Topotecan, pegylated liposomal doxorubicin hydrochloride, paclitaxel, trabectedin and gemcitabine for treating recurrent ovarian cancer

Progression‑free survival

3.10 For platinum‑sensitive network 1 evaluating platinum‑based regimens, the assessment group included 5 trials evaluating progression‑free survival in the network meta‑analysis. Results from the network meta‑analysis found that paclitaxel plus carboplatin, gemcitabine plus carboplatin, and PLDH plus carboplatin statistically significantly improved progression‑free survival compared with platinum alone with hazard ratios (HR) of 0.73 (95% confidence interval [CI] 0.64 to 0.84), 0.71 (95% CI 0.57 to 0.90) and 0.59 (95% 0.50 to 0.71) respectively. PLDH plus carboplatin was found to be statistically significantly more effective at prolonging progression‑free survival than paclitaxel plus carboplatin (HR=0.81, 95% CI 0.71 to 0.92). No other statistically significant differences were identified between platinum‑combination regimens.

3.11 For the platinum‑sensitive network 2 evaluating non‑platinum‑based regimens, the assessment group included 3 trials evaluating progression‑free survival in the network meta‑analysis. Results found that trabectedin plus PLDH statistically significantly improved progression‑free survival compared with PLDH alone, paclitaxel alone and topotecan alone, with hazard ratios of 0.73 (95% CI 0.56 to 0.94), 0.44 (95% CI 0.26 to 0.82) and 0.55 (95% CI 0.38 to 0.82) respectively. No statistically significant differences were identified among the monotherapies evaluated (that is, PLDH, topotecan, and paclitaxel).

3.12 For the platinum‑resistant or platinum‑refractory ovarian cancer group, the assessment group included 3 trials evaluating progression‑free survival for inclusion in the network meta‑analysis. The assessment group also highlighted that trabectedin plus PLDH is outside of the scope for this subgroup, and although the data were included in the network to capture all the available evidence, they were not included in the economic analysis. Results from the network meta‑analysis found no statistically significant differences in progression‑free survival between PLDH, paclitaxel and topotecan alone, and these results were in line with results from the individual trials.

3.13 For the fully platinum‑sensitive ovarian cancer subgroup, the assessment group stated that although 3 trials (OVA‑301, ICON4/AGO‑OVAR, and a study by Pfisterer et al.) included subgroups with fully platinum‑sensitive ovarian cancer, only the OVA‑301 trial reported data, so it was not possible to perform an indirect comparison. In addition, 4 trials (OVA‑301, CALYPSO, ICON4/AGO‑OVAR and the study by Pfisterer et al.) included subgroups with partially platinum‑sensitive recurrent ovarian cancer, but only the OVA‑301 and CALYPSO trials reported data, and as they did not contain a common comparator it was not possible to make an indirect comparison. The OVA‑301 trial reported a statistically significant improvement in progression‑free survival with trabectedin plus PLDH compared with PLDH alone (HR=0.65, 95% CI 0.45 to 0.92; p=0.015). The CALYPSO trial reported a statistically significant improvement in progression‑free survival with PLDH plus carboplatin compared with paclitaxel plus carboplatin (HR=0.73, 95% CI 0.58 to 0.90; p=0.004).

Overall survival

3.14 For platinum‑sensitive network 1 evaluating platinum‑based regimens, the assessment group included 6 trials evaluating overall survival in the network meta‑analysis. Results indicated that PLDH plus carboplatin statistically significantly improved overall survival compared with platinum therapy alone (HR=0.79, 95% CI 0.64 to 0.97). Paclitaxel plus carboplatin was also found to statistically significantly improve overall survival compared with platinum alone (HR=0.77, 95% CI 0.66 to 0.91). No other statistically significant differences in overall survival were identified between platinum‑combination regimens.

3.15 For platinum‑sensitive network 2 evaluating non‑platinum‑based regimens, the assessment group included 4 trials evaluating overall survival in the network meta‑analysis. Results indicated that PLDH alone statistically significantly improved overall survival compared with topotecan alone (HR=0.73, 95% CI 0.56 to 0.97). Trabectedin plus PLDH was also found to statistically significantly improve overall survival compared with topotecan alone (HR=0.60, 95% CI 0.43 to 0.86). No other statistically significant differences were identified between platinum‑combination regimens.

3.16 For the platinum‑resistant or platinum‑refractory ovarian cancer group, the assessment group included 4 trials evaluating overall survival in the network meta‑analysis. The assessment group stated that trabectedin plus PLDH is outside of the scope for this subgroup and although the data were included in the network to capture all the available evidence, the data were not included in the economic analysis. Results from the network meta‑analysis found no statistically significant differences in overall survival among the treatments evaluated. This was in line with results from the individual trials.

3.17 For the fully platinum‑sensitive ovarian cancer group, the assessment group identified 4 trials evaluating overall survival. The assessment group stated that it was not possible to perform a network meta‑analysis because only 2 of the trials reported the necessary data for analysis and these trials did not have a common comparator.

3.18 For the partially platinum‑sensitive ovarian cancer group, the assessment group identified the same 4 trials evaluating overall survival. As before, the assessment group constructed 2 networks. Network 1, evaluating platinum‑based regimens, included only 1 trial (CALYPSO). No statistically significant difference in overall survival was identified for PLDH plus carboplatin compared with paclitaxel plus carboplatin in this trial. For network 2, evaluating non‑platinum‑based regimens, results indicated that trabectedin plus PLDH statistically significantly improved overall survival compared with PLDH alone (HR=0.84, 95% CI 0.667 to 1.032). Trabectedin plus PLDH was also found to statistically significantly improve overall survival compared with topotecan alone (HR=0.60, 95% CI 0.43 to 0.86).

Quality of life

3.19 Of the 16 trials identified, 10 reported data on quality of life. The most commonly used scale in the trials was the EORTC QLQ‑C30 questionnaire. However, the assessment group reported that there were considerable differences in the level of reporting of results, the questionnaires used to evaluate quality of life, and the time points for evaluation. Broadly, improvements in quality of life were identified for PLDH plus platinum compared with paclitaxel plus platinum; paclitaxel compared with oxaliplatin; and trabectedin plus PLDH compared with PLDH alone, in a subgroup of patients with partially platinum‑sensitive ovarian cancer.

Adverse reactions

3.20 The assessment group stated that the most frequently reported adverse reactions in the trials reflected those listed in the individual summaries of product characteristics. Consequently, based on advice from clinical experts, the assessment group limited its network meta‑analyses to the following severe grade 3–4 adverse events, which it considered to be the most problematic: allergic reaction, alopecia, anaemia, fatigue, febrile neutropenia, nausea/vomiting and neuropathy. In many cases a network meta‑analysis was not possible due to lack of available data. The majority of results, supplemented by the individual trial results when a network meta‑analysis was not possible, indicated that the likelihoods of adverse events were not statistically significantly different across treatment regimens. However, in some instances, chemotherapies were estimated as having statistically significantly lower risks of 1 or more adverse events but significantly higher risks of other adverse events. For example, when compared with paclitaxel plus platinum, PLDH plus platinum was associated with statistically significantly lower risks of allergic reaction and alopecia but statistically significantly higher risks of anaemia and nausea/vomiting. Overall, no chemotherapy was consistently associated with either a lower risk or a higher risk of the adverse events assessed.

Company's model – trabectedin

3.21 The company for trabectedin submitted cost‑effectiveness evidence as part of its submission. The company developed a decision analytical model comparing trabectedin plus PLDH with PLDH alone in patients with recurrent platinum‑sensitive ovarian cancer for whom platinum‑based chemotherapy was not suitable because of allergy or intolerance or because they have partially platinum‑sensitive disease. The cohort only had 1 previous platinum‑based chemotherapy regimen, and experienced recurrence or progression.

3.22 The structure was identical to the model developed for NICE technology appraisal guidance 91. Disease was classified into 3 distinct periods: stable disease, progressive disease, and death.

3.23 The company stated that because the OVA‑301 trial was not powered for subgroup analysis within the platinum‑sensitive group, data for the entire platinum‑sensitive population were considered appropriate for the cost‑effectiveness analyses. The company fitted 5 parametric survival distributions, adjusting for potential covariates. Based on the Weibull distributions, the mean progression‑free survival for trabectedin plus PLDH and PLDH alone was 11.26 and 8.25 months respectively. Based on the log‑logistic distributions, the mean overall survival for trabectedin plus PLDH and PLDH alone was 44.69 and 34.97 months respectively.

3.24 The company's base‑case deterministic results, incorporating the patient access scheme for trabectedin, indicated incremental costs of £13,397 and incremental quality‑adjusted life years (QALYs) of 0.49 for trabectedin plus PLDH compared with PLDH alone, resulting in an incremental cost‑effectiveness ratio (ICER) of £27,573 per QALY gained. The corresponding probabilistic results indicated an ICER of £27,761 per QALY gained, and the ICER was most sensitive to the estimate of overall survival. Scenario analyses indicated that the results were also sensitive to the adjustment of the platinum‑free interval as an explanatory variable and alternative survival distributions for progression‑free survival and overall survival.

3.25 The company argued that trabectedin was eligible for consideration under the end‑of‑life criteria. It stated that trabectedin plus PLDH was indicated for women with a life expectancy of less than 2 years without treatment: for patients treated with PLDH alone, median overall survival in the platinum‑sensitive and partially platinum‑sensitive populations was 24.1 months and 16.4 months respectively. Accounting for the imbalance in platinum‑free interval and other prognostic factors in the platinum‑sensitive population reduced the median to 19.4 months. For women with platinum‑sensitive and partially platinum‑sensitive recurrent ovarian cancer, trabectedin treatment increased median survival (after correction of prognostic factors including progression‑free interval) by 4 months, and the estimated mean survival suggested this extension of life could be in excess of 9 months. The company also estimated that trabectedin would be indicated for approximately 500 women with relapsed platinum‑sensitive ovarian cancer in 2014.

Assessment group's model

3.26 The assessment group conducted a systematic review and stated that no cost‑effectiveness analyses including the full range of interventions and comparators were available in the literature. It noted that the majority of analyses available were based on the model developed for NICE technology appraisal guidance 91. The assessment group considered that this model structure, also adopted by the company for trabectedin, was the most appropriate for the decision problem, and therefore used it to develop a de novo model. The model had a lifetime time horizon, which was set as 15 years because at this point over 99.9% of patients in the model would have died. In NICE technology appraisal guidance 91, and other models based on it, the time spent in each health state was based on the estimated mean time to progression (time spent in the stable disease health state) and mean time to death (time spent in the progressed disease health state, after subtracting time spent in the stable disease health state). The assessment group incorporated a similar methodology to estimate the proportion of patients in each health state, but full survival curves rather than mean estimates were derived from the clinical data for each therapy. The assessment group stated that this would appropriately capture time in the economic model, and facilitate the assignment of costs, utilities and discounting.

3.27 The NICE scope for this appraisal specified that the interventions of interest for women with platinum‑sensitive ovarian cancer were paclitaxel alone or paclitaxel plus platinum chemotherapy, PLDH alone or PLDH plus platinum chemotherapy, gemcitabine plus carboplatin, trabectedin plus PLDH, and topotecan. The assessment group explained that although all interventions specified in the scope were considered, 2 independent networks were constructed evaluating platinum and non‑platinum‑based regimens, and therefore interventions were not simultaneously compared with each other.

3.28 The populations with platinum‑sensitive disease and platinum‑resistant or ‑refractory disease were modelled separately and there was no explicit analysis of the full population. The assessment group explained that this was because separation of the results by platinum sensitivity is more clinically relevant because the platinum‑free interval was a key prognostic factor, as confirmed by experts, and this approach was also in line with the data available to inform the analysis. The assessment group stated that data for women with fully or partially platinum‑sensitive disease was insufficient, so these groups were considered in sensitivity rather than base‑case analyses. The assessment group considered that response to non‑platinum‑based therapies would be expected to be consistent between patients with or without an allergy or intolerance to platinum‑based therapy. Therefore, the platinum‑allergic subgroup was included in platinum‑sensitive network 2 and platinum‑resistant and platinum‑refractory subgroups.

3.29 The assessment group noted 3 main concerns with the use of data from the network meta‑analyses in the model. First, due to lack of individual patient data, the network meta‑analyses synthesised data from the whole trial population. Individual patient data would have allowed for differences in baseline characteristics within and between trials to be incorporated. In addition, as discussed in section 3.9, unadjusted hazard ratios were incorporated, which could include potential bias. Second, using hazard ratios based on the literature assumes proportional hazards; that is, the relative treatment effects captured by the hazard ratios hold true across all time points. However, log‑cumulative hazard plots indicated that this assumption may not be appropriate. The assessment group highlighted that when the relative hazard decreases over time for both progression‑free survival and overall survival, the model was likely to overestimate the relative benefit of treatment and vice versa. Third, it was noted that several of the included trials allowed for crossover, which could have confounded overall survival data. The assessment group was unable to assess the degree of crossover bias because of a lack of individual patient data and because none of the trials described the crossover treatment.

3.30 The assessment group included grade 3 and 4 adverse events associated with significant costs in the base‑case analysis – allergic reaction, anaemia, febrile neutropenia, nausea and vomiting. The relative likelihood of an adverse event associated with each therapy was estimated from the network meta‑analysis. Adverse events were not analysed by population because of a lack of data; instead, adverse event data from any population (platinum sensitive or platinum resistant/refractory) were included in the analysis, therefore assuming that the likelihood of an adverse reaction is independent of the platinum‑free interval. Inconsistent reporting between trials led to differences in the networks of treatments available to assess the relative effect of treatment on each adverse event. Consequently, estimates of the impact of treatment on the rates of adverse events were not available for all treatments for all adverse events. Although it was possible to estimate the possibility of each adverse event for the baseline treatment in each network, odds ratios and expert opinion were used to estimate probabilities for the remainder.

3.31 The assessment group conducted a systematic review and identified 22 studies measuring health‑related quality of life. It was noted that the utility values based on the OVA‑301 trial were most relevant, because EQ‑5D utility values in the recurrent ovarian cancer population for the health states needed for the economic model were reported, and were based on a large sample of patients (n=600). The company clarified that these utilities were derived from the platinum‑sensitive population (n=400). The mean estimates of utility in the stable and progressive disease health states were estimated to be 0.718 and 0.649 respectively. These estimates were used in NICE technology appraisal guidance 222, and were identical to the EQ‑5D data identified by the assessment group from the systematic review of the literature. Disutilities associated with adverse events were not included in the base‑case analysis because the estimates identified were based on small samples. This was also to avoid double counting, because the effect of adverse events on quality of life associated with trabectedin plus PLDH and PLDH alone were already included in health state EQ‑5D estimates from NICE technology appraisal guidance 222. This was explored in sensitivity analyses.

3.32 The assessment group model included costs associated with the technologies, administration costs, health state‑related costs and adverse event costs. Chemotherapy costs per cycle were estimated using drug costs in the 'British national formulary' (BNF). The regimens used were as per the summaries of product characteristics (see section 2), except for PLDH, which was assumed to be started at a dose of 40 mg/m² of body surface area in the base case (the licensed dose was considered in a scenario analysis) and paclitaxel, which was assumed to be used in a weekly regimen (80 mg/m²/week) in the platinum‑resistant or refractory subgroup only. These modifications were based on the advice of clinical experts and were intended to reflect the regimens used in clinical practice in the NHS.

3.33 The model assumed that treatments would be administered as infusions in a hospital, and associated administration costs were included in the model. For the base‑case analyses, it was assumed that no vial sharing would happen.

Results of network 1 – platinum‑based regimens in platinum‑sensitive disease

3.34 Both deterministic and probabilistic results indicated that PLDH plus platinum was strictly dominated by (that is, it was more costly and less effective than) paclitaxel plus platinum. Similarly, gemcitabine plus carboplatin was extendedly dominated by paclitaxel plus platinum (that is, its ICER was higher than that of the next, more effective, option when compared with platinum). Therefore, PLDH plus platinum and gemcitabine plus carboplatin were excluded, leaving paclitaxel plus platinum compared with platinum alone as the only relevant comparison for this network. For this comparison, the deterministic ICER was estimated as £24,361 per QALY gained; paclitaxel plus platinum was associated with an estimated incremental cost of £5,694 and an additional 0.23 QALYs when compared with platinum alone. The probabilistic ICER for paclitaxel plus platinum compared with platinum alone was £24,539 per QALY gained. The assessment group also estimated an ICER of £114,410 per QALY gained for gemcitabine plus carboplatin compared with platinum alone and an ICER of £30,188 per QALY gained for PLDH plus platinum compared with platinum alone.

3.35 One‑way sensitivity analyses on various model parameters indicated that the comparisons of paclitaxel plus platinum, and PLDH plus platinum, when compared with platinum alone, were most sensitive to the relative effect of treatment on overall survival. For example:

3.36 Probabilistic sensitivity analyses indicated that at a maximum acceptable ICER of £20,000 per QALY gained, the probabilities of paclitaxel plus platinum or PLDH plus platinum being considered cost effective compared with platinum alone were 13% and 3% respectively. Furthermore, PLDH plus platinum was estimated to be almost as likely to result in greater costs and QALYs as to be dominated by paclitaxel plus platinum. The assessment group highlighted that the costs and QALYs accumulated by the addition of paclitaxel or PLDH to platinum therapy were similar, producing cost‑effectiveness estimates that were sensitive to minor changes in parameter estimates.

Results of network 2 – non‑platinum‑based regimens in platinum‑sensitive disease

3.37 Base‑case results (deterministic and probabilistic) indicated that topotecan was dominated by PLDH. Topotecan was therefore removed from the analysis and the relevant fully incremental comparisons of PLDH compared with paclitaxel and trabectedin plus PLDH compared with PLDH alone were presented. When compared with paclitaxel, PLDH was associated with an incremental cost of approximately £3,900 and approximately 0.16 additional QALYs. This resulted in ICERs of £23,733 and £25,931 per QALY gained in the deterministic and probabilistic analyses, respectively. When compared with PLDH alone, trabectedin plus PLDH was associated with an incremental cost of approximately £13,000 and 0.16 additional QALYs. The resulting ICERs for trabectedin plus PLDH compared with PLDH alone were £85,212 and £81,353 per QALY gained in the deterministic and probabilistic analyses respectively.

3.38 The assessment group did a series of one‑way sensitivity analyses on various model parameters. In network 2, the cost‑effectiveness estimates for all 3 comparisons (PLDH compared with paclitaxel, trabectedin plus PLDH compared with paclitaxel, and trabectedin plus PLDH compared with PLDH alone) were most sensitive to the relative effect of treatment on overall survival:

3.39 The assessment group did a series of scenario analyses and noted that the base‑case results were robust in the majority of the scenarios modelled, with the exception of the scenario in which PLDH monotherapy was assumed to be used at its full licensed dose (50 mg/m² of body surface area). In this scenario topotecan remained dominated, the deterministic ICER for PLDH compared with paclitaxel increased from £23,733 to £31,222 per QALY gained and the deterministic ICER for trabectedin plus PLDH compared with PLDH alone fell from £85,212 to £77,290 per QALY gained. The assessment group stated that apart from this scenario, the ICER for PLDH compared with paclitaxel remained below £30,000 per QALY gained and highlighted that topotecan was dominated by trabectedin plus PLDH in every scenario. The assessment group also carried out an exploratory scenario analysis, using covariate‑adjusted clinical‑effectiveness data from the company for trabectedin's submission, for a head‑to‑head comparison of trabectedin plus PLDH compared with PLDH alone. This resulted in an ICER of £35,646 per QALY gained, compared with ICERs of £85,212 and £27,573 estimated by the assessment group's and the company's base‑case analyses respectively. The assessment group stated that this difference was predominantly a consequence of using adjusted clinical‑effectiveness data.

Results of network 3 – platinum‑resistant and platinum‑refractory group

3.40 The assessment group explained that data for paclitaxel plus platinum were not available from the literature for women with platinum‑resistant or ‑refractory ovarian cancer, so this intervention was not included in the base‑case analysis. Base‑case results showed that paclitaxel was dominated by PLDH alone. In the probabilistic analysis, paclitaxel was associated with an incremental cost of £901 and 0.022 fewer QALYs. Therefore, topotecan compared with PLDH was the only comparison considered in the final cost‑effectiveness analysis. Topotecan was associated with an incremental cost of approximately £7,000 and 0.02 additional QALYs relative to PLDH. The resulting ICERs for topotecan compared with PLDH were £449,553 and £324,188 per QALY gained in the deterministic and probabilistic analyses respectively.

3.41 As with platinum‑sensitive networks 1 and 2, the cost‑effectiveness results were most sensitive to the relative effect of treatment on overall survival:

3.42 The assessment group did a series of scenario analyses and noted that the base‑case results were robust in the majority of the scenarios modelled. It highlighted that the ICER for topotecan compared with PLDH ranged from £374,963 to £503,885 per QALY gained across the scenarios. Paclitaxel was dominated in all scenarios except when the cost associated with a 50 mg/m² dose of PLDH was used and paclitaxel became the least costly treatment, resulting in an ICER of £10,480 per QALY gained for PLDH compared with paclitaxel. In an additional analysis, done after the appraisal had been referred back to the committee after appeal, the assessment group modelled paclitaxel used at its licensed dose of 175 mg/m² of body surface area every 3 weeks (instead of the weekly regimen used in the base case for the network 3 analysis). In this scenario, the costs associated with paclitaxel were lower and it was no longer dominated. The ICERs for PLDH, relative to paclitaxel, were £69,935 per QALY gained at a dose of 40 mg/m² of body surface area and £103,810 per QALY gained at a dose of 50 mg/m² of body surface area.

Assessment group's comments on the additional evidence submitted by the company for trabectedin

3.46 The assessment group considered that the imbalance between the treatment groups in the platinum‑free interval observed in OVA‑301 did not warrant analysing the data by categorical variable. It also considered that the company's new analysis was another post‑hoc analysis and therefore subject to the same concerns as the analysis of the platinum‑free interval as a continuous variable. The assessment group commented that results from a second, independent study would be needed to demonstrate an association between duration of the platinum‑free interval and response to non‑platinum based treatments. It maintained that unadjusted hazard ratios should be used for survival outcomes because a consistent dataset of adjusted hazard ratios was not available to inform the network meta‑analysis.

3.47 The assessment group incorporated the updated patient access scheme (see section 2.14) into its model. The ICER for trabectedin plus PLDH compared with PLDH alone remained above £70,000 per QALY gained. When the assessment group also incorporated the company's hazard ratios that were adjusted for the platinum‑free interval as a categorical variable and covariates (see section 3.44), the ICER fell to £59,772 per QALY gained. Changing the dose of PLDH monotherapy from 40 mg/m² to 50 mg/m² further reduced the ICER to £54,059 per QALY gained.