Rheumatoid arthritis (refractory) - rituximab

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 approached the decision problem by considering the effects of rituximab within two different treatment strategies following failure of an initial TNF-α inhibitor. In the first scenario, patients receive either rituximab in combination with methotrexate or non-biological DMARDs (that is, not TNF-α inhibitors). In the second scenario, patients receive either rituximab in combination with methotrexate or an alternative TNF-α inhibitor to the one they tried initially. In both scenarios the drugs are used sequentially, with patients who do not respond to rituximab, or any other drug, moving on to the next drug in the sequence. Apart from rituximab, the sequences were the same in both treatment and comparator arms. The manufacturer specified that the population under consideration comprised adults with severe active rheumatoid arthritis who have had an inadequate response to or intolerance of other DMARDs, including one or more TNF-α inhibitors. Several outcome measures were outlined in the decision problem, including physical function, pain, mortality, quality of life and inhibition of disease progression, using measures including American College of Rheumatology (ACR) response criteria.

3.2. The manufacturer’s submission presented evidence on the clinical effectiveness of rituximab from a double-blind trial (REFLEX, n = 517) and the open-label extension at 24 weeks to the REFLEX trial (n = 281). The study evaluated the efficacy and safety of rituximab in patients with active rheumatoid arthritis who had experienced an inadequate response to at least one TNF-α inhibitor therapy, because of lack of either efficacy or tolerability. Participants were randomly assigned to either rituximab plus methotrexate or placebo plus methotrexate. The primary efficacy endpoint was a 20% improvement according to ACR response criteria (ACR20) at 24 weeks. Secondary endpoints were 50% and 70% improvements in ACR response criteria (ACR50 and ACR70 respectively), and changes in DAS28 score and EULAR response criteria, at 24 weeks. Additional endpoints included scores on the FACIT-F, health assessment questionnaire (HAQ) and SF-36 instruments, as well as Genant-modified Sharp radiographic scores at 24 weeks.

3.3. The ACR response criteria (ACR20, ACR50 and ACR70) require a specified percentage improvement (20, 50 or 70% respectively) in tender joint count, swollen joint count, global assessments, pain, disability and circulating inflammatory markers (for example, erythrocyte sedimentation rate or C-reactive protein). The disease activity score (DAS) is an alternative scoring system developed in Europe. It is calculated using a formula that includes counts for tender and swollen joints (53 and 44 joints respectively), an evaluation of general health by the patient (on a scale of 0 to 100), and a measure of circulating inflammatory markers. DAS28 is similar to DAS as described above but uses only 28 joints for assessment. A DAS28 score greater than 5.1 is considered to be indicative of high disease activity; between 5.1 and 3.2 indicates moderate disease activity; and less than 3.2 indicates low disease activity. A patient scoring less than 2.6 is defined as being in remission. The European League Against Rheumatism (EULAR) response criteria on drug treatment are based on the DAS measure. A decrease in DAS28 score of 0.6 or less is considered to show a poor response, while decreases greater than 1.2 points indicate a moderate or good response depending on whether a patient’s DAS28 score at the end point is above or below 3.2 respectively. The Stanford Health Assessment Questionnaire (HAQ) is one component of the ACR criteria and scores patients’ ability to perform daily activities from 0 (least disability) to 3 (most severe disability). The modified Sharp score is a measure of joint damage, as assessed radiographically, and is based on joint-space narrowing and erosions.

3.4. The manufacturer’s submission presented additional evidence from two randomised controlled trials, the results of which were pooled with those from the REFLEX trial to demonstrate the long-term efficacy of rituximab and to allow analysis of safety data. Both of these trials were excluded from the primary analyses because they included patients who had no prior exposure to TNF-α inhibitors and who had received unlicensed doses of rituximab.

3.5. In the REFLEX trial at 24 weeks, 51% of patients in the rituximab group reached an ACR20 response compared with 18% of patients in the placebo group. In the REFLEX trial at 48 weeks, including the open-label extension, 19% of patients in the rituximab group reached an ACR20 response compared with 4% of patients in the placebo group. At 24 and 48 weeks, all secondary efficacy outcomes, including ACR50 and ACR70 responses, were significantly different between the two groups in favour of rituximab (p < 0.002). For patients who were randomised to rituximab and were re-treated (n = 279), there was a mean time of 307 days between treatments.

3.6. The manufacturer did not identify any clinical trials that compared rituximab directly with any of the comparator drugs as specified in the decision problem, and therefore reported an indirect comparison. For the indirect comparison the manufacturer identified 44 trials, six of which were included. All of the included trials were considered to have a common comparator (methotrexate). The manufacturer calculated a weighted average of the placebo ACR response rates from all the included trials. For all treatments included in the indirect comparison, this figure was then used to adjust the specific ACR response rates for each of the included trials. The manufacturer noted the differences between the patient populations in the included trials in terms of disease severity

3.7. For the indirect comparison, efficacy values for DMARDs (including TNF-α inhibitors) adjusted by reference placebo (methotrexate, 26%) were as follows: rituximab plus methotrexate (63%); leflunomide (31%); etanercept plus methotrexate (70%); infliximab plus methotrexate (59%); adalimumab plus methotrexate (60%); abatacept plus methotrexate (60%); gold (26%); ciclosporin (26%).

3.8. The manufacturer’s submission presented an economic analysis using a microsimulation Markov model based on the REFLEX trial. All patients entered the model at the start of their next treatment option after initial TNF-α inhibitor therapy had failed. Efficacy estimates were adjusted ACR response rates from the indirect comparison. Patient disease progression was tracked within the model according to their HAQ score. Baseline HAQ scores and changes in HAQ scores relative to ACR responses were taken from the REFLEX trial. HAQ scores were transposed into quality-adjusted life years (QALYs) by using the Health Utilities Index (HUI)-3 transformation.

3.9. The base-case analysis compared rituximab with a return to non-biological DMARDs following the failure of an initial TNF-α inhibitor (scenario 1). It resulted in an incremental cost-effectiveness ratio (ICER) for rituximab of £14,690 per additional QALY. A comparison of rituximab with alternative TNF-α inhibitors used sequentially following the failure of an initial TNF-α inhibitor (scenario 2) resulted in an ICER of £11,601 per additional QALY.

3.10. Univariate sensitivity analyses of the original submitted model demonstrated that the model was most sensitive to variations in patient age (scenario 1) and the assumed time interval between treatments for those patients who respond to treatment (scenario 2). The sensitivity analysis demonstrated that the ICER for rituximab varied from £5000 to £31,500 per QALY gained.

3.11. The ERG reviewed the evidence submitted in relation to clinical and cost effectiveness. The ERG judged that the REFLEX trial was of good methodological quality. The results from the REFLEX trial at 24 and 48 weeks confirmed that rituximab plus methotrexate was more effective than placebo plus methotrexate. Given that the patients in the trial were difficult to treat and had severe disabling disease with marked impairment of quality of life, the results of the REFLEX trial were convincing for this trial population. However, the ERG stated that this evidence cannot be used directly to answer the questions raised in the manufacturer’s analysis of the decision problem. This is because rituximab was not compared directly with the stated comparator (a sequence of DMARDs including, in one of two analyses, a second and a third TNF-α inhibitor). Furthermore, the ERG questioned whether the patients in the REFLEX trial were similar enough to the patients described in the rituximab management strategies in the manufacturer’s decision problem. This is because 40% of the participants in the REFLEX trial had received at least two TNF-α inhibitors before receiving rituximab.

3.12. The ERG considered it appropriate for the manufacturer to conduct the indirect comparison to calculate efficacy rates in the economic model. The ERG stated that although the methodology used by the manufacturer was computationally sound, it questioned the validity of the adjusted ACR scores presented in the manufacturer’s submission, because:

• the inclusion and exclusion criteria for selecting the trials used in the indirect comparison were unclear
• the comparator arms of the included trials were not the same
• no clinical evidence was cited to support the equivalence of methotrexate, gold and ciclosporin
• no rationale was given for the choice of method for the indirect comparison.

3.13. The ERG reviewed the economic model and identified two particular issues of concern regarding its implementation. These were the method of randomisation and the representation of parameter uncertainty in the probabilistic sensitivity analysis. The manufacturer submitted a revised model as requested, which gave very similar results to the original model. Although the revised model did address some of the ERG’s criticisms, the ERG considered the algorithm to be very limited in its coverage of parameter uncertainty. Therefore the ERG stated that the probabilistic sensitivity analyses presented by the manufacturer (both original and revised) were unreliable aids to decision-making.

3.14. In addition, the ERG questioned the validity of the manufacturer’s assumption for long-term HAQ progression. Specifically, the ERG re-examined studies cited by the manufacturer to support long-term HAQ progression rates of 0.034 per annum while on active treatment and 0.13 per annum while receiving palliative care. The ERG concluded that the manufacturer’s analysis of the studies failed both to give an accurate representation of the quoted sources and to recognise the incompatibility of data derived from such studies. The best estimate derived from the studies by the ERG was an average HAQ progression rate for active treatment and palliative care of 0.012 per annum.

3.15. The ERG undertook further analysis to assess predominantly the impact of the alternative interpretation of the evidence on the long-term progression rates for HAQ scores. This additional analysis substantially affected the ICER results. For scenarios 1 and 2, the manufacturer’s ICERs increased from £14,694 to £40,873 and from £11,666 to £32,855 per additional QALY respectively.

3.16. The ERG also identified other key issues that could potentially influence the model results and undertook sensitivity analyses to show their impact. The ‘worse case’ amendment for graduated loss of efficacy (50% reduction in HAQ score), applied in conjunction with the analysis explained in section 3.15, increased the ICER from £40,873 to £80,198 per additional QALY for scenario 1, and from £32,855 to £65,558 per additional QALY for scenario 2. The assumption of a longer mean time between re-treatments resulted in slightly improved ICERs for rituximab (a reduction from £40,873 to £37,002 per additional QALY for scenario 1, and from £32,855 to £28,553 per additional QALY for scenario 2).

3.17. Following the request of the Committee, the manufacturer carried out analyses of four different variations to the model, as described in sections 3.17.1 to 3.17.4.

3.17.1. The first analysis looked at the effect on cost effectiveness of varying the progression rate in HAQ score. Patients receiving rituximab were assumed to have underlying HAQ progression commensurate with the general population (a constant increase of HAQ score indicating worsening functional disability of 0.03 a year). Patients receiving palliative care were assumed to have HAQ progression twice that of the general population, while those on other DMARDs had underlying HAQ progression of 0.045 a year. The manufacturer’s analysis concluded that varying the annual HAQ progression rate as requested produced an ICER for rituximab of £18,823 per additional QALY.

3.17.2. The second analysis looked at the effect on cost effectiveness of using a zero HAQ progression rate for rituximab while using the differential rates for DMARDs and palliative care, as described in section 3.17.1. The manufacturer’s analysis gave an ICER for rituximab of £12,461 per additional QALY.

3.17.3. The third analysis looked at the effect on cost effectiveness of using alternative assumptions about the possible loss of effectiveness of rituximab over time (‘rebound effect’). These alternative assumptions were applied in conjunction with the assumptions for HAQ progression rate as described in section 3.17.1. The manufacturer varied the time over which the rebound effect was considered. In this analysis, the ICER for rituximab varied between £18,823 and £21,423 per additional QALY, depending on the time period over which the loss of effectiveness was assumed to occur.

3.17.4. The fourth analysis looked at the effect on cost effectiveness of changing the definition of initial response to rituximab treatment used in the economic model to include the following.

• The definition of initial response defined as an improvement in DAS28 of 1.2 points or more. The manufacturer reanalysed the REFLEX trial data using the requested DAS28 criteria. This was carried out in conjunction with varying the HAQ progression rate as described in section 3.17.1. The manufacturer’s analysis resulted in an ICER for rituximab of £18,405 per additional QALY.
• The definition of loss of response (in terms of DAS28) for determining time to repeat treatment with rituximab and the impact of varying this definition on the cost effectiveness of rituximab. The manufacturer reported that the design of the REFLEX trial does not allow the examination of a modified mean repeat treatment impact on clinical outcomes. Therefore, the analysis undertaken by the manufacturer considered only modifications to costs. This analysis was carried out in conjunction with varying the HAQ progression rate as described in section 3.17.1. The ICER values for rituximab from the manufacturer’s DAS28 model varied between £12,214 per additional QALY when repeating treatments every 12 months and £29,810 per additional QALY when repeating treatments every 6 months.

3.18. Full details of all the evidence are in the manufacturer’s submission and ERG report, which are available from www.nice.org.uk/TA126 .