Vedolizumab for treating moderately to severely active Crohn's disease after prior therapy

GEMINI II study design

3.3 GEMINI II compared the efficacy and safety of vedolizumab with placebo plus conventional therapy (oral prednisone or budesonide, immunosuppressive agents, mesalazine and antibiotics were permitted) for moderately to severely active Crohn's disease. It comprised an induction trial (weeks 0–6) and a maintenance trial (weeks 6–52), giving an overall study duration of 52 weeks.

3.4 In the blinded induction trial (cohort 1), patients received vedolizumab 300 mg intravenously (n=220) or placebo (n=148) at weeks 0 and 2. Randomisation was stratified by concomitant use of corticosteroids and immunosuppressive agents or previous use of TNF‑alpha inhibitors, or both. The proportion of patients with previous exposure to TNF‑alpha inhibitors was limited to 50% (50.5% and 48.6% in the vedolizumab and placebo arms respectively). To fulfil sample‑size requirements for the maintenance trial, 748 additional patients were assigned treatment in an open‑label group (cohort 2), of whom 747 patients received the same vedolizumab regimen as that in cohort 1 in the blinded induction trial. Before the trial, 58% of all patients had experienced failure of a TNF‑alpha inhibitor.

3.5 In the maintenance trial, patients from both cohorts who had a clinical response with vedolizumab at week 6 (that is, a 70 point or greater decrease in the CDAI score; n=461) were randomly assigned to continue in a blinded fashion to receive vedolizumab every 8 weeks (n=154), vedolizumab every 4 weeks (n=154), or placebo (n=153), for up to 52 weeks. Randomisation was stratified according to (1) participation in cohort 1 or 2 during induction, (2) concomitant use of corticosteroids and (3) concomitant use of immunosuppressive agents or previous use of TNF alpha inhibitors, or both. Patients from either cohort whose disease did not have a clinical response at week 6 to vedolizumab induction therapy (n=412) received maintenance treatment with vedolizumab 300 mg every 4 weeks and were followed to week 52. Patients in the placebo group of cohort 1 who completed induction treatment (n=137) continued to receive placebo and were also followed to week 52. Of 814 patients who received vedolizumab in GEMINI II, 295 completed week 52 assessments and enrolled in GEMINI LTS (an ongoing, single arm, open label safety study).

3.6 In GEMINI II, the primary outcomes during induction at week 6 were clinical remission (CDAI score 150 points or less) and enhanced clinical response (a 100 point or greater decrease in the CDAI score). During maintenance, the primary outcome was clinical remission at week 52. Secondary outcomes included CDAI 100 response and corticosteroid free remission at week 52. Safety outcomes were included and quality of life was assessed using the Inflammatory Bowel Disease Questionnaire (IBDQ), 36 Item Short Form Health Survey (SF 36), and European quality of life 5 domain scale (EQ 5D) questionnaires (at screening and before dosing at weeks 6, 30 and 52).

3.7 The main analyses in the induction study of GEMINI II used the intention‑to‑treat population, which included all patients in cohort 1 who were randomised and received at least 1 dose of blinded study drug (n=148 in the placebo arm and n=220 in the vedolizumab arm). The maintenance study analyses also used the intention‑to‑treat population, which included patients who received vedolizumab whose disease had a clinical response at week 6 and who were then randomised to vedolizumab either every 4 or 8 weeks (n=154 each) or placebo (n=153). The non‑intention‑to‑treat population in the maintenance study was included in the safety assessment, comprising 814 patients who had vedolizumab and 301 patients who received placebo. In addition to patients in the intention‑to‑treat population, it included patients who received placebo in the induction phase and remained on placebo for the maintenance phase, and patients whose disease did not respond to vedolizumab by week 6 of the induction study. The company also presented subgroup analyses for patients in whom a TNF‑alpha inhibitor had failed (vedolizumab every 8 weeks [n=82], vedolizumab every 4 weeks [n=77] and placebo [n=78]) and for patients who had not had a TNF‑alpha inhibitor before (vedolizumab every 8 weeks [n=66], vedolizumab every 4 weeks [n=71] and placebo [n=71]).

GEMINI III study design

3.8 GEMINI III was a 10‑week study that evaluated the efficacy of vedolizumab compared with placebo. Patients were randomised to receive vedolizumab 300 mg (n=209) or placebo (n=207) at weeks 0, 2 and 6 and stratified according to whether they had previously had TNF‑alpha inhibitor treatment (there were 315 patients in whom a TNF‑alpha inhibitor had failed and 101 patients who had not had a TNF‑alpha inhibitor before). They were also stratified according to concomitant use of (a) oral corticosteroids and (b) concomitant use of immunomodulators (mercaptopurine, azathioprine, or methotrexate). At the end of the study, patients could enrol in GEMINI LTS. All randomised patients received at least 1 dose of blinded study drug and were included in the intention‑to‑treat population.

3.9 The primary analysis of GEMINI III focused on the 315 patients in whom a TNF‑alpha inhibitor had failed. The primary outcome was clinical remission at week 6 (CDAI score 150 points or less). A secondary analysis evaluated the overall population including patients who had not had TNF‑alpha inhibitor treatment before. Secondary outcomes included clinical remission at week 6 in the overall population, clinical remission at week 10 in the population in whom TNF‑alpha inhibitor treatment had failed and in the overall population, sustained clinical remission (CDAI score 150 points or less at both week 6 and week 10) in the population in whom TNF‑alpha inhibitor treatment had failed and in the overall population, and safety outcomes. Other outcomes included health‑related quality of life, as shown by change from baseline in IBDQ, SF‑36, and EQ‑5D scores at weeks 6 and 10.

GEMINI II – induction phase results

3.10 The results for the primary outcomes of GEMINI II showed that at week 6, clinical remission rates (CDAI score 150 points or less) were significantly higher in patients having vedolizumab than in patients having placebo (14.5% [95% CI 9.9 to 19.2] and 6.8% [95% CI 2.7 to 10.8] respectively, p=0.02). There was no significant difference in enhanced clinical response (a 100‑point or greater decrease in the CDAI score) at week 6 between the vedolizumab and placebo groups (31.4% [95% CI 25.2 to 37.5] compared with 25.7% [95% CI 18.6 to 32.7] respectively; p=0.23).

3.11 The company carried out pre‑specified subgroup analyses for the primary outcomes, investigating the influence of baseline characteristics on treatment effect. For vedolizumab compared with placebo, the analyses showed a between‑treatment difference in clinical remission at week 6 of 8.2% in the population of patients who had not had a TNF‑alpha inhibitor before (no further details were provided). In the new evidence provided in response to the appraisal consultation document, the company reported that clinical remission at week 6 in the population in whom a TNF‑alpha inhibitor had failed was experienced by 10.5% of patients receiving vedolizumab and 4.3% of patients receiving placebo (95% CI for between‑treatment difference −9.1 to 21.3). It further reported that an exploratory analysis in the same population showed that clinical remission at week 10 was experienced by 16.0% of patients in the vedolizumab group (n=5) and 8.6% of patients in the placebo group (n=70) (95% CI for between‑treatment difference 0.2 to 14.6).

3.12 The company presented results for changes in health‑related quality of life from baseline to week 6 in the vedolizumab group (n=211) and the placebo group (n=146) in the overall population. The company advised that a decrease of at least 0.3 points in the EQ‑5D score represented a clinically meaningful improvement in health‑related quality of life. Adjusted mean change in EQ‑5D score from baseline was −0.5 (95% CI −0.7 to −0.3) in patients who received vedolizumab and −0.3 (95% CI −0.5 to −0.0) in patients who received placebo, giving a difference in adjusted change of −0.2 (95% CI −0.5 to 0.1). The company did not present the change in EQ‑5D scores for the population who had not had a TNF‑alpha inhibitor before. In the new evidence submitted in response to the appraisal consultation document, the company reported that there was no clinically meaningful decrease in EQ‑5D score from baseline to week 6 with vedolizumab in the population in whom a TNF‑alpha inhibitor had failed. The adjusted mean change in EQ‑5D score (95% CI) was −0.2 (−0.5 to 0.1) in the vedolizumab group (n=103) and −0.2 (−0.6 to 0.1) in the placebo group (n=69).

GEMINI II – maintenance phase results

3.13 In the intention‑to‑treat population in the maintenance study, there were statistically significantly higher rates of clinical remission (CDAI score 150 points or less) at week 52 in patients who received vedolizumab every 8 weeks or every 4 weeks, compared with patients who received placebo. In patients receiving vedolizumab every 8 weeks, the treatment difference from placebo was 17.4% (95% CI 7.3 to 27.5, p=0.0007) and in patients receiving vedolizumab every 4 weeks, it was 14.7% (95% CI 4.6 to 24.7, p=0.0042).

3.14 Clinical remission rates were higher for patients who had vedolizumab every 4 or 8 weeks compared with those who had placebo regardless of prior TNF‑alpha inhibitor use.

3.15 The company presented results for the changes in health‑related quality of life from baseline to week 52 in the groups having vedolizumab every 8 weeks (n=79), vedolizumab every 4 weeks (n=92) and placebo (n=81) in the overall population. Adjusted mean change in EQ‑5D score from baseline was −1.5 (95% CI −1.8 to −1.2) in patients having vedolizumab every 8 weeks, −1.4 (95% CI −1.7 to −1.1) in patients having vedolizumab every 4 weeks and −1.0 (95% CI −1.3 to −0.7) in patients having placebo. The mean difference in adjusted change from baseline compared with placebo was –0.5 (95% CI −0.9 to −0.1) for vedolizumab every 8 weeks and –0.4 (95% CI −0.8 to 0.0) for vedolizumab every 4 weeks. The company did not provide results for the changes in health‑related quality of life from baseline to week 52 for the sub‑populations who had not had a TNF‑alpha inhibitor before or in whom a TNF‑alpha inhibitor had failed.

GEMINI III

3.16 For the primary outcome of clinical remission at week 6 in people for whom TNF‑alpha inhibitor treatment had failed, no statistically significant difference was seen between the vedolizumab (15.2% [95% CI 9.6 to 20.8]) and placebo (12.1% [95% CI 7.0 to 17.2]) groups (p=0.433). However, an exploratory analysis found a higher proportion of these patients had clinical remission at week 10 with vedolizumab than with placebo (26.6% [95% CI 19.7 to 33.5]) compared with 12.1% [95% CI 7.0 to 17.2] p=0.0012 [nominal p‑value]).

3.17 An exploratory analysis of the overall population including patients who had not had a TNF‑alpha inhibitor before showed that clinical remission occurred in a higher proportion of patients having vedolizumab than placebo at week 6 (19.1% [95% CI 13.8 to 24.5] compared with 12.1% [95% CI 7.6 to 16.5], p=0.0478 [nominal p‑value]) and week 10 (28.7% [95% CI 22.6 to 34.8] compared with 13.0% [95% CI 8.5 to 17.6], p<0.0001 [nominal p‑value]).

3.18 The company provided results for changes in health‑related quality of life from baseline to weeks 6 and 10 for patients in the population in whom a TNF‑alpha inhibitor had failed. At week 6, adjusted mean change in EQ‑5D score was −0.4 (95% CI −0.6 to −0.2) in patients who had vedolizumab (n=158) and −0.1 (95% CI −0.3 to 0.1) in patients who had placebo (n=149), giving a mean difference in adjusted change from baseline of −0.2 (95% CI −0.5 to 0.1). At week 10, adjusted mean change in EQ‑5D score was −0.6 (95% CI −0.8 to −0.4) in patients who had vedolizumab (n=152) and −0.1 (95% CI −0.4 to 0.1) in patients who had placebo (n=143), giving a mean difference in adjusted change from baseline of −0.5 (95% CI −0.8 to −0.2). Similar results were seen in the overall study population.

Pooled analysis of GEMINI II and III

3.19 In the new evidence submitted in response to the appraisal consultation document, the company described a post‑hoc analysis of pooled data from GEMINI II and III in the population in whom a TNF‑alpha inhibitor had failed. At 6 weeks there was no statistically significant difference between the vedolizumab group (n=263) and placebo group (n=227) in the proportion of patients in clinical remission (13.3% compared with 9.7% respectively, p=0.157). However, at 10 weeks the proportion of patients in clinical remission was statistically significantly higher with vedolizumab than with placebo (21.7% compared with 11.0%, p=0.0008).

GEMINI LTS

3.23 In its new evidence submitted in response to the appraisal consultation document, the company included analyses from GEMINI LTS of clinical remission for up to 2 years. Clinical remission was defined as a Harvey–Bradshaw Index score of 4 points or less. In the population in whom TNF‑alpha inhibitors had failed (n=136), clinical remission was 52% at week 52, 56% at week 80 and 51% at week 104.

Population

3.45 In its first 2 models, the company defined 3 patient groups with moderately to severely active Crohn's disease (CDAI score 220–450 points) who had an inadequate response with, lost response to, or are intolerant to either conventional non‑biological therapy or TNF alpha inhibitors.

3.46 In its third model, the company focused entirely on the subgroup of patients with moderately to severely active Crohn's disease (CDAI score 220–450 points) in whom a TNF‑alpha inhibitor had failed (primary and secondary failure). The cost‑effectiveness results for this subgroup generated using the third model superseded those generated using the first 2 models. The company's third model evaluated the cost effectiveness of vedolizumab compared with conventional non‑biological therapy. Because of data limitations, the company carried out an exploratory cost comparison of vedolizumab compared with other biological treatments (adalimumab and infliximab), which included scenario analyses on comparator dose escalation based on clinical expert opinion.

Clinical parameters and transition probabilities

3.47 Treatment efficacy included response and remission data for the induction phase, and the probability of being in remission or having mild disease at the end of 1 year (the maintenance phase of the GEMINI II study):

3.48 The company estimated the efficacy of each treatment by estimating odds ratios using response and remission data from the network meta‑analyses (the population who had not had a TNF‑alpha inhibitor before) or from pooled trial data (the mixed population and the population in whom TNF‑alpha inhibitor treatment had failed). In the population who had not had a TNF‑alpha inhibitor before, infliximab data were derived from ACCENT I, separate from the network meta‑analysis. The company said this was because the trial by Targan et al. in the network meta‑analysis had a small sample size and did not use a standard infliximab dosage. The odds ratios were then used to estimate the percentage of patients in each health state at the end of the induction and maintenance periods. In the first 2 models, the probability of surgery was assumed to be the same across the different patient populations in the induction phase (2.03%) and maintenance phase (2.7%). In the third model, the probability of surgery was 0.83% during induction then 1.11% in each 8‑week cycle, and the probability of further surgery in the next cycle was 1.11% (amended from 33.75% in the previous models).

3.49 The company based the model's starting annual mortality rate on all‑cause mortality for the UK general population (0.0015). In the first 2 models, relative mortality risks were assumed to be 1.3 for mild disease, 2.3 for moderate–severe disease and 3.2 for surgery. In the third model, the relative risk of mortality was assumed to be 1 in all health states. Patients in remission were assumed to have the same mortality risk as the general UK population.

Adverse events and surgical complications

3.50 The company selected adverse events to be used in its economic model based on clinical expert opinion. They included serious infection, lymphoma, acute hypersensitivity reactions and skin reactions. In the first 2 models, the probability of each adverse event occurring with each treatment was estimated from clinical trial data included in the network meta‑analyses. In the third model, the probabilities of adverse events associated with vedolizumab and conventional non‑biological therapy were obtained from summaries of product characteristics. Surgical complications in the model were also based on clinical expert opinion and the probabilities of these occurring were estimated from pooled data from a systematic literature review on surgical intervention. Annual probabilities of discontinuing biological treatment owing to adverse events were derived from clinical trials.

Utility values

3.51 The company's base case used the observed EQ‑5D scores from GEMINI II and GEMINI III. The company assumed a utility value for the surgery state that was equal to that for the moderate–severe health state because patients in GEMINI II and GEMINI III were not followed for surgery. The utility values used in the model were 0.820 for remission, 0.730 for mild disease and 0.570 for both moderate–severe disease and surgery. The company applied disutilities from published literature for adverse events.

Costs

3.52 Treatment acquisition costs, including the estimated doses and unit costs for conventional non‑biological therapy, were taken from the British National Formulary (2013; costs updated to November 2014 edition in the third model). The patient access scheme was applied to the cost of vedolizumab as a simple discount on the list price (the level of the discount is confidential; the discount was increased in the third model). In the first 2 models, administration costs of £308 per administration in the maintenance phase and £616 in the induction phase were included for vedolizumab and infliximab (adalimumab did not have any administration costs). In the third model, a cost of £328 per administration of vedolizumab was applied in both phases.

3.53 In the first 2 models, health‑state costs were taken from Bodger et al. (2009) and inflated to 2012 prices. The health‑state costs were £110 for remission, £313 for mild disease, £490 for moderate–severe disease and £10,581 for surgery, which included surgical complications. In the third model, health‑state costs were estimated based on a survey of 8 clinical experts in England. These were £100 for remission, £205 for mild disease, £1239 for moderate–severe disease and £7293 for surgery (including post‑surgery complications).

3.54 In all models, surgery‑related complication costs were estimated by applying NHS reference costs to resource use as reported by the company's clinical experts. The company estimated costs of adverse events as weighted averages according to the NHS reference costs and assumed that all affected patients were hospitalised.

Company base‑case results

3.55 Using its second model submitted in response to clarification, which replaced the original model, the company provided base‑case results incorporating the original vedolizumab patient access scheme for 3 populations:

3.56 Using its third model submitted as new evidence in response to the appraisal consultation document, the company provided base‑case results incorporating the updated vedolizumab patient access scheme for the population in whom a TNF‑alpha inhibitor had failed. The deterministic ICER for vedolizumab compared with conventional non‑biological therapy was £21,620 per QALY gained (incremental costs and QALYs are confidential) and the probabilistic ICER was £27,428 per QALY gained (95% CI −7883 to 82,947). The probability of vedolizumab being cost effective compared with conventional non‑biological therapy at a maximum acceptable ICER of £30,000 was 61%.

Company sensitivity and scenario analyses

3.57 For the original model, the company concluded that its deterministic sensitivity analyses showed that the key drivers were transition probabilities (in particular for remission), health state costs and utility values. It noted that assuming a longer time horizon in the original model made vedolizumab more cost effective in all populations. The company did not provide any sensitivity analyses using its second model.

3.58 For its third model, the company concluded that its deterministic sensitivity analyses showed that the ICER was most sensitive to variation in the conventional non‑biological therapy arm transition probabilities from the moderately–severely active state and from the remission state. It noted that the scenario analyses showed that the ICER increased most markedly when the time horizon was reduced to 10 years, when enhanced response was used to determine progression to maintenance treatment (100‑CDAI point decrease), when response was assessed at week 6 and when efficacy data were based solely on GEMINI II. The company's scenario analyses showed that continuing treatment with vedolizumab for 2 or 3 years (instead of 1 year) increased the ICER to £24,695 and £26,207 per QALY gained respectively.

Company exploratory analyses

3.59 The company carried out exploratory cost‑comparison analyses that compared vedolizumab with adalimumab and with infliximab in the population in whom a TNF‑alpha inhibitor had failed. The results of these analyses are confidential.

Model structure

3.61 The ERG expressed concerns about the structure of the company's model in 4 main areas:

3.62 The ERG noted that the duration of induction with the biological therapies was not always in line with UK licensing and clinical practice, meaning not all studies delivered a full induction dose in the second model:

Population

3.63 The ERG was unclear how results from the mixed intention‑to‑treat population could be interpreted. It believed that patients who had previously had TNF‑alpha inhibitors and those who had never had TNF‑alpha inhibitors are 2 distinct patient groups with different characteristics and likelihood of responding to treatment. It considered that the results from the 2 groups should be interpreted separately.

3.64 The ERG was satisfied that analyses according to disease severity could potentially be informative, despite not being defined in the NICE final scope. However, the ERG was unable to confirm the results of these analyses because it could not verify the calibrated transition probabilities and it was unsure how the clinical data had been estimated in the company's second model.

Clinical parameters

3.65 The ERG noted that the company had provided limited details on the network meta‑analyses used in its second economic model for the population who had not had a TNF‑alpha inhibitor before; how the vedolizumab clinical trial data had been pooled; and how the discontinuation rates because of adverse events had been calculated. Although the ERG recognised that the Targan trial comparing infliximab with placebo had limitations, it believed that it should have been included in the network meta‑analysis for infliximab and used in the base case, potentially adjusting for small sample size. The ERG noted that the company had instead used data for infliximab from the placebo‑controlled ACCENT I trial (separate from the network meta‑analysis), but had not discussed the trial's limitations. The ERG noted that including data for adalimumab from the trial by Watanabe et al. as well as CLASSIC I in the primary analysis would likely increase the probabilities of remission and response for adalimumab.

3.66 The ERG was unclear from the company submission and the publication by Bodger et al. how the transition probabilities for patients having surgery had been calculated. It considered the values used by the company for transitioning from surgery to surgery in the next cycle to be high (33.75%), and was not satisfied by the company's explanation. The ERG was not able to predict how correcting the transition matrix for movement between states after surgery would affect the ICERs.

3.67 The ERG expressed concerns about the assumptions about mortality used in the company's model because of a lack of detail in the company's submission. It noted that because mortality is conditional on the current health states in the company's model, the model predicts greater survival for patients who had biological therapy compared with patients who had conventional non‑biological therapy. However, the study by Lichtenstein et al. (2009), used by the company in its model, suggests no statistical differences in the excess mortality rates according to disease severity at baseline, or in mortality between patients who did or did not receive infliximab. The ERG stressed that no increased mortality rate was observed in patients taking placebo in GEMINI II. Given the lack of evidence of a differential mortality rate between treatments, the ERG believed that the same excess risk mortality should be applied to all Crohn's disease health states.

3.68 The ERG considered that the inclusion of adverse events and their impact on costs and health‑related quality of life was flawed. It was unclear if all or only grade 3 or 4 adverse events had been included and noted the selection was based on the opinion of 2 clinical experts. It found the calculations from the company to be simplistic and likely to be incorrect because they did not account for trial duration. Moreover, the ERG was unsure why data from the network meta‑analysis for the incidence of serious adverse events were not used in the company's model.

Utility values

3.69 The ERG was largely satisfied with the company's approach to estimating utility values for the different health states in its model, but had some concerns:

Cost‑effectiveness results

3.70 The ERG noted that the company presented pairwise comparisons rather than a fully incremental analysis for the group who had not had a TNF‑alpha inhibitor before and that it had not provided updated cost‑effectiveness estimates for all of the patient groups covered by the original model. The ERG therefore extracted this information from the company's second model. In the group which had not had a TNF‑alpha inhibitor before, a fully incremental analysis gave the following ICERs:

3.71 In the mixed population and in the population in whom a TNF‑alpha inhibitor had failed, vedolizumab was associated with greater costs and QALYs compared with conventional non‑biological therapy in subgroups according to disease severity, with ICERs of:

3.72 The ERG noted that the company had not re‑run its deterministic sensitivity analyses using the second model. The ERG considered that the parameters that had the largest impact on the ICER would not change between the 2 versions of the model submitted. It agreed with the company that the key drivers of the ICER included many of the transition probabilities, health state costs and utility values. It considered the ranges used by the company for its deterministic and probabilistic sensitivity analyses to be somewhat arbitrary for most input parameters.

3.73 Using the company's second model, the ERG reported the results from the scenario analyses presented in the original company submission. It noted that the ICER was sensitive to all the scenarios considered, especially the time horizon and health state utility values.

3.74 For transparency, the ERG extracted the probabilistic ICERs using the second version of the company's model and noted that these were consistent with the deterministic ICERs. In a fully incremental comparison, the ERG reported that the probability of vedolizumab being cost effective was less than 1% at a maximum acceptable ICER of £20,000 per QALY gained for all 3 populations in the base case of the company's second model. The probability of cost effectiveness increased to about 2% at a maximum acceptable ICER of £30,000 per QALY gained.

3.75 The ERG had concerns about the validity of the predictions made by the company's model, including several discrepancies between the results generated using the model and those from the clinical trials. It noted that for the population in whom a TNF‑alpha inhibitor had failed, the company's model seemed to under‑predict the proportion of patients having conventional non‑biological therapy who were in remission.