Bevacizumab in combination with paclitaxel and carboplatin for first-line treatment of advanced ovarian cancer

The Committee considered that the main source of evidence for the clinical effectiveness of bevacizumab plus paclitaxel and carboplatin was GOG‑0218. The Committee agreed this was a well-designed double-blind randomised trial. It noted the Evidence Review Group's (ERG's) assessment of the quality of this trial, and accepted that the results of the trial are relevant to the first-line treatment of patients with advanced ovarian cancer in the NHS. The Committee noted that there were 3 arms in the trial but only the CPB15+ arm (bevacizumab given with chemotherapy and then for up to 15 months as maintenance therapy) showed a statistically significant improvement in PFS compared with chemotherapy alone. The CPB15 arm of the trial (bevacizumab given only with chemotherapy) showed no statistically significant PFS benefit compared with the control arm. The Committee concluded that GOG‑0218 provided relevant evidence for this appraisal and that bevacizumab was shown to be clinically effective only when it is given within its marketing authorisation, that is, at the same time as paclitaxel and carboplatin and then as a maintenance treatment for up to 15 months.

The Committee noted that PFS results from GOG‑0218 were reported in the manufacturer's submission using both censored data (in which patients with a CA‑125 rise were censored at the time of their previous scan and their results removed from further PFS analysis), and uncensored data (in which a CA‑125 rise indicated progression). The Committee heard from the clinical specialists that a rise in CA‑125 alone was not used as an indication to change treatment because this approach had not been shown to alter prognosis and also approximately a third of patients did not express CA‑125. The clinical specialists also commented that a CA‑125 rise often suggested the disease was progressing but that it could take up to 6 months or so for progression to become apparent. The Committee concluded that a significant rise in CA‑125 is an indicator of progression and might be an early marker, but is not usually used in clinical practice in the UK as the sole indicator of progression.

The Committee discussed the most relevant PFS results for the clinical effectiveness of bevacizumab plus paclitaxel and carboplatin in the UK. The Committee noted that the results from the uncensored data of a 3.8-month difference in median PFS in favour of bevacizumab was less than the 6-month difference in median PFS obtained from the censored data. The Committee noted the ERG's concerns about the lack of consistent reporting of the various PFS assessments at all time points. The Committee noted that clinical advice to the ERG was that the uncensored data are the most relevant to the NHS. However, the Committee heard from the clinical specialists that, in their opinion, the censored data are more relevant because patients in the NHS would not be treated as progressed based on a CA‑125 rise alone (see section 4.5). The Committee was aware that the censored data had excluded patients with raised CA‑125 from the analysis and, because these patients could be regarded as being at high risk of progression, this could have led to bias in the results. The Committee noted that all the analyses of PFS data presented in the manufacturer's submission taken at different time points, both censored and uncensored, showed a difference in median PFS in favour of bevacizumab of between 3.8 and 6 months. The Committee concluded that bevacizumab plus paclitaxel and carboplatin improved PFS compared with paclitaxel and carboplatin alone and that, of the available data, the censored PFS data are more relevant to UK clinical practice, although the Committee was aware of the potential bias introduced by censoring the data from patients with raised CA‑125.

The Committee considered the overall survival results from GOG‑0218 and noted concerns from the ERG that switching patients in the control arm to bevacizumab after progression had confounded the overall survival analysis. The Committee noted the various estimates given in the manufacturer's submission of the percentage of patients switching and concluded that the precise extent of switching was unclear. It heard from the manufacturer that approximately 40% of patients in the chemotherapy arm compared with 20% of patients in the bevacizumab arm subsequently received bevacizumab after disease progression. The Committee considered that patient crossover from the control arm would have an impact on the overall survival results only if second-line treatment with bevacizumab was more effective than other therapies given after progression, and the Committee did not have this information. Nevertheless, the Committee accepted that the interpretation of overall survival figures from GOG‑0218 was problematic, and that the non-statistically significant difference in median overall survival of 3.2 months attributed to bevacizumab should be interpreted cautiously. The Committee concluded that the overall survival benefit of bevacizumab plus carboplatin and paclitaxel is uncertain from the results of GOG‑0218 because of the uncertainty related to the extent to which patients received bevacizumab after progression and the impact of this.

The Committee noted the adverse events reported in GOG‑0218. It understood that these events were as predicted from other studies with bevacizumab and did not raise new safety concerns. The Committee heard from the clinical specialists that most adverse events could be satisfactorily managed. The Committee concluded that adding bevacizumab to a paclitaxel and carboplatin regimen did not lead to unacceptable toxicity compared with paclitaxel and carboplatin alone and that adverse events were manageable.

The Committee discussed the evidence from the supporting ICON7 trial. It heard from the clinical specialists that the strengths of this trial were that it included some UK patients, and used the most appropriate definitions of progression, disease staging and surgical debulking. The clinical specialists stated that patients in the optimal debulking subgroups differed between the GOG‑0218 and ICON7 trials. The clinical specialists considered that this is shown by fewer patients in the group with optimally debulked cancer in GOG‑0218 having complete resection (that is, no visible disease) than in ICON7. The Committee also heard from the clinical specialists that this might be because another trial, conducted at the same time as GOG‑0218, may have enrolled the patients who had complete resection. The Committee noted that the information on the proportion of patients with stage 3 and 4 disease in ICON7 who had completely resected disease was not available, and that the marketing authorisation for bevacizumab did not specify complete or incomplete resection. The Committee also noted that, in ICON7, no patient crossover was allowed after disease progression. The Committee was aware of the disadvantages of ICON7 relative to the NICE decision problem, including its open-label design and inclusion of some patients with stage 1 and 2 cancer, which is not covered by the marketing authorisation for bevacizumab. In addition, the trial used a lower dose and shorter duration of bevacizumab treatment than is now licensed. Nevertheless, 81% of the patients were covered by the marketing authorisation and the Committee considered that the results from the trial contributed to the body of knowledge about the efficacy of bevacizumab plus paclitaxel and carboplatin for advanced ovarian cancer.

The Committee considered the PFS results from ICON7. It noted that the overall difference between the PFS medians in the intention-to-treat population was 2.4 months, which was less than the 6 months in the intention-to-treat population in GOG‑0218 using the censored data. The clinical specialists emphasised that professional opinion was that this difference was related to patient selection and patient characteristics, not to the lower dose and duration of treatment in ICON7. The Committee also noted that, PFS in the chemotherapy comparator arm was worse in the high-risk subgroup from ICON7 than in the intention-to-treat population in GOG‑0218 using the censored data, which could affect interpretation of the results. However, the Committee noted that the high-risk subgroup from ICON7 excluded patients with optimally debulked cancer, whereas the intention-to-treat population in GOG‑0218 included approximately one-third of patients with optimally debulked cancer who might be expected to have a better prognosis, and who experienced the longest PFS of the subgroups in the GOG‑0218 chemotherapy arm. The Committee concluded that the trials were difficult to compare because of different inclusion criteria, bevacizumab dose and duration of treatment, definitions of progression and optimal debulking, and differing baseline factors between the trials.

The Committee considered the results presented for the ICON7 high-risk subgroup. This subgroup was broadly comparable with 2 of the 3 stratified groups in GOG‑0218 but did not represent the whole population covered by the marketing authorisation, which does not specify debulking status. Separate analysis of the high-risk subgroup showed a difference in median PFS of 5.5 months in favour of bevacizumab. This was higher than the 2.4-month difference in the ICON7 intention-to-treat population, and was comparable to the gain in the intention-to-treat GOG‑0218 population using the censored data (6 months; see section 3.4). Hazard ratios were not provided for the non-high-risk subgroup (that is, the intention-to-treat population minus the high-risk subgroup) but the Committee assumed that they would have shown little or no benefit. The Committee also considered the PFS results by cancer stage and debulking status in the population of ICON7 covered by the marketing authorisation. It noted that there was an apparent differential response, with little benefit shown in the stage 3 population with optimally debulked cancer (difference in median PFS 1.6 months in favour of bevacizumab) compared with the population with stage 3 suboptimally debulked cancer (difference in median PFS 6.8 months) or stage 4 cancer (difference in median PFS 3.4 months). The Committee heard from the clinical specialists that these PFS analyses from ICON7 showed benefit in only a proportion of the population covered by the marketing authorisation. The Committee noted that the subgroup data from GOG‑0218 showed a benefit, which was not dependent on debulking status in stage 3 and 4 cancer, with the greatest PFS benefit shown in the optimally debulked subgroup (based on uncensored data from GOG‑0218; corresponding censored subgroup data were not supplied by the manufacturer, see section 3.5). The Committee also noted that, although the results from ICON7 suggested a greater PFS benefit in patients with stage 3 suboptimally debulked or stage 4 cancer, the uncensored subgroup data from GOG‑0218 did not appear to support the manufacturer's suggestion that bevacizumab provides greater benefit for ovarian cancer patients with a poor prognosis. The Committee agreed that the results of the open-label ICON7 trial indicated a smaller PFS benefit in the intention-to-treat population than was seen in GOG‑0218, and also suggested a different benefit based on cancer stage and debulking status that had not been shown in GOG‑0218. The Committee was aware that several hypotheses could explain these differences. The Committee concluded that the ICON7 data contributed to confidence that treatment with bevacizumab could delay progression, but that the reasons for the apparent differential response and differences in PFS suggested by the ICON7 subgroup analysis compared with the analysis of GOG‑0218 uncensored data remained uncertain.

The Committee examined the overall survival data from ICON7 and noted that mature data on the intention-to-treat population were not yet available but an interim analysis of the high-risk subgroup showed a difference in median overall survival of 7.8 months in favour of bevacizumab. The Committee noted that, taking into account the shape of the Kaplan–Meier curve from the interim analysis of the high-risk patients, it is likely that the mean overall survival benefit would be much less than the median. The Committee concluded that the interim data for a subgroup in the trial suggested a difference in overall survival in favour of bevacizumab, but that this should be interpreted with caution.

The Committee discussed the cost-effectiveness estimates and the assumptions on which these were based from the manufacturer's economic model based on GOG‑0218. The Committee concluded that the model adhered to the NICE reference case for economic analysis and was acceptable for assessing the cost effectiveness of bevacizumab plus paclitaxel and carboplatin for first-line treatment of advanced ovarian cancer.

The Committee considered the model inputs including clinical effectiveness, patient outcomes, resource use and costs. It noted that PFS was based on the Kaplan–Meier curve from the censored data of GOG‑0218 until month 28, after which PFS is represented by a log-logistic parametric function. The Committee also noted that the model used an equal post-progression death rate between the arms. It understood the ERG's concerns about the treatment duration of 12 months instead of the15 months specified in the marketing authorisation and the time horizon of 10 years. The ERG did not consider this time horizon to be long enough, although the Committee heard from the clinical specialists that 10 years was probably appropriate because only a very small number of patients are likely to survive beyond 10 years. The Committee noted that the EQ‑5D utilities from the expanded high-risk subgroup of ICON7 (see section 3.15) were used in the model and that the same utilities were assumed in both arms of the model. It also noted that no disutilities associated with adverse events had been incorporated into the model. The Committee concluded that the model inputs used by the manufacturer were reasonable.

The Committee considered the most plausible incremental cost-effectiveness ratios (ICERs) from the model based on the GOG‑0218 trial presented by the manufacturer and by the ERG in their exploratory analyses. It noted that the manufacturer's base-case ICER was approximately £144,000 per quality-adjusted life year (QALY) gained. The Committee considered the ERG's exploratory analyses, which examined the changes in the ICER with a treatment duration of 15 months or a time horizon of 25 years or both, and gave a range of ICERs from £128,000 to £161,000 per QALY gained. The Committee agreed that the range of ICERs obtained from the cost-effectiveness model of bevacizumab plus paclitaxel and carboplatin were outside the range normally considered as a cost-effective use of NHS resources. It therefore concluded that bevacizumab within its marketing authorisation (that is, at a dose of 15 mg/kg), plus paclitaxel and carboplatin, would not be a cost-effective use of NHS resources for first-line treatment of advanced ovarian cancer compared with paclitaxel and carboplatin alone.

After receiving comments from the manufacturer in response to the appraisal consultation document, the Committee further considered the impact of patient crossover on overall survival in the GOG‑0218 study as previously discussed (see section 4.7) and how this had been accounted for in the economic model. The Committee was aware that, because a proportion of patients in the chemotherapy arm from the GOG‑0218 study subsequently received bevacizumab after disease progression, the manufacturer had assumed a similar probability of death in both treatment arms in its base-case analysis. The Committee also noted from the ERG that, by using this approach, the model resulted in a greater survival difference in favour of bevacizumab than was observed in the GOG‑0218 study (see section 3.22). The Committee noted that this would result in a lower ICER than if the overall survival observed in the GOG‑0218 study had been used. The Committee heard from the manufacturer that it did not consider it appropriate to use other alternative approaches to adjust for crossover, such as the rank-preserving structural failure time method. The Committee noted the manufacturer's response to the appraisal consultation document, which attempted to adjust for crossover in the trial by applying the overall survival curves estimated for the control and treatment arms in the expanded high-risk subgroup from the ICON7 study, rather than using post-progression survival data from the GOG‑0218 study. The manufacturer justified this approach on the basis that the expanded high-risk subgroup from the ICON7 study, which did not permit crossover at progression, was broadly comparable to the intention-to-treat population in the GOG‑0218 study. However, the Committee agreed that this was an unconventional approach that lacked credibility because of the significant differences identified between the expanded high-risk subgroup in the ICON7 study and the intention-to-treat population in the GOG‑0218 study (see section 4.10). The Committee concluded that the manufacturer's novel approach to adjust for patient crossover in the GOG‑0218 study was not a robust basis on which to estimate the cost effectiveness of bevacizumab, and agreed that the base-case ICER remained the most plausible one on which to base its decision.

The Committee considered the comments received by the consultees and commentators on the cost-effectiveness estimate for bevacizumab at its unlicensed dose of 7.5 mg/kg, noting that the manufacturer had submitted an economic analysis of bevacizumab at the unlicensed dose based on ICON7. The Committee also noted from 1 consultee comment that an estimate of the cost effectiveness of bevacizumab at its unlicensed dose had been submitted to and presented by the Scottish Medicines Consortium. The Committee noted that the cost-effectiveness estimate presented to the Scottish Medicines Consortium by the manufacturer differed from the one that was submitted to NICE. The Committee was aware that the ERG had not provided a detailed critique of the ICON7 economic model because it was based on the unlicensed dose of bevacizumab and therefore outside the scope of this appraisal. Therefore, the Committee concluded that it was unable to comment on the validity of the cost-effectiveness analysis of bevacizumab for the first-line treatment of advanced ovarian cancer at its unlicensed dose of 7.5 mg/kg.

The Committee also discussed whether it could comment on the use of bevacizumab for the first-line treatment of advanced ovarian cancer at its unlicensed dose of 7.5 mg/kg, noting the request for this from the comments received from consultees and commentators in response to the appraisal consultation document. The Committee noted from the European Medicines Agency's statement that there was insufficient evidence of an acceptable balance of clinically relevant benefit to risk at the lower dose (7.5 mg/kg) used in the ICON7 study. In response to the Committee's question as to whether it was able to recommend a drug outside its licensed dose, NICE reiterated its position that the Committee was only permitted to make a recommendation on the licensed dose of bevacizumab (15 mg/kg). The Committee therefore concluded that it was reasonable not to consider further the cost effectiveness of bevacizumab at its unlicensed dose.

The Committee discussed whether bevacizumab should be considered an innovative treatment. The Committee acknowledged that advanced ovarian cancer is a disease with limited treatment options, and that bevacizumab represented a novel biological approach to therapy. It also noted the clinical specialists' comments (see section 4.2). However, the Committee concluded that all benefits of a substantial nature relating to treatment with bevacizumab plus paclitaxel and carboplatin had been captured in the QALY calculation.