The Appraisal Committee considered evidence submitted by the manufacturer of pixantrone and a review of the submissions by the Evidence Review Group (ERG). The manufacturer submitted additional evidence (about the patient population, trial design and clinical effectiveness, a revised model, and support for consideration of end-of-life criteria) after consultation on the appraisal consultation document. The submission and the additional evidence did not incorporate a patient access scheme. The manufacturer later made a confidential simple discount patient access scheme submission in July 2013, which was superseded by an updated patient access scheme submitted in November 2013. The Committee's considerations and decision-making are based on the November 2013 patient access scheme.
The manufacturer's systematic review identified 1 randomised controlled trial, which was included in its original submission. No other relevant randomised controlled trials or non-randomised controlled trials were identified. The manufacturer also included some supporting cardiotoxicity data from a randomised phase 2 study that did not meet the inclusion criteria of the literature review (because it evaluated pixantrone in combination with other drugs, not as monotherapy).
PIX301 is a randomised, controlled, open-label phase 3 study conducted in 66 centres, including the USA and Europe. Eligible patients were adults with aggressive de novo or transformed non-Hodgkin's lymphoma that had relapsed after 2 or more chemotherapy regimens, including at least 1 standard anthracycline-containing regimen with a response that had lasted at least 24 weeks. Seventy patients were randomised to pixantrone and 70 patients to a physician's choice of single-agent comparators. The full publication of PIX301 described how 67 patients went on to receive vinorelbine (n=11), oxaliplatin (n=30), ifosfamide (n=12), etoposide (n=9), mitoxantrone (n=4) or gemcitabine (n=1). Pixantrone was administered at a dosage of 85 mg/m2 on days 1, 8 and 15 of a 28‑day cycle for up to 6 cycles. Comparators were administered at predefined standard dosages for up to 6 cycles. Follow-up was for 18 months after completing study treatment.
The primary outcome was complete and unconfirmed complete response, which was determined by a blinded independent assessment panel. Secondary outcomes were overall survival, response lasting at least 4 months and progression-free survival. Other predefined end points were overall response rate, time to response, time to complete response, duration of response and relative dose intensity. Health-related quality of life was not assessed. The primary analysis was the intention-to-treat population. Secondary analyses included a prespecified analysis of the response and survival end points for the histologically confirmed intention-to-treat population (that is, if the lymphoma had been classified according to retrospective independent central pathological assessment).
It was initially planned that 320 patients would be recruited to PIX301 but study enrolment was closed early because of slow accrual. The manufacturer's original submission stated that, with a final enrolment of 140 patients, the study was considered to be sufficiently powered (about 80%) to detect a 15% difference in the complete or unconfirmed complete response rate, assuming a rate of at least 18% in the pixantrone arm. In contrast, the full publication of PIX301 reported that the study was originally powered to detect a difference of 10% in the proportion of patients who achieved a complete or unconfirmed complete response. The publication further stated that, according to the original sample size assumptions, a sample of 70 patients per group would have about 40% power. It added that, to achieve 81% power with 70 patients per group, the true proportion of patients with a complete or unconfirmed complete response would have to be 22% in the pixantrone group and 5% in the comparator group.
The manufacturer's original submission reported that baseline demographic and disease characteristics were similar in the 2 arms. Previous treatment for non-Hodgkin's lymphoma was broadly similar for both groups, including number of chemotherapy regimens (median 8 regimens [range 2 to 9] in both arms). Aggressive histological features were identified onsite in all patients before treatment was given and confirmed by central independent pathological review in 54 (77%) of 70 patients in the pixantrone arm and 50 (71%) of 70 patients in the comparator arm receiving treatment of physician's choice. Of the remaining 36 patients, reasons for non-confirmation were low-grade histology (n=13), lack of consensus (n=10), shortage of specimen (n=6), confirmation of a non-aggressive subtype other than non-Hodgkin's lymphoma (n=5), and the specimen reviewed by only 1 pathologist (n=2). Out of 140 patients, 36 patients completed 6 cycles of protocol treatment, and 104 patients discontinued early. The most common reason for early discontinuation in both groups was disease progression or relapse. After completing study treatment, 95 patients entered follow-up and 26 of these completed 18 months of follow-up.
The manufacturer's original submission reported that, at the end of treatment, confirmed and unconfirmed response rates for the intention-to-treat population (70 patients in each arm) were statistically significantly higher for the pixantrone group than the comparator group receiving treatment of physician's choice (20% compared with 5.7%; p=0.021). This was also the case at the end of the study after 18 months of follow-up (24.3% compared with 7.1%; p=0.009).
The manufacturer's original submission described the results for progression-free and overall survival in the intention-to-treat population. Median progression-free survival was statistically significantly longer for the group receiving pixantrone than the comparator group receiving treatment of physician's choice at 5.3 months compared with 2.6 months (hazard ratio 0.60 and 95% confidence interval [CI] 0.42 to 0.82; p=0.005). However, there was no statistically significant difference in median overall survival between the 2 groups (10.2 months in the pixantrone arm compared with 7.6 months in the comparator arm receiving treatment of physician's choice (hazard ratio 0.79 [95% CI 0.53 to 1.18]; p=0.251).
In addition to the results for the intention-to-treat population of PIX301, the manufacturer included clinical-effectiveness data for several post-hoc subgroups in its original submission for patients with aggressive B‑cell lymphoma (classed as diffuse large B‑cell lymphoma, transformed indolent lymphoma or follicular lymphoma [grade 3]):
Disease confirmed by onsite pathological review (all lines of treatment, and third- or fourth-line treatment only).
Disease confirmed by central independent pathological review (all lines of treatment, and third- or fourth-line treatment only).
Disease confirmed by central independent pathological review in patients who had previously received rituximab treatment.
The manufacturer also presented the results for a subgroup of patients with diffuse large B‑cell lymphoma confirmed by onsite pathological review (over 80% of the total number of patients with aggressive B‑cell lymphoma). The results of the subgroup analyses that were incorporated into the manufacturer's original economic model are described below. The results of other subgroup analyses included in the manufacturer's original submission have been previously reported in the appraisal consultation document.
In its original submission, the manufacturer considered the post-hoc subgroup of patients with aggressive B‑cell lymphoma confirmed by onsite pathological review to be similar to the population eligible for treatment according to pixantrone's European marketing authorisation, and indicated that this formed the basis of the population in the original base case of its cost-effectiveness analysis (however, it should be noted that the manufacturer stated that its economic evaluation focused on those who had received 2 or 3 previous therapies; see section 3.20 for details). This subgroup excluded patients with peripheral T‑cell lymphoma not otherwise characterised and other disease subtypes not included in pixantrone's European marketing authorisation. Compared with the group that received treatment of physician's choice (n=62), complete or unconfirmed complete response rates at the end of the study in the pixantrone group (n=64) were statistically significantly higher (23.4% compared with 8.1%; p=0.027). Overall response rates were also statistically significantly higher in the pixantrone group (40.6% compared with 16.1%; p=0.003). Median progression-free survival was statistically significantly longer in patients who had received pixantrone than those who had received a comparator drug (5.7 months compared with 2.5 months, hazard ratio 0.56 [95% CI 0.38 to 0.81]; p=0.002). The manufacturer advised that median overall survival was not included because the aggressive B‑cell lymphoma analyses were exploratory.
In its original submission, the manufacturer presented a further analysis of patients with aggressive B‑cell lymphoma confirmed by onsite pathological review who received pixantrone (n=50) or a comparator (n=49) as third- or fourth-line treatment, which it stated was more closely aligned with pixantrone's marketing authorisation. It is not clear from the manufacturer's submission how this population differs from that in the base case of its cost-effectiveness analyses (see section 3.20 for details). The group receiving pixantrone had a statistically significantly higher complete response or unconfirmed complete response rate (28.0% compared with 4.0%; p=0.002) and overall response rate (48.0% compared with 12.2%; p<0.001), and statistically significantly longer progression-free survival (5.8 months compared with 2.8 months, hazard ratio not stated; p=0.002) than the comparator group receiving treatment of physician's choice. Median overall survival in this population was numerically higher in the pixantrone arm than the comparator arm but this difference was not statistically significant (13.9 months compared with 7.8 months, hazard ratio 0.76 [95% CI 0.47 to 1.24]; p=0.275). The manufacturer's submission did not state whether the results were for end of treatment or end of study.
In addition to the histologically defined subgroups of the PIX301 population in its original submission, the manufacturer also supplied subgroup analyses that showed the influence of previous rituximab treatment on pixantrone's efficacy in the subgroup of patients who had aggressive non-Hodgkin's B‑cell lymphoma confirmed by central independent pathological review. In this subgroup of patients who had previously received rituximab, there was no statistically significant difference between pixantrone (n=30) and the comparator arm (n=26) in the proportion of patients who had a complete or unconfirmed complete response at the end of treatment (16.7% compared with 7.7%; p=0.431). Median progression-free survival was longer in the pixantrone group than in the comparator group receiving treatment of physician's choice for this subgroup of patients but the difference did not reach statistical significance (3.5 months compared with 2.3 months, hazard ratio 0.66 [95% CI 0.38 to 1.14]). Similarly, median overall survival was longer in the pixantrone group than in the comparator group receiving treatment of physician's choice but the between-group difference was not statistically significant (6.0 months compared with 4.6 months, hazard ratio 0.85 [95% CI 0.48 to 1.50]).
The manufacturer's original submission described the adverse events in PIX301 for 68 patients in the pixantrone group and 67 patients in the comparator group who received treatment of physician's choice. One dose reduction was allowed for patients who had neutropenia during treatment, and reductions were similar in the pixantrone and comparator groups (18% compared with 15%). Dose delay was more frequent with pixantrone (40% compared with 22%).
A similar number of patients had an adverse event of any grade but there was a higher incidence of grade 3 and 4 adverse events in the pixantrone group than in the comparator group (76.5% compared with 52.2%). Neutropenia occurred more frequently in the pixantrone group and was the most common adverse event of any grade (50.0% compared with 23.9%) and the most common grade 3 or 4 adverse event (41.2% compared with 19.4%). Grade 3 or 4 febrile neutropenia was also more common in the pixantrone group than in the comparator group (7.4% compared with 3.0%), and more patients in the pixantrone group than in the comparator group received an immunostimulant (51.5% compared with 26.9%). The manufacturer reported that severity of neutropenia did not increase with increasing cycle number and that the overall rates of grade 3 and 4 infections were similar in the 2 groups. It further stated that the common adverse events were similar to those expected in a heavily pretreated patient population, which reflected pixantrone's intended use in clinical practice in England and Wales (that is, third and subsequent lines of treatment).
Approximately 40% of patients in both treatment arms presented with a history of cardiac disease at study enrolment, and cardiac risk factors were also similar in the 2 groups. The manufacturer stated that pixantrone is an innovative treatment because it has been specifically designed to reduce cardiotoxicity associated with anthracyclines without compromising efficacy. More cardiac adverse events occurred in the pixantrone group (24 patients [35.3%] than in the comparator group who received treatment of physician's choice (14 patients [20.9%]). Thirteen (19.1%) patients in the pixantrone group experienced decreased left ventricular ejection fraction compared with 7 patients in the comparator group. The manufacturer provided supporting cardiotoxicity data from the randomised open-label phase 2 PIX203 trial, which closed before enrolment completed. This trial compared the combination of cyclophosphamide, pixantrone, vincristine, prednisone and rituximab with the standard of care (that is, rituximab in combination with a regimen of cyclophosphamide, doxorubicin, vincristine and prednisone) as first-line treatment in patients with diffuse large B‑cell lymphoma. The cardiotoxicity results of PIX203 broadly supported those of PIX301.
In response to consultation on the first appraisal consultation document, the manufacturer requested and received permission from NICE to submit additional evidence. The additional evidence contained the results for 4 subgroups and showed the effect of treatment in patients who had previously received rituximab. Two of the subgroups were patients who had aggressive non-Hodgkin's B‑cell lymphoma confirmed by central independent pathological review for all lines of therapy, and the other 2 subgroups were patients with aggressive B‑cell lymphoma confirmed by central independent pathological review who were receiving third- or fourth-line treatment. During the second Committee meeting, the manufacturer clarified that the subgroups of all patients regardless of rituximab status (those patients who had previously received rituximab plus those who had not) were labelled as 'without rituximab' in its response to consultation.
In its additional evidence, the manufacturer presented amended results for the subgroup of patients who had aggressive non-Hodgkin's B‑cell lymphoma confirmed by central independent pathological review and had previously received rituximab (see section 3.11). Complete or unconfirmed complete response rates were higher in the pixantrone arm than in the comparator arm (20% compared with 11%) but this difference was not statistically significant. Results for progression-free survival and overall survival were as before, except for median progression-free survival in the comparator arm, which the manufacturer confirmed was an error. In its consultation response, the manufacturer also reiterated data for the subgroup of all patients who had aggressive non-Hodgkin's B‑cell lymphoma confirmed by central independent pathological review regardless of whether they had previously received rituximab or not (n=50 in the pixantrone group, n=47 in the comparator group). At the end of the study, there was no statistically significant difference in complete or unconfirmed complete response rates between the pixantrone and comparator groups (9 patients compared with 4 patients; p=0.236). However, the overall response rate was statistically significantly higher in the pixantrone group (18 patients compared with 8 patients; p=0.041). Median progression-free survival was statistically significantly longer in the pixantrone arm than in the comparator arm (5.6 months compared with 2.5 months, hazard ratio 0.51 [95% CI 0.33 to 0.78]; p value not stated) but there was no statistically significant difference in median overall survival between the 2 groups (8.1 months compared with 6.3 months, hazard ratio 0.72 [95% CI 0.45 to 1.13]; p value not stated).
In its additional evidence, the manufacturer reiterated subgroup analyses for all patients (that is, patients who had previously received rituximab plus those who had not) with aggressive B‑cell lymphoma confirmed by central independent pathological review who were receiving third- or fourth-line treatment with pixantrone (n=39) or treatment of physician's choice (n=39). Compared with the group receiving treatment of physician's choice, the pixantrone group had a statistically significantly higher complete or unconfirmed complete response rate (23.1% compared with 5.1%; p=0.047) and overall response rate (43.6% compared with 12.8%; p=0.005). Median progression-free survival was statistically significantly longer with pixantrone than with treatment of physician's choice (5.7 months compared with 2.8 months, hazard ratio 0.44 [95% CI 0.27 to 0.71]) but there was no statistically significant difference in median overall survival between treatment groups (11.9 months with pixantrone compared with 7.0 months with treatment of physician's choice, hazard ratio 0.67 [95% CI 0.40 to 1.12]).
In its additional evidence, the manufacturer also provided results for a subgroup described as patients with aggressive B‑cell lymphoma confirmed by central independent pathological review who were receiving third- or fourth-line treatment who had previously received rituximab (n=25 in both study arms), but confirmed at the second Committee meeting that this population was in fact a subgroup of patients whose disease had been confirmed by onsite (not central) pathological review. The correct population (that is, with pathology confirmed by central independent review), whose results had previously been included as part of the manufacturer's clarification response, showed an increase in complete response or unconfirmed complete response in the pixantrone arm (n=20) compared with the comparator arm (n=18; 30% compared with 5.6%; p=0.093). There was no statistically significant difference between treatment arms in median progression-free survival (5.4 months in the pixantrone group and 2.8 months in the comparator group, hazard ratio 0.52 [95% CI 0.26 to 1.04]). Similarly, there was no statistically significant difference in median or mean overall survival between treatment arms (hazard ratio 0.76 [95% CI 0.38 to 1.55]). Median overall survival was 7.5 months in the pixantrone group and 5.4 months in the comparator group. Mean overall survival was 9.9 months in the pixantrone arm and 7.9 months in the comparator group (difference of 2.0 months).
The manufacturer did not identify any published economic evaluations or costing studies that were relevant to the decision problem. Consequently, it submitted a de novo economic analysis that assessed the cost effectiveness of pixantrone compared with treatment of physician's choice in treating multiply relapsed or refractory aggressive B‑cell lymphoma, which was later revised as part of the manufacturer's response to consultation (see section 3.31 for details of changes in the revised model). Further minor updates were made in the manufacturer's model that formed part of a patient access scheme submission which was submitted in July 2013. In November 2013, this was superseded by another patient access scheme submission, which applied a further simple discount to the model (see sections 3.34 to 3.38 for details).
The manufacturer advised that the base-case model considered patients who had received 2 or 3 prior therapies and whose disease was sensitive to treatment with anthracyclines because this population was consistent with pixantrone's European marketing authorisation for treating multiply relapsed or refractory aggressive non-Hodgkin's lymphoma (the marketing authorisation notes that a treatment benefit has not been established 'when used as fifth line or greater chemotherapy in patients who are refractory to last therapy'). The clinical data for this population were derived from PIX301. The analysis was conducted from an NHS and personal and social services perspective and a lifetime horizon of 23 years was used. Weekly cycles were chosen to capture the 4‑week treatment cycles of pixantrone and 3‑week treatment cycles of some of the comparator treatments and a half-cycle correction was applied. Costs and benefits were discounted at 3.5% per annum.
The manufacturer created a semi-Markov model that contained 3 health states: stable or no progression, progressive or relapsed disease, and death. The stable or no progression health state had 2 distinct subpopulations. The first of these was patients on initial third- or fourth-line treatment. The second was patients who had discontinued third- or fourth-line treatment (because of complete response, adverse event, completion of 6 months' treatment or a non-clinical reason) but had not experienced progression. All patients entered the model in the on-treatment subpopulation within the stable or no progression health state. During each cycle, patients could remain in the on-treatment subpopulation of this health state, discontinue treatment and move into the other subpopulation in this health state, progress and move into the progressive disease health state, or die. Patients who discontinued treatment before progression remained at risk of progression or death. Following progression, patients were at risk of death and unable to return to the stable or no progression health state. It was assumed that the original treatment was stopped following disease progression and patients received further treatment or palliative care. Adverse events were captured as events within the model by applying a utility decrement (disutility).
The manufacturer outlined how the transition between health states was calculated from the clinical data for any given weekly cycle. It noted that semi-Markov models allow the use of a partition approach, which has been used extensively in oncology because it is particularly suited to progressive conditions that have ongoing risks that may vary over time. The distribution of the patient group between the different health states was estimated by calculating the area under the survival curves at each cycle. The progression-free survival curve defined the stable or no progression state, while the progressed state was defined by subtracting those patients who remained progression free from all surviving patients.
Clinical parameters for progression-free survival and overall survival were incorporated into the base case of the manufacturer's economic model by statistical analysis of patient-level data from the aggressive B‑cell population of PIX301. Predictive equations for progression-free survival and overall survival were derived by fitting the patient-level data and extrapolating beyond the data from PIX301 (around 2 years). A log-normal distribution was used in the base case for both progression-free survival and overall survival.
Further clinical parameters were incorporated into the base case of the manufacturer's economic model. The cycle probability of treatment discontinuation distinguished between patients remaining on initial treatment and those who discontinued while stable. The frequency and duration of adverse events (grades 2 to 4) before progression while taking initial treatment were based on PIX301. Grade 3 and 4 adverse events occurring in at least 5% of the total patient population were considered to have cost and utility consequences. Some grade 2, and rarer grade 3 and 4, adverse events were included if considered important by clinical specialists in England. Other data from PIX301 that were used to inform the model were mean dose for the comparator treatments plus sex and body surface area.
There were no patient-reported outcomes in PIX301 and the manufacturer did not identify any utility data for any line of treatment in aggressive non-Hodgkin's lymphoma in its systematic literature review for studies on health-related quality of life. Utility data were identified from published sources for similar patient populations, and for disease areas with similar expected survival, disease progression, nature of the disease and quality of life. These were diffuse large B‑cell lymphoma, chronic myelogenous leukaemia, chronic lymphocytic leukaemia, follicular lymphoma, renal cell carcinoma and melanoma. For its original model, the manufacturer considered the self-reported quality of life in older patients with aggressive diffuse large B‑cell lymphoma to give the estimation closest to the PIX301 trial population and used these values (pre-progression 0.81, post-progression 0.60) in its base-case analysis. The manufacturer did not provide a rationale for this decision. Utility values were assumed to depend only on the health state and any adverse events experienced, but not the treatment arm. Based on expert clinical opinion, the manufacturer assumed no difference in baseline health-related quality of life between the 2 subpopulations in the stable or no progression health state. All stable/no progression patients were assumed to have similar quality of life (that is, there was no difference according to complete response, partial response or stable disease).
The manufacturer determined disutilities associated with each adverse event that was included in the original model from relevant literature from other oncology indications. If no utility decrements were available, the maximum value of the range identified was assumed by the manufacturer to keep the calculations conservative (that is, so that pixantrone was not favoured).
Adverse events were modelled by the manufacturer as events rather than as health states and were assumed to be time independent because adverse events are likely to be experienced at different stages of treatment. Any grade 1 to 4 adverse event that occurred in less than 5% of the trial population was assumed to have no impact on quality of life. After consulting some clinical specialists in England, the manufacturer included some rarer grade 3 and 4 adverse events and some grade 2 adverse events that the clinical specialists considered to be important. Because no disutility values were available specifically for grade 2 and grade 3 or 4 adverse events, they were assumed to be the same for each grade. Within a health state, disutilities relating to an adverse event were applied to the proportion of patients assumed to experience the adverse event as weighted average disutilities. For each treatment, the manufacturer calculated a weighted average of grade-specific disutilities that were weighted by the number of effects of that particular grade. The disutility for each adverse event was then applied for the duration of that specific type of effect. The manufacturer's model limited the consideration of adverse events to patients on original treatment upon entering the model (pixantrone or treatment of physician's choice).
Costs captured in the manufacturer's model included drugs and their administration, plus those associated with health state and disease management, including adverse events. Drug and administration costs in the original model were calculated based on average dose per administration from the PIX301 trial using the BNF edition 62 (published in September 2011) and the NHS reference costs. No patient access scheme was incorporated in the original model. From the second attendance onwards, administration costs were £206 for each attendance for all drugs except etoposide 50 mg (£163). At clarification, the manufacturer corrected an error in the vial price, which had been mistakenly quoted as £343.80 (based on the vial size given for pixantrone base) instead of £553.50 (equivalent to 50 mg pixantrone dimaleate). It advised that this error had a minimal impact on the cost-effectiveness estimates (which increased by 0.3%) because the drug costs in the model had been calculated based on cost per administration. The total number of administrations varied according to the dosing schedule for each drug. Drug wastage was incorporated in the base case. Personal and social services costs were £476.42 per 28 days for stable health state on treatment, £119.10 for stable health state on palliative care and £1,993.89 for progressive health state. Disease management costs (comprising healthcare professional contact, disease follow-up and hospital-related costs) were different for active treatment and palliative care. For active treatment, health professional contact costs were £788.96 on treatment and £220.38 after treatment (per 28 days), disease follow-up costs were £86.63 per 28 days and annual hospital-related costs were £2,357.28. For palliative care, health professional contact costs were £990.74 per 28 days, disease follow-up costs were £18.44 per 28 days and annual hospital-related costs were £1,982.03. End-of-life care was excluded from the calculations because it affected only the last few weeks of life and estimates would be similar for pixantrone and its comparators. Within a health state, costs for managing an adverse event were applied to the proportion of patients assumed to experience the adverse event.
The manufacturer advised that the predicted median progression-free survival and predicted median overall survival were similar to the results reported in PIX301. Compared with the clinical trial results, the manufacturer noted that the original model slightly underestimated the median overall survival with pixantrone (13.1 months compared with 13.8 months) while overestimating it for the comparator (9.2 months compared with 7.6 months). It reported that, conversely, the original model overestimated the median progression-free survival for the pixantrone arm (7.8 months compared with 6.4 months) and slightly underestimated it for the comparator arm (3.2 months compared with 3.5 months).
Using the original model, the manufacturer's base-case analyses for pixantrone compared with treatment of physician's choice in patients with aggressive B‑cell lymphoma confirmed by onsite pathological review (third- or fourth-line treatment) produced a deterministic incremental cost-effectiveness ratio (ICER) of £28,423 per quality-adjusted life year (QALY) gained. Incremental costs were £17,638 and incremental QALYs were 0.62. Using the correct vial price supplied at clarification increased the ICER to £28,503 per QALY gained. No probabilistic base-case ICER was presented. All economic analysis results generated using the manufacturer's original model have been superseded by those using the model provided with the patient access scheme submission submitted in November 2013 (see sections 3.34 to 3.38).
In response to the first appraisal consultation document, the manufacturer provided a revised economic model, which contained these amendments:
The adverse-event disutilities used in the ERG's exploratory analyses were incorporated.
Drug costs for comparator treatments were taken from the NHS Commercial Medicines Unit's Electronic Marketing Information Tool (eMIT) database instead of the BNF in line with NICE's Guide to the methods of technology appraisal (2013).
Utility values were changed from self-reported quality of life in older patients with aggressive diffuse large B‑cell lymphoma (0.81 for the pre-progression health state, 0.60 for the post-progression health state) to those for second- and subsequent-line treatment of renal cell carcinoma (0.76 for the pre-progression health state and 0.68 for the post-progression health state). This was in response to the Committee's conclusion in the first Committee meeting that the original utility values had overestimated quality of life for patients with multiply relapsed or refractory non-Hodgkin's lymphoma.
The revised model did not incorporate a patient access scheme.
Using its revised model, the manufacturer provided cost-effectiveness estimates of pixantrone compared with treatment of physician's choice for several subgroups, including patients with aggressive B‑cell lymphoma confirmed by central independent pathological review who were receiving third- or fourth-line treatment and had previously had rituximab. The deterministic ICER for this subgroup was £45,282 per QALY gained (incremental costs £9,170; incremental QALYs 0.20). No probabilistic ICER was provided. All economic analysis results generated using the manufacturer's revised model have been superseded by those using the model provided with the patient access scheme submission submitted in November 2013 (see sections 3.34 to 3.38).
The manufacturer agreed a patient access scheme in July 2013, which was a confidential simple discount on the list price of pixantrone. It further updated its economic model so that costs for treating adverse events and the cost for methotrexate were in line with the Committee's preferred values decided at the second Committee meeting. All economic analysis results generated using the manufacturer's model provided with the patient access scheme submitted in July 2013 have been superseded by those using the model provided with the patient access scheme submission submitted in November 2013 (see sections 3.34 to 3.38).
In response to the second appraisal consultation document, the manufacturer submitted a patient access scheme in November 2013 that contained an additional discount to the patient access scheme proposed in July 2013. The November 2013 patient access scheme is a simple discount on the list price of pixantrone, and the economic model was further updated with this additional discount as part of this submission. The manufacturer advised that the patient access scheme would apply to patients with histologically confirmed aggressive non-Hodgkin's B‑cell lymphoma who had previously received rituximab and would be receiving pixantrone as a third- or fourth-line treatment. The manufacturer made no other revisions to the economic model.
The manufacturer provided cost-effectiveness estimates incorporating the patient access scheme submitted in November 2013 for pixantrone compared with treatment of physician's choice for patients with aggressive B‑cell lymphoma confirmed by central independent pathological review who were receiving third- or fourth-line treatment and had previously received rituximab. This additional discount reduced the deterministic ICER to £18,462 per QALY gained (incremental costs are commercial in confidence and so cannot be shown here; incremental QALYs 0.20).
The manufacturer tested the robustness of the model using one-way sensitivity analyses and reported that the key drivers of the cost-effectiveness estimates produced using its economic model were the parametric fitting methodology for progression-free survival and overall survival, and the utility estimate for the stable or no progression health state. The manufacturer noted that the ICER was sensitive to changes in the estimates for progression-free survival.
In its base-case patient access scheme submission provided in November 2013, the manufacturer used the same utility values as the revised version of the model, which were for patients receiving second- and subsequent-line treatment for renal cell carcinoma (0.76 for the pre-progression health state and 0.68 for the post-progression health state). The manufacturer provided alternative utility scenarios using data from published sources for similar patient populations, and for disease areas with similar characteristics. These were second-line treatment in patients with chronic myelogenous leukaemia, third-line treatment in patients with chronic lymphocytic leukaemia, first-line maintenance treatment in patients with follicular lymphoma, first-line treatment in patients with metastatic renal cell carcinoma, self-reported quality of life during chemotherapy in elderly patients with aggressive non-Hodgkin's lymphoma and second-line treatment in patients with malignant melanoma. The ICERs ranged from £14,607 per QALY gained to £18,871 per QALY gained.
To explore uncertainty, the manufacturer undertook a probabilistic sensitivity analysis that incorporated the patient access scheme submitted in November 2013. The probabilistic mean ICER was £22,024 per QALY gained (incremental costs are commercial in confidence and so cannot be shown here; incremental QALYs 0.18). However the manufacturer asserted that there was structural uncertainty inherent in the probabilistic sensitivity analysis that reduced the advantage of pixantrone and skewed the probabilistic results. It explained that the model assumed that overall survival and progression-free survival were independent, leading to the survival curves crossing (that is, more patients were in the progression-free survival state than alive) and the model artificially reduced progression-free survival to avoid this, and that this applied to around 30% of the simulations. In order to illustrate the structural uncertainty arising from the probabilistic ICER, the manufacturer provided additional analyses; for example, if it is assumed that overall survival and progression-free survival are not independent (by assuming the same random numbers for progression-free survival and overall survival using the Cholesky decomposition), the probabilistic mean ICER is £9,938 per QALY gained (incremental costs are commercial in confidence and so cannot be shown here; incremental QALYs 0.21). The manufacturer also noted that the joint uncertainty of incremental QALYs and incremental costs did not follow a normal distribution, and reported that the median probabilistic ICER was £14,692 per QALY gained (incremental costs are commercial in confidence and so cannot be shown here; incremental QALYs 0.18). The manufacturer reported that pixantrone was more likely to be cost effective compared with treatment of physician's choice in patients with aggressive B‑cell lymphoma confirmed by central pathological review who were receiving third- or fourth-line treatment at a maximum acceptable ICER of £25,000 per QALY gained.
The ERG considered the evidence included by the manufacturer to be relevant to the decision problem in its analysis. No additional relevant trials were identified and the ERG found that the manufacturer's systematic review followed standard practices.
The ERG had concerns about the generalisability of the PIX301 population to clinical practice in England and Wales, particularly the potential effect of previous rituximab treatment on the response to pixantrone because rituximab is given as part of standard first-line treatment in the UK. The ERG noted that about 50% of patients in PIX301 had previously received treatment with a biological agent (for example, rituximab). The ERG considered the clinical benefit of pixantrone in patients who have previously been treated with rituximab to be a key area of uncertainty, given that there were no statistically significant differences between the pixantrone and comparator arms for complete or unconfirmed complete response, progression-free survival or overall survival in the subgroup of patients with aggressive B‑cell lymphoma confirmed by central independent pathological review who had previously received rituximab.
The ERG considered whether the treatments of physician's choice in PIX301 represented clinical practice in England and Wales. Following input from its clinical specialists, the ERG noted that there is no consensus on which chemotherapy regimens should be used after second-line treatment fails and that there is a lack of comparative data on their clinical effectiveness. The ERG concluded that this meant the choice of treatment in the comparator arm of PIX301 was unlikely to be a key issue. It also concluded that the small number of patients receiving each treatment meant that the choice of treatment in the comparator group could not be reliably analysed.
The ERG was concerned about the statistical power of PIX301 to detect a difference between treatment groups. According to the manufacturer's revised power calculation, 81% power with 70 patients per group (the intention-to-treat population) would be achieved if the true proportion of patients with complete or unconfirmed complete response was 22% in the pixantrone group and 5% in the comparator group. However, the observed proportions of patients with a complete or unconfirmed complete response in the intention-to-treat population were 20.0% in the pixantrone group and 5.7% in the comparator group. The ERG noted that the difference between groups did not always reach statistical significance, and that results of the analyses in the subgroups confirmed by central independent pathological review should be interpreted with caution because they are likely to be underpowered to detect a difference between treatment groups. For these reasons, the ERG had reservations about whether pixantrone had been shown to have superior efficacy in PIX301.
The ERG was concerned about the reliability of the diagnosis of aggressive non-Hodgkin's lymphoma at study entry. It noted that central independent pathological review by consensus was undertaken retrospectively (that is, after the trial), rather than at enrolment, and that aggressive disease was subsequently confirmed in only 104 of the 140 patients who were randomised. Consequently, it felt that results from the full trial population might not reflect the benefit of pixantrone in patients with aggressive B‑cell lymphoma. The ERG acknowledged that the manufacturer said it had not been practical to confirm aggressive disease by central independent pathological review at enrolment, but considered it was important to evaluate data from the subgroup of patients with disease confirmed by central independent pathological review. The ERG noted that, as a subgroup analysis, the statistical power of PIX301 would be less than the intention-to-treat population.
The ERG considered the different patient populations in the subgroup analyses presented by the manufacturer. The ERG viewed the data from the post-hoc subgroup of patients with aggressive non-Hodgkin's B‑cell lymphoma that was histologically confirmed by central independent pathological review to be more relevant to the marketing authorisation and the decision problem in the NICE scope than the other 2 subgroups categorised according to type of lymphoma determined by onsite pathological review (patients with aggressive non-Hodgkin's B‑cell lymphoma and patients with diffuse large B‑cell lymphoma). The ERG noted that retrospective central independent pathological review revealed 23% of patients receiving pixantrone and 29% of patients receiving a comparator in the intention-to-treat population had disease that was not subsequently confirmed as being aggressive. The ERG was aware that disease severity is an important factor in deciding treatment strategy because patients without aggressive disease are likely to have a more favourable response than those with aggressive disease.
The ERG considered the statistical robustness of the subgroup analyses. It observed that the comparative clinical-effectiveness results for most of the subgroups were based on post-hoc subgroup analyses. It also noted that the number of patients in the analysis was generally small, increasing uncertainty around the results. For subgroups based on retrospective histological confirmation of aggressive disease and previous rituximab treatment, the ERG noted the potential for unbalanced groups because randomisation had not been stratified by these factors. The ERG concluded that the results of the subgroup analyses should be interpreted with caution.
The ERG considered the adverse events reported to occur more often in the pixantrone group than the comparator group receiving treatment of physician's choice to be consistent with the common adverse events associated with pixantrone reported in the summary of product characteristics.
Overall, the ERG considered the manufacturer's original model to be in line with current best practice recommendations, generally well constructed and largely transparent. The ERG considered that an important limitation of the manufacturer's original base-case analysis was that it used data from patients whose disease had not been histologically confirmed as aggressive. The ERG indicated that the subgroup of patients with aggressive non-Hodgkin's B‑cell lymphoma confirmed by central independent pathological review for all lines of treatment in PIX301 was the most informative to the decision problem because it excluded patients who were later found to have disease that was irrelevant to the decision problem (for example, indolent disease). However, the ERG noted that the manufacturer's estimate of cost effectiveness in this patient population was highly uncertain because it used post-hoc subgroup data and because the subgroups were not powered to detect a difference in efficacy between treatment with pixantrone and the comparators.
The ERG considered the utility weights used by the manufacturer in its original economic model to be potentially inappropriate. It noted that the utility values were from a population of patients receiving first-line treatment for aggressive non-Hodgkin's lymphoma and were derived from a study that had initially been rejected by the manufacturer in its systematic review. It further noted that the manufacturer's reported utility values were higher than those that have been derived for healthy older patients in the UK.
The ERG noted that the results of the manufacturer's original economic model may potentially be biased towards pixantrone because of an overestimation of pixantrone's relative progression-free survival benefit compared with treatment of physician's choice for the populations with aggressive B‑cell lymphoma (whether confirmed by onsite or central independent pathological review). Clinical specialist opinion received by the ERG expressed concern that the data used in the model may not be sufficient to reach reasonable conclusions about the clinical or cost effectiveness of pixantrone.
The ERG identified other areas of inaccuracy or uncertainty in the assumptions and parameter estimates used in the manufacturer's original model and indicated the most significant of these were structural assumptions made about treatment discontinuation, disutility, and the cost parameters used:
The potential double-counting of treatment discontinuation because of disease progression.
Excluding adverse event disutilities for patients on further lines of treatment.
Discrepancies between the manufacturer's and ERG's interpretation of the literature on disutilities for adverse events.
Using weighted average adverse event rates to inform costs and disutilities associated with adverse events for patients on original treatment.
Missing data from data used to inform average adverse event costs.
Excluding costs associated with treating leukopenia and thrombocytopenia.
Using costs from BNF 62 (published September 2011) rather than BNF 64 (published September 2012).
The ERG critiqued the manufacturer's additional evidence submitted in response to the first consultation. Although there were some uncertainties in the definition and labelling of the manufacturer's subgroups, the ERG was able to validate the terminology used and check that the changes the manufacturer had made to its model were appropriate. The ERG indicated that, based on the European marketing authorisation and clinical practice in England and Wales, it was important to evaluate patients with aggressive B‑cell lymphoma confirmed by retrospective central independent pathological review who had previously received rituximab. However, it noted that it was also appropriate to evaluate a subgroup of these patients receiving third- or fourth-line treatment.
The ERG critiqued the manufacturer's patient access scheme submission. It validated the changes made to the economic model in the July 2013 and November 2013 submissions and agreed that the subgroup of patients covered by the patient access scheme was characteristic of patients in England and Wales who were likely to be eligible for treatment with pixantrone.
The ERG validated the manufacturer's deterministic and probabilistic sensitivity analyses on the updated base case in the patient access scheme submitted in November 2013, which were conducted in the same way as in the manufacturer's original submission. The ERG had commented in its critique of the manufacturer's original submission that the manufacturer's assessment of uncertainty was very detailed and that the probabilistic and one-way sensitivity analyses, including various scenario analyses, were satisfactorily reported.
The ERG commented that the absence of utility data in patients with relapsed or refractory aggressive non-Hodgkin's lymphoma results in a great deal of uncertainty. The ERG stated that although the utility values derived from patients receiving second-line treatment for renal cell carcinoma were less favourable than those used in the manufacturer's original submission, they may still overestimate the utility of patients with aggressive non-Hodgkin's B‑cell lymphoma receiving third- or fourth-line treatment.
The ERG stated that the manufacturer's original base-case cost-effectiveness results (before inclusion of any patient access scheme) were generated deterministically rather than probabilistically (that is, mean values rather than distributions were used to inform the value of each parameter). However, the ERG noted that probabilistic cost-effectiveness results could be assessed using the manufacturer's original model. It noted a wide range in the 95% confidence interval for the mean probabilistic ICER. The ERG considered this showed substantial uncertainty in the manufacturer's cost-effectiveness results.
Using the manufacturer's original model (before inclusion of any patient access scheme), the ERG carried out exploratory sensitivity analyses to investigate the impact of alternative assumptions or parameters on the manufacturer's cost-effectiveness results. The ERG judged the population with aggressive B‑cell lymphoma confirmed by central independent pathological review for all lines of treatment in PIX301 to be the most relevant to the decision problem (because it excluded patients who were later found to have disease that was not relevant to the decision problem [for example, indolent disease]) and used it in all its exploratory analyses with the original model.
Because it had concluded that the utility values used by the manufacturer in its original model (before inclusion of any patient access scheme) may have been inappropriate, the ERG investigated how alternative utility values affected the manufacturer's original base case. The ERG presented a markedly increased ICER for pixantrone compared with treatment of physician's choice when it used utility data from chronic lymphocytic leukaemia patients receiving third- or later-line treatment to inform the utility of progression-free survival and progressive disease (0.428 for the pre-progression health state and 0.279 for the post-progression health state).
The ERG discussed the probabilistic cost-effectiveness results generated by the manufacturer using the patient access scheme submitted in November 2013, and validated the changes by the manufacturer to the economic model. The ERG commented that the manufacturer's probabilistic analysis showed that, after incorporating the patient access scheme, the probability of pixantrone being cost-effective compared with treatment of physician's choice was 49.3% at up to £20,000 per QALY gained and 55.7% at up to £30,000 per QALY gained. The ERG considered it important to note that, although 68% of probabilistic iterations showed a greater benefit for pixantrone than treatment of physician's choice, around 32% of probabilistic iterations indicated that patients treated with pixantrone fared worse than those who received treatment of physician's choice.
Full details of all the evidence are in the evaluation report.