Crizotinib for untreated anaplastic lymphoma kinase-positive advanced non-small-cell lung cancer

The committee discussed the results of PROFILE 1014 and the primary outcome measure of progression-free survival:

• It noted that progression was determined using radiographic criteria, specifically the Response Evaluation Criteria in Solid Tumours (RECIST). The committee heard from the clinical experts that radiographic criteria are the gold standard for monitoring NSCLC.

• The committee noted that crizotinib increased progression-free survival compared with pemetrexed with either cisplatin or carboplatin (hazard ratio [HR] 0.45; 95% confidence interval [CI] 0.35 to 0.60).

• The committee was aware that the company used a Cox proportional hazards model to estimate the hazard ratio for progression-free survival. It noted the ERG's critique that in this case, the proportional hazards assumptions needed to analyse data using a Cox proportional hazards model may not hold because the 2 treatment regimens are given differently (in PROFILE 1014, crizotinib was given until progression whereas platinum-based chemotherapy [the control group] was given for a finite number of cycles). The ERG stated that this did not have a large effect on cost effectiveness, but because patient-level data were available, the company could have modelled the data using separate independent parametric curves with fewer assumptions.
  
  On balance, the committee concluded that crizotinib increases progression-free survival compared with pemetrexed plus either cisplatin or carboplatin in people with ALK-positive NSCLC.

The committee discussed the results of PROFILE 1014 and the secondary outcome measure of overall survival. It noted the ERG's comments that the results for overall survival were based on relatively immature data, that is, that few patients had died at the time of data analysis. It also noted that a high proportion of patients crossed over from chemotherapy to crizotinib. The committee was aware that because crossover occurred at or after disease progression, it would not affect progression-free survival, but would affect overall survival. The committee concluded that it was appropriate for the company to adjust for crossover when estimating the size of the benefit on overall survival associated with crizotinib.

The committee discussed the methods for adjusting for crossover in PROFILE 1014. The company presented evidence using different methods to adjust overall survival for crossover (rank-preserving structural failure time, iterative parameter estimation, and 2‑stage methods) and presented a range of analyses, which accounted for different confounders. The committee recognised that the company had used the 2‑stage method for its cost-effectiveness analyses. It was aware that there was some uncertainty about whether all confounders were measured at the time of crossover, but noted that the ERG agreed this was the most appropriate approach because it did not assume a common treatment effect (that is, that the treatment effect is the same regardless of when a person starts treatment). The committee concluded that the 2‑stage method was appropriate.

The committee discussed the size of the benefits associated with crizotinib on overall survival. The committee noted that crizotinib increased overall survival compared with pemetrexed plus either cisplatin or carboplatin (HR 0.62; 95% CI 0.41 to 0.96), when using the 2‑stage method to account for crossover. It noted that applying different methods to account for crossover did not vary the hazard ratio substantially. It noted the ERG's comments that the results for overall survival were based on relatively immature data (that is, few patients had died at the time of data analysis). The committee recognised that the size of the benefit was uncertain because of relatively immature data and the high proportion of patients crossing over from chemotherapy to crizotinib. On balance, the committee concluded that crizotinib is very likely to increase overall survival compared with pemetrexed plus either cisplatin or carboplatin in people with ALK-positive NSCLC, but the size of the increase is uncertain.

The committee discussed the company's approach to modelling overall and progression-free survival. It noted that, to generate more realistic survival estimates relevant to the UK population, the company had adjusted PROFILE 1014 data to reflect the characteristics of patients in a retrospective cohort study from the US and Canada (Davis et al. 2015). The committee discussed whether the characteristics of patients in the study reflected those of patients in England and noted from the company's sensitivity analyses that the assumptions were conservative. The committee concluded that it was satisfied with the company's approach.

The committee considered whether assuming proportional hazards between treatments was appropriate for extrapolating progression-free and overall survival to estimate how long, on average, crizotinib extended the progression-free period and delayed death. The committee noted consultation comments from the company that the recommended statistical checks in NICE's Decision Support Unit technical support document 14, including log-cumulative hazard plots for overall survival, had shown that the proportional hazards assumptions held. The committee recalled the different methods of administration between treatments (see section 3.7) and noted that the log-cumulative hazard plots for each treatment were not parallel. The committee therefore disagreed with the company, noting that the hazard ratios were likely to change over time and that the assumption of proportional hazards was unlikely to hold. The committee was aware that NICE's Decision Support Unit technical support document 14 suggests using separate parametric curves for each treatment group. The committee concluded that using separate parametric curves for each group was appropriate.

The committee considered whether the parameters used to adjust the parametric curves to the UK population should be the same or different for both treatments (that is, use independent covariate stratification). The committee noted comments received during consultation that there is no evidence suggesting a difference in prognosis between treatments or that covariates (such as age or sex) influence outcomes depending on treatment. The committee heard from the ERG that stratifying covariates independently uses fewer assumptions, and noted that it had a minor effect on the incremental cost-effectiveness ratios (ICERs). On balance, the committee agreed that it was appropriate to adjust each treatment for the population separately and concluded that using independent prognostic covariates was appropriate.

The committee discussed which parametric curves for extrapolating progression-free and overall survival it considered most plausible:

• The committee was aware that the company's base case used a generalised gamma distribution for progression-free survival and a Weibull distribution for overall survival. It noted that the ERG presented exploratory analyses using separate parametric curves for each treatment (that is: progression-free survival – a log-normal distribution for crizotinib and a generalised gamma distribution for chemotherapy; overall survival – a generalised gamma distribution for crizotinib; and an exponential distribution for chemotherapy). The committee recognised that different parametric distributions predicted a range of differences in progression-free and overall survival. The committee noted comments received from the company during consultation that the overall survival increase using the ERG's selected distributions was implausible (0.8 months increased survival with crizotinib compared with chemotherapy). It heard from the clinical experts that they expected no less than a 6‑month increase in overall survival with crizotinib. The committee agreed that curves used in the ERG's exploratory analysis (that is: progression-free survival – a log-normal distribution for crizotinib and a generalised gamma distribution for chemotherapy; overall survival – a generalised gamma distribution for crizotinib; and an exponential distribution for chemotherapy) were not plausible.

• The committee considered statistical tests of model fit and noted that, based on either the Akaike information criterion (AIC) or the Bayesian information criterion (BIC), there were no large differences between the distributions. It noted comments received from the company during consultation that the exponential distribution (for both crizotinib and chemotherapy) had the lowest cumulative AIC and BIC. The committee agreed that it did not consider adding AIC and BIC scores together to be routine statistical practice, and that selecting parametric curves for extrapolation should not be based on statistical methods alone.

• To extrapolate overall survival, the committee noted that the company, in its response to the appraisal consultation document, preferred the exponential distribution for crizotinib, and the exponential or Weibull distributions for chemotherapy. The company provided its 4 criteria for selecting parametric curves:

  • Overall survival with chemotherapy must be less than 24 months; the committee agreed that this was reasonable given the committee's discussion of end-of-life criteria (see section 3.21).

  • The mean overall-survival gain with crizotinib must be more than 7.1 months based on an estimate from PROFILE 1007, a trial used in NICE's technology appraisal guidance on crizotinib for previously treated anaplastic lymphoma kinase-positive advanced NSCLC. The company suggested that crizotinib should be more effective as a first-line treatment for NSCLC than as a second-line treatment. However, the committee recognised that PROFILE 1007 was based on a different population, a different comparator (docetaxel), and a different duration of treatment. The committee was also aware that in NICE's technology appraisal guidance on crizotinib for previously treated NSCLC associated with an ALK-fusion gene, overall survival was uncertain because of crossover between treatments. The committee heard from the company that although several years of additional follow-up data were collected for PROFILE 1014, the next re-analysis will be done when median overall survival is reached (as per protocol). The committee agreed that the dataset for overall survival remained incomplete and that the gain in overall-survival data was uncertain.

  • The mean gain must be more than the median gain in overall survival: the committee heard from the clinical experts that some people on crizotinib live longer than expected, and agreed that this criterion is plausible and appropriate.

  • The mean gain in overall survival should be clinically plausible; the committee agreed that this criterion was appropriate.
    
    The committee acknowledged there was uncertainty about the size of the average gain in overall survival. It was aware that the company planned to do more analyses and emphasised the need for mature overall-survival data. The committee noted that using different parametric curves for overall survival had a major effect on the ICERs. Given the limited data available, the committee agreed to use the same distributions for both treatments to minimise the differences in assumptions between treatments. The committee noted that the exponential, log-normal and log-logistic distributions had a median overall-survival gain of more than 6 months and a mean overall-survival gain higher than the median overall-survival gain, and agreed that these distributions were all plausible and considered these for its decision-making.

The committee discussed the time on treatment assumed in the company's model. The committee was concerned about the way in which the company estimated time on crizotinib treatment using PROFILE 1014. The company assumed that people taking crizotinib stopped treatment at the end of the trial (that is, they were censored), and applied this in the model. The ERG considered that this substantially underestimated the time on treatment after progression. The committee agreed that it was inappropriate to assume that patients in the trial who stopped treatment because the trial ended would also stop treatment in real life, and preferred the ERG's analyses using a parametric survival curve, that accounts for censoring, to estimate the mean duration of treatment. However, it noted that the ERG did not adjust the analyses to reflect the population in England because it did not have access to the relevant data. The committee noted that in response to the appraisal consultation document, the company presented a revised analysis reflecting the population in England. The committee agreed that this was appropriate.

The committee discussed the approach to second-line treatment in the company's model:

• Docetaxel: the committee noted that the company assumed that everyone with progressed disease had docetaxel (as described in NICE's guideline on diagnosing and managing lung cancer), but it heard from the clinical experts that some people are not fit enough for second-line docetaxel. It also noted the ERG's comments that in PROFILE 1014, people went on to have a wide range of therapies (other than docetaxel) after disease progression.

• Second-line crizotinib: the committee noted that the company did not include second-line crizotinib in its model. It heard from the clinical experts that people who have first-line platinum-based chemotherapy may go on to have second-line crizotinib. The committee was aware that crizotinib is not recommended in NICE's technology appraisal guidance on crizotinib for previously treated anaplastic lymphoma kinase-positive advanced NSCLC, and that second-line treatment with crizotinib was only available through the Cancer Drugs Fund.

• Second-line ceritinib: the committee was aware that ceritinib is recommended in NICE's technology appraisal guidance on ceritinib for previously treated anaplastic lymphoma kinase positive NSCLC. The committee noted that ceritinib is considered as an option after treatment with crizotinib.
  
  Therefore, the committee was unclear on whether the company's model accurately reflected second-line treatment for people with ALK-positive NSCLC in England. However, the committee was also aware that the ERG presented analyses without second-line treatment because of sparse data on time on second-line treatment, which suggested that the effect on the ICER would likely be small. On balance, the committee concluded that, in the absence of robust data and because of uncertainty about second-line treatments, excluding second-line treatment from the model was the most robust approach.

The committee discussed the utility values used in the company's model:

• The company applied a lower utility for the progression-free health state in its model for platinum-based chemotherapy (0.72) than it did for crizotinib (0.81). The committee noted that this may underestimate the utility associated with platinum-based chemotherapy because health-related quality-of-life data were collected for patients only during chemotherapy, but not after chemotherapy had finished. The committee was aware that the ERG presented analyses with a higher utility value (0.81) for the progression-free health state with a platinum-based chemotherapy. In its response to the appraisal consultation document, the company accepted that it had underestimated the utility value for chemotherapy and suggested a higher utility value of 0.75. The committee heard from the ERG that the revised utility value also underestimated the utility decrement associated with chemotherapy because it included minor and not major adverse events, and that the data were based only on people on treatment. The committee heard from the clinical experts that people treated with cisplatin can have ongoing peripheral neuropathy and a lower quality of life compared with people taking oral treatments. On balance, the committee concluded that the utility value was closer to 0.75 than 0.81 and that the company's revised utility was appropriate.

• The committee noted that the company applied a utility value (0.74) for the period after a patient's disease has progressed but they continue to take crizotinib. It noted that this value averaged the utility for first-line treatment with crizotinib (before disease progression, 0.81) and the utility for second-line treatment with docetaxel (after disease progression, 0.66). It noted that this was not based on evidence. In its response to the appraisal consultation document, the company used a higher utility value (0.78) estimated from PROFILE 1014. The ERG stated that there was a risk of attrition bias for this estimate (that is, sicker patients were lost to follow-up earlier and not included in the analysis), which reduced the difference in the health-related quality of life of patients before and after progression. The committee heard from the clinical experts that despite progression, a patient on crizotinib may not have symptoms of lung cancer. The committee concluded that the utility value for the period after a patient's disease has progressed but the patient continues to take crizotinib was uncertain but likely to be between 0.74 and 0.78.

The committee discussed the costs used in the company's model:

• The committee considered the appropriate cost for ALK testing in the model and noted that the ERG considered that the company had underestimated the cost of testing. The committee heard from the ERG that a recent Cancer Research UK study reported the cost of ALK-mutation testing as £153 per patient. The ERG estimated that the cost of identifying a person with the ALK mutation was around £4,500, because over 29 people with NSCLC need to be tested to identify 1 person with the mutation. The committee heard from a clinical expert that the cost of immunohistochemistry was between £50 and £100 (excluding laboratory costs). In its response to the appraisal consultation document, the company assumed a cost of £75 for immunohistochemistry, increasing the cost of identifying a person with the ALK mutation to £2,380. The ERG suggested that the company underestimated the cost because it excluded laboratory costs. The committee heard from the company that the sequence of tests is a recent change, so the costs of ALK-mutation testing were still uncertain. Although the true cost for ALK-mutation testing is unknown, the committee considered that the cost of ALK-mutation testing was between £2,380 and £4,500.

• The committee discussed administration costs for crizotinib, and noted that the company did not include them in its model. It heard from the clinical experts that there would be administration costs, and also noted that these costs were included in NICE's technology appraisal guidance on crizotinib for previously treated anaplastic lymphoma kinase-positive advanced NSCLC. The committee recognised that the ERG took this into account in its exploratory analyses. However, the committee agreed with consultation comments from the company that the lower administration cost, based on that used in the NICE technology appraisal guidance on ceritinib for previously treated anaplastic lymphoma kinase positive non-small-cell lung cancer was more appropriate than the cost used by the ERG in its exploratory analyses.

The committee considered 4 results for the cost effectiveness of crizotinib compared with pemetrexed plus either cisplatin or carboplatin for people with advanced ALK-positive NSCLC:

• Company's revised base case: £49,186 per quality-adjusted life year (QALY) gained that:

  • assumes proportional hazards without independent covariate stratification

  • reflects changes to the original company submission:

    • higher patient access scheme discount

    • alternative utility values (see section 3.18)

    • adjusted time on treatment (see section 3.16)

    • adjusted administration costs and ALK-mutation testing costs (see section 3.19).

• Company's independent parametric curve analysis: £47,921 per QALY gained that:

  • no longer assumes proportional hazards between treatment groups, but uses independent exponential curves for both treatments

  • adjusts each treatment for the population separately (independent covariate stratification).

• ERG's revised exploratory analysis: £58,029 per QALY gained that:

  • assumes proportional hazards without independent covariate stratification

  • reflects changes to the company's revised analysis:

    • higher utility for platinum-based chemotherapy for the progression-free health state (0.81 rather than 0.75)

    • lower utility when a patient's disease progresses and they continue crizotinib (0.74 rather than 0.78)

    • higher ALK-mutation testing cost (£4,500 rather than £2,380)

    • higher administration cost for crizotinib (£163.85 rather than £14.40).

• ERG's independent parametric curves analysis: £55,131 per QALY gained that:

  • no longer assumes proportional hazards between treatment groups but uses independent exponential curves for both treatments

  • adjusts each treatment for the population separately (independent covariate stratification).
    
    The committee recalled its preferred assumptions relating to time on treatment (see section 3.16), utilities (see section 3.18), costs (see section 3.19), proportional hazards (see section 3.13), independent covariate stratification (see section 3.14), and extrapolation (see section 3.15). It concluded that the company's independent parametric curve analysis (with an estimated ICER of £47,291 per QALY gained) most closely reflected the committee's preferred assumptions and noted that the ICERs for other alternative curves (such as the log-normal and log-logistic distributions for both treatments) were similar. However, the committee acknowledged that this used some assumptions that the committee did not prefer or that the committee considered to be uncertain (such as the utility value for the period after a patient's disease has progressed but the patient continues to take crizotinib, and the cost of testing for an ALK mutation), and therefore acknowledged that uncertainty on the cost effectiveness of crizotinib remained fairly large. The committee was aware that it needed to be increasingly certain of the cost effectiveness of a technology as the technology's effect on NHS resources increases, as described in NICE's guide to the methods of technology appraisal. It recalled that ALK-positive mutation is rare in people with advanced NSCLC and considered the effect of adopting the technology as relatively small. On balance, the committee concluded that even after accounting for the uncertainty in the cost effectiveness, the most plausible ICER was likely to be at a level at which crizotinib could be considered a cost-effective use of NHS resources when the end-of-life criteria to apply.