Trabectedin for the treatment of advanced soft tissue sarcoma

3.1 The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of trabectedin and a review of this submission by the Evidence Review Group (ERG; appendix B).

3.2 The manufacturer's submission included a phase 2 randomised trial (STS-201) evaluating the efficacy of trabectedin in participants with locally advanced or metastatic soft tissue sarcoma in whom the disease had relapsed or become refractory after treatment with at least 1 anthracycline and ifosfamide, given either in combination or in sequence. All participants had liposarcomas or leiomyosarcomas (L-sarcomas) and an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. The trial randomised participants to 1 of 2 dosing regimens of trabectedin. One group received the dosage of trabectedin specified in the marketing authorisation (1.5 mg/m² every 3 weeks as a 24-hour intravenous infusion, n=136) and the other group (n=134) received trabectedin at a dosage of 0.58 mg/m² every week as a 3-hour intravenous infusion. In addition, the manufacturer's submission presented 3 uncontrolled phase 2 trials of trabectedin. These included a total of 194 participants with soft tissue sarcoma, of whom 104 had L-sarcomas. Participants in all of these studies had an ECOG performance status of 0 or 1. In the absence of relevant comparator data in the included trials, the manufacturer reported historical control data for patients receiving treatments considered to be equivalent to best supportive care (BSC; see sections 3.7 to 3.9). These data were derived from studies in the database of the European Organisation for Research and Treatment of Cancer Soft Tissue and Bone Sarcoma Group (EORTC STBSG).

3.3 The primary outcome of the STS-201 trial was time to progression (time between randomisation and the first documentation of disease progression or death as a result of progressive disease); secondary outcomes included progression-free survival, overall survival and best overall response. According to the manufacturer, treatment with trabectedin continued as long as participants derived therapeutic benefit, until the disease progressed, or for at least 2 courses of therapy beyond a confirmed response. The design of STS-201 permitted crossover for participants in either arm whose disease progressed. The manufacturer acknowledged that the crossover design of the study affected overall survival.

3.4 The median time to progression from intention-to-treat analyses was statistically significantly longer (hazard ratio [HR] 0.734, p=0.032) for the licensed dosage of trabectedin, with a time to progression of 3.7 months (95% confidence interval [CI] 2.1 to 5.4) compared with 2.3 months (95% CI 2.0 to 3.5) for the comparator dosage of trabectedin. Median overall survival was 13.9 months (95% CI 12.5 to 18.6) for the licensed dosage of trabectedin compared with 11.8 months (95% CI 9.9 to 14.9) for the comparator trabectedin regimen. Median progression-free survival at 3 and 6 months was 51.5% (95% CI 43.0 to 60.1) and 35.5% (95% CI 27.1 to 43.9) respectively for the licensed dosage of trabectedin, compared with 44.7% (95% CI 36.0 to 53.3) and 27.5% (95% CI 19.4 to 35.5) for the comparator trabectedin regimen. The manufacturer reported that in a pre-planned subgroup analysis, efficacy outcomes appeared to be more favourable in patients with liposarcomas than in those with leiomyosarcomas, regardless of the study arm.

3.5 The manufacturer reported that the main treatment-related severe (grades 3 and 4) adverse events observed in all studies were transient and reversible, and comprised non-cumulative neutropenia and elevations of hepatic transaminase without clinical consequences. Grade 3 or 4 nausea and vomiting were reported by some participants. The manufacturer stated that unlike with other commonly used cytotoxic agents, no cardiotoxicity or neurotoxicity was observed with trabectedin.

3.6 No health-related quality of life data were presented for patients with advanced soft tissue sarcoma and the manufacturer stated that none were obtained from the trials.

3.7 Historical control data were used to approximate BSC, with the manufacturer acknowledging the limitations of this approach. To estimate overall survival, data for those in whom treatment with ifosfamide had failed, for those receiving dacarbazine, and for those receiving etoposide were taken from an unpublished analysis of 4 phase 2 studies in the EORTC STBSG database of adults with advanced pre-treated soft tissue sarcoma. To estimate progression-free survival, data for the comparators were taken from a publication reporting on phase 2 studies from the EORTC STBSG database. The studies included in the analysis varied in the treatment given to patients during and before entering the trials. Therefore, the manufacturer selected the pre-treated populations that they considered to be most relevant.

3.8 The manufacturer reported that the median overall survival of historical control patients treated with ifosfamide was 6.6 months from start of therapy (95% CI 5.0 to 9.0); a further figure was included by the manufacturer, but was marked as academic-in-confidence and therefore cannot be presented. The manufacturer reported that the median overall survival for those treated with dacarbazine was 6.6 months (95% CI 4.3 to 8.4) and 6.3 months (95% CI 4.4 to 8.9) for those treated with etoposide.

3.9 The manufacturer reported that the mean progression-free survival of historical control patients treated with inactive regimens (n=234) was 21% (standard error [SE] ± 3%) and 8% (SE ± 2%) at 3 and 6 months respectively. Inactive regimens include treatment with mitozolomide, nimustine, fotemustine, miltefosine, liposomal muramyl tripeptide phosphatidylethanolamide, temozolamide, etoposide, Tomudex or gemcitabine. The corresponding figures for historical control patients treated with active regimens comprising ifosfamide and dacarbazine (n=146) were 39% (SE ± 4%) and 14% (SE ± 3%) respectively.

3.10 The manufacturer developed its own economic evaluation, comprising a 2-arm state-transition model. The first arm was designed to capture the costs and outcomes associated with treatment with trabectedin; the second arm was designed to capture the costs and outcomes associated with BSC. Administration of other chemotherapies in addition to BSC was explored in a sensitivity analysis. The model included 4 mutually exclusive health states: progression-free after treatment with trabectedin; progressive disease after treatment with trabectedin; progressive disease with BSC; and death. People treated with trabectedin entered the model in the progression-free state, whereas people treated with BSC entered the model in the progressive disease state. The model cycle length was 1 month with a time horizon of 5 years.

3.11 The model used the effectiveness data from the STS-201 trial of trabectedin, which included only participants with L-sarcomas after they had been treated with a regimen containing at least 1 anthracycline and ifosfamide (combined or sequential). To represent the base case, the manufacturer selected effectiveness data from participants receiving a 24-hour infusion of trabectedin every 3 weeks. In a sensitivity analysis, the manufacturer modelled the pooled effectiveness from the 3 initial phase 2 uncontrolled studies of trabectedin. Transition probabilities for the trabectedin arm were estimated from Weibull parameters derived from the patient-level data for time to progression from the STS‑201 trial. Weibull curves were fitted to Kaplan–Meier curves for time to progression and overall survival. The Weibull estimates were considered by the manufacturer to be comparable to the Kaplan–Meier curves. Following a request by the ERG, arising because of differences in patient characteristics between the trabectedin and BSC arms, Weibull curves for trabectedin were re-calculated using age, gender and severity as covariates.

3.12 The effectiveness data for patients who receive BSC after failure of anthracyclines and ifosfamide were estimated from pooled data from 4 published trials from the EORTC STBSG database. These data were used in the same manner as the STS-201 data to estimate the transition probabilities (in this case, only from progression to death). In response to requests for clarification, the manufacturer submitted a revised model in which the survival curves were adjusted for the differences in patient characteristics between the trabectedin and BSC arms.

3.13 Because no studies of quality of life in patients with soft tissue sarcoma were identified, the manufacturer, following discussion with its clinical experts, used health-state utilities for non-small-cell lung cancer as proxies, assuming comparable prognoses and stages of the 2 diseases. Health-state utilities in progression-free and progressive disease states were assumed to be similar for all patients, irrespective of treatment. The utility values for progression-free and progressive disease health states were assumed to be 0.653 and 0.473 respectively. Admission to hospital as a result of adverse events associated with trabectedin treatment was associated with a utility of 0.610, which was equal to that associated with nausea and vomiting. This was assumed to last 1 month and equated to a quality-adjusted life year (QALY) decrement of 0.004. The utility associated with developing grade 3 or 4 neutropenia was 0.56. This was assumed to last 1 week and equated to a QALY decrement of 0.002. Adverse events were assumed to occur only during the first cycle of trabectedin treatment, and no disutility associated with adverse events was modelled for patients receiving BSC.

3.14 Following concerns raised by the ERG about the calculation of the average cost per patient, the manufacturer revised the methodology used to estimate the acquisition cost of the drug. The manufacturer used patient-level data from the STS-201 trial to calculate the average number of 1-mg and 250-microgram vials used for each patient and the proportion of patients receiving trabectedin in each cycle. The ERG stated that the manufacturer's revised response reported a cost per patient of £23,719 with administration costs excluded, and £25,986 with administration costs and a pre-treatment injection of dexamethasone included. The manufacturer obtained management costs for patients in the progressive disease health state from a cost-of-illness study, and assumed that the costs for the progression-free health state, in the absence of data, were half the costs for the progressive disease health state. Additional costs were included when a patient died. Costs of hospitalisation were average costs and were dependent on patients' diagnoses. The manufacturer did not include costs for treating neutropenia, for treating adverse events in the BSC arm, or for patient monitoring.

3.15 With discounting at 3.5% per annum, the manufacturer's revised base-case cost-effectiveness results gave an incremental cost-effectiveness ratio (ICER) of £56,985 per QALY gained for trabectedin compared with BSC, based on an incremental cost of £27,145 and an incremental QALY gain of 0.476. The manufacturer explored uncertainty in 1-way sensitivity and probabilistic sensitivity analyses. The ICER appeared most sensitive to changes in estimates of utility.

3.16 The manufacturer presented results for additional scenarios:

3.17 The ERG stated that the revised method used to estimate the cost of trabectedin was, in general, appropriate. It noted, however, that the model may have underestimated the cost of trabectedin because a few participants were still being treated at the end of the follow-up period, yet the model assumed no patients incurred costs beyond follow-up. The ERG also identified a number of errors in the revised model submitted by the manufacturer. These errors were corrected by the ERG and were shown to have limited impact on the results. The ERG's corrections to the manufacturer's model resulted in an ICER of £56,949 per QALY gained for the base case (compared with the manufacturer's figure of £56,985, see section 3.15) and £49,992 per QALY gained for the pooled analysis (from the 3 phase 2 uncontrolled studies of trabectedin).

3.18 The ERG expressed strong concerns over the structure of the model in that people treated with trabectedin entered the model in the progression-free health state, whereas those who received BSC entered in the progressive disease health state, which was associated with a lower estimate of utility than the progression-free health state. The manufacturer conducted a revised scenario analysis (based on the revised method to estimate the cost of trabectedin), which assumed that the utility for the progression-free health state (in the BSC arm) was 0.653 for the first cycle and followed a linear decline over the next 4 cycles to reach the utility for progressive disease (0.473). This manufacturer's analysis was associated with an ICER of £61,064 per QALY gained.

3.19 In response to comments received during consultation about the way in which utility was modelled, the ERG presented analyses exploring the impact of different assumptions regarding utility on the ICER. These analyses did not include the model correction (described in section 3.18), which was estimated to increase the ICERs by approximately £5,000 per QALY gained. One analysis assumed the same utility value for the progressive disease and progression-free health states, and varied this value between 0.4 and 0.9. This caused the ICER to vary from more than £80,000 per QALY gained (with the utility value for both states set to 0.4) to approximately £40,000 per QALY gained (with the utility value for both states set to 0.9). Another analysis explored the difference in the utilities of the progressive disease and progression-free health states by setting the utility for progression-free to 0.653 (the manufacturer's base case) and varying the utility of progressive disease between 0.473 and 0.653. This had little impact on the ICER. The reverse analysis, which set the utility for progressive disease to 0.473 (the manufacturer's base case) and varied the utility for the progression-free health state between 0.473 and 0.9, produced an ICER range of approximately £46,000 to approximately £70,000 per QALY gained.

3.20 The ERG also noted the following uncertainties in the cost-effectiveness estimates presented in the manufacturer's submission:

3.21 After the second Appraisal Committee meeting, the manufacturer proposed a patient access scheme for trabectedin for the treatment of advanced soft tissue sarcoma when treatment with anthracyclines and ifosfamide has failed or a person is intolerant of or has contraindications for anthracyclines and ifosfamide. Under this patient access scheme, the acquisition cost of trabectedin to the NHS would be capped at 5 cycles of treatment. The acquisition cost of trabectedin for treatment needed after the fifth cycle (that is, cycle 6 and beyond) would be met by the manufacturer. The Department of Health considered this would not place an excessive administrative burden on the NHS and accepted the consideration of this scheme by NICE.

3.22 The manufacturer submitted an updated cost-effectiveness model incorporating the patient access scheme. The model assumed reimbursement of the acquisition cost of trabectedin from the sixth treatment cycle onwards, and included additional costs to cover the increased operational costs to the NHS of implementing the scheme. The base-case analysis, which assumed equal utility values for the progression-free and progressive disease health states, produced an ICER of £28,712 per QALY gained. This was based on 41% of patients receiving more than 5 cycles of trabectedin, as observed in the STS-201 trial. The manufacturer also presented the scenario analysis with a higher utility value in the progression-free health state (0.653) than in the progressive disease health state (0.473), and incorporating a linear decline (in the BSC arm) of the value in the progression-free health state to the value of the progressive disease health state (see section 3.18). Incorporating the patient access scheme into this scenario reduced the ICER to £34,484 per QALY gained.

3.23 The ERG reviewed the updated analyses from the manufacturer and considered that the model had correctly incorporated the patient access scheme. The ERG reiterated that the scenario analysis that assumed that the utility value in the progression-free health state in the BSC arm followed a linear decline to reach the utility value for progressive disease represented the most appropriate estimation of the ICER. Corrections to minor errors noted in the model resulted in an ICER for this scenario of £34,538 per QALY gained.

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