolizumab monotherapy is recommended only for people whose tumours express PD‑L1 at 50% or more. At consultation, clinical feedback stated that in clinical practice people would be assessed and treatment decisions would be made based on the 3 PD‑L1 tumour proportion scores (that is, PD‑L1 tumour proportion scores of less than 1%,1% to 49% and 50% or more). The committee was aware of the different treatment options for people whose tumours express PD‑L1 at different levels (see section 3.1). It was satisfied that prognosis may be dependent on the 3 PD‑L1 subgroups but that treatment options may not be. It concluded that subgroups based on PD‑L1 tumour proportion scores of 0% to 49% and 50% or more should be considered separately.

3.3 Pembrolizumab monotherapy is the standard first-line treatment option for people whose cancer has a PD‑L1 tumour proportion score of 50% or more. The company, in its original submission, did an indirect treatment comparison of pembrolizumab combination therapy and pembrolizumab monotherapy using data from KEYNOTE‑407 and KEYNOTE‑042. KEYNOTE‑042 was a trial including 1,274 people with PD‑L1‑positive tumours that compared pembrolizumab monotherapy with platinum-based chemotherapy. After consultation the company clarified that it was seeking continued access to pembrolizumab combination only for those groups that had access in the Cancer Drugs Fund. That is,

3.4 The main clinical evidence for pembrolizumab combination therapy came from KEYNOTE‑407, a randomised placebo-controlled trial. It included 559 adults with untreated advanced or metastatic squamous NSCLC with an Eastern Cooperative Oncology Group performance status of 0 or 1. Pembrolizumab combination therapy was compared with placebo plus carboplatin and paclitaxel or nab-paclitaxel (from now, standard chemotherapy) as a first-line treatment. In NHS clinical practice, carboplatin plus gemcitabine is the most commonly used chemotherapy regimen for people whose tumours express PD‑L1 at less than 50%. At technical engagement, the company provided additional overall-survival data from a later follow up of KEYNOTE‑407 (data cut September 2020). Median overall survival was 17.2 months for pembrolizumab combination therapy and 11.6 months for standard chemotherapy (hazard ratio [HR] 0.71, 95% confidence interval [CI] 0.59 to 0.86). Median progression-free survival was 8.0 months for pembrolizumab combination therapy and 5.1 months for standard chemotherapy (HR 0.59, 95% CI 0.49 to 0.71). The committee agreed that overall- and progression-free survival data from the final analysis and additional data cuts were more mature than those from the interim analysis used in the original appraisal. It recognised that pembrolizumab combination therapy improved overall and progression-free survival compared with standard chemotherapy in the intention‑to‑treat population. However, it acknowledged that the results are not generalisable to clinical practice. This was because the treatments used in the study were different to those used in the NHS, depending on the PD‑L1 tumour proportion score. The committee concluded that the clinical results used in the intention-to-treat population did not reflect clinical practice. It further concluded that decisions about clinical effectiveness should be based on PD‑L1 status (that is, PD‑L1 tumour proportion scores of 0% to 49% and 50% or more in those who need urgent clinical intervention).

3.5 The company presented clinical-effectiveness results for the PD‑L1 subgroups in its submission. The committee was aware that, in the protocol for KEYNOTE‑407, people were stratified to treatment arms by a PD‑L1 tumour proportion score of at least 1% and less than 1%. However, people were enrolled regardless of PD‑L1 status and were spread across 3 PD‑L1 tumour proportion score subgroups (less than 1%, 1% to 49%, and 50% or more). The final analysis of KEYNOTE‑407 (data cut May 2019) showed a reduction in risk of death of:

3.6 The company's economic analysis was based on the ERG's preferred analysis that was used in the original appraisal. This used the more conservative clinical estimates around progression-free survival as agreed by the committee (termed the 'ERG's pessimistic analysis 6b'). In its submission to the Cancer Drugs Fund review, the company made several changes to its updated model. These included the following amendments, which it applied to the May 2019 data cut of KEYNOTE‑407 and updated to the September 2020 cut after consultation:

3.7 The committee recalled that its preference was to consider the PD‑L1 subgroups based on those with tumour proportion scores of 0% to 49% and those with 50% or more who need urgent clinical intervention (see section 3.2 and section 3.3). After consultation the company agreed this was appropriate. The committee was aware that because the KEYNOTE-407 trial did not include a prespecified subgroup of people whose tumours express PD‑L1 at less than 50%, there was no evidence on the effectiveness of pembrolizumab combination therapy in this group. Therefore, the ERG weighted the life years gained, quality-adjusted life years (QALYs) and costs from the company's model with the proportions from the KEYNOTE-407 groups with PD‑L1 expression of less than 1% and 1% to 49%. After the first committee meeting, the company noted that recent prescribing data had identified differences in the proportions using pembrolizumab combination therapy in the trial compared with those in NHS clinical practice. It suggested that the market research prescribing data was the most accurate source to weight the 0% to 49% PD‑L1 subgroup. This showed that pembrolizumab combination therapy was used in NHS clinical practice by 22% of those with PD‑L1 tumour proportion scores of less than 1% and 68% of those with tumour proportion scores of 1% to 49%. In KEYNOTE-40, 35.5% of patients receiving pembrolizumab combination therapy had a PD-L1 tumour proportion scores less than 1% and 37.8% had tumour proportion scores of 1% to 49%. At the second committee meeting the clinical lead for the Cancer Drugs Fund explained that 48% of those with PD‑L1 tumour proportion scores of less than 1% and 41% of those with tumour proportion scores of 1% to 49% had accessed pembrolizumab combination therapy in the Cancer Drugs Fund. The committee considered the 3 data sources and agreed that the Cancer Drugs Fund was the most accurate source of prescribing data because it was based upon real-world use of pembrolizumab combination therapy. This data showed the total number of people who had received treatment with pembrolizumab combination therapy for metastatic squamous NSCLC in the NHS. It concluded the Cancer Drugs Fund prescribing data is the most appropriate source to weight the 0% to 49% PD‑L1 subgroup.

3.8 In the original appraisal, the company and the ERG used various modelling approaches to estimate long-term survival in the pembrolizumab combination therapy and the comparator arms. The company fitted a hybrid model using Kaplan–Meier data from the interim analysis of KEYNOTE‑407 and additional data from the Surveillance, Epidemiology, and End Results (SEER) database. The committee had concluded that the company's modelled overall survival in the pembrolizumab combination therapy arm was too optimistic. It preferred the ERG's log-logistic extrapolation in each treatment arm with no cut-off points. This was because data in the SEER database had not included second-line immunotherapy treatments. In its submission to the Cancer Drugs Fund review, the company fitted a log-logistic model to the May 2019 data cut of KEYNOTE‑407 with no cut-off points to the data for each treatment group. The company justified this choice because it was in line with the committee's preferred analyses from the original appraisal. It also had one of the best goodness-of-fit statistics, and the most clinically plausible 5‑year and 10‑year overall-survival estimates. The committee noticed that, overall, the Weibull extrapolation model had one of the best statistical fits. However, it considered that the differences were marginal. For progression-free survival, the company fitted the same hybrid model it used in the original appraisal, updated with the May 2019 data cut off. At the first committee meeting the company provided additional survival follow-up data (based on a cut-off date of September 2020) to reinforce the overall- and progression-free survival estimates. The company did not initially incorporate the Kaplan–Meier estimates using the most recent data cut-off plots into its economic model. The ERG was able to provide these updated overall-survival extrapolations using the latest cut-off data. However, it was not able to do so to update progression-free survival because that had been based on a hybrid model, which the company had not updated. The ERG carried out additional sensitivity analyses using alternative parametric survival models. The committee recognised that all the estimates were subject to uncertainty. It also noted that using the alternative parametric survival models did influence the incremental cost-effectiveness ratio (ICER) for pembrolizumab combination therapy. The committee noted that, with the exception of the log-normal extrapolation model, the ICER would increase if any other extrapolation model was used to model both overall and progression-free survival. However, the ICER was less sensitive to the choice of progression-free survival model. After consultation the company updated its economic model for the 3 PD‑L1 subgroups. It used the September 2020 data cut from KEYNOTE-407, justifying its choice of extrapolation model for the PD‑L1 subgroups with tumour proportion scores of less than 1%, 1% to 49%, and 50% or more based on goodness-of-fit assessments. The ERG noted that the company's consultation response had not considered the plausibility of the chosen model. However, at the second committee meeting, the company clarified that the clinical plausibility of the extrapolation models had been considered in its original submission during the committee's decision making for NICE's original technology appraisal for pembrolizumab combination therapy with carboplatin and paclitaxel for untreated metastatic squamous NSCLC. The committee concluded the company's choice of parametric models to extrapolate overall and progression-free survival was appropriate for decision making.

3.9 The company's updated economic model submitted to the Cancer Drugs Fund review assumed that the costs of subsequent immunotherapies applied to everyone having standard care and subsequent-line treatment. This was in line with the committee's preferred assumptions from the original appraisal about subsequent-line immunotherapy in the standard chemotherapy group. But, the ERG noted that this was inconsistent with the experience of people in KEYNOTE‑407, in which a few people had chemotherapy alone as subsequent treatment. It also noted that the model overestimated the costs of second-line immunotherapy in the standard care group, which would underestimate the ICER for pembrolizumab combination therapy. The ERG used an alternative approach in its preferred base case, in which the costs of chemotherapy were only applied to people who had subsequent-line treatment. This included the people in KEYNOTE‑407 who had subsequent chemotherapy. The clinical experts explained that this did not reflect clinical practice. The committee noted that, although including costs for subsequent-line chemotherapy differed from usual clinical practice, it preferred the consistent approach used by the ERG. At consultation, the company recognised there were problems with each approach taken. It considered that using KEYNOTE-407 data on subsequent treatments might undervalue the cost-effectiveness of pembrolizumab combination therapy. But, it stated, it would accept using the 'within-trial' approach adopted by the ERG. The committee agreed that the costs of subsequent treatment included in the economic model should reflect the treatments in KEYNOTE‑407.

3.10 In the original appraisal, the committee's preferred analysis included a 2‑year treatment stopping rule and a treatment effect lasting between 3 years and 5 years. In its submission to the Cancer Drugs Fund review, the company proposed a 5‑year duration of treatment benefit for overall survival in its base case. It suggested that there was no direct evidence to support the suggestion that the treatment benefit will wane 5 years after stopping treatment. However, it chose this for consistency with previous immunotherapies. The company included scenario analyses exploring the effects of 3‑year and 4‑year durations of treatment effect. The ERG noted that the waning of treatment effect in the company's base case had only been applied to progression-free survival in the subgroup of people with PD‑L1 tumour proportion scores of 50% or more. So, it applied the waning of treatment effect to both overall and progression-free survival in its preferred base case. The committee noted that including waning of treatment effect on progression-free survival for the intention-to-treat population did not have a large effect on the ICER. At its first meeting, the committee considered the company's and ERG's preferred assumptions alongside decisions made in previous appraisals of immunotherapies and agreed there was no new evidence presented to change its position from the original appraisal. So, it concluded that, for consistency with previous appraisals of immunotherapies for NSCLC, a treatment effect lasting between 3 years and 5 years after starting treatment was appropriate for decision making. In its response to consultation, the company stated its new data cut from September 2020 of KEYNOTE-407, showing a follow up beyond 3 years did not support the committee's preference on waning of treatment effect. The company updated its economic analyses to include a treatment effect for overall survival up to 5 years but assumed a lifetime treatment effect for progression-free survival. The ERG noted the Kaplan–Meier data from the September 2020 data cut -off of KEYNOTE-407 and suggested there may be high levels of censoring and both treatment groups had few overall-survival events at later time points. The data did not show whether the treatment effect would continue for longer than 4 years. Although alternative analyses such as using empirical hazard functions and log cumulative hazards may have assessed whether the treatment effects persisted longer term, these had not been used. The committee agreed the longer-term data of KEYNOTE-407 showed the treatment effect of pembrolizumab combination therapy probably did not wane before 4 years. It was satisfied with the company's assumption for overall survival but it was unclear about the company's rationale about whether the treatment effect for progression-free survival would persist for a lifetime duration. It considered that treatment waning is usually an assumption that is applied when the available data is immature, and so cannot show what happens when people stop having treatment. However, it had agreed the KEYNOTE-407 was now more mature and it was reasonable to assume that the treatment effect for progression-free survival would follow a similar trajectory to overall survival. For this reason the committee concluded that a treatment effect lasting 5 years for both overall and progression-free survival is appropriate for decision making.

3.11 The company considered various survival-extrapolation models fitted to the final data cut of KEYNOTE-407 to model time to stopping treatment for the pembrolizumab combination arm. It chose to use the generalised gamma distribution in its base case, based on goodness-of-fit statistics. The company did not fit parametric models for the comparator group. But, in line with the company model for the original appraisal, it used Kaplan–Meier estimates based on the observed cumulative probabilities of staying on treatment. These were taken from the final data cut of KEYNOTE‑407. The ERG had no concerns with this approach for the standard care group. But, it considered the extrapolation modelling approach used for the pembrolizumab arm did not fit the data well. It chose to use the cumulative probabilities from the Kaplan–Meier estimates in its preferred base-case analysis. The committee agreed that the methods used to model time to stopping treatment in the ERG's base case were preferable to fitting survival models that did not represent the data well. For this reason, the committee concluded that time to stopping treatment for both treatment groups should be modelled using cumulative probabilities from the Kaplan–Meier estimates.

3.12 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. This states that a NICE technology appraisal committee should satisfy itself that all of the following criteria have been met:

3.13 For the short-life-expectancy criterion, the company noted that the most recent data from KEYNOTE‑407 (data cut September 2020) reported a median overall survival of 11.6 months for the standard care arm for the intention-to-treat population. The company model predicted a mean overall survival with standard care of 27.1 months. Based on the most recent overall-survival data, it noted that its economic model predicted that only 28.5% of people in the intention-to-treat population having standard chemotherapy would be alive at 24 months. The company noted that survival with current therapies in NICE's ongoing technology appraisal for nivolumab with ipilimumab and chemotherapy for untreated metastatic NSCLC is less than 24 months for people with squamous histology and a PD‑L1 tumour proportion score of less than 50%. This had been supported by the clinical experts involved in that technology appraisal, who confirmed that treatment options in the NHS would vary by histology and PD‑L1 status. At the second committee meeting, the company accepted that NICE's end of life criteria should be considered based on the PD‑L1 subgroups of 0% to 49% and 50% or more in those who need urgent clinical intervention. The committee noted that the company's updated base case using the September 2020 data cut and weighted using the percentage usage from the Cancer Drugs Fund, suggested that people on standard chemotherapy would live approximately 2 years. It concluded that the short-life-expectancy criterion was likely met for both subgroups.

3.14 NICE's guide to the methods of technology appraisal notes that above a most plausible ICER of £20,000 per QALY gained, judgements about the acceptability of a technology as an effective use of NHS resources will take into account the degree of certainty around the ICER. The committee's preferred assumptions included:

3.15 The committee considered the results of its preferred analysis for the 0% to 49% PD‑L1 subgroup, reweighted using the prescribing data from the Cancer Drugs Fund. The ICER was below £50,000 per QALY gained. The committee concluded that pembrolizumab combination is recommended for routine use in people whose tumours express PD‑L1 with a tumour proportion score of 0% to 49%.

3.16 The committee considered the results of its preferred analysis for the 50% or more PD‑L1 subgroup who need urgent clinical intervention. The ICER was below £50,00 per QALY gained. It noted it had not seen any direct or indirect evidence specifically for the 50% or more PD‑L1 subgroup who need urgent clinical intervention. It considered there was a high unmet need for this group because there is a lack of other treatment options available and because this group are likely to have a very short life expectancies. It considered the ICER was likely to be a cost-effective use of NHS resources, despite the lack of a robust evidence in the subgroup. For this reason, the committee concluded that pembrolizumab combination therapy is recommended for routine use in the NHS for people whose tumours have a PD‑L1 tumour proportion score of 50% or more and who need urgent clinical intervention.

3.17 At consultation, the company highlighted that people with squamous NSCLC could have poorer outcomes because of smoking-related comorbidities. It suggested this was likely to have an impact on people in lower socioeconomic groups because rates of smoking are higher in these populations. The committee noted it had no evidence to suggest outcomes would differ for this group of people compared with people who had other cancers. The committee concluded there are no equalities issues and all relevant benefits of the technology were captured in the QALY.