Golimumab for the treatment of rheumatoid arthritis after the failure of previous disease-modifying anti-rheumatic drugs

Two trials with DMARD-experienced participants were included in the submission – a phase 3 randomised controlled trial (RCT) with 4 groups (GO‑FORWARD) and a phase 2 dose-ranging trial with five groups (Kay et al. 2008). The trials investigated the efficacy and the dose effect of golimumab. The manufacturer's submission focused on the groups who had received the licensed dosage of 50 mg golimumab monthly.

GO‑FORWARD was a multicentre randomised double-blind trial that compared 50 mg golimumab every 4 weeks plus methotrexate (15 mg or more every week; n=89) with placebo plus methotrexate (15 mg or more every week; n=133). The trial participants had had active rheumatoid arthritis (defined as persistent disease activity with at least 4 swollen joints and 4 tender joints) for at least 3 months and had received methotrexate for at least 3 months. The trial included a controlled phase to 24 weeks and an open-label extension to 5 years. Participants whose disease was inadequately controlled in the placebo arm could cross over to the golimumab arm at week 14. All other participants in the placebo arm crossed over to the golimumab arm at week 24. Participants whose disease was inadequately controlled on 50 mg golimumab were able to cross over to the 100 mg golimumab arm.

The primary outcome measures were the proportion of participants with an ACR20 response at 14 weeks and an improvement from baseline in the Health Assessment Questionnaire – Disability Index (HAQ-DI) score at 24 weeks. Secondary outcome measures included ACR20 response at 24 weeks, ACR50 response at 14 and 24 weeks, ACR70 response at 14 and 24 weeks, Disease Activity Score (DAS) 28 at 14 and 24 weeks and improvement from baseline HAQ‑DI score at 14 weeks. Health-related quality of life was measured using the SF‑36 tool.

A significantly greater proportion of participants who received 50 mg golimumab plus methotrexate had an ACR20 response at 14 weeks compared with participants who received placebo plus methotrexate (55.1% and 33.1% respectively; p=0.001). Improvement in HAQ-DI score at 24 weeks was significantly greater in the 50 mg golimumab plus methotrexate group compared with the placebo plus methotrexate group (median 0.375 and 0.125 respectively; p<0.001).

Following consultation on the appraisal consultation document, the manufacturer provided long-term outcomes data from 52- and 104‑week follow-up on ACR responses and on the proportion of participants maintaining a HAQ improvement greater than or equal to 0.25. These data suggested that for the people who continued to receive golimumab the response to treatment was maintained.

The manufacturer also reported that for key secondary endpoints a significantly greater proportion of participants in the 50 mg golimumab plus methotrexate group had a response compared with participants in the placebo plus methotrexate group. An ACR20 response at 24 weeks was seen in 59.6% of the participants who received 50 mg golimumab plus methotrexate compared with 27.8% of the participants who received placebo plus methotrexate (p<0.001). More participants in the 50 mg golimumab plus methotrexate group had an ACR50 response at 24 weeks than in the placebo plus methotrexate group (37.1% and 13.5% respectively; p<0.001). An ACR70 response at 24 weeks was seen in 20.2% of the 50 mg golimumab plus methotrexate group compared with 5.3% of the placebo plus methotrexate group (p<0.001).

The manufacturer submitted SF‑36 data from the GO‑FORWARD trial following consultation on the appraisal consultation document. At 24 weeks there was a statistically significant improvement in the physical component summary score in people treated with golimumab compared with placebo (mean change 8.23 and 2.54 respectively, p<0.001). There were statistically significant changes in 6 of the 8 domains, including all physical health domains, in people treated with golimumab compared with placebo. Social functioning and role–emotional domains did not show statistically significant improvements in people treated with golimumab compared with placebo. Data on golimumab from 104‑week follow-up suggested that changes in SF‑36 were maintained.

Following consultation on the appraisal consultation document, 24‑week and 52‑week radiographic progression data from the GO‑FORWARD trial were submitted. This reported no difference in the mean change from baseline in the van der Heijde modified Sharp (vdH-S) score between the 50 mg golimumab group and the placebo group (mean change 0.93 and 1.10 respectively, p=0.855). Median change was reported to be zero in both golimumab and placebo groups. Further 52‑week and 104‑week follow-up data were provided by the manufacturer. The key data were marked as academic in confidence and so cannot be reported here. The manufacturer noted a number of factors that could account for the minimal progression rates observed in both the placebo and golimumab groups in the GO‑FORWARD trial, including the use of radiographic outcomes as a secondary endpoint, the crossing over of all participants treated with placebo at week 24 and a lower baseline disease activity in the trial compared with trials of other biological therapies.

The manufacturer submitted a subgroup analysis that assessed people with moderate (DAS 28 score of between 3.2 and 5.1) and severe (DAS 28 score greater than 5.1) disease activity from the GO‑FORWARD study separately. The analysis reported relative risks for ACR20, ACR50 and ACR70 response at 24 weeks. For people with moderately active rheumatoid arthritis treated with golimumab (n=18) and placebo (n=28), the relative risks of achieving an ACR20, ACR50 and ACR70 response with golimumab compared with placebo were 2.67 (95% confidence interval [CI] 1.30 to 5.48), 1.78 (95% CI 0.78 to 4.05) and 3.89 (95% CI 0.84 to 17.95) respectively. For people with severely active rheumatoid arthritis treated with golimumab (n=71) and placebo (n=104), the relative risks of achieving an ACR20, ACR50 and ACR70 response with golimumab compared with placebo were 2.00 (95% CI 1.39 to 2.87), 3.33 (95% CI 1.75 to 6.32) and 3.81 (95% CI 1.42 to 10.21) respectively.

The manufacturer reported similar rates of adverse events at 16 weeks in the 50 mg golimumab plus methotrexate and the placebo plus methotrexate groups (68.5% and 60.9% respectively). The incidence of serious adverse events at 16 weeks was 5.6% in the 50 mg golimumab plus methotrexate group and 2.3% in the placebo plus methotrexate group. Long-term safety data were provided by the manufacturer following consultation on the appraisal consultation document. These were 52- and 104‑week safety data in trial participants with psoriatic arthritis, rheumatoid arthritis and ankylosing spondylitis who had received treatment with golimumab across all of the original phase 3 studies. These data were marked as confidential and therefore cannot be reported.

The second trial (Kay et al. 2008) was a multicentre randomised double-blind study, 2 arms of which compared 50 mg golimumab (every 4 weeks) plus methotrexate (10 mg or more every week; n=35) with placebo plus methotrexate (10 mg or more every week; n=35). The trial participants had had active rheumatoid arthritis (defined as persistent disease activity with at least 6 swollen joints and 8 tender joints) for at least 3 months and had been treated with methotrexate for at least 3 months. The primary outcome was the proportion of people who had an ACR20 response at 16 weeks. Secondary outcomes included ACR20, 50 and 70 responses over time until 52 weeks, numeric index of the ACR response at 16 weeks and DAS28 at 16 weeks.

Primary outcome data were not presented separately for the 50 mg golimumab group in the manufacturer's submission. However, they were available from a published paper, which showed that an ACR20 response at 16 weeks was seen in 60.0% of people who received 50 mg golimumab plus methotrexate and 37.1% of people who received placebo plus methotrexate.

An ACR20 response at 24 weeks was seen in 74.3% of people in the 50 mg golimumab plus methotrexate group and 45.7% of people in the placebo plus methotrexate group. More participants in the 50 mg golimumab plus methotrexate group had an ACR50 response at 24 weeks than participants in the placebo plus methotrexate group (40.0% and 11.4% respectively). An ACR70 response at 24 weeks was seen in 20.0% of participants in the 50 mg golimumab plus methotrexate group and 5.7% of those in the placebo plus methotrexate group. The ACR20 and 50 responses for the golimumab plus methotrexate group were statistically significantly different from the placebo plus methotrexate group. However, the ACR70 responses were not statistically significantly different between the treatment arms.

In the second trial (Kay et al. 2008) the proportion of participants who experienced at least 1 adverse event was slightly higher in the 50 mg golimumab plus methotrexate group than in the placebo plus methotrexate group (91.9% and 85.3% respectively).

Following consultation on the appraisal consultation document, the manufacturer provided further data from the trial by Kay et al. (2008) to support the dosage frequency used in the marketing authorisation. The manufacturer provided additional data for the 3 groups who received golimumab at unlicensed dosages not included in the submission (50 mg and 100 mg once every 2 weeks, and 100 mg once every 4 weeks). The manufacturer reported that no clear dosage–response relationship was observed, and that the lowest dosage regimen (that is, 50 mg once every 4 weeks) had an ACR response similar to that observed in the higher dosages.

The manufacturer's submission included a single phase 3 randomised double-blind placebo-controlled trial (GO‑AFTER) for the TNF inhibitor-experienced population. The trial had 3 groups and the manufacturer's submission focused on 2 of the groups: the placebo group (n=155) and the group who received 50 mg golimumab (n=153) rather than the group who received the unlicensed dose of 100 mg golimumab. The trial participants had had active rheumatoid arthritis (defined as persistent disease activity with at least 4 swollen joints and 4 tender joints) for at least 3 months and had been treated with at least 1 dose of a TNF inhibitor (etanercept, adalimumab or infliximab). People in the trial were not required to take golimumab in combination with another DMARD. Approximately 66% received golimumab in combination with methotrexate.

The primary outcome was the proportion of participants with ACR20 response at 14 weeks. The duration of follow-up was 24 weeks. The secondary outcomes included ACR50, 70 and 90 at 14 weeks, ACR20, 50, 70 and 90 at 24 weeks and change from baseline in HAQ-DI score at 24 weeks. No data were collected for SF‑36, and no data were provided for radiographic progression.

A significantly higher proportion of the participants who received 50 mg golimumab had an ACR20 response at 14 weeks compared with placebo (35.3% and 18.1% respectively; p<0.001). An ACR20 response at 24 weeks was seen in 34.0% of participants in the 50 mg golimumab group compared with 16.8% of participants in the placebo group (p<0.001). An ACR50 response at 24 weeks was seen in more participants in the 50 mg golimumab group than in the placebo group (18.3% and 5.2% respectively; p<0.001). An ACR70 response at 24 weeks was seen in 11.8% of participants in the 50 mg golimumab group and 3.2% of those in the placebo group (p=0.004). Change in HAQ-DI from baseline was assessed at 24 weeks. For the 50 mg golimumab group there was a median improvement in HAQ-DI of 0.25. For the placebo group there was no change in the median HAQ-DI score.

No major differences in the number of reported adverse events were evident in the GO-AFTER study at 24 weeks. The number of serious adverse events at 24 weeks was slightly lower in the 50 mg golimumab group than in the placebo group.

To provide support for the radiographic data from the GO‑FORWARD trial, the manufacturer also provided data from the GO‑BEFORE trial. The GO‑BEFORE trial compared methotrexate plus placebo with golimumab plus methotrexate in participants who had rheumatoid arthritis not previously treated with methotrexate. Intention-to-treat analyses reported a statistically significant difference in mean change from baseline in radiographic progression at week 52 (1.37 in the methotrexate group [n=160] and 0.74 in the 50 mg golimumab group [n=159; p=0.015]). Analyses of data from the participants who remained on their originally allocated treatment reported a change from baseline in radiographic progression at week 52 of 0.22 in the methotrexate group (n=9) and 0.06 in the 50 mg golimumab group (n=99). At week 104 the change from baseline in radiographic progression was 0.40 (n=10) and −0.10 (n=99) in each group respectively.

No head-to-head trials analysing the efficacy of golimumab compared with other active treatment options were available. Therefore, the manufacturer searched for trials of comparator interventions and completed mixed treatment and indirect comparison analyses to estimate the relative effect of golimumab versus the comparators. The manufacturer included comparators that had been recommended by NICE at the time of submission. For the DMARD-experienced population comparisons were made with placebo, adalimumab, certolizumab pegol, etanercept and infliximab. For the TNF inhibitor-experienced population comparisons were made with placebo and rituximab. Following consultation on the appraisal consultation document, the manufacturer submitted additional cost-effectiveness analyses for the comparison of golimumab, tocilizumab and abatacept. However, separate data on the relative clinical effectiveness of golimumab compared with tocilizumab and abatacept were not provided.

The ERG considered the clinical effectiveness review methods and results to be reasonably clearly presented, with adequate systematic searches conducted. The ERG stated that all the relevant RCTs for golimumab and the comparators appeared to have been included and the golimumab trials were of reasonable methodological quality. The ERG considered that the mixed treatment comparisons and indirect comparisons used appropriate trials.

The ERG commented that the populations in GO‑FORWARD and Kay et al. (2008) were generally representative of the UK population with rheumatoid arthritis, although in the GO‑FORWARD trial the proportion of people who received glucocorticoid therapy was higher than the UK average. Similarly, steroid use in the GO‑AFTER population may have been higher than the average in the UK population with rheumatoid arthritis, and in this study only 66% of the participants had also received methotrexate.

The ERG noted inconsistencies between the data presented for ACR20 and ACR50 responses in Kay et al. (2008). Different values were presented in the original study publication (week 16) and in the efficacy meta-analyses in the manufacturer's submission. The ERG was unclear how the original efficacy data from Kay et al. (2008) had been derived and handled in the meta-analyses.

The ERG commented on the complexities involved in comparing data across the interventions in the mixed treatment and indirect comparison analyses because response rates can be influenced by changes in patient populations over time. It noted that the certolizumab pegol trials had a higher ratio of ACR responses on active treatment compared with placebo, and these trials may not be comparable with the trials of other TNF inhibitors.

The ERG reviewed the additional data provided by the manufacturer. The ERG welcomed the SF‑36 data from the GO‑FORWARD trial and confirmed that SF‑36 data were not collected in the GO‑AFTER trial. The ERG reported that the radiographic progression data from the GO‑BEFORE trial appeared to suggest a reduction in progression of structural damage for participants with rheumatoid arthritis not previously treated with methotrexate. The ERG also stated that the summary of the 24‑week data from the GO‑FORWARD trial was appropriate. The ERG noted that interpretation of the longer-term data from this trial was limited by cross-over between treatments.

The manufacturer submitted 2 decision-analytic Markov models, each with a lifetime horizon. Both models evaluated golimumab as part of a sequence of treatments: 1 evaluated golimumab in a DMARD-experienced population (comparing golimumab with TNF inhibitors and methotrexate in people whose disease had had an inadequate response to 2 DMARDs) and the other evaluated golimumab in a TNF inhibitor-experienced population (comparing golimumab with rituximab and methotrexate in people whose disease had had an inadequate response to 2 DMARDs and a TNF inhibitor). All treatments were given in combination with methotrexate. Methotrexate monotherapy was included as a comparator in each model because it represented the placebo arm in the indirect and mixed treatment comparisons. The manufacturer did not include technologies being appraised by NICE at the time of its submission (tocilizumab, abatacept and the use of etanercept, infliximab and adalimumab after the failure of a first TNF inhibitor) as comparators.

On starting treatment, people could have either an ACR20 response, ACR50 response or no response. The probability of response for golimumab and methotrexate monotherapy was derived from the GO‑FORWARD and GO‑AFTER trials. To derive efficacies for the other comparators the response for golimumab was adjusted using the relative effects estimated from the mixed treatment and indirect comparison analyses. For each ACR response criterion the corresponding change in HAQ‑DI was calculated based on data from the GO‑FORWARD and GO‑AFTER trials. The HAQ‑DI was in turn mapped to EQ‑5D with an equation used in NICE's previous technology appraisal guidance on adalimumab, etanercept and infliximab (TA130, now replaced by TA375). People progressed to the next treatment if they did not have at least an ACR20 response at 6 months, or if they stopped treatment because of a lack of efficacy or an adverse event. In both models, people progressed to leflunomide, gold, azathioprine, ciclosporin and then palliative treatment.

At the start of the models, people were aged 50 years in the DMARD-experienced population and 54 years in the TNF inhibitor-experienced population. HAQ scores for people entering the model were derived from the baseline characteristics of the GO‑FORWARD and GO‑AFTER trials: 1.41 and 1.58 respectively. While people were receiving a treatment, it was assumed that their disease severity increased over time. This was modelled with an annual worsening of HAQ score (that is, an HAQ progression rate). The HAQ progression rate was 0.045 for a person being treated with DMARDs, 0.00 for TNF inhibitors, 0.045 for rituximab and 0.09 for palliative treatment.

Costs relating to treatment, administration, monitoring and hospitalisation were included in the economic models using 2006 reference costs and 2008 unit costs. Following a clarification request, the manufacturer incorporated 2008 reference costs and 2009 unit costs. It was assumed that a course of rituximab was given once every 6 months. The cost of joint replacement was not included in the model. Costs and quality-adjusted life years (QALYs) were discounted at a rate of 3.5%.

The results from the economic model were presented incrementally with all treatments compared with each other, and for each treatment individually in comparison with methotrexate. The deterministic results for the DMARD-experienced population showed that the incremental cost-effectiveness ratios (ICERs) were £31,464 (£34,030 additional costs and 1.082 additional QALYs) and £31,444 (£37,702 additional costs and 1.199 additional QALYs) per QALY gained for infliximab and certolizumab pegol respectively in comparison with methotrexate, and £25,346 (£31,878 additional costs and 1.258 additional QALYs) per QALY gained for golimumab in comparison with methotrexate. The ICERs for adalimumab and etanercept in comparison with methotrexate were £25,353 (£31,006 additional costs and 1.223 additional QALYs) and £24,514 (£38,339 additional costs and 1.564 additional QALYs) per QALY gained respectively. The incremental analysis showed that infliximab and certolizumab pegol were both dominated by golimumab because golimumab was more effective and less costly. However, adalimumab and golimumab were both extendedly dominated by etanercept. Etanercept generated the most QALYs of any strategy, at a lower cost per QALY ratio (£24,514 per QALY gained in comparison with methotrexate).

The results for the deterministic base-case analysis of golimumab in a TNF inhibitor-experienced population show that rituximab was dominated by golimumab because golimumab was less costly and more effective (£31 fewer costs and 0.189 additional QALYs). Golimumab compared with methotrexate had an ICER of £28,286 (£16,502 additional costs and 0.583 additional QALYs) per QALY gained whereas rituximab compared with methotrexate had an ICER of £41,935 (£16,533 additional costs and 0.394 additional QALYs) per QALY gained.

The ERG noted that the model results (total costs and QALYs, time in states, HAQ scores and incremental costs and QALYs) appeared plausible given the parameter inputs. It commented that the model was generally of a high quality. The ERG identified some programming errors in the model that it corrected. However, these errors did not change the conclusion in the manufacturer's submission that, compared with methotrexate, golimumab has an ICER that is comparable to other TNF inhibitors but that golimumab is never the most cost-effective TNF inhibitor treatment.

The ERG considered that it would have been appropriate to include ACR70 response data in the model so that all the available clinical evidence is used to evaluate golimumab. The manufacturer justified the exclusion of these data by stating that there was not a statistically significant difference between golimumab and the comparators and that incorporating this outcome would only add an element of uncertainty to the model inputs. The ERG noted that this reason was not justified because there was also no statistically significant difference in the ACR20 and ACR50 response data for golimumab and the comparators, but these data had been included in the model.

The ERG undertook a number of exploratory analyses to address some of its concerns. The original model used 2006 reference costs and 2008 unit costs. However, after clarification, the manufacturer incorporated 2008 reference costs and 2009 unit costs. The ERG used the updated reference and unit costs and found that they had little impact on the incremental costs for the different treatments in the DMARD-experienced population, and so the resulting ICERs did not change substantially.

The ERG identified an error in the model for infliximab in the DMARD-experienced population, which resulted in a cost being allocated when a person dies. There was also an error in the modelling of HAQ decrements for certolizumab pegol. Correcting the infliximab costs reduced the total cost of infliximab treatment, and it was no longer dominated by adalimumab. Correcting the HAQ decrements for certolizumab pegol meant that it was the optimal intervention instead of etanercept.

The economic model used the response rates from the GO‑FORWARD trial to estimate the probability of ACR response and the probability of stopping treatment because of an adverse event at 6 months in the golimumab and methotrexate groups. However, the model used the mixed treatment comparison to estimate the rates of these events for the comparators; this approach excludes the evidence from Kay et al. (2008). In the exploratory analysis the ERG used the mixed treatment comparison, incorporating the evidence from Kay et al. (2008) to estimate the probability of these outcomes in the placebo group, which is used to populate the methotrexate arm of the economic model. Using the mixed treatment comparison rather than the GO‑FORWARD study alone to inform the golimumab versus methotrexate comparison did not substantially alter the results.

The cumulative impact of the changes described in sections 3.42 to 3.44 reduced all the ICERs for all TNF inhibitors in comparison with methotrexate in the DMARD-experienced group. The ICERs for infliximab and certolizumab pegol in comparison with methotrexate were £24,137 and £20,800 per QALY gained. The ICER for golimumab compared with methotrexate was £24,794 per QALY gained. The ICERs for adalimumab and etanercept in comparison with methotrexate were £24,800 and £23,990 per QALY gained. The incremental analysis suggested that certolizumab pegol including its patient access scheme is the optimal treatment strategy, dominating etanercept and extendedly dominating golimumab, adalimumab and infliximab.

The ERG stated that for the TNF inhibitor-experienced population there was considerable uncertainty in the HAQ progression rate estimates and the re-administration frequency of rituximab. The ERG commented that the manufacturer assumed a HAQ progression rate equal to the rate for DMARDs rather than for TNF inhibitors, which may underestimate the benefit of rituximab. The ERG also commented that the model assumes that rituximab is re‑administered every 6 months but it considered that 9 months would be more reflective of current clinical practice. The ERG amended the model so that rituximab had a zero HAQ progression rate (equal to that of TNF inhibitors) rather than the 0.045 that was assumed in the base-case analysis. The ERG also amended the model so that each person received 2 infusions in the first 6 months and then 1 infusion every 9 months. The costs were updated as described for the DMARD-experienced population.

The cumulative impact of the changes described in 3.46 reduced the ICERs for golimumab and rituximab in comparison with methotrexate (£28,115 and £10,088 per QALY gained respectively). The incremental analysis showed that rituximab dominated golimumab.

Following comments received during consultation on the appraisal consultation document about the inclusion of the TEMPO study and the TNF inhibitor monotherapy studies in the base-case analysis, the ERG performed sensitivity analyses to assess the impact on the ICERs of separately excluding the monotherapy studies and the TEMPO study. In an incremental analysis, when the TEMPO study is excluded etanercept is no longer dominated by certolizumab pegol and it becomes the optimum strategy. When the TNF inhibitor monotherapy studies are excluded, the results do not differ substantially from the base case, with certolizumab pegol remaining the optimum strategy.

In response to a request from NICE, the manufacturer provided additional analyses of the cost effectiveness of golimumab. The analyses included:

• incorporation of ACR70 response data and disease progression on palliative treatment reflected as an increase in HAQ score of 0.06 per year in the economic model

• a sensitivity analysis in which SF‑36 data are included in the economic model using mapping to SF‑6D

• cost-effectiveness results for the comparison of golimumab, abatacept and tocilizumab for the group of people whose disease has responded inadequately to a TNF inhibitor.

The manufacturer did not provide any analyses that reported the estimates of cost effectiveness of including the 100 mg dose of golimumab for people weighing over 100 kg whose rheumatoid arthritis does not respond to the 50 mg dose. The manufacturer submitted a patient access scheme that would provide the 100 mg dose at the same cost as the 50 mg dose in this population. This scheme has been approved by the Department of Health.

In the resubmitted analyses (described in 3.58 and 3.59) the manufacturer corrected the internal inconsistencies previously present in the analyses. The analyses also incorporated the changes made by the ERG in response to the original submission (that is, updated unit costs and corrections to the HAQ decrements for certolizumab pegol and costs for infliximab). The analyses also included ACR70 response data and a progression rate while on palliative treatment of 0.06 HAQ score units a year.

The results from the economic model for the DMARD-experienced population were presented for each treatment in comparison with methotrexate. Including ACR70 response data in the model produced ICERs that were £21,944 and £25,825 per QALY gained for certolizumab pegol and infliximab respectively in comparison with methotrexate, and £26,996 per QALY gained for golimumab in comparison with methotrexate. The ICERs for adalimumab and etanercept in comparison with methotrexate were £25,523 and £27,157 per QALY gained respectively.

A sensitivity analysis was provided that included the SF‑36 data from the GO‑FORWARD study converted to SF‑6D. The SF‑6D scores were calculated only for ACR20 and ACR50 responses. The results showed that the ICERs were £27,413 and £29,484 per QALY gained for certolizumab pegol and infliximab respectively in comparison with methotrexate and £31,046 per QALY gained for golimumab in comparison with methotrexate. The ICERs for etanercept and adalimumab in comparison with methotrexate were £30,936 and £30,893 per QALY gained respectively.

The results from the economic model for the TNF inhibitor-experienced population were presented for golimumab, rituximab abatacept and tocilizumab in comparison with methotrexate. The analyses incorporated a progression rate while on palliative treatment of 0.06 HAQ score units a year, and zero while on treatment with either golimumab, abatacept or tocilizumab. The progression rate for rituximab was 0.045 HAQ score units per year.

The analyses produced ICERs of £35,288, £32,036 and £35,382 per QALY gained for tocilizumab, golimumab and abatacept respectively in comparison with methotrexate, and £59,328 per QALY gained for rituximab in comparison with methotrexate. One-way sensitivity analyses assuming that people on rituximab experienced no disease progression while on treatment reduced the ICER for rituximab compared with methotrexate to £24,683 per QALY gained. An alternative 1-way sensitivity analysis that assumed re-treatment with rituximab every 9 months reduced the ICER for rituximab to £28,047 per QALY gained in comparison with methotrexate.

The ERG stated that the results from the manufacturer's resubmitted analyses were consistent with the electronic models provided. The ERG confirmed that the changes reported to have been implemented by the manufacturer had been completed appropriately and that errors previously identified had been corrected. The ERG noted that the results provided were deterministic and that no incremental analyses were included.

For the DMARD-experienced population, the ERG re-ran the manufacturer's model using a probabilistic analysis and presented the results incrementally. The incremental probabilistic analysis suggested that certolizumab pegol was the most cost-effective option, with an ICER in comparison with methotrexate of £22,693 per QALY gained. All other treatments were either dominated or extendedly dominated by certolizumab pegol. For each of the other treatments in comparison with methotrexate the probabilistic analysis produced ICERs for infliximab, golimumab, adalimumab and etanercept of £25,541, £27,946, £25,951, and £27,129 per QALY gained respectively.

The ERG reviewed the sensitivity analysis provided by the manufacturer that used the SF‑36 values converted to SF‑6D. The ERG noted that the manufacturer had not directly used the SF‑6D values for the placebo and methotrexate group; rather, it had estimated the ratio between the HAQ scores from the 2 groups in the clinical trial and then applied this ratio to the SF‑6D scores for the golimumab group to obtain a value for the methotrexate group. The HAQ adjustment resulted in lower SF‑6D scores associated with ACR20, ACR50 and ACR70 responses for the methotrexate group compared with the golimumab group. The ERG stated that it was unclear why the manufacturer had chosen this method.

The ERG re-ran the manufacturer's model using a probabilistic analysis and presented the results incrementally. The incremental probabilistic analysis suggested that certolizumab pegol was the most cost-effective option with an ICER in comparison with methotrexate of £27,182 per QALY gained. All other treatments were either dominated or extendedly dominated by certolizumab pegol. For each of the other treatments in comparison with methotrexate the probabilistic analysis produced ICERs for infliximab, golimumab, adalimumab and etanercept of £28,990, £31,420, £30,129, and £30,412 per QALY gained respectively.

The ERG checked the revised model for the TNF inhibitor-experienced population and stated that the changes had been implemented appropriately. The ERG cross-checked the cost-effectiveness estimates in the model with the study papers and relevant NICE technology appraisals and reported that the values used in the model corresponded. The ERG stated that a full validation of the model was not possible. However, the model maintained internal consistency and the clinical effectiveness results matched those in previous submissions.

The ERG re-ran the manufacturer's model using a probabilistic analysis. The ICER for golimumab compared with methotrexate was £32,979 per QALY gained. The ICERs for rituximab, abatacept and tocilizumab compared with methotrexate were £68,663, £34,155, and £34,644 per QALY gained respectively. The ERG produced a sensitivity analysis that assumed that the rate of underlying disease progression while on treatment with rituximab was zero (that is, the same assumption as the other biological treatments), and that rituximab was administered every 9 months rather than every 6 months. This reduced the ICER for rituximab in comparison with methotrexate from £68,663 to £12,196 per QALY gained.

Full details of all the evidence are in the manufacturer's submissions and the ERG reports.