Tocilizumab for the treatment of rheumatoid arthritis

3.1 In the submission, the manufacturer presented evidence on the clinical effectiveness of tocilizumab in combination with DMARDs for two populations: people whose rheumatoid arthritis had responded inadequately to previous DMARDs but before treatment with a TNF‑alpha inhibitor (the 'DMARD‑IR' population) and people whose rheumatoid arthritis had responded inadequately to previous TNF‑alpha inhibitors but before treatment with rituximab (the 'TNF‑IR' population). The manufacturer also presented evidence on the clinical effectiveness of tocilizumab as monotherapy. The submission focused on the tocilizumab 8 mg/kg treatment arms of the included studies because this is the recommended dose in the SPC. Some of the studies also included doses other than the licensed dose. Results for doses other than the licensed dose are not considered in this appraisal.

3.16 The manufacturer did not identify any economic evaluations of tocilizumab and developed an economic model for the submission. This was an individual sampling model with a hypothetical homogenous cohort. The model used a lifetime horizon for costs and benefits. It considered the DMARD‑IR and TNF‑IR populations separately. No evidence on the cost effectiveness of tocilizumab monotherapy was presented.

3.17 The manufacturer's initial economic model compared a treatment sequence that included tocilizumab with the same treatment sequence without tocilizumab for two populations. For the DMARD‑IR population, tocilizumab plus methotrexate was the first biological treatment and if the condition did not respond or if the ACR20 response rate was no longer achieved then etanercept plus methotrexate was the next treatment. This was followed by rituximab plus methotrexate, then leflunomide, then gold, then ciclosporin until people withdrew from the last treatment (ciclosporin) and moved on to palliative care. The sequence was the same for the comparator arm, but excluded tocilizumab plus methotrexate at the beginning. For the TNF‑IR population, the sequence was the same as the DMARD‑IR population, except for the omission of etanercept plus methotrexate (that is, the first treatment in the comparator arm was rituximab plus methotrexate).

3.18 The probabilities of response were derived from the adjusted ACR response rates (adjusted for placebo differences across trials) from the base-case mixed treatment comparison. There were four categories of response: non-response, ACR20 response, ACR50 response, and ACR70 response. People were assigned a predefined drop in HAQ score (that is, an improvement in physical function) based on their ACR responses. Data from four RCTs (OPTION, TOWARD, LITHE and RADIATE) were analysed to estimate the relationship between ACR response and HAQ score in the first 24 weeks. People whose condition responded were assumed to have a constant probability of withdrawal owing to lack of efficacy. The probability of withdrawing from treatment was the same for the biological treatments (infliximab, etanercept, adalimumab, rituximab and tocilizumab) and was calculated as the average of two withdrawal rate estimates for etanercept and infliximab. At the point of switching to the next treatment, people were assumed to experience an increase in their HAQ score (rebound) equal to the initial HAQ improvement. After the initial 24‑week period the HAQ score with tocilizumab plus methotrexate was assumed to decrease linearly (improve) based on the observational extensions to the RCTs. Because of substantial uncertainty in the data for weeks 132–156, this continued improvement was only assumed for the first 3 years in the DMARD-IR cohort and 2.5 years in the TNF-IR cohort. Beyond this (3 years after initial treatment in the DMARD-IR cohort and 2.5 years after initial treatment in the TNF-IR cohort), the HAQ score was assumed to stay constant (that is, zero HAQ improvement) with tocilizumab plus methotrexate treatment. After the initial 24‑week treatment period, no change in HAQ score was assumed (zero HAQ improvement) for biological treatments such as etanercept and rituximab. After the initial 24‑week treatment period, an increase in HAQ score (that is, a worsening of physical function) was assumed for traditional DMARDs. The manufacturer also carried out sensitivity analyses using an assumption of zero HAQ progression (no improvement or worsening) while on treatment.

3.19 Tocilizumab plus methotrexate was assumed to be given for a minimum of 6 months and the administration cost of each infusion of tocilizumab was assumed to be £142 (see section 3.25 for subsequent considerations of administration costs). The costs of treating any adverse events were not included in the economic model presented by the manufacturer. The manufacturer reported that EQ‑5D scores from the tocilizumab OPTION and LITHE trials were mapped to HAQ scores using a quadratic regression model. Alternative mapping equations as used in the guidance on adalimumab, etanercept and infliximab for rheumatoid arthritis (NICE technology appraisal guidance 130) and other submissions to NICE were examined in scenario analyses. Utility weights were derived from the EQ‑5D scores using the UK time trade-off tariff. Adverse events associated with tocilizumab treatment were assumed to generate an insignificant burden on people's quality of life, and therefore were not included in the model.

3.20 For the DMARD‑IR population, the treatment sequence including tocilizumab plus methotrexate compared with the sequence without tocilizumab produced incremental costs of £23,253 and incremental quality-adjusted life years (QALYs) of 1.17. This resulted in a base-case incremental cost-effectiveness ratio (ICER) of £19,870 per QALY gained. For the TNF‑IR population, the treatment sequence including tocilizumab plus methotrexate compared with the sequence without tocilizumab produced incremental costs of £26,640 and incremental QALYs of 1.21. This resulted in a base-case ICER of £22,003 per QALY gained. Probabilistic sensitivity analyses suggested that the addition of tocilizumab and methotrexate to the treatment sequences had a 56.4% and 22.4% probability of being cost effective (for the DMARD‑IR and TNF‑IR populations respectively) if the maximum acceptable amount to pay for a QALY gained is £20,000. All scenario analyses presented by the manufacturer resulted in ICERs of less than £30,000 per QALY gained. The ICERs increased to £24,905 and £24,739 per QALY gained for the DMARD‑IR and TNF‑IR populations respectively, using an assumption of no change in HAQ score (that is, no continued improvement on tocilizumab after the initial ACR response).

3.21 The ERG highlighted the following key areas of concern with the manufacturer's submission.

3.22 The ERG explored the combined adjusted ACR response rates for TNF‑alpha inhibitors used in the mixed treatment comparison (DMARD‑IR population) and considered that etanercept appeared less efficacious in the comparison than the literature suggested. The ERG commented that the reason for the apparent low efficacy of etanercept compared with both tocilizumab and the other TNF‑alpha inhibitors was a single large trial with a very high response rate in the placebo arm (the Klareskog trial). The ERG noted that this trial only included people who were likely to benefit from methotrexate and had an aggressive dosing schedule of methotrexate if the signs and symptoms of rheumatoid arthritis reappeared. When the ERG removed the Klareskog trial from the analysis, etanercept appeared more efficacious than tocilizumab and all the other treatments in the comparison. The ERG then questioned the validity of assuming that all TNF‑alpha inhibitors had the same efficacy in the model, because this lowered the estimate of the effectiveness of the TNF‑alpha inhibitor used in the model.

3.23 The ERG commented that the follow-up period of 24 weeks in the five included tocilizumab studies could be considered too short. It noted that the longer-term data on tocilizumab came from single-arm studies with no comparator of placebo, conventional DMARDs or biological agents, so the long-term effectiveness of tocilizumab was unclear. The manufacturer estimated the medium-term HAQ progression (up to 3 years for the DMARD‑IR population and 2.5 years for the TNF‑IR population) using linear functions. However, the ERG suggested that an exponential function was equally plausible. The ERG noted that any functions fitted to the data needed to be constructed carefully because even small changes to the predictions would have a significant impact on the ICER.

3.24 The ERG was also concerned about the way the relationship between HAQ and EQ‑5D was modelled. The manufacturer's submission used a quadratic equation for this. The quadratic model predicted that EQ‑5D scores would be lower at high HAQ scores compared with a linear model. In addition, literature has shown that EQ‑5D and HAQ are closely correlated at baseline and that when quality of life worsened over time the EQ‑5D became more variable (resulting in a weaker correlation). The ERG noted that the modelled relationship between HAQ and EQ‑5D scores resulted in negative utilities for health states (that is, health states that are considered to be worse than death). The ERG stated that using negative utility values is questionable because a certain amount of disability (because of irreversible characteristics such as damaged joints) may remain despite optimal control of inflammatory disease. The ERG concluded that algorithms for modelling the relationship between HAQ and EQ‑5D should only be used when there are no direct utility scores; however, the trials for tocilizumab (OPTION and LITHE) measured EQ‑5D directly.

3.25 The manufacturer assumed the cost of administering each infusion of tocilizumab was £142. This was derived by adjusting for inflation the cost of an infusion as used in the guidance on adalimumab, etanercept and infliximab for rheumatoid arthritis (NICE technology appraisal guidance 130). However, the ERG commented that this cost should have been adjusted for inflation from 2001 and not from 2004 as was presented by the manufacturer.

3.26 The manufacturer's submission assumed that the rebound after withdrawal from treatment was equal to the initial HAQ improvement only. The manufacturer's submission also assumed that the HAQ score for people treated with tocilizumab improved over the course of treatment, but that for other treatments the HAQ score either remained the same (biological treatments) or worsened (conventional DMARDs and palliative care). Therefore, it was assumed that the short- to medium-term HAQ benefit was retained in the long term. The guidance on adalimumab, etanercept and infliximab for rheumatoid arthritis (NICE technology appraisal guidance 130) accepted a similar assumption that people would lose their initial HAQ improvement when treatment was withdrawn, and also that biological treatments delayed disease progression more than conventional DMARDs. However, whereas the HAQ score representing underlying disease progression for all biological treatments in the guidance on adalimumab, etanercept and infliximab for rheumatoid arthritis (NICE technology appraisal guidance 130) remained the same or worsened only slightly while on treatment, the manufacturer assumed that HAQ score improvement was possible for tocilizumab only. The ERG commented that the assumptions about rebound effect and HAQ progression disproportionately favoured tocilizumab by not only allowing the drug to delay disease progression, but also by allowing for a lasting improvement of the condition.

3.27 In addition, the ERG considered that excluding adverse events in the manufacturer's model was questionable because biological treatments are known to be associated with adverse events. It reported that it was unclear whether the adverse-event rate is higher or lower for tocilizumab than for other biological treatments. The manufacturer's submission states that the mean and median duration of treatment with tocilizumab in the clinical trials was 1.08 years. The ERG commented that the risks of longer-term treatment with tocilizumab were unknown.

3.28 In response to three rounds of consultation for the original guidance on tocilizumab for rheumatoid arthritis (NICE technology appraisal guidance 198), the manufacturer presented revised ICERs for the DMARD‑IR and TNF‑IR populations incorporating some of the ERG's suggested changes. The manufacturer also provided ICERs for positioning tocilizumab after an inadequate response to rituximab, and tocilizumab for people who are intolerant to rituximab or for whom rituximab is contraindicated. The clinical-effectiveness data for tocilizumab used in these positions were taken from the RADIATE trial. All of the revised and new ICERs incorporated degraded ACR response rates for tocilizumab, etanercept and rituximab when they are used later in the treatment sequence. Estimates for etanercept were based on treatment response to a second or third TNF‑alpha inhibitor reported from the South Swedish Arthritis Treatment Group. These downgraded the efficacy of etanercept from 62%, 38% and 16% to 49%, 26% and 7% for ACR20, ACR50 and ACR70 response rates respectively when used after one biological treatment. For tocilizumab when used after two biological treatments, degraded rates were based on the subgroup of people from the RADIATE trial whose rheumatoid arthritis had responded inadequately to more than one TNF‑alpha inhibitor. Based on these data, tocilizumab response rates changed from 62%, 31% and 12% to 50%, 31% and 15% for ACR20, ACR50 and ACR70 response rates respectively. For rituximab used after two biological treatments, the manufacturer provided downgraded response rates based on a subgroup of people whose rheumatoid arthritis had responded inadequately to more than one TNF‑alpha inhibitor from a trial comparing rituximab plus methotrexate with placebo plus methotrexate (REFLEX). Based on these data, the rituximab response rates were downgraded from 46%, 23% and 14% to 42%, 22% and 10% respectively.

3.29 The revised ICERs were based on the adjusted ACR rates from the mixed treatment comparison, and included a long-term HAQ improvement for tocilizumab and a stable HAQ score (that is, zero HAQ progression) for all other biological treatments. This was not the case for the ICER for tocilizumab given after rituximab, for which no HAQ improvement for treatment with any biological treatment, including tocilizumab, was assumed. All of the revised ICERs were calculated using the HAQ to EQ‑5D mapping and included negative utilities that represented states worse than death. The ICERs were subject to the assumption that a person would experience the same adverse events during treatment as during palliative care, and that the cost of administration of tocilizumab was £154.

3.30 The manufacturer's revised ICER for the DMARD‑IR population increased from £19,870 to £21,733 per QALY gained and increased from £22,003 to £23,285 per QALY gained for the TNF‑IR population. The ICER for tocilizumab used after rituximab was £23,735 per QALY gained. The ICER for tocilizumab for people who are intolerant to rituximab or for whom rituximab is contraindicated was £20,242 per QALY gained.

3.31 In 2010, the DSU was asked to undertake additional cost-effectiveness analyses to validate the manufacturer's ICERs submitted following the third round of consultation, and to conduct sensitivity analyses to address the Appraisal Committee's concerns about key parameter assumptions. The 2010 report highlighted a key issue with the calculation of the ICERs presented by the manufacturer. This concerned the 'pair-wise' calculation of sequences containing tocilizumab plus methotrexate with the same sequence excluding tocilizumab rather than an 'incremental' comparison of all strategies containing tocilizumab plus methotrexate with each other and with a base-case strategy without tocilizumab. The DSU considered that the incremental approach was the most appropriate, not only to determine whether tocilizumab plus methotrexate was cost effective, but also in what circumstances, given the availability of a number of other treatments that are used sequentially. The DSU's 2010 report explained that an ICER calculated through a pair-wise comparison does not demonstrate that the sequence can be considered cost effective because there are a series of mutually exclusive sequences available and only one can be selected at any one time.

3.32 For etanercept, the mixed treatment comparison analysis combined all TNF‑alpha inhibitors (etanercept, infliximab and adalimumab) but excluded the Klareskog trial of etanercept that the Committee had requested to be removed because of its unusually high placebo response rate. The DSU noted in the 2010 report that the adjusted mixed treatment comparison rates were lower than the unadjusted trial ACR, or point estimate, rates for etanercept. The adjusted etanercept ACR20, ACR50 and ACR70 response rates were 62%, 38% and 16% respectively and the unadjusted ACR20, ACR50 and ACR70 response rates were 71%, 39% and 17% respectively. In 2010 the DSU reported that the unadjusted rates in the model were taken from a single etanercept trial, without justification for the sole use of this particular trial. The DSU provided an alternative set of unadjusted response rates for etanercept, which were based on the two etanercept trials from the mixed treatment comparison (rather than the single trial chosen by the manufacturer). The DSU stated in the 2010 report that this appeared to represent the most robust data. The resulting unadjusted ACR response rates were 73%, 47% and 22% for ACR20, ACR50 and ACR70 respectively. For rituximab, the adjusted mixed treatment comparison ACR response rates were also lower than the unadjusted ACR trial response rates. The percentage of people reaching an ACR20, ACR50 and ACR70 response rate was 51%, 27% and 12% respectively in the unadjusted analysis and 46%, 23% and 14% respectively in the adjusted analysis. The unadjusted data were taken from the REFLEX trial.

3.33 The DSU highlighted in the 2010 report that the opposite effect was observed with the adjusted and unadjusted ACR rates for tocilizumab, that is, the adjusted rates from the mixed treatment comparison were higher than the unadjusted rates. For tocilizumab given as the first biological treatment in the sequence, the adjusted rates were 63%, 41% and 26% for ACR20, ACR50 and ACR70 response rates respectively and the unadjusted rates for tocilizumab, which were based on a separate meta-analysis of OPTION, TOWARD and LITHE (submitted as part of the manufacturer's licence application), were 59%, 37% and 19% respectively. For tocilizumab used as the second biological treatment in a sequence (that is, after a TNF‑alpha inhibitor), the mixed treatment comparison had the same effect of increasing the tocilizumab ACR response rates. The adjusted rates were 62%, 31% and 12%, whereas the unadjusted rates were 50%, 29% and 12% for ACR20, ACR50 and ACR70 response rates respectively. The unadjusted rates for tocilizumab used as the second biological treatment in the sequence were taken from the RADIATE trial.

3.34 The DSU also commented on the degradation rates provided by the manufacturer. These rates were all from single data sources, without justification given for the selection of the sources. The DSU highlighted that the degraded response rates for etanercept were based on the reported ACR rates for the TNF‑alpha inhibitors as a group and may not have been generalisable to etanercept. The DSU also noted that the degraded ACR70 response rate for tocilizumab used after two biological treatments assumed by the manufacturer (15%) was marginally better than when used after a single biological treatment (12%). The DSU stated that this appeared to be counterintuitive and that it would be more appropriate to assume the same ACR70 response rate when tocilizumab is given after two biological treatments as for when it is given after one.

3.35 In the 2010 report the DSU considered four separate approaches that varied the ACR response rates and degradation rates used to calculate the incremental ICERs (approaches to evidence synthesis).

3.36 For each of the four approaches to evidence synthesis, the DSU undertook four sets of sensitivity analyses to assess the robustness of the ICER results to other key parameter assumptions in the 2010 report. These were:

3.37 The DSU in the 2010 report calculated the incremental ICERs for each approach using the four sensitivity analyses and presented the incremental results separately for each of the 16 possible analyses. In each incremental analysis, the treatment strategies compared with each other were:

3.38 For all treatment strategies, the calculation of the ICER included the costs and QALYs associated with treatment with conventional DMARDs and palliative care at the end of the sequence. All treatment strategies were in combination with methotrexate.

3.39 Using the threshold for cost effectiveness (£30,000 per QALY gained), the results of the fully incremental analysis undertaken by the DSU in the 2010 report indicated that using tocilizumab as a first-line treatment before etanercept would not be cost effective for any approach and with any set of parameter assumptions (including the manufacturer's base-case assumptions). Using tocilizumab as a second-line treatment before rituximab would only be cost effective if it is assumed that tocilizumab has long-term HAQ improvement and there is no HAQ improvement assumed with other biological treatments. However, if tocilizumab has zero HAQ improvement, then tocilizumab would only be cost effective when used as a third-line treatment after rituximab. If tocilizumab has zero HAQ improvement and the administration costs of tocilizumab are doubled, then tocilizumab is never cost effective (that is, standard care is the most cost-effective sequence). For people who have an intolerance to rituximab, or for whom rituximab is contraindicated, adding tocilizumab to the current standard care is cost effective. However, if tocilizumab does not have a different effect on long-term HAQ and the administration costs of tocilizumab are doubled, then the current standard care would be more cost effective for this population.

3.40 In the Appraisal Committee's original guidance on tocilizumab for rheumatoid arthritis (NICE technology appraisal guidance 198) tocilizumab plus methotrexate was recommended for the treatment of rheumatoid arthritis that has not responded adequately to one or more TNF‑alpha inhibitors or to rituximab, or in whom rituximab is contraindicated or is withdrawn because of an adverse effect. Following publication of this guidance, the manufacturer submitted a patient access scheme in which a discount was applied to all indications for tocilizumab (see section 2.4) to be considered as a rapid review of the original guidance.

3.41 As part of the rapid review, the manufacturer did not submit any additional clinical-effectiveness data. However, the manufacturer did clarify the ACR and non-response rates for each drug for each position in the treatment sequences. This highlighted that when tocilizumab is the first biological treatment in the sequence, the non-response rate is approximately 40% compared with 27% when etanercept is the first biological treatment in the sequence.

3.42 The manufacturer submitted revised ICERs using the assumptions that the Committee agreed at the final Committee meeting before issuing NICE technology appraisal guidance 198, which included:

3.43 The manufacturer presented the results of an incremental analysis for the DMARD‑IR population in which the following treatment sequences were included:

3.44 The manufacturer was requested to include an additional baseline treatment sequence of tocilizumab, followed by etanercept. In this analysis the ICER for tocilizumab as the first treatment in the sequence was £5716 per QALY gained. As the second treatment in the sequence it was £30,716 per QALY gained, and as the third treatment in the sequence the ICER was £8134 per QALY gained. All ICERs incorporated the discount for tocilizumab agreed as part of the patient access scheme.

3.45 The manufacturer also responded to a request from the DSU as part of this rapid review to provide ICERs for the TNF‑IR population in which the following treatment sequences were included:

3.46 In this analysis the costs and QALYs associated with prior treatment with a TNF‑alpha inhibitor were assumed to be the same for both treatment strategies and were therefore not modelled. The ICER for the tocilizumab sequence compared with the baseline sequence (incorporating the discount for tocilizumab agreed as part of the patient access scheme) was £22,690 per QALY gained.

3.47 In 2011, the DSU was asked to undertake a review of whether the manufacturer had correctly implemented the Department of Health approved patient access scheme within their cost-effectiveness analysis. Additionally the DSU critiqued the changes to the costs of tocilizumab and ensured the Committee's agreed assumptions from the guidance on tocilizumab for rheumatoid arthritis (NICE technology appraisal guidance 198) had been used as the starting point within the economic analysis.

3.48 The DSU confirmed in the 2011 report that these conditions were met. However it raised the following issues with the manufacturer's analyses:

3.49 In 2011, the DSU reported the results of their exploratory analysis for the DMARD‑IR population, which included the same treatment sequences in an incremental analysis as those modelled by the manufacturer (see section 3.42). All ICERs incorporated the discount for tocilizumab agreed as part of the patient access scheme. In this analysis, three sequences were extendedly dominated (first: etanercept followed by rituximab; second: tocilizumab as the first treatment; third: tocilizumab as the second treatment). The ICER for tocilizumab as the third treatment in the sequence was £28,380 per QALY gained compared with £8134 per QALY gained from the manufacturer's analysis.

3.50 The DSU provided an additional exploratory analysis in the 2011 report. This was an exploratory analysis for the rituximab-intolerant DMARD-IR population. All ICERs incorporated the discount for tocilizumab agreed as part of the patient access scheme. In this analysis etanercept alone was extendedly dominated. The ICER for tocilizumab followed by etanercept compared with tocilizumab alone was £10,698 per QALY gained, and the ICER for etanercept followed by tocilizumab compared with tocilizumab followed by etanercept was £30,121 per QALY gained.

3.51 The DSU reported the results of their exploratory analysis for the TNF-IR population, which included the same treatment sequences in an incremental analysis as those modelled by the manufacturer (see section 3.43). All ICERs incorporated the discount for tocilizumab agreed as part of the patient access scheme. In this analysis, tocilizumab followed by rituximab was dominated (was less effective than and at least as costly) by rituximab followed by tocilizumab. The ICER for rituximab followed by tocilizumab was £18,527 per QALY gained compared with the manufacturer's estimate of £22,690 per QALY gained.

3.52 Full details of all the evidence are in the manufacturer's submissions, the ERG report, and the reports from the DSU, which are available from the NICE website.