Bevacizumab in combination with a taxane for the first-line treatment of metastatic breast cancer

The manufacturer's original submission presented clinical-effectiveness data from 1 randomised, open-label, controlled trial (E2100). A total of 722 patients were randomised to either bevacizumab plus weekly paclitaxel (n=368) or weekly paclitaxel alone (n=354). The randomisation was stratified by disease-free interval (less than or equal to 24 months; greater than 24 months), number of metastatic sites (less than 3; greater than or equal to 3), prior receipt of adjuvant chemotherapy (yes; no) and oestrogen receptor status (positive; negative; and unknown). All patients were given intravenous weekly paclitaxel (90 mg/m² over 1 hour) once a week for 3 weeks, with no treatment given at week 4. Patients in the bevacizumab plus weekly paclitaxel arm also received intravenous bevacizumab (10 mg/kg) every 2 weeks, until progression of disease or unacceptable toxicity occurred. There was no limit to the number of cycles of therapy allowed. The patients in the trial had locally recurrent or metastatic breast cancer and over 90% had HER2-negative breast cancer. The primary endpoint of the trial was duration of progression-free survival. Secondary endpoints were overall survival, objective response (complete response and partial response) rate, duration of response and health-related quality of life. Health-related quality of life was measured by the Functional Assessment of Cancer Therapy (FACT-B) questionnaire, which is a scale for measuring quality of life among breast cancer patients.

At the time of the manufacturer's interim analysis most patients had discontinued randomised therapy; for 360 patients (50%) this was because of disease progression, and 131 patients (18%) withdrew from the study because of unacceptable toxicity. The interim analyses consisted of a stratified log rank test where the stratification factors were disease-free interval and prior adjuvant chemotherapy. There was a statistically significant increase in median progression-free survival of 5.5 months, from 5.8 months in the paclitaxel alone arm to 11.3 months in the bevacizumab plus paclitaxel arm. The stratified hazard ratio for progression was 0.48 (95% confidence interval [CI] 0.39 to 0.61; p<0.0001). The stratified hazard ratio for death was 0.87 (95% CI 0.72 to 1.05; p=0.14), indicating a non-statistically significant improvement in median overall survival of 1.7 months, from 24.8 months with paclitaxel alone to 26.5 months with bevacizumab plus paclitaxel.

At baseline, 302 (87.3%) patients in the paclitaxel alone arm and 317 (88.8%) patients in the bevacizumab plus paclitaxel arm completed the FACT-B questionnaire. At week 33, 163 patients in the paclitaxel arm and 205 patients in the bevacizumab plus paclitaxel arm completed the questionnaire. If scores were missing at week 17 or week 33, the patient was not included in the analysis for that respective time point – except when disease progression or death was recorded earlier. For those patients who died or had disease progression, a value of zero (that is, the worst score) for each of the 5 subscales in the FACT-B questionnaire was imputed (rather than the patient being excluded from the analysis). When imputed values were used, the difference in total FACT-B score between the 2 treatment arms was statistically significant (p=0.0046) in favour of the bevacizumab plus paclitaxel arm at week 33. There were no statistically significant differences between treatment arms at week 17, or at week 33 if imputed values were not used. The manufacturer stated that, taken together, these results demonstrated that the addition of bevacizumab to paclitaxel led to a relative improvement in health-related quality of life.

The safety analyses from the E2100 trial reported that the addition of bevacizumab to paclitaxel resulted in a 20% overall increase in the incidence of grade 3 to 5 adverse events. These included neuropathy (25.3%), hypertension (16%), arterial thromboembolic events (3.6%), proteinuria (3%), bleeding (2.2%) and congestive heart failure (2.2%). In addition, adverse event data from a non-randomised single-arm, open-label study (ATHENA, n=2,251) were presented. The most frequent serious adverse events (grade 3 to 5) were febrile neutropenia (5.1%), neutropenia (3.6%) and pyrexia (1.5%).

The manufacturer carried out indirect comparisons for bevacizumab plus weekly paclitaxel compared with docetaxel alone and gemcitabine plus 3-weekly paclitaxel. The comparisons were carried out by an indirect method using 2 comparators to link 3 trials. The manufacturer noted that studies conducted only in patients having first-line treatment for metastatic breast cancer were not always available, so the exclusion criteria specified that trials in which more than 60% of patients were receiving second or later lines of treatment would be excluded. The manufacturer noted that 1 study used a higher docetaxel dosage (100 mg/m² 3-weekly) and a longer duration of treatment (maximum 32 cycles) compared with standard UK practice (considered by the manufacturer to be 75 mg/m² 3-weekly; maximum 6 to 8 cycles). However, based on the similar populations, baseline characteristics and exclusion or inclusion criteria, the manufacturer assumed that heterogeneity would not be significant.

The hazard ratio for progression with bevacizumab plus weekly paclitaxel compared with docetaxel alone was estimated to be 0.56 (95% CI 0.39 to 0.78) and 0.46 (95% CI 0.34 to 0.64) compared with gemcitabine plus 3-weekly paclitaxel. For weekly paclitaxel compared with 3-weekly docetaxel the hazard ratio for progression was 1.15 (95% CI 0.89 to 1.48) and for weekly paclitaxel compared with gemcitabine plus 3-weekly paclitaxel the hazard ratio for progression was 0.96 (95% CI 0.76 to 1.21).

The manufacturer's model was a 3-state Markov model with a cycle length of 1 month. Patients in the model received treatment with either bevacizumab plus weekly paclitaxel or the comparator treatment, that is:

• weekly paclitaxel or

• docetaxel or

• gemcitabine plus 3-weekly paclitaxel.
  
  Although bevacizumab plus docetaxel was included in the scope it was not formally evaluated in the manufacturer's economic analysis.

Patients were assumed to be in 1 of 3 possible discrete health states at any given time: 'progression-free survival', 'progressed' or 'death'. It was assumed that patients would have the same risk of dying after disease progression regardless of the first-line treatment they had received. In addition, the model assumed that patients would have the same sequence of further treatment and resource use after disease progression, regardless of their initial treatment. The number of patients who died while in the 'progression-free survival' state was determined either by the maximum of background mortality or the monthly rate at which patients died (from any cause) while progression-free in the E2100 trial. In the base-case model, the progression-free mortality rates for weekly paclitaxel alone were used as a proxy for the mortality rates for docetaxel and gemcitabine plus paclitaxel.

The manufacturer provided 2 base-case analyses, both incorporating a 10-year time horizon:

• The first used the list prices in accordance with the NICE reference case. The prices for bevacizumab (total average cost per patient, £25,929) and paclitaxel (total average cost per patient, £7,720) were taken from the BNF, edition 58. Drug costs were calculated according to the recommended adult dose and duration of treatment was estimated from the E2100 trial. Dose reductions were not modelled.

• The second used an average NHS cost for paclitaxel (total average cost per patient, £649) based on the average price paid by NHS trusts over a 4-month period, and a patient access scheme for bevacizumab whereby the NHS covers the cost of the first 10 g bevacizumab needed for each patient. Sensitivity analyses were only provided for this case. The patient access scheme (10-g cap) for bevacizumab was not approved by the Department of Health and was not considered by the Committee.

The base-case utility values were taken from 1 study that derived proxy utility values from oncology nurses, using the standard gamble technique. The values were 0.73 for progression-free survival (this was an average of values of 0.81 for response and 0.65 for stable disease), 0.45 for progressive disease, and -0.21 for disutility from febrile neutropenia and peripheral sensory neuropathy (both applied only in month 1 of experiencing the event). It was assumed that the remaining adverse events (hypersensitivity, infection and hypertension) would not have a notable impact on health-related quality of life.

The results based on the NHS list prices indicated incremental costs of £30,469, £31,416 and £27,358 and incremental quality-adjusted life years (QALYs) of 0.259, 0.273 and 0.259 for bevacizumab plus weekly paclitaxel therapy relative to weekly paclitaxel, docetaxel and gemcitabine plus paclitaxel therapy respectively. The cost per QALY was £117,803, £115,059 and £105,777 for bevacizumab plus weekly paclitaxel therapy relative to weekly paclitaxel, docetaxel and gemcitabine plus paclitaxel therapy respectively. The results based on average prices paid by NHS trusts over a 4-month period for paclitaxel and the patient access scheme (not approved by the Department of Health) for bevacizumab were provided and indicated a lower cost per QALY for bevacizumab plus weekly paclitaxel therapy relative to weekly paclitaxel, docetaxel and gemcitabine plus paclitaxel therapy respectively, though remaining above £57,000 per QALY gained. The manufacturer stated that it can be inferred from the high incremental cost-effectiveness ratios (ICERs) in the model that bevacizumab plus docetaxel (the more expensive taxane) is unlikely to provide a more cost-effective outcome than the analysis presented in the submission and, hence, a full economic analysis of bevacizumab plus docetaxel was not presented. The manufacturer carried out sensitivity analyses only on the second base-case scenario, in which the average price of paclitaxel paid by NHS trusts over a 4-month period and the patient access scheme for bevacizumab were used, and it was observed that using different parametric functions for time to progression and alternative assumptions on treatment duration had the largest impact on the ICERs.

The manufacturer conducted further analyses in response to points of clarification requested by the ERG, incorporating into the model a comparison of bevacizumab plus weekly paclitaxel with 3-weekly paclitaxel alone. The ICER for bevacizumab plus weekly paclitaxel compared with 3-weekly paclitaxel was £59,339 per QALY gained using average prices paid by NHS trusts for paclitaxel and incorporating the patient access scheme for bevacizumab. The manufacturer also incorporated the results of the evidence synthesis into the economic model as opposed to assuming that all comparators were equally effective. This was also based on the average price of paclitaxel paid by NHS trusts and the patient access scheme for bevacizumab. This resulted in an ICER for bevacizumab plus weekly paclitaxel compared with docetaxel and gemcitabine plus 3-weekly paclitaxel of £59,310 and £51,795 per QALY gained respectively.

The ERG had several concerns about the selection and quality of the evidence presented in the manufacturer's original submission. The evaluation of the clinical effectiveness of bevacizumab was based primarily on a single trial comparing bevacizumab plus weekly paclitaxel with weekly paclitaxel alone. The ERG highlighted limitations in the methodological quality of the study: for example, lack of blinding and lack of data collection about treatments given after disease progression. The ERG noted concerns that, as the conclusions about health-related quality of life were based primarily on the analyses using extreme imputed values for patients who had died or whose disease had progressed, the significant improvement in the FACT-B score stated by the manufacturer may not be reliable. The ERG noted that the results reported in the manufacturer's submission were derived from interim analyses and suggested that more recent follow-up data would be valuable, particularly for survival outcomes. The ERG also noted that the E2100 trial suggested that overall survival was not statistically significantly different between the treatment arms. However, the ERG was unable to establish whether or not the lack of difference in overall survival was due to crossover between treatment groups or any other post-progression events, because these data were not collected in the trial.

The ERG noted that the manufacturer had not presented any data on the clinical effectiveness of bevacizumab plus docetaxel in its submission. The ERG noted that a trial of bevacizumab plus docetaxel compared with docetaxel plus placebo (the AVADO trial) had been excluded. In addition, the ERG also considered that data from the RIBBON-1 trial could potentially have been included. The RIBBON-1 trial was excluded by the manufacturer because of insufficient statistical power for the relevant docetaxel comparison. Data from both these studies were provided by the manufacturer in response to consultation on the appraisal consultation document (see sections 3.23 and 3.24).

The ERG also highlighted a number of concerns about the indirect comparison conducted by the manufacturer. The ERG noted that an additional study (the Will Weekly Win study comparing weekly paclitaxel with 3-weekly paclitaxel, for which there were some data reported in an abstract) had not been included by the manufacturer. The ERG noted that the inclusion criteria specified that studies could be included as long as fewer than 60% (rather than a strict majority of 50%) of patients were receiving second-line treatments for metastatic breast cancer. The ERG highlighted that the validity of the studies included in the indirect comparison had not been adequately assessed. The ERG also reported concerns about the methods of the indirect comparison, noting differences between patient populations and potentially important methodological limitations among the trials included in this comparison. Given these methodological limitations, the ERG did not consider the findings of the indirect comparison to be reliable.

The ERG had a number of concerns about the economic model submitted by the manufacturer. The ERG considered that the series of pair wise comparisons for bevacizumab plus paclitaxel relative to each separate comparator regimen were inappropriate. It stated that, to establish the correct estimate of the ICER for bevacizumab plus paclitaxel, a fully incremental analysis should have been conducted, comparing all the regimens simultaneously. In addition, the ERG noted that the model assumed that mortality after disease progression was independent of initial treatment. It assumed that the rate of death after progression was constant over time and the same for all initial treatments. This meant that the differences in mean progression-free survival between treatments were maintained in the estimates of mean overall survival. The ERG stated that this was likely to have led to overestimates of overall survival for bevacizumab plus paclitaxel versus paclitaxel alone compared with the results of the E2100 trial.

The ERG highlighted that the base-case model did not include the results from the indirect comparison and that the model made the assumption that all included comparators (docetaxel alone, paclitaxel alone and gemcitabine plus paclitaxel) were equally effective in terms of progression-free survival and overall survival. Additionally, in the second base case, the cost of bevacizumab was based on the NHS paying for a maximum dose of 10 g per patient, and this patient access scheme had not been agreed with the Department of Health. The cost of paclitaxel used in the second base case was based on the average price paid by NHS trusts over a 4-month period, whereas other proprietary prices were taken from the BNF 58. The ERG also reported that the utility values were taken from a non-systematic review of the literature. In addition, the ERG noted that the manufacturer had not attempted to map health-related quality of life data from the E2100 study (measured by the FACT-B instrument) to a preference-based measure or collate alternative values.

The ERG conducted 2 sets of exploratory incremental analyses, both including 3-weekly paclitaxel as a comparator. One was based on the revised results, that incorporated the indirect comparison undertaken by the manufacturer rather than assuming that all comparators were equally effective. The second analysis was based on the original approach employed by the manufacturer where all comparators were assumed to be equally effective. These analyses used the following drug acquisition costs:

• Case 1 (ERG re-analysis) – NHS list prices from BNF 58 excluding the patient access scheme for bevacizumab.

• Case 2 (manufacturer re-analysis) – average prices paid by NHS trusts over a 4-month period for paclitaxel including the patient access scheme for bevacizumab.

• Case 3 (ERG re-analysis) – average prices paid by NHS trusts for paclitaxel over a 4-month period excluding the patient access scheme for bevacizumab.
  
  In both analyses, for Case 1 and Case 3 (excluding the patient access scheme), the ICER for bevacizumab plus paclitaxel versus the next best treatment exceeded £100,000 per QALY gained.

The ERG agreed with the manufacturer's conclusion that bevacizumab plus paclitaxel and bevacizumab plus docetaxel would be expected to be of similar effectiveness. Therefore, the inclusion and exclusion of studies in the indirect comparison would not have a major effect. The analyses by the ERG found that the acquisition cost of docetaxel had very little effect on the ICER of bevacizumab plus paclitaxel compared with docetaxel.

The ERG conducted further exploratory analyses and calibrated the model to the E2100 trial results for overall survival. This was considered important to test the internal validity of the model by comparing the median progression-free survival and overall survival found by the E2100 trial with the model predictions. The pair wise ICER of bevacizumab plus paclitaxel versus weekly paclitaxel was £259,267 per QALY gained (incremental cost of £29,675 and incremental QALY of 0.114) in the exploratory analyses.

After consultation on the appraisal consultation document, the manufacturer provided summaries of the results from the AVADO and RIBBON-1 trials. In addition, new evidence for 2 subgroups was provided:

• patients who had previously received treatment with a taxane and

• patients with disease that was triple negative (that is, oestrogen and progesterone receptor negative, and HER2 negative).

The ERG noted that 25% (615 out of 2,447) of all patients included in the intention-to-treat meta-analysis that was used to address the triple negative subgroup received bevacizumab plus capecitabine, a combination that is outside the scope of this appraisal. The ERG highlighted that the progression-free survival and overall survival hazard ratios for the intention-to-treat population and the triple negative subgroup were almost identical. Furthermore, the median progression-free survival, overall survival and 1-year survival data were not reported for this subgroup. It was noted that these data were not provided for the RIBBON-1 study.

The ERG noted that the meta-analysis addressing the prior taxane-treated subgroup was described as an exploratory meta-analysis. It stated that it included only a small number of patients from trials that individually were insufficiently powered to detect a difference in overall survival. Only 1 trial (AVADO) appeared to stratify for taxane pre-treatment and no interaction tests were conducted. Therefore, although the hazard ratios did show a trend towards being more favourable towards bevacizumab than in the intention-to-treat meta-analysis for both progression-free survival and overall survival, the ERG stated that the analysis cannot be considered as convincing evidence of a subgroup effect.

The ERG had a number of concerns about the economic modelling. The ERG highlighted that rather than using estimates from these subgroup meta-analyses in the subsequent economic model, only the subgroup data from the E2100 trial were used. The ERG noted that this approach may have yielded more favourable ICER estimates because the progression-free survival and overall survival values for E2100 were more favourable than those from the meta-analyses.

The ERG commented that, although the manufacturer selected the log-logistic model as the best fit, the difference in goodness of fit statistics was small, suggesting that the choice between different functions was marginal. However, there was subsequent variation in the ICER estimates based on the different survival functions. At the extreme ends, for the prior taxane-treated subgroup, the ICER using the log-logistic function (the manufacturer's base case), and incorporating the average price for paclitaxel paid by NHS trusts, was £67,714 per QALY gained, whereas if the Weibull function had been used this figure would have risen to £86,854 per QALY gained. When the NHS list prices for paclitaxel were incorporated, the ICERs ranged from £74,640 per QALY gained with the log-logistic function to £95,807 per QALY gained with the Weibull function. The ERG concluded that there was considerable uncertainty surrounding the choice of statistical function used in the cost-effectiveness analysis for the prior taxane-treated subgroup. Equivalent goodness of fit statistics were not reported for the triple negative subgroup so the ERG could not undertake a similar assessment for this subgroup.

The ERG undertook an exploratory analysis in the prior taxane-treated subgroup using the hazard ratio estimated from the individual patient data meta-analysis of the prior taxane-treated group presented by the manufacturer (HR=0.738) and the Weibull function. The results demonstrated an increase in the ICER from £95,807 to £109,242 per QALY gained using the list prices for paclitaxel and from £86,854 to £98,834 per QALY gained using the average price for paclitaxel paid by NHS trusts.

The ERG noted that all the parametric functions investigated by the manufacturer assumed long-term sustained treatment effects for overall survival with bevacizumab plus paclitaxel. Consequently, the ERG undertook an exploratory analysis for both subgroups using the manufacturer's model. This analysis used the Kaplan-Meier survival estimates from the E2100 study up to about 3.2 years, assuming no difference in survival after that point, since the curves suggested that the difference in overall survival may not be sustained over a longer time horizon. For the prior taxane-treated subgroup, the ICERs increased from £74,640 per QALY gained when using the log-logistic function to £129,794 per QALY gained when using the Kaplan-Meier function, both incorporating the list prices for paclitaxel, and from £67,714 per QALY gained when using the log-logistic function to £117,587 per QALY gained when using the Kaplan-Meier function, both incorporating the average price for paclitaxel paid by NHS trusts. For the triple negative subgroup, the ICERs increased from £87,865 per QALY gained when using the log-logistic function to £187,339 per QALY gained when using the Kaplan-Meier function, incorporating the list prices for paclitaxel, and from £82,469 per QALY gained when using the log-logistic function to £175,575 per QALY gained when using the Kaplan-Meier function, incorporating the average price for paclitaxel paid by NHS trusts.

Full details of all the evidence are in the manufacturer's submission and the ERG report.