Rheumatoid arthritis (refractory) - abatacept

3. The manufacturer’s submission

The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of abatacept and a review of this submission by the Evidence Review Group (ERG; appendix B).

3.1. The submission from the manufacturer focused on the use of abatacept in combination with methotrexate for the treatment of rheumatoid arthritis following the failure of a TNF-α inhibitor. The manufacturer identified three possible comparators for abatacept: a return to conventional DMARDs, the use of a second TNF-α inhibitor, and the use of rituximab. Only the first two of these comparators were considered in the assessment of cost effectiveness. A number of outcomes were reported in the submission including symptom measures, measures of physical function and measures of disease activity.

3.2. The manufacturer identified six randomised controlled trials that compared abatacept with placebo, of which one reflected the licensed indication for abatacept. This study (ATTAIN) was a double-blind randomised controlled trial in patients (n = 391) for whom TNF-α inhibitor therapy had previously failed or was currently failing. At the start of the ATTAIN trial, patients had active disease despite currently receiving either conventional DMARDs or TNF-α inhibitors. Conventional DMARDs had to have been given for 3 months before enrolment, with a stable dose given for 28 days and continued throughout the study period. Abatacept or placebo was then added to the ongoing DMARD treatment.

3.3. Patients in the ATTAIN trial had a mean disease duration of approximately 12 years (standard deviation [SD] 8.5), baseline health assessment questionnaire – disability index (HAQ-DI) score of 1.8 (SD 0.6) and disease activity measure (DAS28) of 6.9 (SD 1.0). The median duration of TNF-α inhibitor therapy was approximately 8 months. Data on the number of previous conventional DMARDs used were not collected. The co-primary outcomes for the trial were ACR20 response (the proportion of patients with a 20% improvement in symptoms according to American College of Rheumatology criteria) and the percentage of patients having an initial reduction in HAQ-DI score of 0.3 or more by 6 months.

3.4. At 6-month follow-up there were statistically significant differences favouring the abatacept group over placebo across all outcomes measured in the trial. For example, 50% of patients in the abatacept group compared with 20% of patients in the placebo group achieved an ACR20 response, and 47% of patients in the abatacept group demonstrated an improvement in HAQ-DI score of 0.3 or more compared with 23% of patients in the placebo group. The mean changes in HAQ-DI score from baseline to 6 months for the abatacept and placebo groups were –0.45 and –0.11 respectively. Data from a long-term extension to the study suggested that the proportion of patients responding to abatacept was maintained at 2 years’ follow-up. Data from the supporting trials outside the marketing authorisation supported the efficacy of abatacept compared with placebo when given at a dose of 10 mg/kg.

3.5. In the absence of direct head-to-head comparisons of abatacept with either rituximab or TNF-α inhibitors, the manufacturer included indirect sources of evidence. Data for rituximab were identified from a randomised trial that compared rituximab in combination with methotrexate with placebo plus methotrexate (the REFLEX study). Results from this trial showed that 51% of patients in the rituximab group and 18% of patients in the placebo group achieved an ACR20 response. The mean changes in HAQ-DI score for the rituximab and placebo groups were –0.4 and –0.1 respectively. The manufacturer used these data in an indirect comparison, which showed no significant differences between the treatment with rituximab and the treatment with abatacept. Data for TNF-α inhibitors were taken from observational data from the British Society for Rheumatology Biologics Register (BSRBR). The manufacturer chose a value of –0.15 as reflecting the reduction in HAQ associated with receiving a second TNF-α inhibitor.

3.6. The economic model provided by the manufacturer was a patient-level simulation that estimated the cost effectiveness of abatacept in combination with methotrexate in two scenarios: the first where the comparator was methotrexate alone, and the second where the comparator was a second TNF-α inhibitor. In the model, patients starting abatacept moved to methotrexate either if there was an insufficient response to abatacept in the first 6 months or if they experienced an adverse event. Patients receiving methotrexate remained on methotrexate for the remainder of the 20-year time horizon of the model. The submitted model did not examine the optimal sequencing of abatacept in a strategy containing a range of anti-rheumatic drugs.

3.7. The economic comparison with methotrexate was based on short-term efficacy data from the ATTAIN trial. The comparison with TNF-α inhibitors used data for abatacept from the ATTAIN trial and data from the BSRBR for the second TNF-α inhibitor. Because of limitations in the data, the manufacturer described the comparisons with TNF-α inhibitors as speculative. A comparison with rituximab was not completed principally on the basis that rituximab was not considered to be current standard practice at the time of submission.

3.8. The model was populated with data reflecting the average patient in the ATTAIN trial. On starting treatment with abatacept, patients were assumed to have an initial treatment response modelled using a percentage improvement in starting HAQ score. Patients were then assumed to experience long-term progression of disease, measured as an increase in HAQ score. An HAQ progression rate of 0.06 per year was used for patients on methotrexate, while a rate of 0.015 per year was used for patients on abatacept or a TNF-α inhibitor. The annual rate of treatment failure for abatacept and TNF-α inhibitors was assumed to be 8.2%, while patients starting methotrexate remained on it indefinitely. An initial stopping rule was incorporated into the model: in the first 6 months, patients on abatacept and TNF-α inhibitors who had a reduction in HAQ of more than 0.3 remained on treatment, while those who had a reduction in HAQ of less than 0.3 moved into the methotrexate arm.

3.9. The manufacturer estimated the incremental cost effectiveness of abatacept compared with methotrexate to be £25,395 per quality-adjusted life year (QALY) gained. The corresponding estimate in comparison with a second TNF-α inhibitor was £22,628 per QALY gained. One-way sensitivity analyses suggested that the model was sensitive to assumptions about the time horizon, discounting, rate of underlying disease progression and the cost of abatacept. The manufacturer’s probabilistic sensitivity analyses suggested that there was a high probability that abatacept was cost effective if the acceptable amount to pay for an additional QALY is £30,000. However, at a threshold of £20,000, methotrexate had a higher probability of being cost effective.

3.10. At the request of NICE, the manufacturer completed further analyses for the comparison of abatacept with methotrexate exploring the impact on the incremental cost-effectiveness ratio (ICER) of using different underlying rates of disease progression (measured as increases in HAQ score) that have been used in other NICE technology appraisals of treatments for rheumatoid arthritis. In a scenario where the rates of disease progression were 0.03 per year for patients on abatacept and 0.045 per year for patients on methotrexate, the estimated ICER was £33,567 per QALY. Using an estimate of 0.03 per year for abatacept, and an estimate of 0.06 per year for methotrexate, led to an estimated ICER of £28,445 per QALY.

3.11. The ERG examined the submission from the manufacturer. It considered that the clinical effectiveness section of the submission had used appropriate methodology, that it had identified all relevant trials, and that these were of good methodological quality. The ERG supported the conclusion in the manufacturer’s submission that abatacept was statistically significantly more effective than placebo when added to a regimen of ongoing DMARD therapy.

3.12. The ERG examined the Excel version of the model submitted by the manufacturer. It noted that abatacept was not modelled as part of a sequence of treatments and was considered to be a final active treatment option. The ERG considered that this may not be a true reflection of the likely position of abatacept in clinical practice, and was not consistent with how treatments for rheumatoid arthritis have been modelled in previous NICE appraisals.

3.13. The ERG highlighted a number of issues with the submitted model including errors in the model logic. The ERG suggested alternative estimates of parameter values for treatment costs, disease-related costs, rates of abatacept discontinuation, choice of HAQ mortality multiplier, and the estimated benefit of abatacept therapy on HAQ score. The ERG also constructed an overall mixed gender cohort for the comparisons, as opposed to the female cohort used in the manufacturer’s base case. The ERG used an alternative model for deriving utility values from HAQ scores and investigated the impact of different values for HAQ progression rate. For the comparison with TNF-α inhibitors, the ERG incorporated new parameter values to represent the clinical effectiveness of a second TNF-α inhibitor on HAQ scores, as well as the appropriate treatment costs. The ERG estimated that the cumulative impact of these amendments would increase the ICER in both scenarios (using methotrexate or using further TNF-α inhibitors as comparators) to approximately £50,000 per QALY. For the comparison with methotrexate, the factors having most impact on the estimates were the discontinuation rates while on abatacept, and the equation by which HAQ scores had been mapped to utility estimates. The factors most affecting the estimates for the comparison with a second TNF-α inhibitor were the effectiveness of the TNF-α inhibitor and the process of reconciling mixed- and single-sex data inputs in the model.

3.14. In addition, the ERG examined in detail the evidence relating to underlying disease progression because this was identified as a key driver in the cost-effectiveness modelling. The ERG noted that both the rate of disease progression while on abatacept and the magnitude of the difference in the rate of progression between abatacept and the comparator are important factors in estimating the cost effectiveness of abatacept. The ERG identified a number of limitations in the studies that have been used to calculate underlying disease progression rates, and it reanalysed the available data. The ERG concluded that the rates of progression while on conventional DMARDs or on palliative care could be as low as 0.012 per year, and that a reasonable assumption for the relative progression rate while on biologic treatments such as abatacept would be around 75% of the rate while on conventional DMARDs (0.009). Using these values in the model suggested an estimate of cost effectiveness for abatacept of around £70,000 per incremental QALY gained when the comparator was either methotrexate or a TNF-α inhibitor.

3.15. The ERG also carried out analyses that included all the amendments it had made to the manufacturer’s model as well as the values for underlying disease progression that were consistent with other technology appraisals of treatments for rheumatoid arthritis. Modelling underlying disease progression using increases in HAQ score of 0.03 and 0.045 per year for abatacept and methotrexate respectively, the estimated ICER was approximately £63,000 per QALY gained. Using values of 0.03 per year for abatacept and 0.06 per year for methotrexate, the estimated ICER was approximately £55,000 per QALY gained.

3.16. Full details of all the evidence are in the manufacturer’s submission and the ERG report, which are available from www.nice.org.uk/TA141