Azacitidine for the treatment of myelodysplastic syndromes, chronic myelomonocytic leukaemia and acute myeloid leukaemia

3 The manufacturer's submission

The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of azacitidine and a review of these submissions by the Evidence Review Group (ERG) and the Decision Support Unit (DSU; appendix B).

3.1 The main evidence for the efficacy of azacitidine in patients with high- and intermediate-2 risk myelodysplastic syndromes, chronic myelomonocytic leukaemia or acute myeloid leukaemia in the manufacturer's submission was obtained from a phase III, open-label, multicentre, randomised controlled trial (AZA-001; n = 358). Supplementary data from an open-label extension trial of AZA-001 were also provided. Before randomisation, patients were preselected by the investigator (on the basis of age, general condition, comorbidities and patient preference) for treatment with one of three conventional care regimens: best supportive care alone, low-dose chemotherapy plus best supportive care or standard-dose chemotherapy plus best supportive care. Patients were then randomised to receive either azacitidine or the preselected conventional care regimen. Patients receiving a particular conventional care regimen were compared with patients who had been preselected for the same care regimen but were then randomised to treatment with azacitidine. The manufacturer reported that patients randomised to either azacitidine or one of the conventional care regimens were comparable in terms of age, baseline severity of myelodysplastic syndrome, Eastern Cooperative Oncology Group (ECOG) performance status and time since original diagnosis. However, within the conventional care regimens, patients preselected to receive low- or standard-dose chemotherapy tended to be younger and have a higher ECOG performance status than patients preselected to receive best supportive care alone. Randomisation and subsequent analyses were stratified according to the French–American–British classification (FAB) subtype and IPSS group. Of the 179 patients receiving a conventional care regimen, 105 (59%) were preselected for best supportive care alone, 49 (27%) for low-dose chemotherapy and 25 (14%) for standard-dose chemotherapy. Of the 179 patients receiving azacitidine, 117 (65%) had been preselected for best supportive care alone, 45 (25%) for low-dose chemotherapy and 17 (9%; percentages do not add up to 100% because of rounding) for standard-dose chemotherapy.

3.2 The primary endpoint in AZA-001 was overall survival. Secondary endpoints included time to transformation to acute myeloid leukaemia, haematological response, independence from red blood cell transfusions for 56 consecutive days or more, number of infections needing intravenous antibiotics and occurrence of adverse events.

3.3 The manufacturer's submission stated that the intention-to-treat median overall survival was 24.5 months for patients receiving azacitidine compared with 15.0 months for patients receiving conventional care regimens (p = 0.0001, hazard ratio 0.58, 95% confidence interval [CI] 0.43 to 0.77). The median time to transformation to acute myeloid leukaemia was 17.8 months (interquartile range 8.6 to 36.8, 95% CI 13.6 to 23.6) with azacitidine compared with 11.5 months (interquartile range 4.9 to not reached, 95% CI 8.3 to 14.5) with conventional care regimens (p < 0.0001, hazard ratio 0.50, 95% CI 0.35 to 0.70). The manufacturer also reported results within each of the preselection groups. Treatment with azacitidine led to statistically significant improvements in overall survival in the group preselected for best supportive care alone and in the group preselected for low-dose chemotherapy plus best supportive care, but not in the group preselected for standard-dose chemotherapy plus best-supportive care. Only patients preselected for best supportive care alone had statistically significant improvement with azacitidine in time to transformation to acute myeloid leukaemia. Of the patients who were dependent on red blood cell transfusions at baseline, 45% of patients treated with azacitidine no longer needed transfusions during treatment compared with 11.8% of patients receiving conventional care regimens (p < 0.0001). The manufacturer reported that in a subgroup analysis of patients with the –7/del(7q) chromosomal abnormality, median overall survival was higher in patients receiving azacitidine than in patients receiving conventional care regimens.

3.4 The ERG considered the results from AZA-001 to be robust and to show clinical benefit for patients treated with azacitidine. The ERG noted that the open-label design of the study meant that the results could be subject to bias and that there was an imbalance in the numbers lost to follow-up. The ERG considered that this means that the effectiveness of azacitidine in clinical practice could be lower than that seen in AZA-001. In addition, the ERG noted that the results for the comparison with low- and standard-dose chemotherapy were less robust because of the small numbers of patients included.

3.5 The manufacturer developed an economic evaluation, comprising a two-arm health-state transition model. One arm estimated the costs and outcomes associated with treatment with azacitidine; the other arm estimated the costs and outcomes associated with treatment with the conventional care regimens in AZA-001 (see section 3.1). Patients entered the model in the myelodysplastic syndromes health state at the start of treatment and left the model at death. The model used a 35-day cycle with a lifetime horizon.

3.6 The manufacturer's economic model used data from AZA-001 and its open-label extension trial to estimate effectiveness. The economic model underwent a number of iterations after clarification requests from the Committee, the ERG and the DSU. The manufacturer's base-case analysis used a lognormal parametric function to extrapolate the overall survival from the data observed in the trial. Survival data from a myelodysplastic syndromes registry in Düsseldorf were presented in support of the selection of the lognormal function. Sensitivity analyses explored the use of alternative parametric functions. Time to progression was modelled in such a way that progression to the acute myeloid leukaemia health state occurred eight cycles before death to reflect the mean length of time patients had acute myeloid leukaemia in AZA-001.

3.7 The manufacturer reported that no quality of life data were collected in AZA-001 that could be used to populate the economic model. Utility value estimates for patients treated with azacitidine and best supportive care were taken from the prospective, open-label, multicentre randomised controlled trial CALGB 9221 (n = 191). In this trial, patients with myelodysplastic syndromes were treated with either azacitidine or best supportive care, and European Organisation for Research and Treatment of Cancer (EORTC) quality of life data were collected. This trial was excluded from the clinical-effectiveness analysis because the patient population was of a lower IPSS risk category than the population specified in the marketing authorisation for azacitidine. The manufacturer converted the EORTC quality of life data into EQ-5D values using an algorithm developed using data from a study in patients with oesophageal cancer. Utility value estimates for patients treated with chemotherapy were mapped to the EQ-5D from SF-12 scores published in a report about patients receiving low-dose and standard-dose chemotherapy.

3.8 The manufacturer reported that, when possible, healthcare resource use was determined from AZA-001 protocol regimens. When data were not available from the trial, resource use estimates were based on expert opinion obtained through a questionnaire. Drug costs were taken from the BNF (edition 57). The majority of treatment costs were determined using the NHS 2009/10 tariff. Personal and Social Services Research Unit costs and NHS reference costs (2006/07) were used for resources if a tariff cost was not available. Because azacitidine requires a 7-day continuous treatment cycle, the additional cost of weekend administration was modelled as a two-fold increase in administration cost for 2 days of each treatment cycle. The manufacturer estimated that vial sharing, which involves treating multiple patients on the same day, could be used for 49% of patients. The reduction in unused vials and consequent savings in drug costs resulting from vial sharing were explored in a scenario analysis.

3.9 The manufacturer's base-case results (using the lognormal parametric function to extrapolate overall survival and excluding any patient access scheme; see section 3.6) gave incremental cost-effectiveness ratios (ICERs) for treatment with azacitidine compared with each of the conventional care regimens of £47,432 per quality-adjusted life year (QALY) gained for patients in the best supportive care alone group, £40,754 per QALY gained for patients in the low-dose chemotherapy group, and £37,105 per QALY gained for patients in the standard-dose chemotherapy group. The scenario analysis that explored vial sharing decreased the base-case ICERs to £44,440, £37,929 and £34,366 per QALY gained for the best supportive care alone, low-dose chemotherapy and standard-dose chemotherapy groups respectively. Incorporating the original patient access scheme reduced the base-case ICERs (and those with vial sharing) to £45,538 (£42,756), £38,966 (£36,399) and £35,371 (£32,823) per QALY gained for the best supportive care alone, low-dose chemotherapy and standard-dose chemotherapy groups respectively.

3.10 The manufacturer provided cost-effectiveness analyses for each of the parametric survival functions explored. The ICERs referred to below incorporate the original patient access scheme. For the analyses using the Weibull survival function, the ICERs were £63,177 per QALY gained for the best supportive care alone group, £49,030 per QALY gained for the low-dose chemotherapy group, and £51,252 per QALY gained for the standard-dose chemotherapy group. For the analyses using the exponential survival function, the ICERs were £67,203 per QALY gained for the best supportive care alone group, £58,418 per QALY gained for the low-dose chemotherapy group, and £60,097 per QALY gained for the standard-dose chemotherapy group. For the analyses using the lognormal survival function, the ICERs were £45,538 per QALY gained for the best supportive care alone group, £38,996 per QALY gained for the low-dose chemotherapy group, and £35,371 per QALY gained for the standard-dose chemotherapy group.For analyses using the baseline survival from the Düsseldorf registry data and applying the respective hazard ratios associated with treatment, the ICERs were £71,522 per QALY gained for the best supportive care alone group, £58,282 per QALY gained for the low-dose chemotherapy group, and £85,790 per QALY gained for the standard-dose chemotherapy group.

3.11 The ERG expressed concerns about the analyses of the preselected conventional care groups in AZA-001. It noted that two of the groups, particularly the standard-dose chemotherapy group, consisted of very small numbers of patients, and that to consider the arms of the trial in isolation effectively breaks randomisation.

3.12 The ERG raised concerns about the parametric function selected to model overall survival. It noted that the selection of the lognormal function was not strongly supported by evidence from AZA-001, its open-label extension trial or the Düsseldorf registry data. The ERG reported that when various parametric functions were compared with the individual patient data from the Düsseldorf registry, an exponential survival function underestimated long-term survival, while log-logistic and lognormal survival functions overestimated long-term survival. The ERG noted that the use of log-logistic and lognormal functions estimated a percentage of patients would survive into their nineties, which the ERG considered unrealistic, given the nature of the condition. The ERG reported that of the functions explored, the Weibull survival function provided the best fit to the Düsseldorf registry data.

3.13 The ERG commented that the time to transformation to acute myeloid leukaemia in AZA-001 was subject to considerable censoring from loss of patients to follow-up. It therefore considered that the modelled time to transformation was subject to uncertainty.

3.14 The ERG noted several issues with the conversion of EORTC quality of life data into utility values. The ERG reported that the algorithm used to derive the utility values was considered by its developers to be less reliable for patients in more severe health states than alternative algorithms that were explored and rejected by the manufacturer. The ERG noted that this could bias the results. It also reported that the algorithm had been developed using data from patients with oesophageal cancer, and that patients eligible for azacitidine were of a similar age to these patients. However, the underlying conditions and comorbidities were potentially very different. The ERG stated that the utility values resulting from the algorithm should be treated with caution.

3.15 The manufacturer explored the impact of adjusting the utility values to account for the differences in the baseline patient characteristics. This was shown to have little impact on the ICERs.

3.16 After the appeal, the manufacturer submitted information on the proportions of patients with high-risk myelodysplastic syndromes, chronic myelomonocytic leukaemia or acute myeloid leukaemia receiving low-dose chemotherapy plus best supportive care and the proportions of patients receiving best supportive care alone. This information included:

3.17 The DSU provided a critique of the three sources of information. The survey of 72 UK haematologists was based on clinicians' estimates of patients receiving each conventional care regimen, rather than the proportions of patients who were eligible for each regimen. Because it was not clear whether the data provided were from case note review or from clinician self-reporting, the DSU considered that the survey was subject to a high risk of bias. For this reason the DSU considered that this survey did not provide reliable data to inform the Committee's considerations about the conventional care regimens. Regarding the survey from 23 UK hospitals, the DSU commented that because the data had been presented graphically it was difficult to estimate the exact proportions of patients treated with each care regimen. The survey was considered to offer limited information because:

3.18 After the appeal, the manufacturer also submitted information about the clinical characteristics of patients receiving each of the conventional care regimens in routine clinical practice. This included the results of an 'informal literature review' by the manufacturer and the clinical characteristics of patients receiving low-dose chemotherapy from the survey of UK haematologists. The manufacturer reported that among patients who were eligible for azacitidine according to the marketing authorisation, low-dose chemotherapy was most widely used in the UK in patients with the following characteristics:

3.19 The DSU commented that only two of the eleven studies provided by the manufacturer after the appeal were conducted in the UK and only one had been published since 1991. It was not clear how the characteristics had been selected from these studies; in particular toxicity and administration had been identified in the literature review but did not appear in the final list of characteristics. However, the HCTCI score, which the manufacturer previously reported was not in routine clinical use in the UK, was included in the list. In summary, the DSU considered that the literature review offered limited evidence on eligibility criteria for patients receiving the conventional care regimens.

3.20 After the appeal, the manufacturer also submitted additional cost-effectiveness analyses incorporating health-related quality of life data from the patient group MDS UK. The manufacturer provided analyses:

3.21 The manufacturer's analyses submitted after the appeal indicated that azacitidine compared with best supportive care alone was associated with an incremental cost of £63,756 and an incremental QALY gain of 1.01, producing an ICER of £63,177 per QALY gained. Azacitidine compared with low-dose chemotherapy was associated with an incremental cost of £65,671 and an incremental QALY gain of 1.34, giving an ICER of £49,030 per QALY gained. Azacitidine compared with usual care (a weighted average of all the conventional care regimens together) was associated with an incremental cost of £61,801 and an incremental QALY gain of 1.09, giving an ICER of £56,945 per QALY gained. The manufacturer's submission stated that this weighted average was calculated using the proportion of patients receiving each of the conventional care regimens in AZA-001 (that is, 62% for best supportive care alone, 26% for low-dose chemotherapy plus best supportive care and 12% for standard-dose chemotherapy plus best supportive care).

3.22 The DSU commented that the utility values used in the revised economic evaluation were based on descriptions of health states that included dependence or independence from transfusion as a feature and did not separate the specific utility value of dependence or independence from transfusion from that of associated symptoms. It noted the small numbers of patients in the study (n = 47) and patients from the UK (n = 21). The DSU further noted that the data did not capture adverse events, in contrast with the utility value estimates used in the manufacturer's original base case. The DSU commented that constant utility values had been applied over the time horizon of the model, assuming that a patient's dependence on transfusion would remain constant throughout their treatment period. It considered that this was an unreasonable assumption. The DSU concluded that the MDS utility data did not provide more appropriate information for the economic evaluation than the data used in the manufacturer's base case. However, the DSU ran exploratory analyses using data from all patients in the MDS UK study and then with only data from the UK patients. None of these exploratory analyses resulted in significant changes to the cost-effectiveness estimates provided by the manufacturer.

3.23 The DSU considered that the analysis using a weighted average (see section 3.21) was not an appropriate measure of the cost effectiveness of azacitidine. The DSU took the view that the appropriate approach would be to consider all the treatment options in a single incremental analysis, comparing each treatment with the next most effective alternative, and excluding any dominated (that is, more costly and less effective) treatments from the analysis.

3.24 In response to the appraisal consultation document produced after the appeal, the manufacturer submitted updated cost-effectiveness analyses. Each analysis was deterministic, used a Weibull parametric function to extrapolate overall survival, assumed no vial sharing, and included the revised patient access scheme (see section 2.4). These analyses estimated the following ICERs for azacitidine, compared with three alternative weighted averages of the conventional care regimens:

3.25 The DSU considered that the most generalisable estimates for conventional care patterns from AZA-001 should be taken from the pre-randomisation proportions (that is, the whole trial population), instead of the proportions of patients (50% of the trial population) randomised to one of the conventional care regimens (as presented by the manufacturer). Before randomisation, 62% of patients were preselected to receive best supportive care alone, 26% received low-dose chemotherapy and 12% received standard-dose chemotherapy. The weighted average ICER associated with these proportions was £49,808 per QALY gained.

3.26 The DSU further considered that probabilistic analyses (instead of the deterministic analyses presented by the manufacturer) would be more robust, and therefore more appropriate for the Committee's consideration. The probabilistic ICERs associated with each of the weighted averages were:

3.27 Full details of all the evidence are in the manufacturer's submissions, the ERG report and the DSU reports.