3 The manufacturer's submission

The Appraisal Committee (section 8) considered evidence submitted by the manufacturer of bevacizumab and a review of this submission by the Evidence Review Group (ERG; section 9).

3.1 The key evidence for the clinical effectiveness of bevacizumab plus gemcitabine and carboplatin came from 1 randomised controlled trial (OCEANS). This double-blind, randomised, placebo-controlled trial assessed the safety and efficacy of bevacizumab plus gemcitabine and carboplatin in 484 adults with platinum-sensitive recurrent epithelial ovarian cancer, primary peritoneal cancer or fallopian tube cancer, with a first recurrence of ovarian cancer and who had not previously received vascular endothelial growth factor (VEGF) receptor-targeted agents. The trial was a multicentre study conducted in 96 centres in the USA. Patients were randomised to 1 of the following 2 treatment arms:

  • Bevacizumab plus gemcitabine and carboplatin (n=242) (bevacizumab 15 mg/kg body weight on day 1 every 3 weeks, carboplatin at a dose corresponding to an area under the curve of concentration versus time of 4 mg/ml•min on day 1 every 3 weeks, and gemcitabine 1000 mg/m2 on days 1 and 8 every 3 weeks for 6–10 cycles; followed by bevacizumab 15 mg/kg body weight alone on day 1 every 3 weeks until disease progression or unacceptable toxicity).

  • Placebo plus gemcitabine and carboplatin (n=242) (placebo on day 1 every 3 weeks, carboplatin area under the curve 4 mg/ml•min on day 1 every 3 weeks, and gemcitabine 1000 mg/m2 on days 1 and 8 every 3 weeks for 6–10 cycles; followed by placebo alone on day 1 every 3 weeks until disease progression or unacceptable toxicity).

    Randomisation was stratified by platinum-sensitive category (platinum sensitive or partially platinum sensitive) and incidence of cytoreductive surgery for recurrent disease.

3.2 The primary outcome was progression-free survival (PFS), defined as the period from randomisation to disease progression or death (from any cause). Progression was assessed by investigators using radiological evaluation according to the Response Evaluation Criteria for Solid Tumours (RECIST) criteria. Progression could also be determined by symptomatic progression, but not by cancer antigen-125 (CA‑125) elevation alone. Sensitivity analysis of PFS included an assessment by an Independent Review Committee (IRC) using RECIST criteria. For the IRC analysis, the definition of PFS was the period from randomisation until disease progression or on-study death (that is, death occurring within 9 weeks of the last dose of chemotherapy or study drug). All patients needed to undergo CT scans every 9 weeks from day 1 of cycle 1. Secondary outcomes were overall survival, objective response rate and duration of objective response. Objective response rate and duration of objective response were also assessed by the IRC using RECIST criteria as exploratory analyses. Safety outcome measures were frequency and severity of adverse events.

3.3 Analysis of the primary outcome, PFS in the intention-to-treat population, was based on a cut-off date of 17 September 2010, once 338 (70%) patients had experienced disease progression or died (62.4% of patients in the bevacizumab arm and 77.3% in the placebo arm). The median follow-up was 24 months. Data for patients whose disease had not progressed or who had not died at the time of the last tumour assessment were censored (that is, excluded from the analysis from that point onwards). Data for patients who received non-protocol therapy before disease progression were also censored at the time of the last tumour assessment before therapy was initiated. At 29.8 months, all patients still at risk in the bevacizumab arm had experienced disease progression or had died, and at 24.9 months, 2 patients remained at risk in the placebo arm. Results of the investigator-assessed analysis showed that there was a statistically significant difference of 4 months between the median PFS in the bevacizumab arm compared with the placebo arm (bevacizumab 12.4 months, placebo 8.4 months). In the stratified analysis, there was a 51.6% reduction in disease progression in patients in the bevacizumab arm compared with those in the placebo arm (hazard ratio [HR] 0.48, 95% confidence interval [CI] 0.39 to 0.61, p<0.0001). An unstratified analysis showed a reduction in disease progression of 50.8% with bevacizumab compared with placebo (HR 0.49, 95% CI 0.40 to 0.61, p<0.0001). An IRC analysis of PFS on the same data and a sensitivity analysis without censoring patients for receiving non-protocol therapies were also conducted. The IRC analysis results of PFS were consistent with the primary analysis showing a reduction in disease progression in patients in the bevacizumab arm compared with the placebo arm (bevacizumab 12.3 months, placebo 8.6 months; HR 0.45, 95% CI 0.35 to 0.58, p<0.0001). Results from the sensitivity analysis that did not censor for non-protocol specified therapy were also consistent with the primary analysis results (bevacizumab 12.4 months, placebo 8.4 months; HR 0.52, 95% CI 0.43 to 0.65).

3.4 PFS results for subgroups based on the predefined stratification factors (platinum-sensitive classification and incidence of cytoreductive surgery for recurrent disease) showed that there was a statistically significant reduction in PFS observed for patients in the bevacizumab arm, irrespective of whether they had undergone cytoreductive surgery for recurrent disease or not. Patients whose disease was partially platinum sensitive showed a median PFS of 11.9 months and 8.0 months with bevacizumab and placebo respectively (HR 0.41, 95% CI 0.29 to 0.58). There was also an increase in PFS in patients whose disease was fully platinum sensitive seen in the bevacizumab arm (HR 0.55, 95% CI 0.41 to 0.73).

3.5 Three interim analyses of overall survival were conducted, 2 of which were protocol specified. None of the interim analyses found a statistically significant difference between bevacizumab and placebo in the duration of overall survival. The first interim analysis was carried out at the time of final PFS analysis (17 September 2010), when approximately 29% of patients had died (median overall survival: 35.5 and 29.9 months in the bevacizumab and placebo arms respectively; HR 0.75, 95% CI 0.54 to 1.05). The second analysis was carried out on 29 August 2011, when approximately 49% of the patients had died (median overall survival: 33.3 and 35.2 months in the bevacizumab and placebo arms respectively; HR 1.03, 95% CI 0.79 to 1.33). The third analysis, using a data cut-off date of 30 March 2012 (required by the European Medicine Agency), was conducted when approximately 59% of the patients had died (median overall survival: 33.4 and 33.7 months in the bevacizumab and placebo arms respectively; HR 0.96, 95% CI 0.76 to 1.21). The manufacturer stated that patients in both study arms in third and subsequent lines of therapy received post-progression bevacizumab (at least 18.1% of patients in the bevacizumab arm and 34.7% in the placebo arm received bevacizumab), and therefore confounding may have occurred.

3.6 Objective response rate, according to investigator assessment, was statistically significantly different between the 2 arms (78.5% in the bevacizumab arm compared with 57.4% in the placebo arm, p<0.0001). Median duration of response was 10.4 and 7.4 months in patients in the bevacizumab and placebo arms respectively. IRC assessment of objective response rate was consistent with the results of the investigator-assessed analysis (bevacizumab 74.8%, placebo 53.7%, p<0.0001).

3.7 All patients in the OCEANS trial experienced an adverse event. More patients in the bevacizumab arm experienced a serious adverse event compared with patients in the placebo arm (34.8% and 24.9% respectively). Adverse events for which the incidence was more than 10% higher in the bevacizumab arm than in the placebo arm were hypertension, nose bleeds, headache and proteinuria. Adverse events of special interest (grades 3–5) that occurred with an incidence of at least 2% higher in the bevacizumab arm compared with the placebo arm were hypertension, proteinuria and non-central nervous system bleeding. The proportion of patients who experienced an adverse event that led to discontinuation was larger in the bevacizumab arm (19.8%) compared with the placebo arm (4.7%). However, the absolute number of patients stopping treatment because of adverse events was unclear.

3.8 The manufacturer carried out a literature review and identified 4 randomised controlled trials (CALYPSO, ICON4, AGO‑OVAR‑2.5 and OCEANS) that had assessed the comparative clinical effectiveness of the following comparators:

  • paclitaxel plus platinum-based treatment compared with pegylated liposomal doxorubicin hydrochloride plus platinum-based treatment

  • platinum-based treatment (monotherapy) compared with paclitaxel plus platinum-based treatment

  • gemcitabine plus platinum-based treatment compared with platinum-based treatment (monotherapy)

  • bevacizumab plus gemcitabine and carboplatin compared with gemcitabine plus carboplatin treatment.

    After assessing the feasibility of conducting an indirect comparison of bevacizumab plus gemcitabine and carboplatin with the comparators listed in the final scope, the manufacturer decided against carrying out a network meta-analysis.

3.9 The manufacturer submitted a de novo economic analysis that assessed the cost effectiveness of bevacizumab plus carboplatin and gemcitabine compared with placebo plus carboplatin and gemcitabine for treating people with advanced, recurrent, platinum-sensitive ovarian cancer. The model was a 3‑state semi-Markov model with health states consisting of PFS, progressed disease and death. Data from the OCEANS trial were used to guide model inputs. Because the drug dose is dependent on characteristics (such as body weight, body surface area and creatinine clearance rates) that are influenced by age, demographic data from a UK study were used by the manufacturer in their base case to calculate the dose of bevacizumab, carboplatin and gemcitabine. The cost-effectiveness analysis was conducted from an NHS and personal social services perspective, costs and outcomes were discounted at 3.5% per annum and a 10‑year time horizon was used. The cycle length was 1 week.

3.10 PFS in the model used the Kaplan–Meier survival curves from the OCEANS trial based on the (intention-to-treat population) investigator-assessed analysis (data cut-off date September 2010). The manufacturer examined the fit of various parametric functions to the PFS data and considered a log-logistic model as the best fit to estimate and extrapolate the proportion of patients in the PFS health state. The overall survival from the OCEANS trial (data cut-off date September 2010) was used in the model to estimate the proportion of people in the progressed-disease health state and, implicitly, the death state. The manufacturer also applied a log-logistic distribution to the Kaplan–Meier curves. The incidence of adverse events adopted in the model was derived from adverse events (cut-off September 2010) of at least grade 3 that occurred in more than 2% regardless of the study arm. The manufacturer used the number of patient events to assign a cost associated with each adverse event.

3.11 Health-related quality of life and utilities applied in the model were obtained from Trabectedin for the treatment of relapsed ovarian cancer (NICE technology appraisal guidance 222). The data used in this guidance were taken from the OVA‑301 trial using EQ‑5D. The utility values used in the model for PFS and progressed-disease health states were 0.718 and 0.649 respectively. The manufacturer assumed in the model that health-related quality of life remained constant during PFS and reduced once disease progressed but remained constant after that. The manufacturer did not apply disutilities caused by adverse events in the model.

3.12 Drug costs were estimated using the dose and frequency of administration in the summary of product characteristics. Data from a UK cohort study (Sacco et al. 2010) were used in the dose calculations. The base case assumed that any unused carboplatin and paclitaxel from a vial was reallocated and not wasted, whereas for bevacizumab, it was assumed that any unused drug in a vial was wasted. For bevacizumab and carboplatin, the manufacturer used public list prices from the 'British national formulary', and the price of gemcitabine (£12.57 for a 1000 mg vial) was obtained from the Commercial Medicines Unit (CMU) 2012 electronic Market Information Tool (eMit). Costs of drug administration were taken from the Unit Costs of Health and Social Care and NHS reference cost data, and included in the model. The weekly costs of supporting patients in the PFS and progressed health states were also included. Costs of palliative care were applied to patients as they moved to the death state. Costs of post-progression therapies were taken from the OCEANS trial (cut-off date September 2010) and included other chemotherapy drugs, radiotherapy or surgery. These costs were added together and applied as a one-off cost in the model, and so were not subject to discounting. Costs associated with adverse events that occurred at grade 3 or 4 severity in more than 2% of patients from the OCEANS trial (cut-off date September 2010) were incorporated into the analysis. NHS reference costs were utilised when possible; all adverse events were assumed to occur in cycle 1 of the model, so costs were not discounted.

3.13 The base-case results estimated that adding bevacizumab to carboplatin and gemcitabine provides an additional 0.42 life years and 0.298 quality-adjusted life years (QALYs). These benefits are achieved with an incremental cost of £44,428, resulting in an incremental cost-effectiveness ratio (ICER) of £149,050 per QALY gained for bevacizumab plus carboplatin and gemcitabine compared with carboplatin and gemcitabine alone. The manufacturer's deterministic sensitivity analysis suggested that the cost-effectiveness results were most sensitive to assumptions around the extrapolation of overall survival, the duration of treatment and the utility of patients in PFS. The manufacturer's probabilistic sensitivity analyses concluded that the probability of bevacizumab plus carboplatin and gemcitabine being cost effective compared with carboplatin and gemcitabine alone at a threshold of £30,000 per QALY gained was 0%. The manufacturer identified the key drivers of the cost-effectiveness results to be the cost and duration of treatment with bevacizumab and the time horizon of the analysis.

3.14 The ERG considered the OCEANS trial to be well designed and agreed that, except for baseline weight, the characteristics of the patient population enrolled in the trial were representative of people with first recurrence of ovarian cancer in England and Wales. The ERG noted that the course of treatment assumed in the OCEANS trial (allowing up to a maximum of 10 cycles of bevacizumab plus carboplatin and gemcitabine) may not fully represent clinical practice in the UK (where a maximum of 6 cycles of chemotherapy would be administered). The ERG also noted that the main comparator in the manufacturer's submission was gemcitabine plus carboplatin, whereas this may not be the treatment routinely used in the NHS.

3.15 The ERG highlighted the differences between the number of recorded events in terms of PFS in the investigator-assessed analysis and the IRC-determined analysis. The ERG also highlighted that the number of patients censored in each group at the time of final PFS analysis (September 2010), and the mean PFS and the number of patients lost to follow-up at the time of the final analysis were unknown. The ERG also noted that the absolute number of patients stopping treatment because of an adverse event varied in the manufacturer's submission, and the correct number remained unclear after seeking clarification from the manufacturer.

3.16 The ERG considered the literature search and the reasons given by the manufacturer for not performing an indirect comparison between bevacizumab plus carboplatin and gemcitabine, and the other comparators listed in the scope. The ERG considered that the differences between trials were sufficiently minor such that their inclusion would have a minimal impact on clinical heterogeneity, and decided to perform a network meta-analysis for the primary outcome measure (PFS). Results from the network meta-analysis performed by the ERG suggested that bevacizumab plus carboplatin and gemcitabine is associated with a statistically significant improvement in duration of PFS compared with all comparators listed in the scope (bevacizumab plus carboplatin and gemcitabine compared with: paclitaxel plus carboplatin, HR 0.47, 95% credible interval [CrI] 0.33 to 0.66; pegylated liposomal doxorubicin hydrochloride plus carboplatin, HR 0.58, 95% CrI 0.39 to 0.82; platinum monotherapy, HR 0.35, 95% CrI 0.25 to 0.47; gemcitabine plus carboplatin, HR 0.48, 95% CrI 0.38 to 0.60). Results from the network meta-analysis also suggested that there were no statistically significant differences between most of the other comparators.

3.17 The ERG considered the manufacturer's model structure was appropriate to describe the decision problem and was well constructed and transparent. The ERG highlighted and agreed with the manufacturer that the main criticism of the submitted economic evaluation was the use of the September 2010 OCEANS clinical-effectiveness, cost and adverse-event incidence data. The ERG suggested that the use of data from September 2010, when 29% of the patients had died (rather than data from March 2012, when available, when 59% of the patients had died), may have introduced unnecessary uncertainty into the estimate of the ICER and may have overestimated the overall survival benefit associated with bevacizumab because the analysis of overall survival in September 2010 showed a non-statistically significant overall survival increase for patients in the bevacizumab group, which was not sustained in the 2 later interim analyses. The ERG noted that overall survival was a key driver in the model and estimated that approximately 90% of the QALYs gained in the model were a function of the overall survival. The ERG conducted a scenario analysis assuming that overall survival was the same for patients in both treatment groups. The result of the analysis was an increase in the ICER to over £1.7 million per QALY gained.

3.18 The ERG noted that the manufacturer applied a parametric log-logistic function to the Kaplan–Meier PFS data (cut-off date September 2010) from the OCEANS trial to estimate and extrapolate the proportion of patients in the progression-free health state. At a median follow-up of 24 months (final PFS analysis), 70% of the patients had either experienced disease progression or died. Patients in the bevacizumab arm reached 0% PFS at month 29.8, whereas 2 patients remained at risk at month 24.9 in the placebo arm. The ERG assumed in their exploratory analysis that, by 29 months, all patients would have had disease progression or died according to the last Kaplan–Meier data available, and suggested that mean values for PFS might be available, rather than only medians. The ERG also had concerns about fitting a parametric distribution for PFS given the Kaplan–Meier data available and undertook a scenario analysis using only the Kaplan–Meier data, although this did not have a significant impact on the ICER.

3.19 The ERG noted that adverse events experienced by patients in the model were not subject to estimates of disutility and suggested that this was likely to favour the cost effectiveness of bevacizumab because a larger proportion of patients in the bevacizumab treatment group experienced a serious adverse event compared with the placebo group in the OCEANS trial. The ERG conducted a scenario analysis and assessed a range of average duration of adverse event disutilities. It concluded that, for example, for an average event duration of 1 week, the ICER increased to £149,391 per QALY gained and, for an average adverse event duration of 1 month, the ICER increased to £150,544 per additional QALY gained.

3.20 The ERG explored the impact of the network meta-analysis results in terms of cost effectiveness. The ERG assumed, based on the these results, that overall survival and PFS estimates for patients in every comparator group were the same as for patients in the placebo group in the manufacturer's model. Cost-effectiveness results from the ERG exploratory analysis were:

  • ICER for bevacizumab plus gemcitabine and carboplatin compared with carboplatin £159,273 per QALY gained

  • ICER for bevacizumab plus gemcitabine and carboplatin compared with paclitaxel plus carboplatin £148,014 per QALY gained

  • ICER for bevacizumab plus gemcitabine and carboplatin compared with pegylated liposomal doxorubicin hydrochloride plus carboplatin £145,621 per QALY gained.

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

  • National Institute for Health and Care Excellence (NICE)