Ceritinib for previously treated anaplastic lymphoma kinase positive non-small-cell lung cancer

Clinical effectiveness

3.1 The company presented efficacy data from 2 phase I or II single-arm studies identified by a systematic review: ASCEND‑1 and ASCEND‑2. These were multicentre, open-label studies of people with anaplastic lymphoma kinase (ALK) positive locally advanced or metastatic non‑small‑cell lung cancer (NSCLC), whose disease had progressed after chemotherapy. All patients were treated with ceritinib.

3.2 The phase I ASCEND‑1 study (n=304) enrolled people with a range of treatment histories and explored several different doses of ceritinib. All patients had an Eastern Cooperative Oncology Group (ECOG) performance status of 2 or less and a life expectancy of at least 12 weeks. The company's analysis included only the subgroup of 163 adults who had previously been treated with crizotinib and who had the licensed dose of ceritinib (750 mg). This subgroup had ALK‑positive, locally advanced or metastatic NSCLC that had progressed despite standard therapy and a mean age of 51.5 years. People continued treatment with ceritinib until unacceptable toxicity or disease progression, or at the discretion of the investigator, or by patient request.

3.3 The 2 primary outcomes were overall-response rate (defined as complete or partial response using the Response Evaluation Criteria in Solid Tumours [RECIST]) and duration of response, both assessed by the investigator. The secondary outcomes included overall-response rate assessed by a blinded independent review committee rather than by the investigator, overall survival, progression-free survival (defined as the time from starting treatment to the time of disease progression or death), and adverse events.

3.4 The phase II ASCEND‑2 study enrolled 140 patients previously treated with crizotinib. The mean age of patients was 51.2 years. It included adults:

3.5 The primary outcome was overall-response rate measured by the investigator. Secondary outcomes included overall-response rate assessed by a blinded independent review committee, progression-free survival, overall survival, and safety.

3.6 The results of ASCEND‑1 and ASCEND‑2 are in table 1. The company also presented a pooled analysis using individual patient data assessed by the blinded independent review committee in ASCEND‑1 and ASCEND‑2. The pooled median progression-free survival was 7.0 months and the pooled median overall survival was 15.6 months.

Naive indirect comparison

3.8 The ASCEND‑1 and ASCEND‑2 studies did not include control groups, so the company could not directly compare ceritinib with best supportive care (BSC). The company searched the literature to find evidence of outcomes for patients who had BSC. It then did a naive indirect comparison of ceritinib with BSC (meaning the comparison was not adjusted for differences in patient or study characteristics between the studies). To assess whether people lived longer with ceritinib than BSC, the company compared the ASCEND studies with Ou et al. (2014). The study by Ou et al. was a retrospective analysis of people with advanced ALK‑positive NSCLC, whose disease had progressed after initial treatment and who had crizotinib as a second or subsequent treatment while in a clinical trial (PROFILE 1001 and PROFILE 1005). Ou et al. analysed data from 3 groups of patients whose disease had progressed after treatment with crizotinib, those who:

Cost effectiveness

3.12 The company's Markov model compared the cost effectiveness of ceritinib with BSC for people with advanced ALK‑positive NSCLC that had been previously treated with crizotinib. The model contained 3 mutually exclusive health states:

3.13 For ceritinib, the company took data from the blinded independent review committee's assessment of progression-free survival and the pooled results for overall survival from the ASCEND‑1 and ASCEND‑2 studies. For ASCEND‑1, it used data from the relevant patient population (that is, people who had previous crizotinib treatment and who had 750 mg of ceritinib). To extrapolate beyond the study period, the company fitted several parametric models to the data and selected the best-fitting curve based on visual inspection, statistical tests and external validity. The company chose a Weibull curve for overall survival and a log-logistic curve for progression-free survival.

3.14 To compare ceritinib with BSC, the company took overall-survival data from Ou et al. (2014) and progression-free survival data from Shepherd et al. (2005). For BSC, the company chose a Weibull curve for overall survival and a log-logistic curve for progression-free survival as it did for ceritinib.

3.15 The company used an 'area under the curve partitioned survival analysis' technique in which the number of patients in each health state was based on the survival curves described in sections 3.13 and 3.14. Patients entered the model in the progression-free health state and had ceritinib or BSC until progression, when they moved to the progressed-disease health state. Patients could move to the death state from either the progression-free or the progressed-disease health state.

3.16 For ceritinib, the company included the cost of grade 3 and 4 drug-related adverse events that had happened in at least 5% of patients in the pooled analysis of ASCEND‑1 and ASCEND‑2. The included events were:

3.17 The one-off cost associated with adverse events was £71.11. In its base case, the company did not include a decrease in utility for patients who had adverse events. In a scenario analysis, the company applied utility decrements for adverse events based on Nafees et al. (2008). The company assumed that patients having BSC did not experience adverse events.

3.18 The company estimated utility values by mapping EORTC QLQ-C30 data from ASCEND‑2 to the EuroQol EQ‑5D questionnaire. The mapping algorithm was developed in the UK for multiple myeloma (Proskorovsky et al. 2014). The company stated in its submission that, for the progression-free health state, it used the same utility value for both ceritinib and BSC based on patients with stable disease in ASCEND‑2. The value is academic in confidence and cannot be reported here. The ERG advised that, for the progression-free health state, the utility values in the company's model did not match the description in the company's submission (see section 3.28).

3.19 For the progressed-disease health state, the company stated that it was not appropriate to use the data on quality of life from ASCEND‑2, so instead it used published EQ‑5D data from patients with advanced NSCLC (Chouaid et al. 2013). The company's rationale was that, in ASCEND‑2, no data were collected on quality of life after disease progression. The ASCEND‑2 data therefore represented people whose disease had progressed recently and their quality of life was likely to be higher than for people at a later stage of progression. The utility value in the model for the progressed-disease health state was 0.460 for both ceritinib and BSC. The company's scenario analyses used alternative utility values.

3.20 The model included the costs of treatment with ceritinib and BSC.

3.21 The resource use in the model included clinic appointments, scans and laboratory tests. The model did not include the cost of diagnostic testing for the ALK mutation; the company assumed that this testing would already have been done because the modelled population had previously had crizotinib (for which ALK testing is needed). The company based its assumptions on resource use from NICE's technology appraisal guidance on erlotinib for non-small-cell lung cancer and on EGFR-TK mutation-positive non-small-cell lung cancer. The total cost per month for the progression-free health state was £180.88 (excluding medication costs), for progressed disease £313.70 (including medication costs) and for death £6,079.40 (including palliative care only).

3.22 The company's deterministic base case resulted in an incremental cost-effectiveness ratio (ICER) of £62,456 per quality-adjusted life year (QALY) gained for ceritinib compared with BSC (see table 3). The company stated in its submission that the key drivers of cost effectiveness were the cost of ceritinib, the discount rate and the utility values.

Evidence review group's critique

3.23 The ERG noted that only a small number of patients in Ou et al. (2014) were directly relevant to this appraisal (those who had only BSC after crizotinib, n=37), and there was limited information about what the authors considered to be BSC or systemic chemotherapy. The ERG also noted that the company's submission gave baseline patient characteristics only for the combined BSC and chemotherapy subgroups in Ou et al. (2014), so the characteristics of the BSC group (which in the ERG's opinion is the relevant subgroup for the appraisal) were not presented to the committee.

3.24 The ERG noted that when indirectly comparing the ASCEND studies, Ou et al. (2014), and Shepherd et al. (2005), the company did not adjust for differences in baseline patient characteristics, so the validity of the modelled results relied on assuming that the study populations were the same. However, the ERG noted that the ASCEND and Ou et al. studies differed in their inclusion and exclusion criteria, specifically previous treatment and ECOG performance status. Also, the ERG's clinical adviser stated that Ou et al. excluded patients with symptomatic brain metastases, whereas the ASCEND studies included these patients if their symptoms were stable. The ERG noted that Shepherd et al. recruited patients with all types of NSCLC, whereas the ASCEND and Ou et al. studies only recruited patients with the ALK‑positive mutation.

3.25 Regarding the populations, the ERG noted that there were differences in ECOG performance status and previous treatments between the patients in ASCEND and those in the combined BSC and chemotherapy subgroups in Ou et al. (2014), but, based on small numbers, the differences were not statistically significant. The ERG advised that, because the choice of treatment for patients in Ou et al. was based on clinical advice rather than a study protocol, the patients in the BSC group may have had more severe disease than the patients in the active treatment groups. The ERG advised that the BSC arm of the model may be informed by data from patients who were more ill than the patients in the ASCEND studies, potentially underestimating survival with BSC.

3.26 Regarding extrapolating overall survival with BSC beyond that seen in Ou et al. (2014), the ERG noted that the company's choice of a Weibull curve was the worst-fitting curve (based on both Akaike and Bayesian information criteria), and it suggested that a log‑normal curve should have been used instead (as the ERG did in its scenario analysis). The ERG conducted sensitivity analyses to test the impact of treatment benefits from ceritinib, in terms of overall survival and progression-free survival, stopping at 18 and 24 months from treatment initiation, by switching to the progression-free-survival and overall-survival curves of the BSC arm of the model.

3.27 The ERG noted that the company included only those adverse events that occurred in more than 5% of people, and it noted that the company may have excluded rare but serious adverse events. By contrast, the ERG's exploratory analyses included all grade 3 and 4 events from the ASCEND studies.

3.28 The ERG noted an inconsistency between the model and the company's submission in the utility values for the progression-free health state. The company's submission stated that the same utility value was used for both ceritinib and BSC. However, the ERG noted that, in the model, a weighted average utility value was calculated separately for ceritinib and BSC, based on the proportion of patients whose disease responded to treatment in ASCEND‑2 and Shepherd et al. (2005) respectively. The ERG advised that the company did not justify its method in its submission and it may not be appropriate. Accordingly, the ERG used the same utility value for both ceritinib and BSC, based on data from ASCEND‑2. Because patients having BSC would not have the adverse reactions associated with ceritinib, the ERG increased the utility value for the progression-free health state for BSC using utility values from Nafees et al. (2008).

3.29 The ERG noted that in ASCEND‑2, patients continued ceritinib treatment after disease progression for a median of 1.6 months. However, the company's base case assumed that ceritinib treatment would continue only until disease progression. The ERG's clinical expert advised that, in clinical practice, it is likely that patients would continue treatment beyond progression. Therefore, the ERG's exploratory analyses included extra treatment costs for ceritinib.

3.30 The ERG changed the following in the company's model:

3.31 In further exploratory analyses, the ERG reduced the duration of treatment benefit with ceritinib from 10 years (as assumed in the company's base case) to between 2 and 9 years. Beyond any given time point reflecting the end of benefit, the ERG set the probabilities of progressing or dying on ceritinib to be the same as for BSC. The ERG scenario that assumed 2 years' duration of treatment benefit reduced the ICER from £79,528 to £76,066 per QALY gained (see table 4). This reduction was mainly driven by lower treatment costs.

Patient access scheme and updated economic model

3.32 The company submitted a revised economic model after agreeing a patient access scheme, consisting of a simple discount, with the Department of Health (the amount of discount and net price are commercial in confidence). The company updated its economic model to address the concerns of the committee and the ERG. These updates included all changes done by the ERG outlined in section 3.30. In response to the ERG's comments about the duration of treatment benefits (see section 3.31), the company highlighted that for all scenarios in the model, this had been modelled by extrapolating the Kaplan–Meier curves over a time horizon of 10 years, and this did not imply that the benefits from ceritinib continue indefinitely after stopping treatment. To explore this further, the company provided 2 scenario analyses in which it assumed that the benefits and costs associated with ceritinib (as modelled in the curves for progression-free survival and overall survival) arbitrarily stop either 18 months or 2 years after starting treatment. The company assumed that from those points onwards, treatment benefits no longer follow the ceritinib curves for extrapolated progression-free survival and overall survival, but switch to the respective progression-free-survival and overall-survival curve of the BSC arm of the model. The company highlighted that these scenarios resulted in minimal changes to the ICERs.

3.33 Combining all of the ERG's changes to the parameters described in section 3.30 increased the ICER for ceritinib compared with BSC from £62,456 per QALY gained to £86,364 per QALY gained (not including the patient access scheme discount). But, incorporating the patient access scheme discount reduced the ICER to a level that the company stated was cost-effective compared with current treatment alternatives (the ICER cannot be presented here because the discount is commercial-in-confidence).

3.34 The company did a series of one-way sensitivity analyses, which showed that ceritinib drug costs have the largest impact on the results. The company did not vary estimates of the effectiveness of certinib compared with BSC. The company also submitted several scenario analyses and a cost-effectiveness acceptability curve, the details of which are commercial in confidence and cannot be presented here.