Infliximab, adalimumab and golimumab for treating moderately to severely active ulcerative colitis after the failure of conventional therapy (including a review of TA140 and TA262):clinical effectiveness systematic review and economic model

Interventions

Any of the following interventions were included:

1. For adults (defined by the Assessment Group as aged ≥ 18 years):

   • ADA

   • IFX

   • GOL.

2. For children and adolescents aged 6–17 years (inclusive):

   • IFX.

Biosimilar versions of IFX (Remsima and Inflectra) are also licensed for the same indications as Remicade and have been considered as part of the evidence base for IFX within this assessment.

Studies in which the interventions were assessed in line with licensed indications were included in the systematic review.

Populations

1. Adults aged ≥ 18 years with moderately to severely active (non-systemic) UC (defined as patients with moderately active disease according to the modified Truelove and Witts’ criteria23) whose disease has responded inadequately to conventional therapy including corticosteroids and mercaptopurine or AZA, or who are intolerant of, or have medical contraindications to, such therapies.

2. Children and adolescents aged 6–17 years with severely active (non-systemic) UC (as classified by the PUCAI measure25) whose disease has responded inadequately to conventional therapy including corticosteroids and mercaptopurine or AZA, or who are intolerant to, or have medical contraindications to, such therapies.

Comparators

Relevant comparators included in the final NICE scope were (1) interventions as defined in the protocol for this assessment (i.e. IFX, ADA or GOL compared with each other) and (2) standard clinical management, which may include a combination of aminosalicylates (sulfasalazine, mesalazine/mesalamine, balsalazide or olsalazine), corticosteroids (beclomethasone, budesonide, hydrocortisone or prednisolone), thiopurines (mercaptopurine or AZA), calcineurin inhibitors or elective surgical intervention.

It should be noted that although calcineurin inhibitors (tacrolimus and ciclosporin) were included as potential comparators in the final NICE scope, these options were excluded from the assessment for two reasons:

1. They are treatments typically reserved for patients with acute severe disease. This differs to the population within the final NICE scope. Studies were specifically excluded if they related to these patients (see Exclusion criteria).

2. There are no direct data comparing biologicals versus calcineurin inhibitors for the population under investigation. Altogether, there are very limited data on the efficacy of either ciclosporin or tacrolimus in this indication.

Outcomes

Eligible outcomes for consideration were:

• mortality

• measures of disease activity

• rates of and duration of response, relapse and remission

• rates of hospitalisation

• rates of surgical intervention (both elective and emergency)

• time to surgical intervention (both elective and emergency)

• AEs of treatment (including leakage and infections following surgery)

• HRQoL.

Following discussions during the NICE appraisal scoping process, data relating to mucosal healing were not considered eligible for this assessment.

Study design

Randomised controlled trials were eligible for inclusion in the systematic review of clinical effectiveness. Long-term extension studies associated with included RCTs were also included in the review.

Studies published as abstracts or conference presentations were eligible for inclusion only if sufficient details were presented to allow an assessment of the trial methodology and results to be undertaken.

Population: adults aged ≥ 18 years with moderately to severely active ulcerative colitis who have had an inadequate response to conventional therapy including corticosteroids and mercaptopurine or azathioprine, or who are intolerant to, or have medical contraindications against, such therapies

A total of nine relevant RCTs were identified which were performed in adult populations. Four RCTs evaluated the use of IFX [Active Ulcerative Colitis Trial (ACT)1,49 ACT2,49 Probert et al. 50 and UC-SUCCESS51], three RCTs were of ADA [Ulcerative colitis Long-Term Remission and maintenance with Adalimumab treatment of moderate to severe ulcerative colitis (ULTRA)1,44 ULTRA2,45 and Suzuki et al. , 201446] and two RCTs were of GOL [Program of Ulcerative Colitis Research Studies Utilizing an Investigational Treatment (PURSUIT)-SC,47 PURSUIT-Maintenance48]. Four of these RCTs (ACT1,49 ACT2,49 ULTRA1,44 and ULTRA245) had long-term open-label extension studies associated with them (ACT1 and ACT2 extension studies,55 ULTRA354) that were also included as part of the evidence for these interventions. All of the included RCTs for adults were undertaken against a comparator of PBO, with the exception of UC-SUCCESS51 which assessed the use of IFX against active comparators of AZA and combination IFX/AZA. No head-to-head RCTs comparing interventions of interest against each other were identified for adults. All RCTs were Phase III (if stated), with the exception of Suzuki et al. 46 (Phase II/III) and PURSUIT-SC47 (Phase II/III). If stated, all included adult population trials were powered for the primary end points of clinical remission (ULTRA1,44 ULTRA2,45 Probert et al. ,50 UC-SUCCESS51) or clinical response (ACT1,49 ACT2,49 PURSUIT-SC,47 PURSUIT-Maintenance48). Where the geographical location(s) of study sites were reported, all trials were multicentre, international studies, with the exception of Probert et al. ,50 which was performed in the UK and Germany, and Suzuki et al. ,46 which was conducted exclusively in Japan. All trials were at least partly industry funded.

Eight trials included time points for the assessment of the use of interventions in achieving induction of clinical response or remission, of which four assessed IFX (ACT1,49 ACT2,49 Probert et al. 50 and UC-SUCCESS51), three assessed ADA (ULTRA1,44 ULTRA245 and Suzuki et al. 46) and one assessed GOL (PURSUIT-SC47). Six trials reported outcomes at time points for the evaluation of the use of interventions in the maintenance of clinical response or remission, consisting of three IFX trials (ACT1,49 ACT249 and UC-SUCCESS51), two ADA trials (ULTRA245 and Suzuki et al. 46) and one GOL trial (PURSUIT-Maintenance48).

None of the included RCTs applied Truelove and Witts’ disease severity criteria23 in their eligibility criteria (as referred to in the final NICE scope for this appraisal36 and as specified in the protocol). All included trials applied the Mayo score [except Probert et al. 50 for which the score was specified simply as Ulcerative Colitis Symptom Score (UCSS)] to classify the disease severity of potential participants (note: the UCSS was confirmed to be equivalent to Mayo score by Professor C Probert, University of Liverpool, 2014, personal communication). The included trials required a Mayo score of 6–12 points (with evidence of endoscopic disease) for participant eligibility. Mayo scores of 6–12 points were described in the included trial literature as moderate to severe disease and were also subsequently confirmed following clinical advice as representing moderate to severe disease (note: ad-hoc searches were performed to attempt to identify evidence relating to the relationship between the Truelove and Witts’ and Mayo disease severity indices; however, no evidence published in full text in English could be identified). Included trials required a varying range of prior use of conventional therapy for eligibility, as described in Tables 6 and 7. The UC-SUCCESS trial,51 which specified patients to be either AZA-naive or free from AZA treatment for at least 3 months before enrolment, was a borderline inclusion in the clinical effectiveness systematic review because the wording of the population in the scope and the licensed indications required prior use of AZA or 6-MP. However, as the trial reported the use of a stated (albeit low) proportion of prior immunosuppressant use, this trial was included in the clinical effectiveness systematic review for completeness. However, this trial was not eligible for subsequent inclusion in meta-analyses or NMAs. Suzuki et al. 46 included Japanese patients aged ≥ 15 years (ADA is not licensed in the paediatric population), but the mean ages of participants across treatment arms at baseline was 41.3–42.5 years.

The Core Outcome Measures in Effectiveness Trials (COMET) initiative,56 which promotes the use of core outcome sets in clinical trials, referenced the work by Cooney et al. 22 in classifying the use of outcome measures in UC clinical trials. Although acknowledging the very broad range of available disease severity/activity measures available, all adult population trials included in the assessment were consistent in their utilisation of the Mayo score as a measure of clinical response and/or remission. The included trial by Probert et al. 50 applied the UCSS in the evaluation of clinical remission at induction. This score is equivalent to the full Mayo score: the components of the UCSS are consistent with the elements assessed within the Mayo score (i.e. stool frequency, rectal bleeding, sigmoidoscopic appearance and physician’s global assessment) and also is referenced using the citation quoted for the Mayo score. 24 None of the included studies utilised the modified Truelove and Witts’ criteria23 (as referred to in the NICE appraisal scope36 and as specified in the assessment protocol) in their outcome assessments.

As recommended in the Committee for Human Medicinal Products (CHMP) guideline57 on the development of new medicinal products for the treatment of UC patients with confirmed UC were eligible for the included trials. Severity of disease was defined by clinical and endoscopic evaluation, as recommended in the CHMP guideline. Although the interventions of interest in this assessment were developed for the treatment of patients not responding/intolerant to previous immunomodulatory therapy, the Assessment Group did not consider that adequate definitions of inadequate response/intolerance were included in trials, as recommended by the CHMP guideline. The guideline recommended that, for refractory populations, a minimum duration and dose of previous baseline medication should be defined, but this was not the case in the included trials. In addition, intolerance was not defined by minimum criteria of severity in the trials. In terms of study duration, it was recommended that induction studies should be 8–12 weeks, but could be shorter based on the pharmacodynamic properties of the study drug. All induction trials assessed efficacy at 8 weeks, with the exception of the PURSUIT-SC GOL trial,47 which was a 6-week study. All maintenance studies were at least 1 year in length, as recommended in the CHMP guideline.

Population: children and adolescents aged 6–17 years (inclusive) with severely active ulcerative colitis who have had an inadequate response to conventional therapy including corticosteroids and mercaptopurine or azathioprine, or who are intolerant to, or have medical contraindications against, such therapies

A single Phase III open-label RCT was identified for the paediatric population (T7252) which evaluated the use of IFX in maintenance therapy. All patients received the licensed IFX induction regimen before being randomised to one of two IFX maintenance regimens. Outcomes were reported at week 30 and week 54. No PBO-controlled or head-to-head RCTs were identified for children and young people. The absence of a PBO or non-IFX control group in the included RCT made it difficult to consider the effectiveness of IFX in paediatric patients compared with conventional UC therapies. This industry-funded trial had the primary end point of clinical response and was conducted in the USA, the Netherlands, Canada and Belgium. Eligible patients were 6–17 years of age with moderately to severely active UC and a Mayo score of 6–12 points with endoscopic evidence of disease. Therefore, although IFX is licensed in this age group with severe disease only (as reflected in the scope population), this trial was included in consideration of limited paediatric RCT evidence.

Population: adults aged ≥ 18 years with moderately to severely active ulcerative colitis who have had an inadequate response to conventional therapy including corticosteroids and mercaptopurine or azathioprine, or who are intolerant to, or have medical contraindications against, such therapies

Mean and median reported ages of participants were considered consistent across included adult population trials, ranging from 37 to 42.5 years. Mayo scores at baseline were also consistent across trials and spanned from 8.1 to 8.9. Average proportions of male participants ranged from 41% to 73% and the majority of included patients (when reported) were Caucasian in ethnicity (79.5–95.9%), with the exception of the Suzuki et al. 46 study, which included exclusively Japanese patients. Mean and median disease duration of participants ranged from 59 months (4.9 years) to 8.5 years. Conventional UC medications at baseline were variable between the included trials. In none of the included studies had all participants previously been trialled on corticosteroids and AZA or 6-MP, as required by the wording used in the final NICE scope36 population and the wording of the European Medicines Agency (EMA) licensing for IFX, ADA and GOL. Although it is noted that AZA and 6-MP may be used more typically in clinical practice as maintenance therapies owing to their longer initiation of effect, it is debatable whether or not the included trial populations would represent patients who had failed or were intolerant to previous conventional therapies. All trials related to anti-TNF-α-naive populations, with the exception of ULTRA245 (which permitted the inclusion of anti-TNF-α experienced patients) and PURSUIT-Maintenance48 (in which patients responding to prior GOL induction therapy were randomised to GOL maintenance regimens or PBO). Data at induction were reported according to anti-TNF-α experience. Data relating to patients who were anti-TNF-α-naive for maintenance time points were requested and received from the manufacturer of ADA (AbbVie). Weight and smoking status were both relatively poorly reported across included studies.

Population: children and adolescents aged 6–17 years (inclusive) with severely active ulcerative colitis who have had an inadequate response to conventional therapy including corticosteroids and mercaptopurine or azathioprine, or who are intolerant to, or have medical contraindications against, such therapies

The included trial population52 averaged 15 years of age in both treatment groups, was 43.5–45.5% male and mostly Caucasian in ethnic origin (82.6–90.9%) with average disease duration of 1.1–1.8 years. Patients had a median Mayo score of 7.5 to 8.0 points and a median PUCAI score of 50–57.5 points, for which a PUCAI of score of ≥ 65 would indicate severe disease (and, therefore, were a mixture of patients with moderate and severe disease, while IFX is licensed in paediatric patients with severe disease only). Participants were required to have had prior use of at least one conventional therapy (with 61% to 64% receiving corticosteroids, 50% to 57% immunosuppressants and 46% to 52% aminosalicylates at baseline).

Rates of and duration of response, relapse and remission

Clinical response and remission data were well reported across the included trials for the adult population. It was assumed by the Assessment Group that the numbers of patients who were reported in the trial publications as being in clinical response also included those patients who were in clinical remission. Data relating to transitions of patients between no response, response and remission categories at maintenance time points were requested and received from the manufacturers (MSD and AbbVie). The induction trial data (as reported in the trial publications) and maintenance transition data (received from the manufacturers) from eligible trials were analysed using NMA methods (see page 69). Definitions of clinical response and clinical remission used in the included trials are presented in Table 10.

Measures of disease activity

Mortality

Reported deaths for the included trials are presented in Appendix 4.

Rates of hospitalisation

A total of four included trials reported hospitalisation data for the adult population (ULTRA144 and ULTRA245 for ADA, ACT149 and ACT249 for IFX, no trials for GOL).

Rates of surgical intervention (both elective and emergency)

Six included trials in the adult population included information on rates of surgical intervention (ULTRA144 and ULTRA245 for ADA, PURSUIT-Maintenance48 for GOL, and ACT1,49 ACT249 and Probert et al. 50 for IFX). No trials reported whether surgical outcomes were elective or emergency in nature.

Adverse events of treatment (including leakage and infections following surgery)

The included trials report data relating to AEs associated with the interventions under assessment only (i.e. IFX, ADA and GOL) and do not report safety outcomes (e.g. leakage and infections) post surgery. However, although the clinical effectiveness systematic review does not take these factors into account, these factors are relevant to the economic analysis (see Chapter 4). p-values are provided when available; however, the statistical significance of observed differences in safety outcomes was poorly reported across the included trials.

Adalimumab

Injection site reactions were slightly more frequent at week 8 of ULTRA144 among patients receiving 160 mg/80 mg of ADA (13/223, 5.8%) than PBO (7/223, 3.1%). 44 Injection site reactions were also more frequent in the 160 mg/80 mg of ADA group at week 52 of ULTRA245 (31/257, 12.1%) than for PBO (10/260, 3.8%). 45 Patients receiving ADA through week 52 of ULTRA3 experienced injection site reactions at a rate of 10.5 per 100 PYs. 62 Injection site reactions were more frequent through week 8 of the Suzuki et al. 46 trial in the 160 mg/80 mg of ADA group (7/90, 7.8%) than for PBO (2/96, 2.1%). 46

No serious allergic reactions were described as having occurred in the included ADA trials.

Golimumab

At week 6 of the PURSUIT-SC47 trial, injection site reactions were more common in patients receiving 200/100 mg of GOL induction (11/331, 3.3%) than PBO (5/330, 1.5%). 47 The number of patients reporting one or more injection site reactions through week 54 of PURSUIT-Maintenance48 was higher in the 100 mg of GOL maintenance treatment arm (11/154, 7.1%) than 50 mg of GOL (3/154, 1.9%) and PBO (3/156, 1.9%). 47,48

No serious allergic reactions were reported.

Rates of and duration of response, relapse and remission

Table 12 presents the definitions of clinical response and remission in the included paediatric population RCT.

All enrolled patients received induction therapy with 5 mg/kg of IFX. At week 8, clinical response was reached by 44 out of 60 patients (73.3%), while 24 out of 60 (40.0%) of patients achieved clinical remission.

The PUCAI remission rates were evaluated at weeks 30 and 54. A greater proportion of patients in the 5 mg/kg of IFX every 8 weeks treatment group achieved PUCAI remission at week 30 (40.0% vs. 19.0%; p-values not reported) and week 54 (38.1% vs. 18.2%, p-values not reported) than the 5 mg/kg of IFX every 12 weeks group. At week 54, PUCAI remission without the use of corticosteroids was reported for 38.5% of the every 8 weeks group and 0% of the every 12 weeks group. 52

The absence of a PBO or non-IFX control group limits the comparative evaluation of the efficacy of IFX in induction and maintenance of clinical response and remission in paediatric patients. A briefing document68 by Centocor Ltd to the FDA Gastrointestinal Drugs Committee was produced in June 2011 and considered the evidence available from the Hyams et al. 2012/T72 study52 and compared this with the ACT1 and ACT2 trials49 of IFX previously conducted in the adult UC population. The briefing document considered efficacy to be similar between T72 and the ACT1 and ACT2 studies during (1) induction (with clinical response and Mayo remission at week 8 induced in 73.3% and 40.0% of paediatric patients and 66.9% and 36.4% of pooled 5 mg/kg adult patients from ACT1 and ACT2, respectively) and (2) maintenance (with PUCAI remission at week 54 in 38.1% of paediatric subjects in the every 8 weeks group and 34.7% at week 54 of ACT1) (with reported good correlation of 0.75–0.88 between PUCAI and Mayo scores described at baseline and week 8).

Measures of disease activity

At week 8 of the Hyams et al. 52 study, the median reductions in partial Mayo scores were 4 points for both the 5 mg/kg of IFX every 8 weeks group and 5 mg/kg of IFX every 12 weeks group. 52 By week 30, the median reduction in partial Mayo score was approximately 3 points for the every 8 weeks group and 1 point for the every 12 weeks group. 52

Mortality

No deaths were reported in the Hyams et al. 52 trial.

Rates of hospitalisation

No hospitalisation-related outcome data were reported in Hyams et al. 52

Rates of surgical intervention (both elective and emergency)

One of 22 patients (4.5%) in the 5 mg/kg of IFX every 8 weeks group required colectomy through week 54 in the Hyams et al. 52 trial as compared with 2 out of 23 (8.7%) patients in the 5 mg/kg of IFX every 12 weeks treatment arm.

Colectomy rates during maintenance in children were reported by one trial evaluating the licensed dose of IFX every 8 weeks or every 12 weeks (Hyams et al. ,52 45 participants). The between-group at week 54 was not significant [RR = 0.52 (random effects), 95% CI 0.05 to 5.36; p = 0.59; Figure 12].

FIGURE 12.

Forest plot of comparison: colectomy – children. Outcome: IFX maintenance. M–H, Mantel–Haenszel; q8w, every 8 weeks; q12w, every 12 weeks.

Time to surgical intervention (both elective and emergency)

No data were reported in the paediatric population for the outcome of time to surgical intervention.

Health-related quality of life

No HRQoL data were included in the Hyams et al. 52 trial.

Adverse events of treatment (including leakage and infections following surgery)

Base case: induction phase

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the induction phase. Data were available from five studies comparing two treatments. 44,45,47,49 Figure 13 presents the network of evidence for the base-case induction phase.

FIGURE 13.

Base case: network of evidence for the induction phase. Note: solid line indicates a two-arm trial.

Figure 14 presents the effects of each treatment relative to PBO on the probit scale for the base-case induction phase. Figure 15 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 18.16, being close to the total number of data points included in the analysis, 20. The between-study SD was estimated to be 0.12 [95% credible interval (CrI) 0.01 to 0.50], which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 14.

Base case: comparative effect of anti-TNF-α treatment on clinical response/remission in the induction phase [SD ∼ HN(0,0.32²)].

FIGURE 15.

Base case: ranking probability histograms for the induction phase.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX. All treatment effects were statistically significant at a conventional 5% level. IFX was associated with the greatest effect –0.92 (95% CrI –1.27 to –0.56) and was most likely to be the most effective treatment (probability of being the best = 0.93).

Table 13 presents the probabilities of achieving each of the following categories: no response, response and remission for the base-case induction phase. IFX was associated with the highest probability of moving from no response to response and no response to remission respectively. The effects of ADA and GOL on each transition probability were comparable.

Base case: maintenance phase 8–32 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 8–32 weeks for patients starting in response. Data were available from four studies comparing two or three treatments. 45,48,49 Figure 16 presents the network of evidence for the base-case maintenance phase at 8–32 weeks for patients starting in response.

FIGURE 16.

Base case: network of evidence for the maintenance phase at 8–32 weeks for patients starting in response. Note: solid line indicates a two-arm trial and dashed line indicates a three-arm study.

Figure 17 presents the effects of each treatment relative to PBO on the probit scale for the base-case maintenance phase at 8–32 weeks for patients starting in response. Figure 18 presents the probabilities of treatment rankings for this analysis. There was some suggestion that the model did not represent the data well with the total residual deviance, 11.73, being smaller than would be expected given the total number of data points included in the analysis, 18. The probability of observing a value < 11.73 was 0.139, which means that it could be a chance event. All four studies had smaller residual deviances than expected (ULTRA2: deviance 3.0 compared with four data points;45 ACT1: deviance 2.1 compared with four data points;49 ACT2: deviance 2.66 compared with four data points;49 and PURSUIT: deviance 4.0 compared with six data points). 48 The between-study SD was estimated to be 0.17 (95% CrI 0.01 to 0.61), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 17.

Base case: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 18.

Base case: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in response. Note: the horizontal axis represents the rank of each treatment, i.e., from the best rank (left hand side) to the worst rank (right hand side) within each treatment.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with 100 mg of GOL. However, none of the treatment effects were statistically significant at a conventional 5% level. 100 mg of GOL was associated with the greatest effect –0.42 (95% CrI –0.78 to 0.29) and was most likely to be the most effective treatment (probability of being the best = 0.47).

Table 14 presents the probabilities of achieving each of the following categories: no response, response and remission for the base-case maintenance phase at 8–32 weeks for patients starting in response. 100 mg of GOL was associated with the highest probability of moving from response to remission and staying in the response state at 8–32 weeks. It was also associated with the smallest probability of moving from response to no response. The probabilities of staying in response were comparable among all treatments at 8–32 weeks.

TABLE 14 - Base case: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.524	0.525	0.426 to 0.622	0.270	0.270	0.198 to 0.341	0.206	0.202	0.117 to 0.311
ADA	0.512	0.512	0.230 to 0.782	0.261	0.267	0.140 to 0.354	0.227	0.211	0.055 to 0.493
50 mg of GOL	0.403	0.399	0.173 to 0.660	0.283	0.285	0.176 to 0.374	0.313	0.303	0.108 to 0.588
100 mg of GOL	0.368	0.360	0.149 to 0.619	0.285	0.288	0.176 to 0.377	0.347	0.338	0.129 to 0.623
IFX	0.432	0.430	0.220 to 0.659	0.282	0.283	0.189 to 0.371	0.286	0.276	0.109 to 0.518

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 8–32 weeks for patients starting in remission. Data were available from four studies comparing two or three treatments. 45,48,49 Figure 19 presents the network of evidence for the base-case maintenance phase at 8–32 weeks for patients starting in remission.

FIGURE 19.

Base case: network of evidence for the maintenance phase at 8–32 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 20 presents the effects of each treatment relative to PBO on the probit scale for the base-case maintenance phase at 8–32 weeks for patients starting in remission. Figure 21 presents the probabilities of treatment rankings for this analysis. The model fitted the data well, with the total residual deviance, 18.20, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.18 (95% CrI 0.01 to 0.64), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 20.

Base case: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 21.

Base case: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in remission.

All treatments except ADA were associated with beneficial treatment effects relative to PBO with the greatest effects being associated with 50 mg of GOL (–0.63, 95% CrI –1.36 to 0.11) and 100 mg of GOL (–0.61, 95% CrI –1.32 to 0.11). However, none of the treatment effects was statistically significant at a conventional 5% level. 50 mg and 100 mg of GOL was most likely to be the most effective treatments (probability of being the best = 0.47 and 0.42 respectively).

Table 15 presents the probabilities of achieving each of the following categories: no response, response and remission for the base-case maintenance phase at 8–32 weeks for patients starting in remission. 50 mg and 100 mg of GOL were associated with the highest probability of staying in remission and the smallest probability of moving from remission to response or remission no response at 8–32 weeks.

TABLE 15 - Base case: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in remission

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.353	0.347	0.168 to 0.572	0.180	0.174	0.070 to 0.316	0.467	0.466	0.225 to 0.708
ADA	0.428	0.420	0.099 to 0.803	0.166	0.164	0.053 to 0.297	0.406	0.392	0.083 to 0.804
50 mg of GOL	0.177	0.152	0.027 to 0.457	0.136	0.131	0.028 to 0.283	0.687	0.708	0.321 to 0.933
100 mg of GOL	0.182	0.158	0.029 to 0.469	0.138	0.134	0.030 to 0.285	0.680	0.700	0.322 to 0.929
IFX	0.325	0.309	0.084 to 0.648	0.169	0.165	0.057 to 0.304	0.506	0.509	0.178 to 0.829

Base case: maintenance phase 32–52 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 32–52 weeks. Data were available from three studies comparing two or three treatments. 45,48,49 Figure 22 presents the network of evidence for the base-case maintenance phase at 32–52 weeks for patients starting in response.

FIGURE 22.

Base case: network of evidence for the maintenance phase at 32–52 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 23 presents the effects of each treatment relative to PBO on the probit scale for the base-case maintenance phase 32–52 weeks for patients starting in response. Figure 24 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 12.88, being close to the total number of data points included in the analysis, 14. The between-study SD was estimated to be 0.21 (95% CrI 0.01 to 0.71), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 23.

Base case: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 24.

Base case: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in response.

All treatments except 100 mg of ADA and GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with 50 mg of GOL; however, none of the treatment effects was statistically significant at a conventional 5% level. IFX was associated with the greatest effect –0.36 (95% CrI –1.33 to 0.62) and was most likely to be the most effective treatment (probability of being the best = 0.56).

Table 16 presents the probabilities of achieving each of the following categories: no response, response and remission for the base-case maintenance phase at 32–52 weeks for patients starting in response. IFX was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 32–52 weeks. The probabilities of staying in the response state were comparable among treatments at 32–52 weeks.

TABLE 16 - Base case: probabilities of being in each category for the maintenance phase at 32–52 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.338	0.319	0.066 to 0.711	0.370	0.378	0.122 to 0.604	0.292	0.259	0.027 to 0.717
ADA	0.450	0.440	0.063 to 0.889	0.327	0.340	0.067 to 0.562	0.223	0.167	0.005 to 0.716
50 mg of GOL	0.295	0.258	0.025 to 0.750	0.353	0.363	0.081 to 0.616	0.352	0.319	0.021 to 0.842
100 mg of GOL	0.410	0.393	0.055 to 0.852	0.342	0.353	0.083 to 0.581	0.248	0.199	0.009 to 0.741
IFX	0.250	0.205	0.013 to 0.716	0.341	0.353	0.065 to 0.621	0.409	0.385	0.029 to 0.892

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 32–52 weeks for patients starting in remission. Data were available from three studies comparing two or three treatments. 45,48,49 Figure 25 presents the network of evidence for the base-case maintenance phase at 32–52 weeks for patients starting in remission.

FIGURE 25.

Base case: network of evidence for the maintenance phase at 32–52 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 26 presents the effects of each treatment relative to PBO on the probit scale for the base-case maintenance phase at 32–52 weeks for patients starting in remission. Figure 27 presents the probabilities of treatment rankings for this analysis. The model fitted the data well, with the total residual deviance, 18.46, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.21 (95% CrI 0.01 to 0.72), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 26.

Base case: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 27.

Base case: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in remission.

All treatments except 50 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with ADA. However, only the treatment effects of ADA were statistically significant at a conventional 5% level. ADA was associated with the greatest effect –1.04 (95% CrI –1.93 to –0.12) and was most likely to be the most effective treatment (probability of being the best = 0.84).

Table 17 presents the probabilities of achieving each of the following categories: no response, response and remission for the base-case maintenance phase at 32–52 weeks for patients starting in remission. ADA was associated with the highest probability of staying in remission and the smallest probability of moving from remission to response or from remission to no response at 32–52 weeks.

TABLE 17 - Base case: probabilities of being in each category for the maintenance phase at 32–52 weeks for patients starting in remission

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.301	0.296	0.174 to 0.449	0.164	0.147	0.029 to 0.449	0.536	0.548	0.237 to 0.734
ADA	0.081	0.059	0.005 to 0.288	0.084	0.061	0.005 to 0.337	0.834	0.874	0.447 to 0.985
50 mg of GOL	0.329	0.314	0.080 to 0.664	0.155	0.141	0.024 to 0.415	0.515	0.523	0.135 to 0.851
100 mg of GOL	0.266	0.245	0.052 to 0.604	0.147	0.132	0.020 to 0.417	0.587	0.604	0.169 to 0.894
IFX	0.247	0.220	0.033 to 0.613	0.140	0.126	0.017 to 0.413	0.613	0.634	0.174 to 0.928

Sensitivity analysis 1: induction phase

Sensitivity analysis 1 involved replacing ULTRA2 anti-TNF-α-naive data with ULTRA2 ITT data. A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the induction phase. Data were available from five studies comparing two treatments. 44,45,47,49 Figure 28 presents the network of evidence for the sensitivity analysis 1 induction phase.

FIGURE 28.

Sensitivity analysis 1: network of evidence for the induction phase. Note: solid line indicates a two-arm trial.

Figure 29 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 1 induction phase. Figure 30 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 17.08, being close to the total number of data points included in the analysis, 20. The between-study SD was estimated to be 0.11 (95% CrI 0.01 to 0.47), which implies mild heterogeneity between studies in treatment effects.

FIGURE 29.

Sensitivity analysis 1: comparative effect of anti-TNF-α treatment on clinical response/remission in the induction phase [SD ∼ HN(0,0.32²)].

FIGURE 30.

Sensitivity analysis 1: ranking probability histograms for the induction phase.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.91, 95% CrI –1.25 to –0.57). All treatment effects were statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.94).

Table 18 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 1 induction phase. IFX was associated with the highest probability of moving from no response to response and from no response to remission.

Sensitivity analysis 1: maintenance phase 8–32 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 8–32 weeks. Data were available from four studies comparing two or three treatments. 45,48,49 Figure 31 presents the network of evidence for the sensitivity analysis 1 maintenance phase at 8–32 weeks for patients starting in response.

FIGURE 31.

Sensitivity analysis 1: network of evidence for the maintenance phase at 8–32 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 32 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 1 maintenance phase at 8–32 weeks for patients starting in response. Figure 33 presents the probabilities of treatment rankings for this analysis. There was some suggestion that model did not represent the data well with the total residual deviance, 11.54, being smaller than would be expected given the total number of data points included in the analysis, 18. The probability of observing a value < 11.54 was 0.130, which means that this could be a chance event. Similar to the base-case analysis, all four studies had smaller residual deviances than expected. The between-study SD was estimated to be 0.17 (95% CrI 0.01 to 0.63), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 32.

Sensitivity analysis 1: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 33.

Sensitivity analysis 1: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in response.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with 100 mg of GOL (–0.41, 95% CrI –1.06 to 0.22); however, none of the treatment effects was statistically significant at a conventional 5% level. A treatment of 100 mg of GOL was most likely to be the most effective (probability of being the best = 0.43).

Table 19 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 1 maintenance phase at 8–32 weeks starting in response. A treatment of 100 mg of GOL was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 8–32 weeks. The probabilities of staying in the response state were comparable among treatments.

TABLE 19 - Sensitivity analysis 1: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.548	0.548	0.447 to 0.649	0.269	0.267	0.199 to 0.361	0.183	0.181	0.085 to 0.282
ADA	0.468	0.467	0.210 to 0.744	0.279	0.283	0.158 to 0.391	0.252	0.240	0.059 to 0.525
50 mg of GOL	0.427	0.423	0.190 to 0.688	0.289	0.289	0.176 to 0.412	0.284	0.273	0.081 to 0.552
100 mg of GOL	0.390	0.384	0.162 to 0.649	0.293	0.292	0.182 to 0.421	0.318	0.310	0.098 to 0.591
IFX	0.453	0.451	0.237 to 0.685	0.286	0.287	0.186 to 0.403	0.260	0.252	0.082 to 0.490

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 8–32 weeks for patients starting in remission. Data were available from four studies comparing two or three treatments. 45,48,49 Figure 34 presents the network of evidence for the sensitivity analysis 1 maintenance phase at 8–32 weeks for patients starting in remission.

FIGURE 34.

Sensitivity analysis 1: network of evidence for the maintenance phase at 8–32 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 35 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 1 maintenance phase at 8–32 weeks for patients starting in remission. Figure 36 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 15.29, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.19 (95% CrI 0.01 to 0.65), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 35.

Sensitivity analysis 1: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 36.

Sensitivity analysis 1: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in remission.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effects being associated with 50 mg of GOL (–0.62, 95% CrI –1.36 to 0.11) and 100 mg of GOL (–0.61, 95% CrI –1.34 to 0.13). However, none of the treatment effects was statistically significant at a conventional 5% level. A treatment of 50 mg and 100 mg of GOL were most likely to be the most effective (probability of being the best = 0.44 and 0.41 respectively).

Table 20 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 1 maintenance phase at 8–32 weeks starting in remission. A treatment of 50 mg and 100 mg of GOL were associated with the highest probability of staying in remission and the smallest probability of moving from remission to no response and from remission to response at 8–32 weeks.

TABLE 20 - Sensitivity analysis 1: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in remission

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.392	0.389	0.217 to 0.584	0.180	0.175	0.078 to 0.309	0.428	0.426	0.218 to 0.650
ADA	0.379	0.365	0.096 to 0.736	0.167	0.164	0.06 to 0.3	0.454	0.450	0.128 to 0.807
50 mg of GOL	0.204	0.182	0.036 to 0.493	0.143	0.139	0.036 to 0.285	0.653	0.669	0.300 to 0.914
100 mg of GOL	0.207	0.188	0.038 to 0.494	0.144	0.140	0.037 to 0.287	0.648	0.662	0.303 to 0.911
IFX	0.359	0.347	0.107 to 0.679	0.170	0.166	0.065 to 0.301	0.471	0.470	0.165 to 0.793

Sensitivity analysis 1: maintenance phase 32–52 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 32–52 weeks. Data were available from three studies comparing two or three treatments. 45,48,49 Figure 37 presents the network of evidence for the sensitivity analysis 1 maintenance phase at 32–52 weeks for patients starting in response.

FIGURE 37.

Sensitivity analysis 1: network of evidence for the maintenance phase at 32–52 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 38 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 1 maintenance phase 32–52 weeks for patients starting in response. Figure 39 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 12.32, being close to the total number of data points included in the analysis, 14. The between-study SD was estimated to be 0.21 (95% CrI 0.01 to 0.72), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 38.

Sensitivity analysis 1: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 39.

Sensitivity analysis 1: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in response.

All treatments except 100 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.37; 95% CrI –1.30 to 0.59); however, none of the treatment effects was statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.52).

Table 21 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 1 maintenance phase at 32–52 weeks starting in response. IFX was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 32–52 weeks. The probabilities of staying in the response state were comparable among treatments at 32–52 weeks.

TABLE 21 - Sensitivity analysis 1: probabilities of being in each category for the maintenance phase at 32–52 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.338	0.317	0.064 to 0.718	0.373	0.378	0.118 to 0.628	0.29	0.259	0.019 to 0.714
ADA	0.332	0.300	0.031 to 0.790	0.354	0.363	0.084 to 0.625	0.314	0.276	0.013 to 0.812
50 mg of GOL	0.302	0.269	0.024 to 0.769	0.355	0.364	0.077 to 0.632	0.344	0.308	0.015 to 0.844
100 mg of GOL	0.417	0.401	0.053 to 0.854	0.341	0.352	0.077 to 0.595	0.242	0.192	0.006 to 0.742
IFX	0.249	0.201	0.012 to 0.730	0.343	0.353	0.063 to 0.643	0.408	0.387	0.022 to 0.898

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 32–52 weeks for patients starting in remission. Data were available from three studies comparing two or three treatments. 45,48,49 Figure 40 presents the network of evidence for the sensitivity analysis 1 maintenance phase at 32–52 weeks for patients starting in remission.

FIGURE 40.

Sensitivity analysis 1: network of evidence for the maintenance phase at 32–52 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 41 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 1 maintenance phase at 32–52 weeks for patients starting in remission. Figure 42 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 17.73, being close to the total number of data points included in the analysis, 14. The between-study SD was estimated to be 0.22 (95% CrI 0.01 to 0.72), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 41.

Sensitivity analysis 1: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 42.

Sensitivity analysis 1: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in remission.

All treatments except 50 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with ADA (–0.86; 95% CrI –1.71 to 0.00); however, only the treatment effects of ADA were statistically significant at a conventional 5% level. ADA was most likely to be the most effective treatment (probability of being the best = 0.78).

Table 22 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 1 maintenance phase at 32–52 weeks starting in remission. ADA was associated with the highest probability of staying in remission and the smallest probability of moving from remission to response or from remission to no response at 32–52 weeks.

Sensitivity analysis 2: induction phase

Sensitivity analysis 2 involved including Suzuki et al. 46 A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the induction phase. Data were available from six studies comparing two treatments. 44–47,49 Figure 43 presents the network of evidence for the sensitivity analysis 2 induction phase.

FIGURE 43.

Sensitivity analysis 2: network of evidence for the induction phase. Note: solid line indicates a two-arm trial.

Figure 44 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 2 induction phase. Figure 45 presents the probabilities of treatment rankings for this analysis. The model fitted the data well, with the total residual deviance, 24.36, being close to the total number of data points included in the analysis, 24. The between-study SD was estimated to be 0.10 (95% CrI 0.01 to 0.41), which implies mild heterogeneity between studies in treatment effects.

FIGURE 44.

Sensitivity analysis 2: comparative effect of anti-TNF-α treatment on clinical response/remission in the induction phase [SD ∼ HN(0,0.32²)].

FIGURE 45.

Sensitivity analysis 2: ranking probability histograms for the induction phase.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.92, 95% CrI –1.24 to –0.60). All treatment effects were statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.96).

Table 23 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 2 induction phase. IFX was associated with the highest probability of moving from no response to response and from no response to remission.

Sensitivity analysis 2: maintenance phase 8–32 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase response at 8–32 weeks for patients starting in response. Data were available from five studies comparing two or three treatments. 45,46,48,49 Figure 46 presents the network of evidence for the sensitivity analysis 2 maintenance phase at 8–32 weeks for patients starting in response.

FIGURE 46.

Sensitivity analysis 2: network of evidence for the maintenance phase at 8–32 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 47 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 2 maintenance phase at 8–32 weeks for patients starting in response. Figure 48 presents the probabilities of treatment rankings for this analysis. There was some suggestion that model did not represent the data well with the total residual deviance, 14.80, being smaller than would be expected given the total number of data points included in the analysis, 22. The probability of observing a value < 14.8 is 0.129, which means that this could be a chance event. Similar to the base-case analysis, all five studies had smaller residual deviance than expected. The between-study SD was estimated to be 0.16 (95% CrI 0.01 to 0.58), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 47.

Sensitivity analysis 2: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 48.

Sensitivity analysis 2: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in response.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with 100 mg of GOL (–0.43; 95% CrI–1.03 to 0.19); however, none of the treatment effects was statistically significant at a conventional 5% level. A treatment of 100 mg of GOL was most likely to be the most effective (probability of being the best = 0.44).

Table 24 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 2 maintenance phase at 8–32 weeks for patients starting in response. A treatment of 100 mg of GOL was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 8–32 weeks. The probabilities of staying in the response state were comparable among treatments.

TABLE 24 - Sensitivity analysis 2: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.525	0.525	0.437 to 0.609	0.274	0.274	0.220 to 0.329	0.201	0.199	0.130 to 0.286
ADA	0.441	0.441	0.238 to 0.647	0.286	0.288	0.209 to 0.354	0.273	0.263	0.116 to 0.485
50 mg of GOL	0.398	0.393	0.174 to 0.649	0.289	0.292	0.199 to 0.357	0.313	0.305	0.116 to 0.569
100 mg of GOL	0.363	0.356	0.158 to 0.617	0.291	0.293	0.200 to 0.360	0.346	0.338	0.132 to 0.598
IFX	0.429	0.425	0.222 to 0.657	0.287	0.289	0.204 to 0.352	0.284	0.276	0.110 to 0.504

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 8–32 weeks for patients starting in remission. Data were available from five studies comparing two or three treatments. 45,46,48,49 Figure 49 presents the network of evidence for the sensitivity analysis 2 maintenance phase at 8–32 weeks for patients starting in remission.

FIGURE 49.

Sensitivity analysis 2: network of evidence for the maintenance phase at 8–32 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 50 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 2 maintenance phase at 8–32 weeks for patients starting in remission. Figure 51 presents the probabilities of treatment rankings for this analysis. The model fitted the data well, with the total residual deviance, 21.05, being close to the total number of data points included in the analysis, 22. The between-study SD was estimated to be 0.17 (95% CrI 0.01 to 0.60), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 50.

Sensitivity analysis 2: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 51.

Sensitivity analysis 2: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in remission.

All treatments except ADA were associated with beneficial treatment effects relative to PBO with the greatest effects being associated with 50 mg of GOL (–0.61; 95% CrI –1.30 to 0.09) and 100 mg of GOL (–0.60; 95% CrI –1.29 to 0.09); however, none of the treatment effects was statistically significant at a conventional 5% level. Treatments of 50 mg and 100 mg of GOL was most likely to be the most effective (probability of being the best = 0.46 and 0.44 respectively).

Table 25 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 2 maintenance phase at 8–32 weeks starting in remission. Treatments of 50 mg and 100 mg of GOL were associated with the highest probability of staying in remission and the smallest probability of moving from remission to no response and from remission to response at 8–32 weeks.

Sensitivity analysis 2: maintenance phase 32–52 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 32–52 weeks. Data were available from four studies comparing two or three treatments. 45,46,48,49 Figure 52 presents the network of evidence for the sensitivity analysis 2 maintenance phase at 32–52 weeks for patients starting in response.

FIGURE 52.

Sensitivity analysis 2: network of evidence for the maintenance phase for patients starting in response at 32–52 weeks. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 53 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 2 maintenance phase 32–52 weeks for patients starting in response. Figure 54 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 15.92, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.20 (95% CrI 0.01 to 0.67), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 53.

Sensitivity analysis 2: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 54.

Sensitivity analysis 2: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in response.

All treatments except ADA and 100 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.36; 95% CrI –1.29 to 0.58); however, none of the treatment effects was statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.57).

Table 26 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 2 maintenance phase at 32–52 weeks for patients starting in response. IFX was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 32–52 weeks. The probabilities of staying in response were comparable among treatments at 32–52 weeks.

TABLE 26 - Sensitivity analysis 2: probabilities of being in each category for the maintenance phase at 32–52 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.322	0.318	0.189 to 0.472	0.393	0.398	0.200 to 0.544	0.286	0.278	0.107 to 0.520
ADA	0.371	0.363	0.130 to 0.661	0.371	0.378	0.176 to 0.524	0.259	0.238	0.05 to 0.576
50 mg of GOL	0.283	0.265	0.066 to 0.606	0.370	0.378	0.153 to 0.545	0.347	0.332	0.069 to 0.713
100 mg of GOL	0.404	0.395	0.128 to 0.730	0.359	0.368	0.153 to 0.521	0.237	0.211	0.035 to 0.579
IFX	0.230	0.202	0.030 to 0.575	0.352	0.363	0.115 to 0.542	0.418	0.410	0.081 to 0.820

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 32–52 weeks for patients starting in remission. Data were available from four studies comparing two or three treatments. 45,48,49 Figure 55 presents the network of evidence for the sensitivity analysis 2 maintenance phase at 32–52 weeks for patients starting in remission.

FIGURE 55.

Sensitivity analysis 2: network of evidence for the maintenance phase at 32–52 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 56 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 2 maintenance phase at 32–52 weeks for patients starting in remission. Figure 57 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 21.07, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.18 (95% CrI 0.01 to 0.65), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 56.

Sensitivity analysis 2: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 57.

Sensitivity analysis 2: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in remission.

All treatments except 50 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with ADA (–0.93, 95% CrI –1.59 to –0.25); however, only the effect of ADA was statistically significant at a conventional 5% level. ADA was most likely to be the most effective treatment (probability of being the best = 0.84).

Table 27 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 2 maintenance phase at 32–52 weeks starting in remission. ADA was associated with the highest probability of staying in remission and the smallest probability of moving from remission to response or no response at 32–52 weeks.

Sensitivity analysis 3: induction phase

Sensitivity analysis 3 involved replacing ULTRA2 anti-TNF-α-naive data with ULTRA2 ITT data and including data from Suzuki et al. 46 A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the induction phase. Data were available from six studies comparing two treatments. 44–47,49 Figure 58 presents the network of evidence for the sensitivity analysis 3 induction phase.

FIGURE 58.

Sensitivity analysis 3: network of evidence for the induction phase. Note: solid line indicates a two-arm trial.

Figure 59 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 3 induction phase. Figure 60 presents the probabilities of treatment rankings for this analysis. The model fitted the data well, with the total residual deviance, 23.63, being close to the total number of data points included in the analysis, 24. The between-study SD was estimated to be 0.09 (95% CrI 0.00 to 0.38), which implies mild heterogeneity between studies in treatment effects.

FIGURE 59.

Sensitivity analysis 3: comparative effect of anti-TNF-α treatment on clinical response/remission in the induction phase [SD ∼ HN(0,0.32²)].

FIGURE 60.

Sensitivity analysis 3: ranking probability histograms for the induction phase.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.91; 95% CrI –1.21 to –0.62). All the treatment effects were statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.97).

Table 28 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 3 induction phase. IFX was associated with the highest probability of moving from no response to response and no response to remission.

Sensitivity analysis 3: maintenance phase 8–32 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 8–32 weeks. Data were available from five studies comparing two or three treatments. 45,46,48,49 Figure 61 presents the network of evidence for the sensitivity analysis 3 maintenance phase at 8–32 weeks for patients starting in response.

FIGURE 61.

Sensitivity analysis 3: network of evidence for the maintenance phase at 8–32 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 62 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 3 maintenance phase at 8–32 weeks for patients starting in response. Figure 63 presents the probabilities of treatment rankings for this analysis. There was some suggestion that model did not represent the data well with the total residual deviance, 14.05, being smaller than would be expected given the total number of data points included in the analysis, 22. The probability of observing a value < 14.05 is 0.100, which means that this could have occurred by chance. Similar to the base-case analysis, all 5 studies had smaller residual deviance than expected. The between-study SD was estimated to be 0.15 (95% CrI 0.01 to 0.55), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 62.

Sensitivity analysis 3: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 63.

Sensitivity analysis 3: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in response.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with 100 mg of GOL; however, none of the treatment effects was statistically significant at a conventional 5% level. A treatment of 100 mg of GOL was most likely to be the most effective (probability of being the best = 0.40).

Table 29 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 3 maintenance phase at 8–32 weeks starting in response. A treatment of 100 mg of GOL was associated with the highest probability of moving from response to remission and the smallest probability of moving from response to no response at 8–32 weeks. The probabilities of staying in response were comparable among treatments.

TABLE 29 - Sensitivity analysis 3: probabilities of being in each category for the maintenance phase at 8–32 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.545	0.545	0.459 to 0.630	0.270	0.270	0.223 to 0.32	0.185	0.183	0.120 to 0.260
ADA	0.427	0.425	0.243 to 0.626	0.292	0.293	0.223 to 0.353	0.280	0.274	0.129 to 0.470
50 mg of GOL	0.425	0.422	0.202 to 0.669	0.289	0.290	0.204 to 0.354	0.287	0.276	0.107 to 0.523
100 mg of GOL	0.386	0.382	0.176 to 0.628	0.293	0.294	0.213 to 0.356	0.321	0.313	0.129 to 0.560
IFX	0.451	0.449	0.246 to 0.664	0.287	0.289	0.211 to 0.349	0.262	0.255	0.109 to 0.465

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 8–32 weeks for patients starting in remission. Data were available from five studies comparing two or three treatments. 45,46,48,49 Figure 64 presents the network of evidence for the sensitivity analysis 3 maintenance phase at 8–32 weeks for patients starting in remission.

FIGURE 64.

Sensitivity analysis 3: network of evidence for maintenance phase at 8–32 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 65 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 3 maintenance phase at 8–32 weeks for patients starting in remission. Figure 66 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 17.77, being close to the total number of data points included in the analysis, 22. The between-study SD was estimated to be 0.16 (95% CrI 0.01 to 0.59), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 65.

Sensitivity analysis 3: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 8–32 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 66.

Sensitivity analysis 3: ranking probability histograms for the maintenance phase at 8–32 weeks for patients starting in remission.

All treatments were associated with beneficial treatment effects relative to PBO with the greatest effects being associated with 50 mg of GOL (–0.62, 95% CrI –1.33 to 0.06) and 100 mg of GOL (–0.61, 95% CrI –1.29 to 0.07); however, none of the treatment effects was statistically significant at a conventional 5% level. A treatment of 50 mg and 100 mg of GOL were most likely to be the most effective treatments (probability of being the best = 0.46 and 0.44 respectively).

Table 30 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 3 maintenance phase at 8–32 weeks starting in remission. A treatment of 50 mg and 100 mg of GOL were associated with the highest probability of staying in remission and the smallest probability of moving from remission to no response and from remission to response at 8–32 weeks.

Sensitivity analysis 3: maintenance phase 32–52 weeks

Patients starting in response

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase for patients starting in response at 32–52 weeks. Data were available from four studies comparing two or three treatments. 45,46,48,49 Figure 67 presents the network of evidence for the sensitivity analysis 3 maintenance phase at 32–52 weeks for patients starting in response.

FIGURE 67.

Sensitivity analysis 3: network of evidence for the maintenance phase at 32–52 weeks for patients starting in response. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 68 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 3 maintenance phase 32–52 weeks for patients starting in response. Figure 69 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 15.21, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.18 (95% CrI 0.01 to 0.64), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 68.

Sensitivity analysis 3: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in response [SD ∼ HN(0,0.32²)].

FIGURE 69.

Sensitivity analysis 3: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in response.

All treatments except 100 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with IFX (–0.38, 95% CrI –1.27 to 0.55). However, none of the treatment effects was statistically significant at a conventional 5% level. IFX was most likely to be the most effective treatment (probability of being the best = 0.55).

Table 31 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 3 maintenance phase at 32–52 weeks for patients starting in response. IFX was associated with the highest probability of moving from response to remission, and the smallest probability of moving from response to no response at 32–52 weeks. The probabilities of staying in response were comparable among treatments at 32–52 weeks.

TABLE 31 - Sensitivity analysis 3: probabilities of being in each category for the maintenance phase at 32–52 weeks for patients starting in response

Treatment	No response			Response			Remission
	Mean	Median	95% CrI	Mean	Median	95% CrI	Mean	Median	95% CrI
PBO	0.344	0.340	0.218 to 0.484	0.393	0.395	0.252 to 0.514	0.263	0.257	0.114 to 0.449
ADA	0.314	0.304	0.109 to 0.580	0.382	0.385	0.219 to 0.520	0.305	0.290	0.086 to 0.605
50 mg of GOL	0.309	0.293	0.081 to 0.625	0.374	0.380	0.191 to 0.514	0.317	0.302	0.067 to 0.661
100 mg of GOL	0.436	0.431	0.149 to 0.759	0.354	0.363	0.172 to 0.490	0.210	0.187	0.031 to 0.520
IFX	0.240	0.213	0.041 to 0.575	0.358	0.369	0.151 to 0.513	0.402	0.392	0.088 to 0.771

Patients starting in remission

A NMA was used to compare the effects of ADA, GOL and IFX relative to PBO on clinical response in the maintenance phase at 32–52 weeks for patients starting in remission. Data were available from four studies comparing two or three treatments. 45,46,48,49 Figure 70 presents the network of evidence for the sensitivity analysis 3 maintenance phase at 32–52 weeks for patients starting in remission.

FIGURE 70.

Sensitivity analysis 3: network of evidence for the maintenance phase at 32–52 weeks for patients starting in remission. Note: solid line indicates a two-arm trial; dashed line indicates a three-arm study.

Figure 71 presents the effects of each treatment relative to PBO on the probit scale for the sensitivity analysis 3 maintenance phase at 32–52 weeks for patients starting in remission. Figure 72 presents the probabilities of treatment rankings for this analysis. The model fitted the data reasonably well, with the total residual deviance, 20.55, being close to the total number of data points included in the analysis, 18. The between-study SD was estimated to be 0.18 (95% CrI 0.01 to 0.65), which implies mild to moderate heterogeneity between studies in treatment effects.

FIGURE 71.

Sensitivity analysis 3: comparative effect of anti-TNF-α treatment on clinical response/remission in the maintenance phase at 32–52 weeks for patients starting in remission [SD ∼ HN(0,0.32²)].

FIGURE 72.

Sensitivity analysis 3: ranking probability histograms for the maintenance phase at 32–52 weeks for patients starting in remission.

All treatments except 50 mg of GOL were associated with beneficial treatment effects relative to PBO with the greatest effect being associated with ADA (–0.85, 95% CrI –1.49 to –0.16); however, only the effect of ADA was statistically significant at a conventional 5% level. ADA was most likely to be the most effective treatment (probability of being the best = 0.80).

Table 32 presents the probabilities of achieving each of the following categories: no response, response and remission for the sensitivity analysis 3 maintenance phase at 32–52 weeks for patients starting in remission. ADA was associated with the highest probability of staying in remission and the smallest probability of moving from remission to response or no response at 32–52 weeks.

Study CT-P13 1.1, PLANET-AS

PLANET-AS was a prospective Phase I, randomised double-blind multicentre study, in which 250 patients were randomised (CT-P13 n = 125, Remicade n = 125). Patients received CT-P13 (5 mg/kg) or Remicade (5 mg/kg) at weeks 0, 2 and 6 and then every 8 weeks to week 54. The primary objective of the study was to demonstrate comparable pharmacokinetics of CT-P13 and Remicade at steady state (between weeks 22 and 30). The primary parameters evaluated were the area under the plasma concentration time curve from time zero to time t (AUC_T) and maximum serum concentrations (C_max) after dose 5 (weeks 22–30), with secondary parameters being average concentration at steady state (C_{av, ss}), minimum concentration immediately before next dose at steady state (C_{min ss}), terminal elimination half-life (T_1/2), total-body clearance at steady state (CL_ss) and volume of distribution at steady state (V_ss). Additional observed parameters were maximum serum concentrations (C_max), minimum concentration immediately before next dose (C_min) and time to reach maximum serum concentrations C_max (T_max) after each dose. Efficacy parameters included the proportion of patients achieving clinical response according to the assessment in AS response criteria (ASAS20 and ASAS40). The EPARs reported that PLANET-AS demonstrated that (at 5 mg/kg) pharmacokinetic behaviour between CT-P13 and Remicade was similar, a view that was also supported by pharmacokinetic data from RA patients in study CT-P13 3.1 (PLANET-RA). Furthermore, the EPARs also stated that the proportions of patients experiencing clinical response according to the ASAS20 and ASAS40 criteria at weeks 14 and 30 were similar across the CT-P13 and Remicade groups.

Study CT-P13 3.1, PLANET-RA

PLANET-RA was a prospective Phase III, randomised, double-blind, multicentre study, in which 302 patients were randomised to CT-P13 and 304 to Remicade (randomisation was stratified by geographical region and baseline CRP level). Patients received CT-P13 or Remicade at 3 mg/kg at weeks 0, 2, 6 and then every 8 weeks up to 54 weeks (administered in combination with a stable dose of methotrexate and folic acid). The primary objective of the trial was to demonstrate that CT-P13 was equivalent to Remicade up to week 30 in efficacy as measured by ACR20. Secondary objectives were ACR20, ACR50 and ACR70 responses at weeks 14 and 30, Disease Activity Score 28 at weeks 14 and 30, European League Against Rheumatism response at weeks 14 and 30, Simplified Disease Activity Index (SDAI) and Clinical Disease Activity Index at weeks 14 and 30 and SF-36 at weeks 14 and 30.

Fewer patients randomised to the CT-P13 arm (n = 69, 22.8%) discontinued PLANET-RA by week 54 than patients in Remicade arm (n = 82, 27.0%). Patients received CT-P13 and Remicade at the RA dose of 3 mg/kg. It was stated in the EPARs that a similar proportion of patients at week 30 in the CT-P13 (184/302, 60.9%) and Remicade (178/304, 58.6%) arms achieved ACR20 response (Table 33).

Furthermore, at week 30, the findings for the secondary end points (including ACR50, ACR70 and decrease in Disease Activity Score 28) were also described as being consistent with the results of the primary end point. Efficacy results were reported to be comparable between treatment arms up to week 54. It was concluded in the EPAR that PLANET-RA provided that robust evidence of therapeutic equivalence between CT-P13 and Remicade. ACR responses between CT-P13 and Remicade remained comparable through the 12-month PLANET-RA extension study. 73

The safety profile of CT-P13 was evaluated in the clinical studies described above. A total of 871 patients were included in the safety population. It was reported in the EPARs that the type and incidence of adverse drug reactions with CT-P13 and Remicade were broadly similar and that no new safety concern was identified. Additionally, it was stated that no marked differences in immunogenicity between CT-P13 and Remicade were observed up to 54 weeks, with comparable effects of antibodies on efficacy and safety. Although there was a numerical imbalance described in SAEs observed in study CT-P13 3.1 (PLANET-RA) (with a higher number of serious infections), reported numbers were stated to be low and, therefore, the CHMP concluded that this observed difference was likely to be due to chance.

In summary, the EMA considered CT-P13 to be biosimilar to the reference product Remicade and judged that the submitted data in the submissions for Remsima and Inflectra allowed for extrapolation to all other indications of Remicade.

An ECCO position statement was presented by Danese and Gomollon75 stating that the use of biosimilars in patients with IBD requires clinical trials in the IBD patient population to allow comparison between the biosimilar and reference products, on the basis of potential differences in manufacturing and structure that could lead to important differences in immunogenicity and efficacy. However, a subsequent statement issued on behalf of the Working Party on Similar Biological (Biosimilar) Medicinal Products of the CHMP argued that no pharmacokinetic and safety issues are known to be particular to IBD, the most responsive known population (RA) was assessed for immunogenicity and that the data submitted allow extrapolation to patients with IBD. 76

Park et al.83

Park et al. 83 report the methods and results of an economic analysis of early colectomy plus ileal pouch anal anastomosis (IPAA) versus standard medical therapy in patients with severe UC in the USA. The model population is intended to reflect 21-year old patients with newly diagnosed pancolitis UC confirmed by colonoscopic biopsies. The economic analysis compares two sequences of treatments: (1) immediate colectomy with IPAA; and (2) standard medical therapy, which is assumed to comprise a sequence of (i) 2 g of mesalamine per day; (ii) 125 mg of AZA per day; (iii) 5 mg/kg/dose of IFX every 8 weeks; (iv) 1.5 mg of tacrolimus b.i.d.; and (v) colectomy + IPAA. The analysis does not consider the comparative cost-effectiveness of alternative sequences of medical treatments. The authors purport to have adopted a societal perspective; however, it does not appear that any indirect costs borne outside the health sector (e.g. lost productivity or out-of-pocket expenses) have been included in the economic analysis. It would be more accurate to describe the adopted perspective as that of the health-care payer. Health economic results are presented in terms of the incremental cost per quality-adjusted life-year (QALY) gained over a lifetime horizon. In line with the reference case set out in Gold et al. ,85 costs and health outcomes were discounted at a rate of 3% per year. Costs were valued at 2009 prices.

The economic analysis uses a Markov approach to evaluate relevant events, costs and health outcomes. The duration of each Markov cycle is not entirely clear from the paper;83 the text indicates that the first cycle is 3 months in duration and the cycle length appears to be 8 weekly thereafter. However, the table of parameter values presented within the paper suggests that probability parameters are defined according to various time intervals. In addition, the text does not mention whether or not a half-cycle correction has been applied. The precise health states adopted in the model are also not entirely clear from the text; while a model diagram is presented in the paper, this details the sequences of treatments in each group but does not specify the relevant clinical events that patients may experience. It appears that separate states are assigned for patients who are on treatment, in remission, experiencing UC flare, postcolectomy and experiencing death. With respect to medical treatment, the model appears to separate response and remission based on the Simple Colitis Activity Index (SCAI) score. 86 The model also includes the possibility of patients developing colorectal cancer, which is assumed to result in colectomy. It appears that the model does not include an excess risk of death due to colorectal cancer. Patients enter the model at the point of being hospitalised during their initial flare definitively diagnosing them of pancolitis UC through endoscopic biopsies. Although this may indicate a more severe population than that stated within the scope of this appraisal, the model uses RCT evidence from studies that relate to a moderate to severe UC population49 and assumes that the flare is resolvable without surgery. Following diagnosis, patients are then assumed to receive i.v. methylprednisolone and subsequently mesalamine as maintenance therapy once they are able to tolerate oral medicines. Patients progress along the treatment pathway to IFX, tacrolimus and potentially colectomy + IPAA if remission is not achieved. Patients in the intervention group within the model bypass all medical treatments and immediately undergo surgery. Different cost and HRQoL estimates are applied to each health state.

Treatment benefits are defined differently for surgery and medical treatment. For the standard medical therapy group, treatment benefits are characterised as response, remission and UC flare rates. For the surgery group, treatment benefits are determined according to colectomy success rates, the avoidance of AEs (pouchitis and infertility), the requirement for antibiotics, and remission rates for antibiotics. Effectiveness data were drawn from a number of sources including RCTs and non-randomised studies. 49,87–99 The effectiveness of IFX in inducing and maintaining response and remission was based on the results of the ACT1 and ACT2 studies. 49 The rate of developing colorectal cancer was derived from an observational study. 100 The approach for determining the effectiveness of medical and surgical treatment options is essentially a naive indirect comparison and, as such, the results of the analysis should be interpreted with a degree of caution.

Health utilities were assigned for the following events and states: UC flare (0.48), remission (0.91), post-colectomy (0.87), and infertility following IPAA (0.74). Health utilities were drawn from a variety of sources. 101–104 The elicitation methods from which these estimates were derived were not consistently clear. Although several studies used the time trade-off (TTO) method, the study reportedly used to derive a disutility for female infertility is not a quality-of-life valuation study and actually relates to the medical costs of epididymitis and orchitis in men;104 it is unclear how the information contained within this paper has been used to inform the economic analysis.

The model includes resource costs associated with diagnosis of pancolitis UC, drug therapy, colectomy + IPAA, managing pouchitis and the diagnosis of colorectal cancer. The costs associated with hospitalisations, outpatient visits, procedures and laboratory costs were estimated using national reimbursements from the Centers for Medicare and Medicaid Services and average reimbursement rates from all patient billing records in 2009 at Stanford University Medical Center. The Office of Statewide Health Planning and Development tables were used to validate institutional rates with the intention of reflecting national average cost estimates. Wholesale costs of medical therapies were estimated by prices from online pharmacies and were validated against the drug costs at Lucile Packard Children’s Hospital pharmacy.

The analysis includes deterministic sensitivity analysis and probabilistic sensitivity analysis (PSA). Results are presented as mean costs and QALYs, incremental cost-effectiveness ratios (ICERs), one-way deterministic sensitivity analyses and summary results of the probability of achieving the greatest net benefit at a given willingness-to-pay threshold.

Table 36 presents the headline results of the economic analysis. The analysis indicates that standard medical treatment produces more health (0.06 QALYs) at a considerably greater cost than colectomy + IPAA (US$88,607). The incremental cost-effectiveness of standard medical treatment versus colectomy + IPAA is estimated to be US$1,476,783 per QALY gained.

Assuming a willingness-to-pay threshold of US$50,000 per QALY gained, the probability that colectomy + IPAA produces the greatest net benefit is approximately 1.0. Assuming a willingness-to-pay threshold of US$100,000 per QALY gained, the probability that colectomy + IPAA produces the greatest net benefit is approximately 0.96. The sensitivity analysis indicates that the utility of the cure state after receiving colectomy + IPAA was the only variable which reduced the ICER to below US$100,000 per QALY gained. The authors state that the level of HRQoL for patients with UC would need to be very low in order for exhaustive medical therapy in severe UC to be cost-effective.

The Park et al. 83 study clearly addresses the question of whether or not colectomy + IPAA is cost-effective in comparison with medical management over a lifetime horizon. However, the description of the mathematical model is unclear, hence the assumptions underpinning the analysis are not transparent and their credibility is difficult to judge. The population reflected in the economic analysis is only partially relevant to the scope of this appraisal as the patients considered within the model are by definition hospitalised for UC flare. However, the model also appears to assume that the flare can be resolved; hence, patients may go on to receive biological therapy in a non-acute setting. Given the absence of head-to-head trials comparing medical and surgical management options, the need for an indirect comparison is inevitable and may lead to bias in the model results. It is also noteworthy that the study relates to a US setting and, therefore, its relevance to UK clinical practice may be questionable.

Tsai et al.82

Tsai et al. 82 report the methods and results of an economic analysis that compares two IFX-based strategies with standard care in patients with moderate to severe UC from the perspective of the UK NHS. The model structure and parameter values appear to be very similar to the economic analysis submitted to NICE to inform technology appraisal TA140,28 although it should be noted that the total cost estimates for each group reported by Tsai et al. 82 differ to those reported within the manufacturers’ submission. 80 Patients within the model were assumed to have a mean body weight of 73.1 kg. The base-case scenario evaluated a treatment strategy of 5 mg/kg of IFX every 8 weeks only for patients achieving response while a secondary analysis evaluated 5 mg/kg of IFX every 8 weeks for patients achieving and maintaining remission following induction. Standard care was assumed to include colectomy + IPAA and other medications (5-ASAs, corticosteroids and immunosuppressants). Cost-effectiveness was assessed in terms of the incremental cost per QALY gained over a 10-year time horizon for each comparison of IFX versus standard care. A fully incremental analysis was not undertaken between the two responder/remission stopping rule treatment approaches. The perspective adopted was that of the NHS. Costs borne by Personal Social Services (PSS) were not included in the model. The authors note that although productivity costs are substantial for patients with UC, these were not included in the economic analysis. Costs and health outcomes were discounted at a rate of 3.5% per year. Costs were valued at 2006–7 prices.

The economic evaluation takes the form of a Markov model, as shown in Figure 74. The model appears to include eleven health states: (1) mild (responders); (2) moderate-severe (responders); (3) remission (responders); (4) mild (non-responders); (5) moderate-severe (non-responders); (6) remission (non-responders); (7) temporary discontinuers; (8) surgery (tunnel state); (9) post-surgery remission; (10) post-surgery complications; and (11) death. The cycle length used within the model was specified according to the time intervals of the assessment visits in the ACT1 and ACT2 studies. 49 The first cycle was 8 weeks in duration, followed by 6 weeks in cycle 2, followed by 8 weeks for all subsequent cycles. It should be noted that within the ACT trials, the assessments at these time points were based on partial Mayo scores and may not correspond to full Mayo scores (the latter of which includes endoscopic visualisation). 105 The paper does not mention whether a half-cycle correction was applied to account for the timing of events within the model.

FIGURE 74.

Model diagram presented by Tsai et al. 82 Reproduced with permission from Tsai HH, Punekar YS, Morris J, Fortun P, A model of the long-term cost effectiveness of scheduled maintenance treatment with infliximab for moderate-to-severe ulcerative colitis, Aliment Pharmacol Ther 2008;28:1230–9, © 2008 The Authors. Journal compilation © 2008 Blackwell Publishing Ltd.

All patients enter the model in the moderate to severe health state. At the end of each cycle, patients achieving a Mayo score of 0–2 and 3–5 transit to the remission state and mild health state, respectively, and continue to receive the same treatment. Patients who do not achieve remission or response are classified as non-responders. A ‘temporary discontinuers’ state is included for patients experiencing temporary AEs, which is a tunnel state which is applied for one 8-week cycle. After resolution of AEs, these patients return to their prior health state. Non-responders and patients permanently discontinuing active treatment (e.g. owing to AEs) transit to the corresponding non-responder states and cannot restart IFX treatment. Patients in the moderate to severe states can undergo surgery, which may result in complications. Different costs and utilities are applied to each health state. The model does not include any survival difference between the competing treatments hence the differences in QALYs are driven entirely by differences in sojourn time in each health state.

Transition probabilities in each group were estimated using data from the ACT1 and ACT2 studies. 49 No details are provided within the paper with respect to how these studies were pooled. Transition probabilities for patients in the responder states for IFX and standard care were drawn from the treatment and PBO arms of these trials, respectively. Transition probabilities for non-responders for both groups were drawn from PBO arms of the ACT studies. 49 As ACT1 employed a longer study duration than ACT2, the former trial alone was used to estimate transition probabilities beyond 30 weeks. The ACT1 and ACT2 studies were also used to estimate the probabilities of temporary discontinuation based on the observed AE rates. Transition probabilities for patients undergoing surgery were derived from the literature. 106–108 None of the transition probabilities applied within the model are reported in the paper.

Health-related quality-of-life values are assigned for remission (0.88), mild (0.76), moderate-severe (0.42), temporary discontinuers (0.42), surgery (0.61), post-surgery remission (0.61), post-surgery complications (0.55). Utility values for UC states are stated to have been drawn from an EQ-5D survey of UC patients; however, this appears to be misreferenced as the publication title relates to resource use in patients with CD. 109 Health utility values for patients who were temporary discontinuers and for those with post-surgical complications were drawn from Arseneau et al. 110 The text states that these utilities were indexed to the Woehl utility set to avoid any implausible results using regression analysis. No further details are provided regarding this regression analysis.

The model includes treatment- and state-specific costs associated with drug acquisition and administration, consultant visits, hospitalisations, blood tests and endoscopy. The sources used to value the costs of drug acquisition and administration are unclear. The model includes the costs of concomitant medications based on the baseline characteristics of patients in the ACT1 and ACT2 studies,49 and assumes that the use of immunosuppressants and 5-ASAs remain constant while corticosteroid use declines linearly over time for patients responding and achieving remission. The costs associated with non-SAEs were calculated separately but are not detailed. The costs associated with severe AEs were assumed to be subsumed within the costs of hospitalisation. The costs of colectomy + IPAA were based on NHS Reference Costs. 111 Hospitalisation rates for the IFX and standard care groups were based on the ACT1 and ACT2 trials49 and were valued using NHS Reference Costs. In addition, health-care resource use associated with pre-surgical UC states was estimated from a panel of six UK gastroenterologists; these resource-use estimates were valued using national published cost estimates.

The model results are presented as mean costs and QALYs for each treatment group. The economic analysis includes one-way deterministic sensitivity analysis and PSA. Decision uncertainty is represented using cost-effectiveness planes.

Table 37 presents the results of the economic analysis. Within the base-case analysis, in which medical treatment is assumed to be continued only for those patients in whom response is achieved, IFX is estimated to produce an additional 0.75 QALYs at an additional cost of £20,662, which corresponds to an ICER of £27,424 per QALY gained. Within the secondary analysis, in which patients are assumed to continue treatment only if they achieve and maintain remission, IFX is estimated to produce an additional 0.39 QALYs at an additional cost of £7615, which corresponds to an ICER of £19,696 per QALY gained. It should be noted that the estimates of absolute costs and absolute QALYs for the standard care group differ between the two analyses. Although the alternative treatment stopping rules clearly influence which patients continue to receive IFX and the duration over which patients would receive biological therapy, it is unclear why this would affect outcomes for the standard care group.

In the responders sensitivity analysis, the ICER ranged from £21,066 per QALY gained (lower patient weight) to £86,320 per QALY gained (1-year time horizon). The results of all other deterministic sensitivity analyses in the responder comparison produced an ICER below £32,000 per QALY gained. In the more stringent remission only analysis, the deterministic sensitivity analyses produced ICERs in the range £14,728 per QALY gained (lower patient weight) to £46,765 per QALY gained (1-year time horizon). The results of all other deterministic sensitivity analyses in the remission only comparison produced ICERs below £23,000 per QALY gained. The results of the probabilistic analysis are presented in the form of cost-effectiveness planes. The authors state that ‘The PSA showed that the results were robust with [the] majority of simulations clustered together. In both responder and remission treatment strategies, IFX SMT resulted in additional QALYs at an additional cost compared to standard care.’82 The probabilistic results for the remission only scenario appear somewhat dubious as the samples appear to be truncated at the y-axis of the cost-effectiveness plane (the estimates of incremental QALYs for IFX vs. standard care cannot drop below zero). The underlying reason for this within the model is unclear.

As noted earlier, the Tsai et al. 82 analysis appears to be based on the same model submitted to NICE as part of TA140 (not reviewed here). 28 While the QALY estimates reported within Tsai et al. 82 are virtually the same as those reported within the manufacturers’ submission to NICE,80 the incremental costs reported by Tsai et al. 82 are lower than those contained within the manufacturers’ submission. Consequently, the ICERs presented by Tsai et al. 82 are lower than those reported within the NICE submission (responders only analysis ICER = £27,424 per QALY gained82 vs. £33,866 per QALY gained;80 remission only analysis £19,696 per QALY gained82 vs. £25,044 per QALY gained80).

Overall, the Tsai et al. 82 model appears to follow a plausible model structure and includes the majority of costs and outcomes relevant to the decision problem. It is also noteworthy that this is the only published UK analysis included in this review. In general, the paper performs well against the Drummond checklist. The two notable issues relate to the absence of other biological therapies (this is reasonable as GOL and ADA did not have a UK marketing authorisation at the time of publication) and immediate colectomy as comparators and the use of a short time horizon.

Xie et al.84

Xie et al. 84 report a cost-effectiveness analysis comparing IFX plus ADA versus usual care in patients with moderate to severe refractory UC in Canada. Patients were assumed to be 40 years of age with a mean body weight of 80 kg. The model adopts a Markov approach and costs and outcomes are evaluated over a 5-year time horizon. Three options were compared within the economic analysis: (1) ‘strategy A’ – ‘usual care’, which includes conventional medical treatment (5-ASAs plus immunosuppressants) without anti-TNF-α drugs; (2) ‘strategy B’ – ‘5 mg/kg of IFX plus ADA initial and maintenance therapy’, which includes 5 mg/kg of IFX followed by a switch to ADA if there is no response to initial therapy or if response is lost during maintenance therapy; and (3) ‘strategy C’ – ‘5 mg/kg and 10 mg/kg of IFX + ADA’, which involves initial therapy using 5 mg/kg of IFX and, if there is no response, the dose is escalated to 10 mg/kg of IFX, then if response is lost during maintenance therapy, switch to ADA (Figure 75). Surgery is included in the pathway for all three treatment groups within the model but is not included as a comparator. Costs and health outcomes were discounted at a rate of 5% per year. All costs were valued at 2008 prices.

FIGURE 75.

Treatment sequences evaluated by Xie et al. 84 This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The model includes five mutually exclusive health states: (1) remission, which is defined as a total Mayo score of ≤ 2 points without individual subscores exceeding 1 point; (2) active UC, which includes those patients who do not respond and those who do respond but do not achieve remission; (3) surgery, which is a tunnel state; (4) surgical remission; and (5) surgical complication (Figure 76). The model adopts a variable cycle length according to the timing of full Mayo score assessments adopted in the ACT1 and ACT2 studies (0–8 weeks, 9–30 weeks, 31–54 weeks, then 27 weekly thereafter). 49 Patients enter the model in the active moderate to severe state and following initial therapy either achieve remission or not. Those patients who achieve remission may subsequently lose response and transit to active UC or they may maintain remission. For responders in the active UC state, patients may achieve remission or remain in the active UC state with maintenance. Patients who are non-responders can undergo colectomy + IPAA, switch to ADA (strategy B) or receive an increased dose of IFX (strategy C). Following surgery, patients may experience complications, which may or may not be resolved. These complications may arise immediately after surgery or at a later timepoint. Death is not included as an event in the model owing to the short time horizon. Different costs and utilities are applied to each model health state.

FIGURE 76.

Model diagram presented by Xie et al. 84 This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The majority of clinical parameters within the model were drawn from the ACT1 and ACT2 studies,49 derived using a fixed-effects meta-analysis. Remission rates in responders and non-responders were estimated as time-independent parameters based on the ACT1 and ACT2 studies. 49 The authors note that remission rates drawn from the PBO arms of the ACT trials reflect the use of active concomitant medications. Rates of early and late surgery were drawn from a RCT reported by Jarnerot et al. 112 and a non-randomised cohort study reported by Hoie et al. 107 Rates of complications and the probability of their resolution were derived from non-randomised studies. 113,114 Remission rates, the probabilities of maintaining remission over time and the proportion of non-responding patients in those with active UC for each treatment group were modelled as time-dependent parameters based on the ACT1 and ACT2 studies. 49 Owing to the absence of randomised evidence at the time of the analysis, remission rates for ADA were assumed to be equivalent those for 5 mg/kg of IFX. The model did not incorporate the effects of AEs on health outcomes or costs as the ACT1 and ACT2 studies49 reported that the proportions of patients with any AE were similar among the IFX and standard care groups.

The model includes four utility values: remission (0.79), active UC (0.32), surgical remission (0.68) and surgical complications (0.49). The model makes no distinction between the HRQoL outcomes for patients who achieve response but not remission and patients who do not achieve response; this may be considered as a very pessimistic assumption that could bias against IFX and ADA therapy, particularly given the low valuation of HRQoL for patients with active UC. All health utilities were drawn from a previous economic modelling study reported by Arseneau et al. 110 The method of utility elicitation within this study appears to be TTO.

The model includes the costs associated with drug acquisition, drug administration, colectomy and IPAA, medical examination and the management of surgical complications. Drug acquisition costs were drawn from provincial drug benefit lists (including an 8% mark-up). The costs of medical examinations were derived from the Ontario Schedule of Benefits. The costs of surgery were derived from the literature. 115

The results of the economic analysis are presented as mean costs and QALYs and ICERs for each treatment group based on point estimates of parameters. Pairwise comparisons are presented for the IFX and ADA options versus usual care. A fully incremental analysis between all options in the model is reported in the text. PSA was also conducted with decision uncertainty represented using cost-effectiveness acceptability curves (CEACs).

Table 38 presents the headline cost-effectiveness results reported by Xie et al. 84 The model analysis suggests that the strategy B is expected to produce more health gain than strategies A and C. Strategy C is dominated by strategy B. The incremental cost-effectiveness of strategy B versus usual care was estimated to be approximately CA$358,823 per QALY gained.

The PSA suggests that assuming a willingness-to-pay threshold of CA$150,000 per QALY gained, the probability that usual care is optimal is approximately 1.0. The deterministic sensitivity analysis suggests that the lowest ICER is achieved by increasing the utility for remission (ICER = CA$273,081 per QALY gained for strategy B vs. strategy A and CA$428,676 per QALY gained for strategy C vs. strategy A), while the highest ICER is achieved by lowering the utility for remission (ICER = CA$527,236 per QALY for strategy B vs. strategy A, CA$889,227 per QALY gained for strategy C vs. strategy A).

Overall, the analysis reported by Xie et al. 84 appears to adequately address the decision problem using a generally appropriate model. However, the analysis is limited by the use of a short time horizon, the absence of surgery as a comparator and questionable assumptions regarding the health gains associated with achieving response without remission.

Cost of non-biological ‘background’ therapies

The usage and per-cycle costs of non-biological background therapies assumed within the model are presented in Tables 44 and 45. Patients in the standard non-biological treatment group are assumed to receive 4 g of mesalazine daily (if acute) or 2 g of mesalazine daily (if chronic), 2–2.5 mg/kg of AZA daily, 1–1.5 mg/kg of 6-MP daily, 500 mg of ciprofloxacin twice daily and 25–50 mg of prednisolone per day. The same use of background therapies is assumed for all biological treatment arms. However, it should be noted that in the colectomy group, patients are assumed to undergo immediate colectomy, so the actual drug acquisition cost for the colectomy group is zero within the model. Patients in the biological treatment groups are assumed to receive the same ‘background therapies’ with the exception of ciprofloxacin (although as described above, this is applied as a zero cost in the colectomy group). Resource-use estimates for these therapies appear to be based on the PBO arm of the PURSUIT trial47 within the GOL model and from the PBO arm of the ACT1 and ACT2 trials49 within the IFX model. These are similar, but not the same (see Tables 44 and 45); as with the utility values, the justification for using different assumptions concerning resource use in each model is not clear. The source of the unit costs is not reported within the submission, but estimates appear to be drawn from the British National Formulary (BNF37).

Deviations from NICE reference case and final NICE scope

The extent to which the economic analyses reported in the MSD submissions adhere to the NICE reference case is presented in Table 51.

Overall, the MSD economic analyses are generally in line with the NICE reference case. However, the analysis does make one important deviation from the final NICE scope with respect to the options included in the economic analysis; non-biological treatment is assumed to be immediate colectomy. Standard clinical management, which is defined in the NICE scope as ‘a combination of aminosalicylates (sulfasalazine, mesalazine, balsalazide or olsalazine), corticosteroids (beclomethasone, budesonide, hydrocortisone or prednisolone), and thiopurines (mercaptopurine or AZA), calcineurin inhibitors and surgical intervention’,36 is not included in the MSD model. The omission of non-biological non-surgical treatment options from the MSD model is neither discussed nor justified in the MSD submissions. 64,66

It is also noteworthy that the MSD model adopts a 10-year time horizon; at this point around 96% of patients in the model are still alive in each treatment group. The MSD submissions state that the ‘. . . time horizon of 10 years can be considered sufficiently long to capture differences in the distribution of health states between the compared biologics; after 10 years of follow-up all patients are expected to have discontinued biologic treatment’. The modelled profiles of incremental costs and benefits are slightly different when a longer time horizon is adopted. It is reasonable to suggest that the manufacturer should have examined the impact of using different time horizons within their economic analysis.

Assumption that treatment failure is equivalent to severe exacerbation

Related to the issue regarding the lack of conventional drug therapies as the comparator for the economic analysis (see Deviations from NICE reference case and final NICE scope), the MSD economic models appear to also confuse the severity of the patient populations and the associated treatment pathway included in the model. Although the scope of the appraisal relates to patients with moderate to severe UC who have failed conventional treatment, the modelled pathway after failure of biological therapy, and the choice of non-biological comparators included in the analysis, appear to relate to a population with more severe disease and no further medical treatment options are considered. The pathway represented by the model after failure of biological therapy (or in its absence) appears to assume that failure to achieve an induction response, or that the loss of response during maintenance therapy, is synonymous with an acute UC exacerbation. As noted by the MSD submissions, surgery for UC is typically indicated for (1) patients with life-threatening complications (e.g. toxic megacolon or colonic perforation); (2) dysplasia or proven cancer; or (3) severe disease characterised by treatment refractoriness, frequent flare-ups, extracolonic manifestations, chronic corticosteroid dependence, side effects/intolerance/complications from medications (in particular corticosteroids), or according to clinical judgement. 64,66 After failing biological treatment, the MSD model assumes that all patients who have failed biological therapy will receive i.v. steroids and rapidly progress to colectomy (27% of all relapsing patients during each 2-month cycle). This fails to reflect the possibility that patients may continue to receive, and may still obtain clinical benefit from, non-biological medical treatment options as defined in the NICE scope (5-ASAs, immunotherapies and/or steroids).

After removing mortality, the MSD model suggests that within 1 year (the approximate duration of the maintenance trials45,47,49), 15–20% patients are in the colectomy/post-colectomy health states (Figure 83).

FIGURE 83.

Proportion of patients in post-colectomy states over time (excluding mortality).

This contrasts with the colectomy rates observed within the RCTs included in the systematic review [see Chapter 3, Assessment of effectiveness, Rates of surgical intervention (both elective and emergency) 0.7% to 5.8% in in individual trial arms at approximately 1 year]. The manufacturer’s model also suggests that for patients receiving biological treatments, 59–70% will have undergone surgery within 5 years and 89–93% will have undergone surgery within 10 years (note: the precise values differ by biological treatment group). These rates are very high and fail to reflect both the possibility of benefit from further medical therapies and the element of patient choice in deciding whether or not to undergo colectomy. If surgery really was the only remaining treatment option for these patients, it would not have been possible (or ethical) to undertake any of the trials included in this assessment.

Further to this point, the study reported by Turner et al. ,126 which is used to inform the probability of requiring surgery for active UC, is a systematic review of studies describing the short-term outcome of adult and paediatric patients treated with i.v. corticosteroids, with or without ciclosporin, for exacerbations of UC. Within this published analysis, retrospective and prospective studies evaluating adult or paediatric UC patients admitted for first or subsequent exacerbation, who were severe enough to require i.v. corticosteroid therapy, were included if the short-term outcome and/or analysis of predictors of response were reported. This appears to confuse treatment failure with acute exacerbation of UC. The Assessment Group do not believe that either the narrow choice of remaining viable comparators or the treatment pathways assumed within the MSD models are representative of the clinical management of patients with moderate to severe UC in England and Wales.

Questionable use of evidence concerning improving and worsening of inflammation

The NMA model uses separate models to produce information on the probability of sustained remission and the probability of sustained response. Within the health economic models, the ORs estimated using the NMA models are applied to the probability of remaining in the states of remission and response, respectively. The NMA logistic regression models treat these data as binomial – a patient either stays in their existing state or they do not. However, the data are multinomial – the observed data from the trials indicates that some patients who lost remission transited to response while others transited to no response, and some patients in response subsequently achieved remission, some achieved sustained response and some lost response. The structural assumptions employed within the transition matrix (see Table 42) do not reflect this, with some plausible transitions being assigned probabilities of zero. Although this problem is a likely consequence of the limitations of the published data from the ULTRA2 trial,45 it poorly reflects the characteristics of the actual observed data.

Questionable validity of use of methods for including non-randomised data from PURSUIT

The MSD submissions state that ‘PURSUIT used a non-conventional trial design, and thus, conventional NMA techniques would not have sufficed for producing comparative effect estimates between GOL, IFX, and ADA. This NMA employed novel techniques of optimising the use of all available data.’64,66 This approach was used to ‘downgrade’ the available evidence for PBO within the PURSUIT maintenance trial as patients randomised to PBO were prior GOL induction responders. Based on the information provided in the manufacturers’ submission (see MSD GOL submission,66 table 13 footnotes and text on pp. 55–7), the Assessment Group was unable to logically follow or replicate the calculations used to generate hypothetical values for the PBO group. However, the Assessment Group does believe that the manufacturer’s ‘novel’ method involves omitting the randomised data and instead uses a manipulation of the non-randomised PBO arm data as an input into the NMA. Such manipulation of observed trial data should be viewed with considerable caution. The Assessment Group believe that it would have been more appropriate to use more established methods of bias adjustment (e.g. the methods adopted by Turner and Spiegelhalter131) and/or to use the published ITT data and examine the likely impact of the bias using sensitivity analyses.

Lack of clarity regarding the network meta-analysis model

The NMA model is not reported in detail within either of the MSD submissions and the WinBUGS code was not reported (although this was provided to the Assessment Group during the clarification process). In addition, the baseline model is not described, although the health economic model indicates that baseline probabilities of achieving induction response/remission and maintaining response/remission were derived from ‘Average study effect of PBO controlled trials in random effects NMA of induction response’. The appropriateness of these values is unclear.

Inconsistencies between results of the Merck Sharp & Dohme Ltd golimumab and infliximab models

The IFX model and GOL model are based on the same structure and the same decision problem. However the results are different between the models. In response to a request for clarification on the cause of this discrepancy, the manufacturer stated that the two models use different inputs for health utilities; the IFX model uses utility data from the ACT1 and ACT2 trials while the GOL model uses utility data from the PURSUIT trial. The IFX model also uses different assumptions about the use of conventional non-biological therapies compared with the GOL model. The justification for using different utility and resource-use assumptions in two models that are attempting to reflect exactly the same decision problem is inappropriate. It should also be noted that when the PURSUIT utility vector and resource-use assumptions were inserted into the IFX model, the results still did not coincide.

Lack of clarity regarding the identification, selection and use of certain model parameters

In several instances, the justification for selecting particular parameter sources is unclear. In particular, the justification of the dosing and frequency of background therapies, and the justification for unit costs is not described within the MSD submissions. 64,66

Complex implementation of the model

Conceptually, the submitted MSD models are simple Markov models employing eleven health states and four treatment groups. However, the implementation of these models is complex. The model employs 30 worksheets, many of which were locked as read-only. This limited the ability of the Assessment Group to verify the inputs and formulae used in the model.

Failure to undertake an incremental analysis

The MSD submissions do not include an incremental analysis in which each treatment option is compared against its next best non-dominated alternative. Instead, pairwise comparisons are made using NMB given a willingness-to-pay threshold of £30,000 per QALY gained. The IFX submission states that:

The ICER for infliximab versus standard non-biologic treatment (colectomy) is £37,682. The positive impact of infliximab in terms of reducing the Burden of Illness and mitigating the Wider Societal Impact of the condition represents additional value for consideration by the committee. Taking into account the shortfall in quality of life, and in the ability of people to contribute to society as a result of their experience with moderately to severely active UC, it is likely that infliximab represents a cost-effective treatment in first-line biologic treatment of UC.

MSD. Manufacturer submission of evidence: IFX (Remicade)64

The GOL submission states that ‘At £27,322, the ICER for golimumab falls under a £30,000 threshold, and thus golimumab can be considered a cost-effective treatment option for patients with moderately to severely active UC’. 66

Importantly, both of these economic conclusions are based on a comparison of biological therapy versus immediate colectomy. A fully incremental analysis is presented in Tables 49 and 50. Given the ordering of QALY gains across all treatment options, IFX should be compared against GOL, thus resulting in a considerably higher ICER of approximately £75,000–80,000 per QALY gained (note: the discussion around the discrepancy between model results, see Inconsistencies between results of the Merck Sharp & Dohme Ltd golimumab and infliximab models).

Inclusion of a Patient Access Scheme for golimumab which has not yet been agreed by the Department of Health

Both MSD submissions include a PAS in which 100 mg of GOL will be made available at the same price as 50 mg of GOL (see MSD GOL submission,66 p. 8). However, at the time of this assessment, the proposed PAS had not been agreed with the Department of Health. Although the MSD submissions include a secondary analysis in which the PAS is not included, the absence of fully incremental comparisons by the manufacturer (see Failure to undertake an incremental analysis) clouds the correct interpretation of the economic analysis. The amended results of this fully incremental analysis, which excludes the PAS, are shown in Tables 52 and 53.