3 The manufacturer's submission
3.1 The manufacturer identified best supportive care as the comparator for rituximab monotherapy at second and subsequent relapse (the indication appraised in TA 37). No new evidence was provided for this indication. For the use of rituximab with chemotherapy for induction of remission in relapsed follicular non-Hodgkin's lymphoma, the main comparator identified was cyclophosphamide, hydroxydaunomycin, vincristine and prednisolone (CHOP) chemotherapy, and fludarabine-containing regimens were also considered appropriate. For the use of rituximab as maintenance therapy, the appropriate comparator was considered to be observation only. For the latter two indications the manufacturer identified two trials.
3.2 The European Organisation for Research and Treatment of Cancer (EORTC) trial was an open-label study conducted in patients with follicular non-Hodgkin's lymphoma, in first and subsequent relapse, who had not previously received rituximab. Patients (N = 465) were randomised to induction with six cycles of CHOP chemotherapy plus rituximab (N = 234) or CHOP without rituximab (N = 231). Those patients in remission after six cycles (N = 334) were subject to a second randomisation: to observation only (N = 167) or eight doses of maintenance therapy with rituximab, given over 2 years (N = 167). For induction of remission, there was a statistically significant higher overall response rate following combination therapy with CHOP plus rituximab compared with CHOP alone (85% versus 72%, respectively; p < 0.0001). The median progression-free survival was also statistically significantly longer for patients who received combination therapy (33 months versus 20 months; p = 0.0003).
3.3 For patients on rituximab maintenance, the median progression-free survival was 52 months compared with 15 months for patients being observed only, and this was statistically significant (p < 0.0001). When CHOP plus rituximab was used for induction, the median progression-free survival for patients who received rituximab maintenance therapy was 52 months compared with 23 months for patients being observed only (p = 0.0043), a risk reduction of 46%. When CHOP only was used for induction, the corresponding figures were 42 months and 12 months, respectively (p < 0.0001), and the risk reduction was 70%. Median overall survival could not be calculated because fewer than half the patients in each group had died at last reported follow-up. For patients who received CHOP plus rituximab for induction, the adverse effects reported with a 5% higher incidence than in the control group (CHOP only) were skin problems, infections, allergies and neutropenia. During the maintenance phase, patients on rituximab experienced a 5% higher incidence of flu-like symptoms, neurological problems, infections, blood and bone marrow problems, pulmonary problems and allergies, than those who were observed only.
3.4 The German Low-Grade Lymphoma Study Group-Fludarabine, Cyclophosphamide, Mitoxantrone (GLSG-FCM) trial was an open-label study conducted in patients with indolent lymphomas. Patients (total N = 137; follicular non-Hodgkin's lymphoma N = 65) were randomised to induction with four cycles of fludarabine, cyclophosphamide and mitoxantrone (FCM) with or without rituximab. This randomisation was stopped early when patients in the FCM plus rituximab group had a statistically significant better outcome. In the maintenance period, patients (total N = 176; follicular non-Hodgkin's lymphoma N = 105) were randomised to rituximab therapy of two 4-weekly treatment blocks of rituximab at 3 and 6 months or observation only.
3.5 For induction of remission in patients with follicular non-Hodgkin's lymphomas, the combined complete and partial response rates were statistically significantly higher following combination therapy with rituximab (94% versus 70%; p = 0.011) compared with FCM alone. For patients with follicular non-Hodgkin's lymphoma treated with FCM plus rituximab at induction, the median progression-free survival was not reached, whereas median progression-free survival was 21 months in those who received induction with FCM only; this difference was statistically significant (p = 0.0139). Fewer than half the patients in each group had died at last follow up, and there was no statistically significant difference in the proportions surviving to 2 years. For patients with follicular non-Hodgkin's lymphoma who received FCM plus rituximab induction, the median time to progression was 26 months in patients under observation only, but fewer than half of those receiving maintenance rituximab therapy had progressed; this was a statistically significant difference (p = 0.035). Adverse effects with a 5% higher incidence in the rituximab maintenance therapy group compared with the observation only group were blood and bone marrow problems, infection, fever, diarrhoea, pulmonary toxicity and liver enzyme elevation.
3.6 The cost-effectiveness models submitted by the manufacturer were based on the EORTC trial. A four-arm model allowed for use of rituximab at induction compared with chemotherapy alone, with the responders in each of these arms being further randomised to rituximab maintenance or observation only. This allowed for comparison of four treatment strategies: rituximab and chemotherapy induction followed by rituximab maintenance; rituximab and chemotherapy induction followed by observation; chemotherapy induction followed by rituximab maintenance; and chemotherapy induction followed by observation. The four treatment strategies allowed for the comparison of six pairs of between-strategy comparisons. The model consisted of five states: progression-free in induction setting; progression-free in maintenance setting; progression-free not in induction/maintenance setting; progressive disease; and death. The model assumed maintenance treatment continued for 2 years or until disease progression. The time horizon for the model was 30 years and each cycle was 1 month. A two-arm model with similar structure and assumptions, comparing rituximab maintenance therapy with observation, was also submitted.
3.7 The hazard rates were taken from the trial up to 24 months and extrapolated to 30 years using a Weibull function. In order to fit the parametric model to the survival data, data from the clinical trial were limited to 1500 days because this was the point at which the survival curves flattened. The hazard for death and progression for rituximab (that is, the duration of benefit) was assumed to be equivalent to baseline risk after 5 years. Quality of life scores for the health states were derived from a study commissioned by the manufacturer. A utility of 0.805 was attached to the progression-free states and of 0.618 to the progressive disease state. Each patient was assumed to relapse every 2 years and undergo further treatment. The cost of post-protocol treatments was based on the average costs observed in the trial. Patients received an average of 5.93 cycles of maintenance rituximab. Administration costs were calculated as being equal to an outpatient visit (£86). Follow-up costs were equal to an outpatient visit every 3 months for the progression-free state and an outpatient visit every month in the progressive disease state. The model was subjected to univariate sensitivity analysis, and a probabilistic sensitivity analysis was also conducted.
3.8 From the four-arm model, the most effective treatment intervention, using rituximab for induction and maintenance, compared with the next most effective, using rituximab for maintenance alone, gave an incremental cost-effectiveness ratio (ICER) of £16,749 per quality-adjusted life year (QALY) gained. Comparing the use of rituximab for maintenance alone with no use of rituximab gave an ICER of £9076 per QALY gained. Decreasing the duration of treatment benefit to 2 years increased the ICER for using rituximab in induction and maintenance compared with rituximab as maintenance alone to £36,497 per QALY gained. Decreasing the time horizon to 4 years increased the ICER for the comparison of the same treatment strategies to £48,116 per QALY gained. The model was not sensitive to the utility values of the health states.
3.9 The ERG considered that the manufacturer had not adequately described the methods for the systematic review of rituximab as monotherapy for induction at second or subsequent relapse (the indication appraised in TA 37) or explicitly reported on its results. However the ERG was confident that no relevant studies for this or any other indications had been missed. The ERG also confirmed that CHOP and fludarabine-containing regimens were the appropriate comparators for rituximab when used with chemotherapy for induction in relapsed follicular non-Hodgkin's lymphoma. The main clinical trial was conducted in patients who were rituximab-naive. In view of previous NICE guidance (TA 110 [Replaced by TA243]), patients in routine practice in the NHS can be expected to have received rituximab with first-line chemotherapy. The ERG considered that the effectiveness of rituximab in patients who had previously received the drug was not certain.
3.10 The ERG considered the four-arm model appropriate. The model assumed further treatment at relapse with attached costs, but it did not assume any health benefits from these treatments despite incurring the costs. The ERG noted that the best way to overcome this deficiency was to limit the gains to the observed period only. The ERG did not consider the two-arm model suitable because it did not differentiate between patients on the basis of treatment at induction.
3.11 The ERG noted that the administration of rituximab was assumed to occur in the outpatient setting and was costed accordingly. Given the duration of infusion it was considered more appropriate for this to occur in the day-case setting and the ERG calculated a cost of £504. The ERG also recalculated post-progression treatment costs by aggregating treatments into a small number of meaningful categories. This avoided the wide variation in treatment costs, some of which are very expensive, skewing the average post-progression treatment costs of individual strategies. The new categories were 'chemotherapy', 'rituximab' and 'other'. It was assumed that 25% of patients in each treatment strategy would receive other treatments, and 75% would be split between chemotherapy and rituximab treatments depending upon previous use of rituximab. The ERG also added an estimated cost of £5000 towards terminal-care costs.
3.12 The ERG considered that the choice of the Weibull model for the analysis had not been sufficiently justified. In addition, the manufacturer had assumed that the pairs of patient groups to be compared shared common parameters, estimating only three parameters rather than the four required to fit the functions independently. This approach made a proportional hazard assumption, which was not substantiated by reference to trial data. The ERG also noted the lack of an initial period of non-zero hazards for those groups that went on to be randomised at maintenance as these groups would have a protocol-driven event-free period. The ERG repeated the analyses overcoming the problems with model projections by limiting the extrapolation to use of the Kaplan-Meier estimates derived from the data to 1500 days.
3.13 The ERG exploratory analyses with changes to costs and with outcomes limited to 1500 days using the Kaplan-Meier estimates gave the following results. The single use of rituximab, as either induction or maintenance when compared with no use gave ICERs of approximately £16,000 and £13,000 per QALY gained, respectively. Dual use of rituximab at induction and maintenance compared with no use was associated with an ICER of approximately £26,000 per QALY gained. The dual-use strategy compared with single-use strategies was associated with ICERs of about £43,000 per QALY gained for the use of rituximab at maintenance only and about £42,000 per QALY gained for the use at induction only. The ERG suggested that a comprehensive probabilistic sensitivity analysis in the form of a combined cost-effectiveness acceptability probability plot would indicate which strategy would have the highest probability of being preferred across a range of willingness to pay thresholds.
3.14 The manufacturer provided additional analyses relating to the cost effectiveness of using rituximab for induction and maintenance compared with the use of rituximab for maintenance only, as requested by the Committee. All the following ICERs relate to the comparison of these two treatment strategies only. Changing the cost for administration and aggregating post-progression treatment costs as suggested by the ERG (see 3.11) and using the original four-arm model (see 3.6 and 3.7) increased the ICER to £24,161 per QALY gained. Excluding the event-free period when fitting the Weibull model (see 3.12) decreased the ICER to £21,379 per QALY gained. However, with the event-free period excluded, the exponential model gave the best fit and resulted in an ICER of £16,183 per QALY gained. When the proportional hazards assumption was relaxed and the Weibull functions were fitted independently with the event-free period excluded (see 3.12), the ICER decreased to £15,775 per QALY gained. One-way sensitivity analysis showed that these ICERs were sensitive to the assumed time horizon and duration of benefit. A time horizon of 4 years increased the ICERs to above £56,000 per QALY gained, and reducing the duration of treatment benefit to 3 years increased the ICERs to £24,000–£34,200 per QALY gained, depending on the survival model used. A probabilistic sensitivity analysis was conducted with the scenario that resulted in a deterministic ICER of £21,379, and this suggested that, at a threshold above £18,700, the use of rituximab for induction and maintenance had the greatest probability of being the most cost-effective option.
3.15 Full details of all the evidence are in the manufacturer's submission and the ERG report.