3.1 The key evidence for the clinical effectiveness of bevacizumab plus paclitaxel and carboplatin came from 1 randomised controlled trial (GOG‑0218). The trial assessed the efficacy and safety of bevacizumab (at its licensed dose of 15 mg/kg body weight) plus paclitaxel and carboplatin in people with previously untreated stage III (incompletely resected) or stage IV epithelial ovarian, fallopian tube or primary peritoneal cancer who had undergone debulking surgery. This evidence was supported by results from a randomised open-label trial (ICON7) that assessed the efficacy and safety of bevacizumab at an unlicensed dose (7.5 mg/kg body weight) plus paclitaxel and carboplatin in people with high-risk early stage or advanced epithelial ovarian, fallopian tube or primary peritoneal cancer.
3.2 GOG‑0218 was a double-blind randomised placebo-controlled multicentre trial conducted in North America and Asia, and included 1873 patients with previously untreated stage III or stage IV epithelial ovarian, fallopian tube or primary peritoneal cancer who had undergone debulking surgery. The trial was for up to 15 months and patients were randomised to 1 of 3 treatment arms:
The CPP (carboplatin, paclitaxel and placebo) control group (n=625) received standard chemotherapy (carboplatin at a target area under the curve of 6 mg/ml•min and paclitaxel 175 mg/m2 every 3 weeks for 6 cycles), plus placebo for cycles 2 to 22.
The CPB15 (carboplatin, paclitaxel and bevacizumab [15 mg/kg]) group (n=625) received the same standard chemotherapy as the CPP group, plus bevacizumab (15 mg/kg) for cycles 2 to 6 and placebo as monotherapy for cycles 7 to 22.
The CPB15+ group (n=623) received the same standard chemotherapy as the CPP group, plus bevacizumab (15 mg/kg) for cycles 2 to 22.
3.3 Cycles lasted 3 weeks and treatment was discontinued at the onset of disease progression, unacceptable toxic effects, completion of all 22 cycles or withdrawal. Patients in the control arm were allowed to cross over to receive bevacizumab after disease progression. Randomisation was stratified for Gynaecologic Oncology Group (GOG) performance status (0, 1 or 2), and cancer stage and debulking status (optimally debulked stage III [with maximum residual lesion diameter of 1 cm or less], suboptimally debulked stage III [with maximum residual diameter of more than 1 cm] or stage IV). The primary outcome was progression-free survival (PFS), defined as the period from randomisation to disease progression or death. Progression was assessed by the investigator based on any of the following measures: global clinical deterioration, Response Evaluation Criteria in Solid Tumours (RECIST) or rising serum cancer antigen‑125 (CA‑125). CA‑125 progression was defined as at least twice the nadir or upper limit of normal. Secondary outcomes included overall survival, objective response rate and health-related quality of life measured using the Functional Assessment of Cancer Therapy‑Ovarian (FACT‑O) questionnaire, the Ovarian Cancer Subscale measure and abdominal discomfort score.
3.4 The primary efficacy analysis of PFS used censored data from September 2009 in which patients with disease progression based on rising serum CA‑125 alone and patients who received non-protocol therapies before progression were censored at the time of their previous scan and excluded from the analysis. Based on an investigator assessment, the censored data showed a statistically significant improvement of 6 months in the difference between the median PFS of the CPB15+ arm and the CPP arm (CPP 12 months, CPB15+ 18 months; hazard ratio [HR] 0.645, 95% confidence interval [CI] 0.551 to 0.756, p<0.001). There was a 0.7-month difference in median PFS in favour of the CPB15 arm compared with the CPP arm (CPP 12 months, CPB15 12.7 months; HR 0.84, 95% CI 0.71 to 0.99, p=0.0204). An Independent Review Committee assessment of these data showed similar results: a 6-month difference in median PFS in favour of the CPB15+ arm compared with the CPP arm (CPP 13.1 months, CPB15+ 19.1 months; HR 0.62, 95% CI 0.50 to 0.77, p<0.0001) but only a non-statistically significant 0.1-month difference in median PFS in the CPB15 arm compared with the CPP arm (CPP 13.1 months, CPB15 13.2 months; HR 0.93, 95% CI 0.76 to 1.13, p=0.222). A GOG protocol-specified analysis of PFS was undertaken in February 2010 and the results were presented without censoring for CA‑125 progression or use of non-protocol therapy before disease progression. The difference in the median PFS was 3.8 months in favour of the CPB15+ arm compared with the CPP arm (CPP 10.3 months, CPB15+ 14.1 months; HR 0.717, 95% CI 0.625 to 0.824, p<0.0001) and 0.9 months in favour of the CPB15 arm compared with the CPP arm, although this difference was not statistically significant (CPP 10.3 months, CPB15 11.2 months; HR 0.908, 95% CI 0.795 to 1.040, p=0.16).
3.5 A subgroup analysis by cancer stage and debulking status using the uncensored data from February 2010 suggested that the improvement in PFS between CPB15+ and CPP was maintained across all subgroups: patients with stage III optimally debulked cancer showed a 5.1-month improvement in PFS in the CPB15+ compared with the CPP arm (CPP 12.4 months, CPB15+ 17.5 months; HR 0.66, 95% CI 0.5 to 0.86); patients with stage III suboptimally debulked cancer showed a 3.8-month improvement in PFS in the CPB15+ compared with the CPP arm (CPP 10.1 months, CPB15+ 13.9 months; HR 0.78, 95% CI 0.63 to 0.96); patients with stage IV cancer showed a 3.3-month improvement in PFS in the CPB15+ compared with the CPP arm (CPP 9.5 months, CPB15+ 12.8 months; HR 0.64, 95% CI 0.49 to 0.82).
3.6 The overall survival analysis was calculated in August 2011 when 46.9% of patients had died. The median overall survival was 3.2 months longer in the CPB15+ arm than in the CPP arm (CPP 40.6 months, CPB15+ 43.8 months; HR 0.88, 95% CI 0.75 to 1.04, p=0.0641). However, this was not statistically significant at the p value boundary of 0.0116. The manufacturer stated that significant patient crossover from the control arm after progression would have confounded the data. The manufacturer's submission contained 2 estimates of the proportion of patients in the control arm receiving bevacizumab after progression: 27.7% and up to 40%.
3.7 ICON7 was a randomised open-label multicentre study conducted in Europe, and included 1528 patients with high-risk early stage or advanced stage IV epithelial ovarian, fallopian tube or primary peritoneal cancer. The trial was for up to 12 months and patients were randomised to 1 of 2 treatment arms:
The CP (carboplatin and paclitaxel) control group (n=764) received standard chemotherapy (carboplatin at a target area under the curve of 5 or 6 mg/ml•min and paclitaxel 175 mg/m2 every 3 weeks for 6 cycles).
The CPB7.5+ arm (n=764) received the same standard chemotherapy as the CP group plus bevacizumab (7.5 mg/kg) every 3 weeks for 6 cycles, and continued for an additional 12 cycles or until disease progression.
3.8 Randomisation was stratified for cancer stage and residual disease post surgery (category 1 – FIGO stage I–III with residual disease less than 1 cm; category 2 – FIGO stage I–III with residual disease more than 1 cm; category 3 – FIGO stage IV and inoperable FIGO stage III) and the time of initiation of chemotherapy (intention-to-start chemotherapy 4 weeks after surgery or sooner, or intention-to-start chemotherapy more than 4 weeks after surgery). Patients received treatment until disease progression, unacceptable toxicity or completion of 6 or 18 cycles of therapy as appropriate. No crossover was permitted. A pre-specified, but not stratified, subgroup included 31% (n=462) of patients with high-risk disease (defined as stage III suboptimally debulked or stage IV debulked ovarian cancer). The primary outcome of the trial was PFS based on RECIST on the basis of radiological, clinical and symptomatic indicators of progression. Secondary outcome measures included overall survival and quality of life. Health-related quality of life was measured using the European Organization for the Research and Treatment of Cancer (EORTC) QLQ‑C30 and QLQ‑OV28 questionnaires.
3.9 The manufacturer's submission presented analysis of PFS from ICON7 based on a data cut-off of November 2010. For the intention-to-treat population, the difference in median PFS was 2.4 months in favour of bevacizumab (CP 17.4 months, CPB7.5+ 19.8 months; HR 0.87, 95% CI 0.77 to 0.99, p<0.04). For the high-risk subgroup there was a statistically significant difference in median PFS of 5.5 months in favour of bevacizumab (CP 10.5 months, CPB7.5+ 16.0 months; HR 0.73, 95% CI 0.60 to 0.93, p=0.002). A pre-planned exploratory analysis of PFS using subgroups defined by cancer stage and debulking status was also reported. These analyses showed a statistically significant improvement in favour of bevacizumab for patients with stage III suboptimally debulked cancer (difference in median PFS of 6.8 months based on CP 10.1 months [n=154] and CPB7.5+ 16.9 months [n=140]; HR 0.67, 95% CI 0.52 to 0.87). However, no statistically significant differences between treatment arms were observed for stage III patients with optimally debulked cancer (difference in median PFS of 1.6 months based on CP 17.7 months [n=368] and CPB7.5+ 19.3 months [n=383]; HR 0.89, 95% CI 0.74 to 1.07) or patients with stage IV cancer (difference in median PFS of 3.4 months based on CP 10.1 months [n=97] and CPB7.5+ 13.5 months [n=104]; HR 0.74, 95% CI 0.55 to 1.01).
3.10 The protocol-specified final overall survival analysis for ICON7 has not yet been reported. An exploratory overall survival analysis of the intention-to-treat population, conducted when approximately 25% of patients had died, could not calculate the median duration of overall survival because of low numbers, but gave a hazard ratio of 0.85 (95% CI 0.69 to 1.04). An interim overall survival analysis was conducted in the high-risk subgroup when approximately 47% of patients had died in the CP arm and 34% had died in the CPB7.5+ arm. There was a statistically significant difference in the median overall survival of 7.8 months in favour of bevacizumab (CP 28.8 months, CPB7.5+ 36.6 months; HR 0.64, 95% CI 0.48 to 0.85, p=0.002).
3.11 The manufacturer did not consider that a meta-analysis was appropriate because GOG‑0218 and ICON7 used different doses and durations of bevacizumab, and different study populations.
3.12 Almost all patients in GOG‑0218 experienced at least 1 adverse event. Incidences of stomatitis, dysarthria, headache, epistaxis and hypertension were more than 10% higher in the bevacizumab arms than in the placebo arm. Incidences of hypertension, gastrointestinal perforation and non-central nervous system bleeding (adverse events of special interest, grade 3–5) were at least 1% higher in the CPB15+ arm than in the CPP arm.
3.13 The manufacturer submitted a de novo economic analysis that assessed the cost effectiveness of bevacizumab plus carboplatin and paclitaxel compared with carboplatin and paclitaxel only for first-line treatment in women with stage III or IV ovarian cancer. The model was a 3-state semi-Markov model with health states consisting of PFS, progressed disease and death. Data from GOG‑0218 were used to inform model inputs for dosing, survival and safety. Both the intervention and comparator in the model were used in accordance with their marketing authorisations. The manufacturer also presented an economic analysis of bevacizumab at its unlicensed dose of 7.5 mg/kg based on ICON7, the results of which are not presented here. The analysis was conducted from an NHS and personal social services perspective, the costs and outcomes were discounted at 3.5% per year and a 10-year time horizon was used.
3.14 PFS in the model uses the Kaplan–Meier survival curves from the GOG‑0218 trial data up to the convergence of the intervention and comparator arms at month 28. The data are from the updated PFS analysis (February 2010), which includes censoring of patients who were presumed to experience progression based on rising CA‑125 levels or who switched to non-protocol therapies. The manufacturer examined the fit of various parametric survival models to the progression-free data and considered a log-logistic model the best fit to extrapolate survival times beyond month 28. In the progressed disease state, the weekly probability of death was assumed to be constant and the same for both arms of the model.
3.15 The model incorporates patients' health-related quality-of-life outcomes using health-state utility values for the PFS and progressed disease states. The manufacturer applied EQ‑5D utility values from an expanded high-risk subgroup in the ICON7 study, which included all patients with stage III disease with suboptimal debulking or stage IV disease or patients with unresectable disease (n=495). In the PFS state, a log-rank test showed that there was no difference in the utility values across the intervention and comparator arms; therefore, the same utility values were used in both arms of the model. In the PFS state, the values varied with time and in the progressed state, a constant value was used because of the limited data available. The disutilities associated with adverse effects were assumed to have been captured in the assessment of health-related quality of life in ICON7.
3.16 Drug costs were estimated using the dose and frequency of administration in the summary of product characteristics. Data from a UK cohort study were used in the dose calculations. The base case assumed that any unused carboplatin or paclitaxel from a vial is reallocated and not wasted, whereas for bevacizumab it assumed that any unused drug in a vial is wasted. The costs per patient per cycle were £2229 for bevacizumab, £21.80 for paclitaxel and £18.51 for carboplatin. The costs associated with pharmacy preparation of the infusion and its outpatient administration in hospital (based on NHS reference costs 2010/11) were included in the model. The weekly costs of supporting patients in the PFS and progressed health states were included in the model. Post-progression drug acquisition costs were not included in the model because this information was not available in sufficient detail from GOG‑0218. Costs associated with adverse events that occurred at grade 3 or 4 severity in more than 2% of patients were incorporated into the analysis.
3.17 The base-case results estimated that adding bevacizumab to standard chemotherapy provides an additional 0.228 life years (0.188 quality-adjusted life years [QALYs], resulting from a 0.243 QALY gain in the PFS state and a 0.055 QALY loss in the progressed disease state) to patients with an expected survival of approximately 4 years. This benefit is achieved with an incremental cost of £27,089, resulting in an incremental cost-effectiveness ratio (ICER) of £144,066 per QALY gained for the licensed dose of bevacizumab plus carboplatin and paclitaxel, compared with carboplatin and paclitaxel alone. The manufacturer's deterministic sensitivity analysis suggested that the cost-effectiveness results are influenced by the parametric functions used for the PFS extrapolation and the time horizon used in the model. The manufacturer's scenario analyses identified the key drivers of the cost-effectiveness results as the dose and duration of bevacizumab treatment.
3.18 The ERG considered that GOG‑0218 provided evidence of the clinical effectiveness of bevacizumab plus carboplatin and paclitaxel for the first-line treatment of people in the NHS with advanced ovarian cancer, as defined in the scope. It noted that the population from the trial is generally representative of patients treated in secondary care in the UK, although it may not fully represent patients with comorbidities.
3.19 The ERG was concerned that the different assessments of PFS (by investigator and Independent Review Committee, CA‑125 censored, CA‑125 not censored) were not consistently reported for all time points. It commented that there may have been selective reporting of data and it is not clear what impact this may have on conclusions. In response to a request for clarification, the manufacturer stated that updated PFS data censored for CA‑125 are not available; also, exploratory analyses were not updated because they were intended only to confirm the validity of investigator-assessed PFS. The ERG considered that, although the direction of the evidence is consistent, the size of effect varies with the different analyses and over time. For example, the difference in median PFS varied from 4 to 6 months; the hazard ratio varied from 0.62 (assessed by Independent Review Committee, data censored) to 0.77 (updated investigator assessed, without data censoring). Clinical advice to the ERG suggested that CA‑125 is used routinely in UK clinical practice; therefore, the ERG considered that results not censored for CA‑125 rises were most relevant to the UK.
3.20 The ERG considered that the structure adopted for the economic model based on GOG‑0218 was reasonable, and consistent with previous economic evaluations developed for advanced cancer. The methods of analysis were generally appropriate and conformed to the NICE reference case. The ERG noted that a time horizon of 10 years was used in the model. However, the ERG considered a longer time horizon would have been more appropriate because approximately 10% of patients were still alive after 10 years. The ERG agreed that the parameters used for the model were generally appropriate.
3.21 The ERG highlighted that the clinical-effectiveness data used in the model included censoring for patients with rising CA‑125 levels and for patients who switched to non-protocol therapies. It considered that the hazard ratio from these data was relatively favourable compared with other PFS hazard ratios from the trial and this may have produced a more favourable cost-effectiveness estimate. The ERG also noted that the treatment duration was 12 months rather than the 15 months specified in the summary of product characteristics.
3.22 The ERG highlighted that, in the trial, overall survival between the arms was similar, with median values of 39.8 months for the bevacizumab arm and 39.4 months for the chemotherapy-only arm. However, in the model, there is a 2-month difference in mean overall survival between the arms (bevacizumab: 47 months, chemotherapy-only: 45 months).
3.23 The ERG also considered that the uncertainty around the model results had not been fully examined. Not all model parameters were considered in either the deterministic or probabilistic sensitivity analyses. Key parameters missing from the probabilistic sensitivity analysis included the variability in the clinical-effectiveness estimates based on the Kaplan–Meier survival data taken from the trial and variability in the cost of bevacizumab. In the deterministic sensitivity analysis, input parameters that might be expected to be highly influential on the cost-effectiveness results were omitted, such as the cost of bevacizumab, treatment duration and variation in effectiveness.
3.24 The ERG undertook several exploratory deterministic sensitivity analyses that examined the impact of changes to treatment duration, treatment cost and time horizon. Using the trial discontinuation rates in GOG‑0218 and with treatment for a maximum of 15 months instead of the 12 months in the base case, the ICER for bevacizumab increased from the base case of £144,066 per QALY gained to £160,788 per QALY gained. The ERG investigated the effect of changing the 10-year time horizon to the maximum permitted in the model of 25 years; this reduced the ICER to £127,701 per QALY gained. Finally, the ERG combined the analyses for a treatment duration of 15 months and a time horizon of 25 years, which produced an ICER similar to the base case of £142,477 per QALY gained.