The Appraisal Committee considered evidence submitted by Novartis and a review of this submission by the Evidence Review Group. See the Committee papers for full details of the evidence.
3.1 The company included 1 randomised controlled trial, PANORAMA‑1, which compared panobinostat, bortezomib and dexamethasone with bortezomib and dexamethasone in patients with relapsed or relapsed and refractory multiple myeloma, and who have had 1–3 previous treatments. The trial spanned 34 countries and 215 centres (30 of which were in the UK). Patients (n=768) were randomly assigned 1:1 to either panobinostat (n=387) or placebo (n=381) (both in combination with bortezomib and dexamethasone) and were stratified by number of previous treatments and previous bortezomib treatment. Approximately one third (35% in the intervention group and 37% in the comparator group) of patients in the trial had relapsed and refractory multiple myeloma and approximately half had received more than 2 lines of treatment (48.8% for the intervention group and 48% for the comparator group). A subgroup of patients who had at least 2 previous lines of treatment, including 1 immunomodulatory drug (for example thalidomide) and bortezomib (n=147, 19% of the trial population), was a post hoc subgroup whereas other subgroups were pre‑specified in the trial. This is the subgroup which received the marketing authorisation and will be the only group described in this section.
3.2 Treatment allocation in the trial was blinded and no crossover occurred. The trial was divided into phase 1 (24 weeks; 8 cycles of 21 days' each) and phase 2 (24 weeks; 4 cycles of 42 days' each). During phase 1, in week 1 and 2 of each cycle patients had either panobinostat (20 mg) or placebo 3 times a week, bortezomib (1.3 mg/m2) twice a week and dexamethasone (20 mg) 4 times a week. There was no treatment in the third week of the cycle. Patients moved onto phase 2 if they experienced clinical benefit, defined as at least no disease progression on day 1 of cycle 8 (as assessed by the modified European Group for Blood and Marrow Transplantation criteria).
3.3 The primary outcome was progression‑free survival with response assessed at 3‑week intervals during the treatment phases and at 6‑week intervals thereafter. Progression‑free was defined as the time from randomisation until documented disease progression, relapse from complete response or death, whichever came first. The final analysis was done at median follow‑up of 31 months. Progression‑free survival observations were censored at the date of the last response assessment for people who had either not progressed or had a different treatment. In the PANORAMA‑1 trial, patients in the subgroup having panobinostat plus bortezomib and dexamethasone (Pano‑Bort‑Dex) had a median progression‑free survival extension of 7.8 months compared with placebo, representing a 53% reduction in the risk of progression.
3.4 The key secondary outcome was overall survival, which was defined as the time from randomisation to death from any cause. Other secondary outcomes included overall response rate (complete response, near complete response and partial response), time to progression, time to response and duration of response, safety and health‑related quality of life.
3.5 The company performed an indirect comparison for the subgroup of people who had at least 2 previous lines of treatment, including 1 immunomodulatory drug plus bortezomib, to compare Pano‑Bort‑Dex with bortezomib and lenalidomide. The indirect comparison included: PANORAMA‑1, MM‑009 and MM‑010 for lenalidomide plus dexamethasone (Len‑Dex); DOXIL‑MMY‑3001for bortezomib plus dexamethasone (Bort‑Dex); and APEX for bortezomib. The company considered Len‑Dex to be the only relevant comparator for the subgroup.
3.6 Three different methods were used for the indirect treatment comparison for the subgroup who had at least 2 previous therapies: naive comparison, unadjusted Cox regression and matching adjusted indirect treatment comparison.
3.7 For the matching adjusted indirect treatment comparison, patient‑level data from the PANORAMA‑1 trial were used for the panobinostat group whereas data from the pooled analysis of the MM‑009 and MM‑010 studies and a subgroup from Stadtmauer et al. (2009) were used for the Len‑Dex group. Individual patient‑level data from the PANORAMA‑1 trial were reweighted such that the median baseline characteristics matched those reported from the MM‑009 and MM‑010 trials. These variables included age, sex, time since diagnosis, ECOG score, number and type of previous treatments (immunomodulatory drugs and bortezomib) and serum beta‑2 microglobulin level. The hazard ratios for progression‑free survival and overall survival were 1.108 and 1.413 respectively.
3.8 Adverse events were reported for the PANORAMA‑1 trial. The numbers of patients in the Pano‑Bort‑Dex group who needed at least 1 dose change were 194 (51%) for panobinostat, 231 (61%) for bortezomib and 93 (24%) for dexamethasone. In the placebo plus bortezomib and dexamethasone group, the equivalent numbers were 86 (23%) for placebo, 158 (42%) for bortezomib and 65 (17%) for dexamethasone. The most frequent (≥2%) adverse events leading to treatment discontinuation were diarrhoea, fatigue, asthenia and peripheral neuropathy in the Pano‑Bort‑Dex group, and fatigue and pneumonia in the placebo plus bortezomib and dexamethasone group. The incidence of adverse events was much lower during phase 2, when bortezomib and dexamethasone were administered less frequently.
3.9 The company developed 2 models – 1 for the full population in PANORAMA‑1 and 1 for the subgroup who had at least 2 previous treatments including an immunomodulatory drug and bortezomib. This section relates only to the subgroup.
3.10 The company developed a decision analytic semi‑Markov model consisting of 3 health states: pre‑progression, post‑progression and death. The time horizon of the model was 25 years and the cycle length was 3 weeks with a half‑cycle correction applied. Discounting of 3.5% was incorporated for both effects and costs and the analysis was done from an NHS and personal social services perspective.
3.11 Transition probabilities for Pano‑Bort‑Dex were derived from post hoc patient‑level data from PANORAMA‑1, and included progression‑free survival, treatment exposure and overall survival.
3.12 The probabilities for risk of progression or pre‑progression death (based on progression‑free survival data), risk of treatment discontinuation (based on exposure to treatment data) and risk of death (based on overall survival data) were generated by fitting parametric curves to the Kaplan–Meier data. The time between randomisation and progression, death or censoring was considered to be the length of treatment exposure.
3.13 To determine the proportion of patients who were on or off treatment, patient‑level discontinuation data from the PANORAMA‑1 trial were used to estimate the risk of treatment discontinuation in a 3‑week cycle. In this analysis, the length of treatment exposure for a patient was considered the time to treatment discontinuation.
3.14 For the overall survival analysis, time between randomisation and death or censoring was considered as treatment exposure. Patients were censored at the last contact date if they were lost to follow‑up for survival status measurements.
3.15 Patients in the PANORAMA‑1 trial completed an EORTC QLQ‑C30 questionnaire, which was mapped to obtain the corresponding EQ‑5D utility value. Cycle‑specific as well as overall average and median utility values were estimated for the treatment arms.
3.16 No utility data were available for Len‑Dex so 2 scenarios were explored. In the first, the utility value for Len‑Dex was assumed to be the same as that for Bort‑Dex. In the second scenario, it was assumed to be the same as the utility value associated with the progression‑free no treatment health state. The first scenario was considered for the base‑case analysis.
3.17 The cost of lenalidomide applied in the model was calculated as a weighted average of daily doses across all patient days in the MM‑010 study. The resulting weighted average 28‑day cycle cost for lenalidomide was £3773, which translated into a 3‑weekly (21‑day) cycle cost of £2830 (taking into account the patient access scheme for lenalidomide). The cost for dexamethasone was £2.59 per 28‑day cycle (£1.94 per 3‑weekly cycle). The panobinostat costs included in the model are confidential because a patient access scheme has been agreed between the company and the Department of Health. The patient access scheme for bortezomib was not included in the company's analyses, because it only applies to bortezomib monotherapy in people whose multiple myeloma has relapsed for the first time after having one treatment (see NICE technology appraisal guidance on bortezomib for relapsed multiple myeloma).
3.18 The company considered that the unadjusted Cox method was most appropriate to derive the relative efficacy of Pano‑Bort‑Dex compared with that of Len‑Dex.
3.19 The company provided a number of different scenarios, which were: changes to the discount rate, how overall and progression‑free survival were calculated, time to discontinuation, distribution of post‑progression treatments, utility values associated with Len‑Dex, how hazard ratios were generated, and threshold analyses.
3.20 The ERG considered that the population in the PANORAMA‑1 trial generally reflected relapsed and refractory multiple myeloma patients in the UK, although it noted that with a median age of 63 years, the trial population was younger than most UK patients. It also considered that people in the trial had bortezomib up to cycle 16, but in UK clinical practice patients do not have bortezomib beyond cycle 8, with a stopping rule at 4 cycles if no response is seen. The ERG noted that patients in the trial were administered bortezomib intravenously but that in UK clinical practice it is becoming more common to administer bortezomib subcutaneously.
3.21 The ERG considered the company's use of parametric curves fitted to the Kaplan–Meier data to be appropriate to extrapolate beyond the trial time horizon, and noted that the use of logistic regression was particularly appropriate because of the binary nature of the responses (progressed or not progressed). However, the ERG noted that the Len‑Dex overall survival curve had not been compared with the underlying trial data.
3.22 The ERG also observed that the hazard ratios for progression‑free survival and overall survival were calculated using 2 methods of indirect comparison: unadjusted Cox regression and matching adjusted indirect treatment comparison. For the unadjusted Cox regression, the proportional hazards assumption was not consistent with the shape of the Kaplan–Meier curves for progression‑free survival or overall survival for patients having either treatment. The ERG noted that the curves crossed, suggesting that hazard ratios were likely an invalid method of estimating relative effectiveness. The ERG therefore considered that the matching adjusted indirect treatment comparison approach was a more potentially valid method of obtaining point estimates of relative effectiveness, although it reduced the effective sample size and may have increased unobserved confounding and bias.
3.23 The ERG considered that the costs and resources used in the model were generally acceptable. The company included a cost for lymphopenia, but the clinical experts advising the ERG had suggested that the cost of lymphopenia should be 0. The clinical experts also confirmed that most patients have bortezomib subcutaneously because of better tolerance.
3.24 In response to consultation on the appraisal consultation document, the company provided a revised economic analysis that contained all of the ERG's revisions (see sections 3.22 and 3.23). The company submitted 2 new cost‑effectiveness analyses, one incorporating the following:
Time‑dependent hazard ratios derived using the matching adjusted indirect comparison method after independently fitting parametric curves to the Pano‑Bort‑Dex and Len‑Dex data
a comparison of Pano‑Bort‑Dex with Bort‑Dex for patients with relapsed and/or refractory multiple myeloma who have had at least 2 prior regimens including bortezomib and an immunomodulatory agent.
The second analysis incorporated the points above and also included an updated patient access scheme for panobinostat. This used a confidential simple discount on the list price of panobinostat and final 5 year overall survival data from the PANORAMA‑1 trial (academic in confidence and cannot be presented). The patient access scheme for bortezomib was not included in the company's analyses because it only applies to bortezomib monotherapy in people whose multiple myeloma has relapsed for the first time after having one treatment (see NICE technology appraisal guidance on bortezomib for relapsed multiple myeloma). Only the results from this second analysis are presented in this final appraisal document (see sections 3.25 to 3.27).
3.25 For the comparison of Pano‑Bort‑Dex with Len‑Dex, the company's new analysis resulted in a deterministic incremental cost‑effectiveness ratio (ICER) of £11,527 per quality‑adjusted life year (QALY) gained and a probabilistic ICER of £11,883 per QALY gained. For Pano‑Bort‑Dex compared with Bort‑Dex, Pano‑Bort‑Dex dominated (that is, was more effective and less expensive than) Bort‑Dex for both the deterministic and probabilistic ICERs.
3.26 The company also provided several new scenario analyses, all of which incorporated the updated patient access scheme. For the comparison with Len‑Dex, the company presented the following scenarios:
When using a Weibull parametric curve, the ICER increased from £11,527 to £33,385 per QALY gained.
When using the Kaplan–Meier data and extrapolating using a Gompertz model, the ICER increased from £11,527 to £17,891 per QALY gained.
When different models were used for the progression‑free survival data, the ICER decreased and was more favourable to panobinostat.
When the company assumed no active treatment after disease progression, Pano‑Bort‑Dex dominated (that is, was less expensive and more effective than) Len‑Dex.
3.27 The company also carried out a number of scenarios for the comparison of Pano‑Bort‑Dex with Bort‑Dex. In all scenarios Pano‑Bort‑Dex dominated (that is, was less expensive and more effective than) Bort‑Dex except when different approaches were used to extrapolate the overall survival data (which increased the ICERs to over £100,000 per QALY gained).
3.29 The ERG raised concerns about the use of the matching adjusted indirect comparison method and therefore the comparison of Pano‑Bort‑Dex and Len‑Dex. It noted that the comparative effectiveness analysis using this method was more questionable in terms of survival outcomes, because these were likely to be affected by subsequent lines of treatment across the trials included in the comparison (PANORAMA‑1 for Pano‑Bort‑Dex and MM‑009 and MM‑010 for Len‑Dex), although the company had considered them to be equivalent. The ERG was concerned that the company only matched 2 baseline characteristics between the patient groups (time since diagnosis and beta‑2‑microglobulin levels).
3.30 The ERG also noted that in the scenario analysis assuming no treatment after disease progression, the company had only removed the subsequent costs and not the clinical effectiveness associated with these treatments. The ERG was concerned that overall survival gain for Pano‑Bort‑Dex and Bort‑Dex was likely to be driven by differences in the subsequent treatments given after disease progression.
3.31 The ERG was unsure why the company had used a Weibull parametric curve to model overall survival when this was no better a fit than the exponential curve. The ERG considered that this should have been included in the company's sensitivity analyses.
3.32 The ERG noted that in all the company's analyses it had extrapolated the Pano‑Bort‑Dex and Bort‑Dex curves beyond 55 cycles of treatment, which was the post‑progression phase. The ERG noted that the survival curve for Pano‑Bort‑Dex crossed the survival curve for Bort‑Dex at this point and yet the company had extrapolated until cycle 61 (using small numbers; Pano‑Bort‑Dex n=15 and Bort‑Dex n=21), at which point it considered there was no difference in the survival of the 2 groups. The ERG noted that the company did not include a scenario where no survival difference was incorporated from cycle 55 onwards.
3.33 The ERG repeated the company's probabilistic sensitivity analyses. For Pano‑Bort‑Dex compared with Len‑Dex, it found ICERs between £20,000 and £25,000 and so considered the true ICER to be greater than the company's ICER of £11,527 per QALY gained. For Pano‑Bort‑Dex compared with Bort‑Dex, the ERG found that Pano‑Bort‑Dex dominated in all scenarios.
Full details of all the evidence are in the Committee papers.