Rituximab in combination with glucocorticoids for treating anti-neutrophil cytoplasmic antibody-associated vasculitis

The manufacturer's systematic review identified 2 relevant randomised controlled trials for inclusion in its submission: RAVE and RITUXVAS. Seven non-randomised controlled trials were identified but the manufacturer judged that they contained insufficient data to be useful to the decision problem. The manufacturer explained that its submission focused on efficacy data from RAVE, complemented by the RITUXVAS results. Both RAVE and RITUXVAS compared rituximab with cyclophosphamide in patients with anti-neutrophil cytoplasmic antibody (ANCA)-associated vasculitis (granulomatosis with polyangiitis or microscopic polyangiitis). RAVE recruited both newly diagnosed and relapsed patients, whereas RITUXVAS recruited newly diagnosed patients with renal involvement.

In response to consultation, the manufacturer clarified the definition of severe disease. In RAVE, severe ANCA-associated vasculitis was defined as disease activity that threatened the function of the affected organ and had the potential to cause permanent organ damage or to threaten the patient's life unless effective therapy was implemented quickly. Severe disease had previously been referred to as 'generalised', 'generalised organ-threatening' or 'organ-threatening' disease. The manufacturer stated that when RAVE began, the standard of care for inducing remission in people with severe disease was cyclophosphamide and glucocorticoids.

The manufacturer summarised the 18-month follow-up results from RAVE. The proportion of patients who achieved complete remission (BVAS/WG of 0 on a prednisone dose of 0 mg) at 6 months and who maintained complete remission at 12 and 18 months was similar in the rituximab group and the cyclophosphamide group. The rates of severe and limited flares at 6, 12, and 18 months did not differ significantly between the rituximab group and the cyclophosphamide group. Slightly more flares occurred after 6 months in the rituximab group.

The manufacturer provided remission rates after re-treatment with rituximab. In RAVE, 16 patients received a second course of rituximab, of whom 7 (44%) entered complete remission.

The manufacturer provided information about the maximum cumulative dose of cyclophosphamide and its mode of administration in UK clinical practice. The manufacturer advised that the maximum cumulative dose of cyclophosphamide is 25 g and that intravenous therapy is preferred to oral administration. These statements are consistent with the 2013 draft guidelines from the British Society for Rheumatology on the management of ANCA-associated vasculitis. The manufacturer indicated that 2 courses of intravenous therapy would equate to approximately 23 g of cyclophosphamide, based on a body weight of 78.89 kg.

The manufacturer defined a subgroup of people for whom it is desirable to avoid cyclophosphamide:

• Women who wish to preserve their fertility.

• People at a higher risk of severe infection, tuberculosis, or chronic infection such as bronchiectasis.

• People with uroepithelial malignancy or dysplasia.

• People with cytopenia or bone marrow insufficiency.

• People with cyclophosphamide allergy or intolerance.

The manufacturer provided evidence about the long-term safety of rituximab when used as a treatment for rheumatoid arthritis. A global clinical trial programme studied 3595 patients for over 11 years. The patients received up to 20 courses of rituximab. The manufacturer reported there was no evidence of an increased safety risk, an increased risk of malignancy, or increased rates of adverse events after prolonged exposure to rituximab.

The manufacturer provided evidence that rituximab does not prevent women from conceiving children. A retrospective audit in the USA identified 157 women who received rituximab for ANCA-associated vasculitis. The audit identified 7 women who wanted to have children, of whom 6 became pregnant.

The ERG noted that restricting the systematic review of clinical-effectiveness studies to the population and intervention in the marketing authorisation meant that it did not fulfil the scope or decision problem specified by NICE. The ERG did not identify any further randomised controlled trials directly comparing rituximab with the comparators in the NICE scope and decision problem in patients with ANCA-associated vasculitis. However, it did identify 5 ongoing or published trials that could potentially have enabled an indirect comparison or mixed treatment comparison of rituximab with the comparators other than cyclophosphamide that were specified in the NICE scope and decision problem.

The ERG broadly agreed with the treatment pathway described by the manufacturer but noted some uncertainties:

• A high cumulative dose of cyclophosphamide indicates increased risk of adverse events. The ERG noted that giving the drug intravenously rather than orally may offer the opportunity to reduce the cumulative dose, or allow more courses to be given. A complete course of oral cyclophosphamide (2 mg/kg/day for 6 months) would be 31 g for a patient weighing 85 kg (the mean weight in RAVE). Conversely, a complete course of intravenous cyclophosphamide (15 mg/kg × 10 over a 6-month period) for a patient weighing 85 kg would be 12.75 g. The ERG judged this method of reducing the cumulative dose of cyclophosphamide to have been inadequately explored by the manufacturer.

• The ERG believed that the manufacturer's submission did not adequately consider alternative treatments to cyclophosphamide that may be used to induce remission.

• The ERG observed that the European Vasculitis Study Group guidelines recommend maintenance treatment after remission, and received clinical specialist advice that not receiving any maintenance treatment after remission with rituximab was unrealistic. The ERG also noted that relapse is not inevitable with appropriate maintenance treatment.

• The ERG stated that a 2 × 1000 mg dosage of rituximab is used more often in UK clinical practice to treat ANCA-associated vasculitis than the 4 × 375/mg² dosage recommended in the marketing authorisation.

In the ERG's view, the evidence suggested that rituximab was superior to oral cyclophosphamide (p=0.01) in inducing remission in the subgroup of patients with relapsed severe ANCA-associated vasculitis (who had previously received at least 1 dose of cyclophosphamide, methotrexate or azathioprine) and non-inferior to oral cyclophosphamide in patients with newly diagnosed disease. The ERG also highlighted that longer-term efficacy and safety end points of rituximab in treating ANCA-associated vasculitis are unknown, and that there are some potential questions concerning effects on fertility and certain adverse events, especially rates of mortality and malignancies.

The ERG acknowledged that the manufacturer's definition of severe disease was helpful. However, the ERG noted that the clinical evidence submitted by the manufacturer is not relevant for allpeople with severe ANCA-associated vasculitis. RAVE excluded patients who needed mechanical ventilation because of alveolar haemorrhage or had a serum creatinine level greater than 4.0 mg/100 ml attributed to ANCA-associated vasculitis. RITUXVAS did include patients with more severe disease, but the treatment was rituximab plus cyclophosphamide.

The ERG reviewed the 18-month follow-up results from RAVE. The ERG advised that, for patients with relapsed disease at baseline, rituximab was superior to cyclophosphamide followed by azathioprine at 6- and 12-month follow-up, but at 18 months the difference in remission rates was not statistically significant.

The ERG noted that the estimate of remission rates after re‑treatment with rituximab provided by the manufacturer was based on small numbers of patients and was at risk of selection bias.

The ERG acknowledged the value of the 18-month safety data from RAVE, which was submitted by the manufacturer in response to consultation, but noted that these data did not indicate an advantage of rituximab compared with cyclophosphamide. The ERG accepted the relevance of long-term data in rheumatoid arthritis, which suggest that rituximab is well tolerated. The ERG also acknowledged data which indicate that rituximab allows people with ANCA-associated vasculitis to maintain fertility. The ERG advised that the duration of the RAVE study was limited and longer follow-up may be needed to evaluate the safety of rituximab plus glucocorticoids.

The manufacturer's systematic review did not identify any studies that reported on the cost effectiveness of treatment for ANCA-associated vasculitis. The manufacturer therefore submitted a de novo model, which it subsequently revised in its clarification response, evaluating the cost effectiveness of rituximab compared with cyclophosphamide in people with ANCA-associated vasculitis. In line with its marketing authorisation, the manufacturer restricted its analysis to inducing remission only and did not look at treating flares or maintenance therapy. The original base case included the population from RAVE, and subgroup analyses investigated people with newly diagnosed disease and with relapsed disease. A separate subgroup analysis estimated the cost effectiveness of rituximab in people for whom cyclophosphamide was not considered to be the standard of care (because this group was not represented in RAVE). The analysis was conducted from an NHS and personal and social services perspective. A lifetime time horizon was used and a 3.5% discount rate was adopted for health benefits and costs.

The manufacturer developed a Markov model with a similar design to that used in NICE's technology appraisal guidance on tocilizumab for the treatment of rheumatoid arthritis. It consisted of 4 different health states: non-remission, complete remission, uncontrolled disease and death. 'Complete remission' reflected treatment success as assessed in RAVE, 'non-remission' reflected non-attainment of remission and 'uncontrolled disease' reflected a state of worse health that patients enter after the simulated treatment options have been exhausted.

Patients entered the model in the non-remission health state, received induction therapy and either moved to the complete remission health state (if they went into remission) or remained in the non-remission health state (if they did not go into remission). During each 6-month cycle, moving from 1 treatment to the next in each arm's sequence was triggered either by failing to attain complete remission or by the patient eventually relapsing. After receiving all possible treatment options, patients entered the uncontrolled disease health state. The original base-case analysis was designed to compare 2 sequences of treatments:

• In the 'standard of care' sequence, patients received cyclophosphamide as induction therapy. Patients who went into remission with cyclophosphamide switched to azathioprine during remission. Patients who did not go into remission, or who relapsed, received another course of cyclophosphamide. Clinical specialist advice to the manufacturer was that a maximum of 2 courses of cyclophosphamide would be used in standard clinical practice. The manufacturer assumed that 72% of patients received cyclophosphamide intravenously, with the remainder receiving it orally.

• In the 'intervention' sequence, patients received rituximab as a first-line induction treatment. Patients who went into remission did not receive any further treatment until relapse. Patients who did not go into complete remission received a further course of rituximab (this is based on expert opinion, because RAVE did not investigate the effects of re‑treatment). Patients whose disease responded to rituximab could not have re‑treatment on relapse because this is outside the scope of the marketing authorisation. After relapse following 1 or 2 cycles of rituximab, patients received 1 course of cyclophosphamide (it was assumed that 72% of patients received cyclophosphamide intravenously, with the remainder receiving it orally).
  
  If patients received all available induction treatments in the treatment sequence and relapsed, they entered the 'uncontrolled disease' health state and received best supportive care.

The transition probabilities in the manufacturer's original base-case model were based on the primary endpoints from RAVE. A constant rate of relapse was applied in the model and it was assumed that the second course of treatment was associated with a lower probability of achieving remission than the first course. The manufacturer estimated the probability of achieving remission with the second course of treatment using RAVE results from the subgroup of patients with relapsed disease. The same probability of remission was used for re‑treatment with rituximab and with cyclophosphamide. Transition probabilities for adverse events were also based on RAVE data. Disease-specific mortality risks in the manufacturer's economic model were derived from a retrospective cohort study of UK patients with ANCA-associated vasculitis.

The costs used in the manufacturer's original economic model comprised treatment-associated costs plus health-state costs. Cost data (excluding drug costs) were largely derived from National reference costs. Drug costs were derived from the British national formulary (BNF) edition 64. Average drug costs per cycle were £4689.78 for rituximab, £99.15 for oral cyclophosphamide, £110.84 for intravenous cyclophosphamide, £44.17 for azathioprine, £28.01 for methylprednisone, £1497.96 for prednisone and £21.38 for trimethoprim. Treatment administration costs per cycle were £721.16 for rituximab and £1802.89 for intravenous cyclophosphamide, and it was assumed that these included monitoring costs. Monitoring costs for oral cyclophosphamide and azathioprine were £108. The per-cycle cost of best supportive care for patients with uncontrolled disease was £4415.73. Health-state costs were £778.10 for the remission health state and £6309.01 for the non-remission and uncontrolled disease health states.

The manufacturer's systematic review did not identify any relevant studies that reported usable utility values. Health-related quality of life data were collected in RAVE using the SF‑36 questionnaire, which was administered at baseline and at 6 months. The SF‑36 scores were converted from the non-remission and remission health states to the EQ‑5D in a post-hoc analysis using a published model (Ara and Brazier 2008) and adjusted for age. Disutility adjustments were applied for adverse events.

The manufacturer's original base-case results, provided after the request for clarification, showed that treating ANCA-associated vasculitis with rituximab increased the cost of treatment but was associated with more quality-adjusted life years (QALYs) than cyclophosphamide. The manufacturer's incremental cost-effectiveness ratio (ICER) for the comparison of rituximab with cyclophosphamide in patients with ANCA-associated vasculitis was £8544 per QALY gained (incremental costs £1391; incremental QALYs 0.1628). In its response to clarification, the manufacturer provided the results of scenario analyses, one-way deterministic sensitivity analyses, and probabilistic sensitivity analyses. These original analyses have been superseded by the manufacturer's response to consultation (see sections 3.54 to 3.62).

The ERG found that the manufacturer's economic model generally followed NICE's reference case, but noted that not all comparators had been included, and that it may have been more appropriate to consider intravenous cyclophosphamide as the primary comparator because of its lower adverse-event risk, and because its lower cumulative dose could potentially allow additional courses of treatment. The ERG described some uncertainties in the population in the manufacturer's base case. It considered the manufacturer's decision to focus on severe granulomatosis with polyangiitis and microscopic polyangiitis to be appropriate given that this is the population specified in the marketing authorisation and given the populations in RAVE and RITUXVAS. However, the ERG was aware that there is no clear definition of severe disease, and that the definition of severity used in RAVE was closer to that classified as generalised disease in treatment guidelines. The ERG also noted that RAVE excluded patients with severe renal disease and other life-threatening forms of the disease, so the clinical evidence submitted by the manufacturer did not cover the full population with severe disease. The ERG was also concerned that the manufacturer had used values for weight and body surface area that would be likely to underestimate those of the UK population with ANCA-associated vasculitis.

The ERG noted that treatment sequences depend on the patient population under consideration (for example, previous treatment with cyclophosphamide will limit its further use). Consequently, different sequences are available for newly diagnosed patients, patients with relapsed disease, and patients who cannot receive or cannot tolerate cyclophosphamide. The ERG expressed concerns about the treatment sequences used in the manufacturer's economic model:

• The ERG questioned the assumption in the manufacturer's model that all patients in the standard care group would receive 2 courses of cyclophosphamide, given that 28% of cyclophosphamide treatment was given orally, which would result in a high cumulative dose.

• The ERG had concerns about the assumption that after receiving 2 courses of cyclophosphamide, patients would receive only best supportive care.

• The ERG was unsure why rituximab was only considered as the firstinduction treatment in the manufacturer's economic model. It believed it was relevant to consider the relative cost effectiveness of rituximab used before and after cyclophosphamide in the treatment pathway. It noted that the NHS Commissioning Board recommended rituximab as first-line treatment in newly diagnosed patients only when avoiding cyclophosphamide is desirable.

• Clinical specialist advice received by the ERG suggested that it would be unlikely that patients who did not respond to an initial course of rituximab would receive a second course (because of a lack of evidence) and they would instead receive an alternative treatment.
  
  Based on clinical specialist advice, the ERG believed that the results presented by the manufacturer should be approached with considerable caution because other more appropriate treatment sequences exist, and these have not been modelled by the manufacturer.

Clinical specialist advice to the ERG suggested that it was very unlikely that patients who go into remission after treatment with rituximab would not receive subsequent maintenance therapy. The ERG noted that it would seem appropriate to assume that patients who go into remission after rituximab would then receive maintenance therapy with azathioprine or methotrexate. However, in its economic model the manufacturer did not include maintenance treatment for patients who go into remission after receiving rituximab.

The ERG had concerns about how the relapse rates used in the manufacturer's model had been derived from RAVE, and believed they had been poorly estimated. It noted that exponential model distributions were fitted to data from patients who went into complete remission at 6 months in order to estimate the time-to-event for relapse, and was aware that the manufacturer had used summary statistics rather than individual patient-level data. It noted that the Kaplan–Meier time to relapse curves for the rituximab and cyclophosphamide groups crossed, indicating that the proportional hazards assumption did not hold and that applying a constant relapse rate to each treatment group was unlikely to be appropriate. It further noted that the relapse rate for the cyclophosphamide group had potentially been overestimated. The ERG concluded that it appeared highly likely that an alternative parametric model (for example, Weibull, Gompertz, log normal or log-logistic) would have provided a better fit to the relapse data, but that these would not be suitable for use with the standard Markov model structure, so the standard Markov model may not have been an appropriate choice. The ERG was unable to assess the relative fit of the exponential models for the subgroup relapse data, and noted the manufacturer's statement that these were less precise than the all-patient data.

The ERG was aware that the manufacturer had not modelled different severities of relapse, despite the availability of data from RAVE for minor and severe flares. The ERG's clinical specialists advised that treatment options and the subsequent disease pathway depend critically upon severity of relapse. The ERG noted that the manufacturer had assumed that all relapses lead to immediate re‑treatment with cyclophosphamide or rituximab because it believed almost all minor relapses would lead to severe relapses needing re‑treatment. However, the ERG received clinical specialist advice that minor relapses may be controlled in other ways (for example, an increase in glucocorticoid dose) and that not all patients would progress to a severe relapse. The ERG anticipated that modelling severe relapse rates for the subgroups of patients with newly diagnosed or relapsed disease would be likely to be highly uncertain because of very low event numbers, and suggested it may be preferable to assume similar relapse rates in these 2 subgroups.

The ERG believed it would be more appropriate to have included a health state for non-complete remission (that is, when glucocorticoids and other less immunosuppressive treatments are still used). It considered that the failure to model different levels of treatment response and unrealistically high relapse rates may have led to patients in both treatment sequences entering the uncontrolled disease state too quickly. The ERG noted that patients in the standard of care sequence spent 70.7% of their discounted mean life expectancy in this health state, compared with 63.2% of patients in the intervention sequence. However, clinical specialist advice to the ERG suggested that it is very rare for patients with severe ANCA-associated vasculitis to be in this health state because a treatment strategy can usually be identified that offers some disease control. The ERG stated that ideally the manufacturer's model would have included additional lines of treatment, such as mycophenolate mofetil, leflunomide, azathioprine and methotrexate, in line with clinical specialist advice received by the ERG. The ERG believed that patients in the uncontrolled disease health state would have some disease control, so the health state would have a higher utility score than that assumed by the manufacturer. The ERG indicated that costs for this health state would be lower than those assumed by the manufacturer because it was unlikely patients would have outpatient appointments to receive specialist palliative care every 1.5 weeks.

The ERG described several concerns about the costs used in the manufacturer's economic model. It stated that health-state costs were the largest proportion of total costs generated by the manufacturer's economic model (93% for the cyclophosphamide group and 89% for the rituximab group in the manufacturer's base-case analysis) and noted the impact of these on the cost-effectiveness results. The ERG noted that certain costs (including some tests and the total number of outpatient appointments) were not realistic and believed that these costs were substantially overestimated by the manufacturer, creating a significant bias in favour of rituximab. The ERG also considered that the manufacturer's approach to estimating the drug costs may be biased in favour of the rituximab group (by overestimating the amount of oral cyclophosphamide used in a typical treatment course), and noted that wastage costs from part-used vials had not been included in the manufacturer's base-case analysis.

The ERG corrected several apparent technical errors in the manufacturer's economic model, which included using costs of prednisolone instead of prednisone in line with UK clinical practice. Other cost changes were for cyclophosphamide, trimethoprim and blood tests. The ERG also adjusted the utility value for pneumonia, adjusted the numbers at risk of relapse, used normal distributions for cost parameters, included distributions for standardised mortality rates and outpatient appointments in the probabilistic sensitivity analyses, and adjusted the mortality risk for patients aged 91 years and older in the uncontrolled disease health state. Cumulatively, these changes decreased the ICER for the comparison of rituximab with cyclophosphamide for all patients with ANCA-associated vasculitis. The ERG's corrected ICER was £6006 per QALY gained (incremental costs £986; incremental QALYs 0.1642) compared with the manufacturer's base-case ICER of £8544 per QALY gained (incremental costs £1391; incremental QALYs 0.1628). Replacing the cost of prednisone with the cost of prednisolone had the greatest impact.

In further exploratory analyses, the ERG altered several parameter values in the manufacturer's economic model:

• Body surface area and weight were increased to better reflect patients in RAVE.

• It was assumed that patients who went into remission after receiving rituximab would receive azathioprine maintenance treatment at the same dosage as patients who went into remission after receiving cyclophosphamide.

• Relapse rates were re‑estimated based on data from patients who had severe flares after receiving cyclophosphamide in RAVE, to reflect the assumption that only severe flares would lead to renewed induction treatment. Given the assumption that patients receiving rituximab induction treatment also received azathioprine maintenance, the same relapse rate was applied to patients in the rituximab group and patients in the cyclophosphamide group.

• Costs and utility values in the uncontrolled disease state were amended to reflect that patients in this state are likely to have some disease control.

• The number and costs of routine tests were amended to reflect recommendations in published guidelines.

• Methylprednisolone administration costs were increased.

• The costs of X-rays and CT scans were taken from NHS reference costs.

• Wastage costs were included.

• The number of outpatient appointments was reduced.
  
  When these changes in the manufacturer's economic model were added to those described in section 3.45, the ERG's cumulative ICER increased to £26,347 per QALY gained (incremental costs £5704; incremental QALYs 0.2165) for the comparison of rituximab with cyclophosphamide for the full population of patients with ANCA-associated vasculitis. The ERG noted that reducing the number of outpatient appointments (especially in the uncontrolled disease health state) substantially decreased the benefits associated with the rituximab treatment sequence.

The ERG modelled several treatment sequences that it considered to be more appropriate than those in the manufacturer's submission for the different populations (described in sections 3.48–3.52):

• the full population in the manufacturer's economic model

• patients with newly diagnosed ANCA-associated vasculitis

• patients with relapsed ANCA-associated vasculitis who could have further treatment with cyclophosphamide

• patients with relapsed ANCA-associated vasculitis who could not have further cyclophosphamide treatment

• patients who are unable to tolerate cyclophosphamide.

The ERG investigated how different treatment sequences could impact on the cost-effectiveness estimates for the full patient population with ANCA-associated vasculitis in the manufacturer's economic model:

• Adding rituximab to the treatment sequence after 2 courses of cyclophosphamide gave an ICER of £12,075 per QALY gained (incremental costs £3894; incremental QALYs 0.32).

• Using rituximab after 1 course of cyclophosphamide increased the ICER to £69,710 per QALY gained (incremental costs £355; incremental QALYs 0.0051) compared with using it after 2 courses.

• Using rituximab as first-line treatment further increased the ICER to £127,456 per QALY gained (incremental costs £579; incremental QALYs 0.0045) compared with using rituximab as second-line treatment.

• At £30,000 per QALY gained, the probability of rituximab being cost effective after 2 courses of cyclophosphamide was 58.3%. The probability that excluding rituximab from the treatment sequence was cost effective was 11.7%.

The ERG did exploratory analyses for the population with newly diagnosed ANCA-associated vasculitis:

• Adding rituximab to the treatment sequence after 2 courses of cyclophosphamide gave an ICER of £12,851 per QALY gained (incremental costs £3783; incremental QALYs 0.29).

• Using rituximab after 1 course of cyclophosphamide increased the ICER to £81,604 per QALY gained (incremental costs £364; incremental QALYs 0.0045) compared with using rituximab after 2 courses of cyclophosphamide.

• The ICER for using rituximab as a first-line treatment further increased the ICER to £317,038 per QALY gained (incremental costs £843; incremental QALYs 0.0027) compared with using rituximab as second-line treatment.
  
  At £30,000 per QALY gained, the probability that using rituximab after 2 courses of cyclophosphamide was cost effective in patients with newly diagnosed disease was 59.7%. The probability that excluding rituximab from the treatment sequence was cost effective was 13.9%.

The ERG did exploratory analyses on the population of patients with relapsed ANCA-associated vasculitis who could have further treatment with cyclophosphamide:

• Adding rituximab to the treatment sequence after 1 course of cyclophosphamide gave an ICER of £11,129 per QALY gained (incremental costs £4702; incremental QALYs 0.4225).

• The ICER for rituximab as first-line treatment was £51,842 per QALY gained (incremental costs £325; incremental QALYs 0.0063) compared with rituximab as second-line treatment.
  
  The probability of rituximab being cost effective after 1 course of cyclophosphamide was 51.3% at £30,000 per QALY gained. The probability that excluding rituximab from the treatment sequence was cost effective was 10.4%.

The ERG did exploratory analyses on the population of patients with relapsed ANCA-associated vasculitis who could not have further cyclophosphamide treatment. Using rituximab instead of best supportive care gave an ICER of £10,699 per QALY gained (incremental costs £5385; incremental QALYs 0.5033). The ERG assumed that patients who could not tolerate further cyclophosphamide treatment and were receiving best supportive care moved directly to a low-grade disease health state (with partial disease control), and explained that this assumption limited the analysis because active comparators were excluded. At £30,000 per QALY gained, the probability of rituximab being cost effective was 90.4%. The probability that excluding rituximab from the treatment sequence was cost effective was 9.6%.

The ERG did an exploratory subgroup analysis on patients who were unable to tolerate cyclophosphamide. This subgroup did not necessarily have relapsed disease, but could not take cyclophosphamide for a reason other than exceeding the maximum recommended lifetime cumulative dose. Model parameter inputs were based on the full patient population in RAVE. Using rituximab instead of best supportive care gave an ICER of £11,277 per QALY gained (incremental costs £5437; incremental QALYs 0.48). The ERG assumed that patients who could not tolerate further cyclophosphamide treatment and were receiving best supportive care moved directly to a low-grade disease health state (with partial disease control), and explained that this assumption limited the analysis because active comparators were excluded. At £30,000 per QALY gained, the probability of rituximab being cost effective in patients who cannot tolerate cyclophosphamide was 90.5%. The probability that excluding rituximab from the treatment sequence was cost effective was 9.5%.

After receiving feedback from clinical specialists on its exploratory analyses, the ERG did other scenario analyses on the data from the full patient population to further explore uncertainty associated with some parameters used in the economic model. The parameters tested were: reduced administration costs for methylprednisone and cyclophosphamide (because of shorter infusion time); substituting co-trimoxazole for trimethoprim; fewer cyclophosphamide infusions (6 instead of 10); and increased weight and body surface (to reflect the UK population with ANCA-associated vasculitis). These amendments had little cumulative impact on the ICER associated with adding rituximab to the treatment sequence after 2 courses of cyclophosphamide treatment compared with best supportive care after 2 courses of cyclophosphamide treatment, which increased slightly from £12,075 per QALY gained (ERG's base-case ICER) to £12,670 per QALY gained. However, the cumulative ICERs for using rituximab earlier in the treatment sequence increased more markedly because of reduced costs for intravenous cyclophosphamide and increased costs for rituximab (owing to higher body surface area). The ICER for using rituximab after 1 course of cyclophosphamide was £117,545 per QALY gained compared with after 2 courses of cyclophosphamide, and the ICER for using rituximab as first-line treatment was £191,013 per QALY gained compared with using it as second-line treatment. The ERG anticipated that these findings using the full patient population would be mirrored in the subgroups of patients who were newly diagnosed or had relapsed disease.

In response to consultation, the manufacturer provided 2 updated economic models; one for patients who can have cyclophosphamide and one for patients who cannot have cyclophosphamide. Both models incorporated the following changes:

• The minor technical changes proposed in section 3.45.

• The mean body surface area of patients was increased to 1.90 m² and the mean weight of patients was increased to 78.89 kg, based on data from 30 patients with vasculitis treated at Manchester Royal Infirmary.

• The model assumed that only severe relapses would be treated with induction therapy.

• The utility value in the uncontrolled disease health state was increased from 0.671 to 0.710.

• The cost of administering methylprednisolone was included. The cost was assumed to be equivalent to the cost of delivering rituximab and cyclophosphamide.

• The cost of an X-ray was updated to £18.56 and the cost of a CT scan was increased to £100.00. It was assumed that 80% of scans received in the modelled population would be X-rays and 20% would be CT scans.

• The model included wastage costs associated with drug delivery.

• There were 4 outpatient visits every 6 months in the uncontrolled disease health state.

• The model included only intravenous administration of cyclophosphamide (whereas the original model assumed 28% of patients would receive oral cyclophosphamide).

• In the uncontrolled disease health state, patients were assumed to receive the recommended dosage of mycophenolate mofetil, methotrexate, or azathioprine. The average cost of the 3 therapies was used in the model. The model assumed no difference in efficacy between treatment arms once patients entered the uncontrolled disease heath state.

• The models did not include any maintenance therapy after induction treatment with rituximab.

In the manufacturer's updated model for patients who can have cyclophosphamide, the base-case analysis was designed to compare 2 sequences of treatments:

• In the 'standard of care' sequence, patients received intravenous cyclophosphamide as induction therapy. Patients who went into remission with cyclophosphamide received azathioprine as maintenance therapy during remission. Patients who did not go into remission, or who relapsed, received a second course of intravenous cyclophosphamide.

• In the 'intervention' sequence, patients received 1 course of intravenous cyclophosphamide as induction therapy. Patients who went into remission with cyclophosphamide received azathioprine as maintenance therapy during remission. Patients who did not go into remission, or who relapsed, received 2 courses of rituximab. Patients who went into remission with rituximab did not receive any maintenance therapy.

In the manufacturer's updated model, the base-case transition probabilities were based on data from RAVE. In both the 'standard of care' sequence and the 'intervention' sequence, the probability of achieving remission with the first course of cyclophosphamide was estimated using data from the subgroup of patients in RAVE who had newly-diagnosed disease and were treated with cyclophosphamide.

• In the 'standard of care' sequence, the probability of achieving remission with the second course of cyclophosphamide was estimated using data from the subgroup of patients in RAVE who had relapsed disease and were treated with cyclophosphamide.

• In the 'intervention' sequence, the probability of achieving remission with rituximab following a course of cyclophosphamide was estimated using data from the subgroup of patients in RAVE who had relapsed disease and were treated with rituximab. The probability of achieving remission with the subsequent course of rituximab was estimated using data from the subgroup of patients in RAVE who were re-treated with rituximab.

In the manufacturer's updated base-case model, the estimate of relapse rates was based on data from patients who had severe flares after receiving cyclophosphamide in RAVE. The same relapse rate was applied to patients in the rituximab group and patients in the cyclophosphamide group. The relapse rate was assumed to be identical after subsequent lines of therapy.

The manufacturer's updated model for people with ANCA-associated vasculitis who can have cyclophosphamide produced an ICER for the comparison of rituximab with cyclophosphamide of £18,556 per QALY gained (incremental costs £6117; incremental QALYs 0.330).

The model for people with ANCA-associated vasculitis who cannot have cyclophosphamide was the same as the updated base-case model for patients who can have cyclophosphamide, except that it compared the following 2 sequences of treatments:

• In the 'standard of care' sequence, patients received a 6-month course of either mycophenolate mofetil or methotrexate. These treatments were assumed to have the same complete remission rates as cyclophosphamide. Patients who went into remission with mycophenolate mofetil or methotrexate received azathioprine as maintenance therapy during remission. The probability of relapse was assumed to be higher than that with cyclophosphamide or rituximab and was set at 0.103.

• In the 'intervention' sequence, patients received 2 courses of rituximab. Patients who went into remission with rituximab did not receive any maintenance therapy. The probability of relapse was 0.086, based on data from RAVE.

The manufacturer's updated model for people with ANCA-associated vasculitis who cannot have cyclophosphamide produced an ICER for the comparison of rituximab with mycophenolate mofetil or methotrexate of £35,003 per QALY gained (incremental costs £10,186; incremental QALYs 0.291).

The manufacturer conducted one-way sensitivity analyses to explore the effect of assumptions about key parameters on the results of the base-case model. The following changes, when implemented independently, gave ICERs that were higher than the base case: a higher relapse rate in the rituximab arm, treating both minor and severe relapses with induction therapy, reducing the number of outpatient appointments in the uncontrolled disease health state, and assuming that no patients received a second course of rituximab. When it was assumed that there was less wastage of rituximab, the ICERs were lower than the base case.

The committee had requested analyses that incorporated the costs and disutility of the cumulative long-term toxicity of cyclophosphamide. The committee had also requested analyses that incorporated the inpatient costs associated with non-remission, and separate analyses of the benefit of rituximab for patients who wished to have children. The manufacturer stated that they did not provide these analyses because of time constraints and a lack of data.

The ERG advised that the manufacturer's model submitted in response to consultation was incorrect because of several apparent errors:

• The cost of treatment in the uncontrolled disease health state was incorrectly multiplied by 4.

• There were coding errors in the sensitivity analyses that examined re-treatment with rituximab.

• The unit costs for mycophenolate mofetil and methotrexate were incorrect.

The ERG consulted clinical specialists to assess the plausibility of the treatment sequences in the manufacturer's model for patients who can have cyclophosphamide. The model assumed that only 1 course of cyclophosphamide would be provided in the rituximab arm. The ERG advised that some patients may receive a second course of cyclophosphamide even if rituximab was available. The ERG observed that, in the manufacturer's model, all patients in the rituximab arm received 2 courses of rituximab regardless of the effect of the first course of rituximab. The ERG advised that, at the first committee meeting, the committee agreed this assumption was not plausible. The ERG noted that the manufacturer had not modelled all possible treatment sequences as requested by the committee.

The ERG acknowledged there is a lack of consensus about the use of maintenance therapy after remission induced by rituximab. The 2013 draft guidelines from the British Society for Rheumatology include 4 options for maintenance therapy. These are, to wait for relapse and then re-treat, to use an immunosuppressive agent (azathioprine or methotrexate), or to use rituximab as maintenance therapy (2 rituximab regimes are described). Only 1 of the options (wait for relapse and then re-treat) was modelled by the manufacturer.

In the manufacturer's model for patients who cannot have cyclophosphamide, the 'standard of care' arm included only 1 course of mycophenolate mofetil or methotrexate. The ERG stated this may not be realistic. The ERG received clinical advice that cumulative glucocorticoid use is likely to be higher with mycophenolate mofetil or methotrexate than with rituximab, yet this was not reflected in the manufacturer's model.

In the uncontrolled disease health state, the ERG considered 3 outpatient appointments every 6 months to be a reasonable assumption. The manufacturer's model assumed 4 appointments every 6 months.

The ERG noted that in RAVE the rate of severe relapse was lower at 18 months than at 6 or 12 months in the cyclophosphamide group, but was increasing in the rituximab group. The ERG advised that it would be relevant to consider scenarios in which the relapse rate was higher in the rituximab group in the long term. The manufacturer's model did not allow relapse rates to alter over time.

In response to consultation, the manufacturer stated that there were additional QALY gains associated with rituximab because of the preservation of fertility, but these gains were not included in the economic model. The ERG advised that, in the model that included the subgroup of patients who wished to maintain fertility, the comparators were mycophenolate mofetil and methotrexate. The ERG understood that mycophenolate mofetil and methotrexate do not impair long-term fertility. Thus, in the view of the ERG, the manufacturer's argument about QALY gains was not relevant because no fertility advantage had been demonstrated for rituximab compared with mycophenolate mofetil and methotrexate.

The ERG noted that 2 additional changes, which were included as scenario analyses in the ERG's original report, had not been implemented by the manufacturer. First, cyclophosphamide can be infused more quickly than rituximab and therefore may have a lower administration cost. Second, some patients receive fewer than 10 infusions of intravenous cyclophosphamide. The ERG advised that the ICER associated with rituximab would increase if these 2 amendments were made to the manufacturer's economic model.

The ERG made the following changes to the manufacturer's model:

• The apparent errors listed in section 3.63 were amended.

• The ERG incorporated uncertainty around the remission rate after re-treatment with rituximab.

• Only patients who entered remission with rituximab were given a second course of rituximab.

The ERG ran probabilistic sensitivity analyses. The ERG also ran the following scenario analyses for the subgroups who can and cannot have cyclophosphamide:

• The rituximab arm included maintenance therapy with azathioprine.

• There was no re-treatment with rituximab.

The ERG's exploratory analyses examined the cost effectiveness of rituximab for patients who can have cyclophosphamide.

• Assuming only patients who entered remission with rituximab would be given a second course of rituximab and no maintenance treatment after rituximab, the ICER for the comparison of rituximab with cyclophosphamide was £20,879 per QALY gained (incremental costs £5075; incremental QALYs 0.24). The probability of rituximab being cost effective compared with cyclophosphamide was 40.7% at £20,000 per QALY gained and 56.7% at £30,000 per QALY gained.

• Assuming only patients who entered remission with rituximab would be given a second course of rituximab and azathioprine as maintenance treatment after rituximab, the ICER for the comparison of rituximab with cyclophosphamide was £23,444 per QALY gained (incremental costs £5698; incremental QALYs 0.24). The probability of rituximab being cost effective compared with cyclophosphamide was 34.8% at £20,000 per QALY gained and 52.8% at £30,000 per QALY gained.

• Assuming no re-treatment with rituximab and no maintenance treatment after rituximab, the ICER for the comparison of rituximab with cyclophosphamide was £20,080 per QALY gained (incremental costs £2790; incremental QALYs 0.14). The probability of rituximab being cost effective compared with cyclophosphamide was 42.0% at £20,000 per QALY gained and 53.7% at £30,000 per QALY gained.

The ERG's exploratory analyses also examined the cost effectiveness of rituximab for patients who cannot have cyclophosphamide.

• Assuming that only patients who entered remission with rituximab would be given a second course of rituximab and no maintenance treatment after rituximab, the ICER for the comparison of rituximab with mycophenolate mofetil or methotrexate was £60,569 per QALY gained (incremental costs £8345; incremental QALYs 0.14). The probability of rituximab being cost effective compared with mycophenolate mofetil or methotrexate was 13.8% at £20,000 per QALY gained and 25.3% at £30,000 per QALY gained.

• Assuming only patients who entered remission with rituximab would be given a second course of rituximab and azathioprine as maintenance treatment after rituximab, the ICER for the comparison of rituximab with mycophenolate mofetil or methotrexate was £65,700 per QALY gained (incremental costs £9052; incremental QALYs 0.14). The probability of rituximab being cost effective compared with mycophenolate mofetil or methotrexate was 10.9% at £20,000 per QALY gained and 22.4% at £30,000 per QALY gained.

• Assuming no re-treatment with rituximab and no maintenance treatment after rituximab, the ICER for the comparison of rituximab with mycophenolate mofetil or methotrexate was £118,154 per QALY gained (incremental costs £5463; incremental QALYs 0.05). The probability of rituximab being cost effective compared with mycophenolate mofetil or methotrexate was 14.7% at £20,000 per QALY gained and 23.1% at £30,000 per QALY gained.

In response to the second consultation, the manufacturer provided a weighted-average threshold analysis. The aim was to calculate an ICER for rituximab for treating the entire population of people with severely active granulomatosis with polyangiitis and microscopic polyangiitis (including both people who can and people who cannot have cyclophosphamide). For the subgroup of people who can have cyclophosphamide, the manufacturer used an ICER of £12,100 per QALY gained (see section 3.48). Based on the opinion of clinical specialists, the manufacturer assumed that 10% of patients cannot have cyclophosphamide. For the subgroup of people who cannot have cyclophosphamide, the manufacturer used a range of ICERs from £80,000 to £200,000 per QALY gained. The weighted-average ICERs for the entire population of people with severely active granulomatosis with polyangiitis and microscopic polyangiitis ranged from £18,890 to £30,890 per QALY gained.

In response to the second consultation, the ERG provided illustrative analyses based on the manufacturer's updated model for people who cannot have cyclophosphamide; this model compared 2 courses of rituximab with 1 course of mycophenolate mofetil or methotrexate (see section 3.74). The following changes were made to the 'standard of care' sequence in the model:

• The utility in the remission health state was decreased from 0.84 to 0.79.

• The cost of glucocorticoids in the remission health state was increased from £293 to £439 per 6-month treatment cycle.

• The remission rate was decreased from 0.65 to 0.52.

• Mycophenolate mofetil was the only active treatment in the 'standard of care' sequence.
  
  The changes resulted in an ICER for rituximab compared with mycophenolate mofetil of £26,406 per QALY gained.