3 The manufacturer's submission
The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of rituximab and a review of this submission by the Evidence Review Group (ERG; appendix B).
3.1 The manufacturer presented information addressing the decision problem; that is, whether rituximab maintenance treatment is a clinically effective and cost-effective use of NHS resources, compared with standard management without rituximab for people with follicular lymphoma that has responded to first-line induction chemotherapy combined with rituximab. The outcomes defining effectiveness included progression-free survival, overall survival, response rates, adverse effects of treatment and health‑related quality of life.
3.2 The manufacturer undertook a systematic literature review and identified only one trial, the Primary Rituximab and Maintenance (PRIMA) trial, that met its inclusion criteria. The manufacturer presented evidence analysed after a median follow-up of 25 months (at the close of the PRIMA trial), but also provided data from two post-study observational periods with median follow-up periods of 36 and 38 months. In the absence of long-term data from the PRIMA trial, the manufacturer used data from the European Organisation for Research and Treatment of Cancer (EORTC) 20981 study to model the expected longer term outcomes that may have been experienced by participants in the PRIMA trial, had the trial gone on longer. The patient population in the EORTC 20981 study (see section 3.4), however, was different from that of the PRIMA trial.
3.3 The PRIMA trial was a phase III, open-label, multicentre, randomised trial with two treatment phases. The trial included 1193 people with previously untreated advanced follicular lymphoma with an Eastern Cooperative Oncology Group (ECOG) status of 0 or 1 and a median age of 57 years. In the first phase (induction phase or first-line treatment), participants had one of three different regimens, all of which included rituximab: R-CVP (rituximab with cyclophosphamide, vincristine and prednisolone [n = 268]), R-CHOP (rituximab with cyclophosphamide, doxorubicin, vincristine, and prednisolone [n = 881]) or R-FCM (rituximab with fludarabine, cyclophosphamide and mitoxantrone [n = 44]). People whose disease had either a partial or complete response to first-line treatment (n = 1019) entered the second phase of the trial, and were randomised to receive either rituximab maintenance treatment (n = 506) or no treatment (that is, observation; n = 513). People in the maintenance arm received 375 mg/m² rituximab intravenously: one dose every 8 weeks for 2 years, for a total of 12 doses or until disease progression, whichever occurred first. The trial was designed initially to estimate event-free survival as the primary outcome. However, during the course of the trial, the manufacturer amended the protocol in line with recommendations from regulatory authorities and changed the primary outcome to progression-free survival. Length of follow-up was increased from 5 to 7 years, and the study population was increased from 900 to 1200 participants. Secondary clinical outcomes included overall survival, event-free survival, time to next anti-lymphoma treatment, overall response rate at the end of the maintenance observation phase, the proportion of people with histological transformation at first progression, quality of life and safety. Quality-of-life data were collected in the trial using the Functional Assessment of Cancer Therapy – General (FACT-G) questionnaire and the EORTC Quality of Life Questionnaire – Core 30 (EORTC QLQ-C30). After a median follow-up of 25 months, a Data and Safety Monitoring Committee judged that the trial had met its primary objective at the pre-specified interim analysis and recommended closure of the trial. However, investigators continued to follow patients during an observational post-study period to collect longer term data.
3.4 The EORTC 20981 study was a phase III, open-label randomised trial that included people with relapsed or resistant follicular non-Hodgkin's lymphoma (n = 465) who had not previously been treated with CHOP (cyclophosphamide, doxorubicin, vincristine, and prednisolone) or R-CHOP. Instead, patients had been treated with at least 2 months of single-agent therapy (such as chlorambucil) and/or at least two consecutive cycles of combination chemotherapy (such as CVP [cyclophosphamide, vincristine and prednisolone]) or purine analogues. Therefore, these patients differed from those in the decision problem in that their disease had progressed after first-line therapy, and they had not been previously treated with R‑CHOP, R-CVP or R-FCM as in the PRIMA trial. Participants were randomised to treatment with R‑CHOP or CHOP alone after enrolment. People whose disease responded to second-line therapy (n = 334) were then randomised to either second-line maintenance treatment with 375 mg/m² rituximab (one dose every 3 months for 2 years or until relapse) or observation until relapse.
3.5 The results of the PRIMA trial showed that after 25 months' median follow-up, the risk of disease progression was halved in the rituximab maintenance arm compared with the observation arm (hazard ratio [HR] 0.50; 95% confidence interval [CI] 0.39 to 0.64; p < 0.0001) as assessed by the study investigators. This result was based on 18.4% (93 of 505) of people in the rituximab maintenance arm and 33.9% (174 of 513) of people in the observation arm having experienced an event (disease progression, relapse or death). The estimate of the hazard ratio associated with treatment was slightly higher when assessed by an independent review committee. After a median follow-up of 36 months, progression-free survival was statistically significantly improved in people randomised to the rituximab maintenance arm (PFS 74.9%; 95% CI 70.9 to 78.9) compared with those randomised to the observation arm (PFS 57.6%; 95% CI 53.2 to 62.0). The risk of disease progression was significantly reduced for people in the rituximab maintenance arm (HR 0.55; 95% CI 0.44 to 0.68; p < 0.0001). However, too few patients randomised to the rituximab maintenance arm had progressed during the study period to estimate the median time to progression. For patients randomised to observation, the median time to progression was estimated to be 48.3 months (95% CI 38.0 to not reached). After 36 and 38 months' median follow-up, a statistically significant difference in overall survival could not be established between the two arms because of the low number of deaths that had occurred.
3.6 During first-line induction therapy, most people in the PRIMA trial experienced an adverse event and 25% experienced a serious adverse event, consistent with the known safety profiles of these induction regimens. After the first-line maintenance phase of the trial, significantly more grade 3 or 4 adverse events occurred in people in the rituximab maintenance arm (24%) compared with people in the observation arm (17%; risk ratio 1.46; 95% CI 1.14 to 1.87; p = 0.0026).
3.7 In the EORTC 20981 study, second-line maintenance treatment with rituximab significantly improved progression-free survival compared with observation (median 3.7 years versus 1.3 years). Five-year overall survival was not significantly different between the arms (74% in the rituximab maintenance arm and 64% in the observation arm). Second-line maintenance treatment with rituximab was associated with statistically significant increases in grade 3 and 4 infections compared with observation.
3.8 The manufacturer also provided a summary of other studies, as supporting evidence, of the efficacy and safety of rituximab maintenance in people with previously untreated or relapsed or refractory follicular lymphoma. In the ECOG 1496 study, rituximab maintenance led to longer progression-free survival compared with observation (HR 0.4; p < 0.0001) in people who had previously received CVP as induction therapy. Interim data from another study (SAKK 35/98), which is still ongoing to compare a short course of rituximab maintenance (375 mg/m² every 2 months for a total of four doses) and prolonged rituximab maintenance (375 mg/m² every 2 months for a maximum of 5 years) with observation, demonstrated a longer event-free survival after prolonged rituximab maintenance compared with observation in people with either previously treated or previously untreated follicular lymphoma after induction with rituximab monotherapy. Preliminary safety results from this study also indicate that prolongation of maintenance therapy beyond 2 years does not lead to an obvious increase in toxicity.
3.9 The manufacturer produced a Markov economic model to estimate all costs and benefits over a lifetime resulting from the treatment of follicular lymphoma with rituximab compared with observation after first-line induction with different regimens of rituximab and chemotherapy (R-CHOP, R-CVP or R-FCM). Although listed as a comparator in the decision problem for this appraisal, ibritumomab tiuxetan was considered by the manufacturer to not be an appropriate comparator for inclusion in the economic model because of limited evidence available to support its benefits, and because data for local use suggested that it is seldom prescribed in the UK. The model had four distinct health states: progression-free survival while in the first-line maintenance phase (PF1), progression-free survival after receiving second-line induction treatment with rituximab in combination with chemotherapy (PF2), progressive disease (PD) and death. The manufacturer assumed that all people enter the economic model in the PF1 health state after successfully completing induction treatment (that is, the start of the model reflected the second phase of the PRIMA trial). The model had a cycle of 1 month and a time horizon of 25 years. A half cycle correction was applied to the model.
3.10 Data from the PRIMA trial (after 38 months' median follow-up) and the EORTC 20981 study were used by the manufacturer to estimate the transition probabilities between the health states in the economic model. To estimate median progression-free survival, which could not be estimated from the PRIMA trial directly, the manufacturer used the Gompertz function to extrapolate progression-free survival data beyond the end of the PRIMA trial. The manufacturer considered that this function provided a better fit than alternative functions. Based on the results from the PRIMA trial, the EORTC 20981 study and expert opinion, the manufacturer assumed that people in the PF1 health state retain a clinical benefit from rituximab maintenance treatment for 6 years; that is, 4 years beyond the end of treatment in the PRIMA trial. After this time, the risk of disease progression for people in the PF1 health state was assumed to be equal in both the rituximab maintenance and observation arms of the model (that is, both groups progress at the same rate after 6 years). Data from the EORTC 20981 study were used to derive the long-term outcomes, including death, for people according to the treatment they received after progressing from the PF1 health state.
3.11 In the PRIMA trial, participants did not routinely complete EQ-5D questionnaires. Instead, health-related quality-of-life data were collected in the PRIMA trial using the FACT-G and EORTC QLQ-C30 questionnaires developed to assess the quality of life of people with cancer. Overall, no differences in health-related quality-of-life data were observed between the rituximab maintenance and observation arms.
3.12 The manufacturer conducted a systematic literature review to identify studies addressing quality of life, but it considered that only one study (Pettengel et al. 2008) met the inclusion criteria. In this study, 215 adults with follicular lymphoma and an ECOG score of 0−2 completed EQ-5D questionnaires during outpatient appointments across eight sites in the UK. Patients were placed in one of five categories according to the stage of their disease: 'active disease – newly diagnosed', 'active disease – relapsed', 'partial response', 'complete response to therapy (or remission)' and 'disease free (no detectable disease)'. Mean utility values from this study were 0.88 (from 'disease free' category), 0.79 (from 'complete response to therapy' category) and 0.62 (from 'active disease – relapsed' category). These utility values were assigned to the PF1, PF2 and PD health states respectively in the manufacturer's economic model. The economic model did not include values for the disutility associated with grade 3 and 4 adverse events, or with receiving chemotherapy.
3.13 The manufacturer included costs associated with drug acquisition and administration, supportive care, management of adverse events and monitoring for each health state in the economic model. The primary sources of these costs were the BNF (edition 56 was used by the manufacturer, and edition 59 by the ERG; however, the costs were the same in both editions), the NHS Reference Cost Schedule 2008/09 and the Personal Social Services Research Unit 2009 (unit costs of health and social care). The manufacturer assumed that grade 3 and 4 adverse events incur equivalent costs, as estimated from the PRIMA and EORTC 20981 studies. Costs and benefits were discounted at 3.5% per year.
3.14 In the base-case analysis from the manufacturer's original submission, which assumed that the clinical benefit of rituximab is sustained over 6 years, the incremental cost‑effectiveness ratio (ICER) of rituximab maintenance treatment compared with observation was £15,978 per quality-adjusted life year (QALY) gained (incremental costs = £18,681; incremental QALYs = 1.169). Sensitivity analyses explored the impact of varying costs of adverse events (±50%), monthly supportive care (±50%), and administering rituximab (for example, nursing time; upper = £267, lower = £176). In sensitivity analyses, the manufacturer also tested the impact of varying the time horizon (20 years and 30 years), using other types of parametric functions to extrapolate progression-free survival data from the PRIMA trial, and assuming that people who progress from the PF1 health state die (probability of death was 100%, extreme scenario) rather than experience disease progression. From the analyses, the manufacturer concluded that the model was not sensitive to assumptions around the type of parametric extrapolation fitted to the PRIMA data, around costs of supportive care and administration, or around the time horizon. The model was sensitive to assumptions regarding the duration of treatment effect; when the manufacturer assumed that the effect of treatment stopped after 47 months (instead of after 72 months as in the base case), the ICER increased to £21,151 per QALY gained. When the manufacturer assumed that all people died after progressing from the PF1 health state (extreme scenario), the ICER decreased to £13,901 per QALY gained.
3.15 The manufacturer conducted probabilistic sensitivity analyses of all major parameters in the model except age, weight and height. The mean ICER from this analysis was £15,770 per QALY gained, and the manufacturer estimated that the probability of rituximab maintenance treatment being cost effective at £20,000 per QALY gained or less was 84.2%, and at £30,000 per QALY gained or less was 99.7%, compared with observation. The manufacturer concluded that these results demonstrated that the ICER was robust even under a wide range of variation in the model parameters.
3.16 The ERG considered that the PRIMA trial was well designed and, although it was an open-label trial, the results of progression-free survival could be considered robust because the trial used a blinded independent review committee. In the ERG's view, the rituximab chemotherapy regimens used in the induction phase of the PRIMA trial (that is, R-CHOP, R-CVP and R-FCM) were appropriate and in line with the rituximab chemotherapy regimens used in UK clinical practice. Overall, the ERG considered the results of the PRIMA trial to be generalisable to the UK setting, and that the manufacturer's decision not to include ibritumomab tiuxetan as a comparator in the economic analysis was justified.
3.17 The ERG was concerned that follow-up data were not available beyond 4 years and that the manufacturer could not estimate the median time to progression or to death by treatment group. The ERG cautioned that the data were immature (few events), which might have led the results to overestimate the clinical benefits of rituximab maintenance treatment. The ERG noted a meta-regression analysis (Bassler et al. 2010), which found large differences in the size of treatment effects between trials that were stopped early (regardless of the reason) and similar trials that ran for their originally specified time period. Using data from this study, the ERG adjusted the progression-free survival hazard ratio from the PRIMA trial to account for early reporting bias and noted that the hazard ratio increased by 30.7% from 0.55 (manufacturer's base case) to 0.719 (95% CI 0.575 to 0.889). The ERG suggested that sensitivity analyses that include the adjusted progression-free survival hazard ratio should be considered by the Committee.
3.18 The ERG noted that treatments administered after participants' disease had progressed may have affected the rate of overall survival in the PRIMA trial. The ERG stated that the post-progression treatments in the manufacturer's submission were in line with those used in UK clinical practice. However, from the data provided, the ERG was unsure whether the time at which these treatments were offered in the trial reflected the time that they would be offered in routine practice.
3.19 Although the ERG identified a number of problems with the structure and implementation of the manufacturer's model, the ERG did not expect these problems to have a major impact on the cost-effectiveness results. The ERG noted that the manufacturer did not model the disutilities associated with grade 3 and 4 adverse events. The ERG stated that this omission would favour the rituximab maintenance arm because people treated with rituxmab experience more adverse events than those not treated with rituximab (observation). The ERG was concerned that the manufacturer may have underestimated the costs of adverse events experienced in the PF2 health state because most of the people in the PRIMA trial had not progressed beyond the first-line maintenance or observation phase at the time of data analysis (up to 38 months).
3.20 The ERG noted the low proportion of patients censored (less than 3%) during the first 800 days of the PRIMA trial, and that this proportion increased greatly (70% for rituximab maintenance and 50% for observation) by 1600 days. Consequently, the ERG believed that the Kaplan–Meier estimate of progression-free survival becomes uncertain after 800 days. The ERG had concerns about the use of long-term modelling to inform the duration of treatment benefit estimated in the economic model. The ERG noted that the manufacturer used the Gompertz parametric function in their original analyses, which generated the highest overall estimate compared with other algorithms (such as an exponential function), to model progression-free survival.
3.21 The ERG was concerned about the manufacturer's use of data from the EORTC 20981 study to inform the economic model. The ERG noted that the participants in this study received different induction treatments to those in the PRIMA study (the EORTC 20981 study included people if their disease had relapsed after two previous non-anthracycline-containing chemotherapy regimens) and that only half of the patients in the EORTC 20981 study received induction treatment with a rituximab-containing regimen. Therefore, the ERG questioned whether the manufacturer could reliably use the outcomes from the EORTC 20981 study to predict future outcomes for participants in the PRIMA trial.
3.22 The ERG noted that, in the base case, the manufacturer assumed rituximab maintenance treatment had a clinical benefit for 6 years (that is, the hazard ratio from the PRIMA trial was applied for 6 years). In addition, the ERG noted that a large proportion of the gain in progression-free survival in the model arises beyond 4 years. Therefore, the ERG cautioned that if the gain in progression-free survival progressively declines, the ICER could substantially increase depending on the time period over which one assumes a difference between the treatment arms (that is, the time until the progression-free survival curves for each arm converge). The ERG also noted that the ICERs in the manufacturer's analyses were sensitive to the age at which a patient is assumed to start treatment and suggested that the manufacturer should have adjusted this variable in its sensitivity analyses.
3.23 The ERG noted that the manufacturer's model projects future benefits associated with the increased time that a person's disease remains progression free. The manufacturer's base-case modelling estimated that the mean time before a person's disease progresses is 8.64 years for the observation arm and 10.65 years for people who receive rituximab maintenance therapy; a gain of 2.01 years. The ERG noted that the manufacturer's analysis assumed that almost all of this gain in progression-free survival occurs in the PF1 health state. This implied that the majority (89.2%) of the gains in progression-free survival achieved directly by extending the first-line induction response translated into overall survival gains. The ERG cautioned that this should represent the 'best possible' scenario and would require strong supportive evidence from clinical trials before it could be accepted. The ERG explored how different conversion rates of progression-free survival gain to overall survival gain affect the estimated ICER, and predicted that at least 50% of progression-free survival gain would need to be converted into overall survival gain to achieve an ICER below £30,000 per QALY gained. The ERG further noted that if a function other than the Gompertz parametric function were used to extrapolate and convert progression-free survival into overall survival, then the conversion rate would need to be even higher for the ICER to remain below £30,000 per QALY gained.
3.24 The ERG noted that the manufacturer's model included utility values of 0.88 and 0.79 for the PF1 and PF2 health states respectively. The ERG considered that the same utility value should be used in both health states, because people in both states are in remission or have a full response. The ERG conducted a sensitivity analysis assuming that both health states have a utility value of 0.79, and noted that the QALY gain associated with maintenance treatment with rituximab dropped by more than 10% and the ICER increased by 11%.
3.25 In response to comments on the first and second appraisal consultation documents, the manufacturer provided revised cost-effectiveness analyses that modelled the effect on the ICER of different assumptions including progression-free survival translating into overall survival in a range from 50% to 100%; the clinical benefit from rituximab lasting for 28 months, 36 months or 48 months; and the mean age of a patient at induction being 62.5 years. Although the Committee requested analysis of potential utility gains associated with delaying the need for chemotherapy after relapse, the manufacturer was unable to incorporate these because of limitations in the structure of the model.
3.26 In response to the ERG's concern during the second Appraisal Committee meeting that the duration of clinical benefit (that is, the period during which rituximab is better than, rather than equal to, observation) may last only 28 months (based on the cumulative hazard plots from the PRIMA trial), the manufacturer provided an alternative method to model the rituximab treatment effect stopping at 28 months. This entailed using an exponential function (instead of the Gompertz function from the original submission) to extrapolate the hazard ratio observed in the rituximab arm for 28 months to the observation arm (HR 0.48; 95% CI 0.377 to 0.613). From 28 months onwards a HR of 1.00 was then applied to both arms in the model. The manufacturer considered that this alternative modelling approach was a more accurate method for this particular sensitivity analysis, but emphasised that it represented the worst-case clinical scenario and was not in line with available clinical evidence or expert opinion and therefore should be treated with caution.
3.27 The manufacturer noted that the cost of first-line rituximab induction therapy was included incorrectly in the original economic model and that once this had been amended, and the age at the start of treatment adjusted to 62.5 years, the revised base-case ICER (which assumed that rituximab maintenance treatment had a clinical benefit for 72 months) decreased to £15,404 per QALY gained (incremental costs = £16,918; incremental QALYs = 1.10). The manufacturer considered that the assumptions in its revised base-case analysis allowed for an undiscounted conversion rate of 89.2% from progression-free survival to overall survival. The manufacturer explained that the conversion rate is not a specific input in the model, and therefore, to analyse the effect of assuming different conversion rates, other parameters in the model had to be altered. When a conversion rate of 70%, 80% or 90% was assumed in the manufacturer's sensitivity analysis, the ICERs ranged from £17,349 to £18,615 per QALY gained when the duration of clinical benefit from rituximab maintenance was 28 months (using an exponential function); £25,038 to £27,397 per QALY gained when the duration of clinical benefit was 36 months and £21,507 to £23,355 per QALY gained when the duration of clinical benefit was 48 months (both using the Gompertz function). The manufacturer questioned the plausibility of these revised analyses because they were based on assumptions that the manufacturer considered worse than those observed in clinical practice and in the clinical trials.
3.28 The ERG provided an additional critique of the sensitivity analyses conducted by the manufacturer in response to the two appraisal consultation documents (sections 3.25 to 3.27). It noted that the manufacturer had not corrected the model for certain errors that the ERG had previously identified. After revising these errors, the ERG noted that the new base-case ICER increased slightly to £17,136 per QALY gained. The ERG considered that the structure of the manufacturer's model did not allow sufficient flexibility to enable the sensitivity analyses, which the Committee had requested from the manufacturer, to be robustly undertaken.
3.29 The ERG conducted its own exploratory sensitivity analyses of the clinical scenarios requested by the Committee; that is, assuming rituximab maintenance has a clinical benefit of 28 months, 36 months or 48 months and a conversion rate of progression-free survival to overall survival of 70%, 80% and 90%. The ERG considered that it was not possible to adjust the parameters in the manufacturer's model to assess the impact of different assumptions on the proportion of progression-free survival gain which may be expected to result in overall survival gain. Instead, the ERG adjusted the outcomes and costs generated by the model to reflect long-term outcome scenarios, and calculate the post-progression survival rate per patient. This estimate was subsequently discounted using a simple linear regression equation and used to revise the estimated overall discounted cost per patient in the model.
3.30 After adjusting the mean age of the population in the model at the start of treatment to 62.5 years, and using a hazard ratio for progression-free survival of 0.55 (in line with the manufacturer's base case), the ICERs in the ERG's analyses ranged from £24,595 to £27,558 per QALY gained, if the duration of clinical benefit was 48 months; £31,067 to £35,327 per QALY gained, if the duration of clinical benefit was 36 months; and £38,234 to £43,934 per QALY gained, if the duration of clinical benefit was assumed to be only 28 months. The ERG also presented sensitivity analyses using its revised hazard ratio of 0.719 (adjusted for early reporting bias) and noted that the ICERs ranged from £39,319 to £66,870 per QALY gained. The ERG also explored the effect on the ICER of adjusting the hazard ratios for progression-free survival for specific patient ages to reflect a reduction in clinical effect with the increase of age, however this did not change the ICER substantively.
3.31 Full details of all the evidence are in the manufacturer's submission and the ERG report; these and the responses to consultation on the first and second appraisal consultation documents (ACD) from consultees and commentators are available from www.nice.org.uk/guidance/TA226