A systematic review and economic evaluation of the use of tumour necrosis factor-alpha (TNF-a) inhibitors, adalimumab and infliximab, for Crohn's disease

Description of Crohn’s disease

Inflammatory bowel disease (IBD) refers to a group of chronic intestinal diseases characterised by inflammation of the gastrointestinal mucosa. The most common types of IBD are ulcerative colitis and Crohn’s disease (CD). CD can affect any part of the gastrointestinal tract, from mouth to anus, but most commonly the terminal ileum (35%) or the ileocaecal region (40%) are affected. 8

The main symptoms of CD are dependent on disease location and include chronic or nocturnal diarrhoea, abdominal pain, anal lesions, rectal bleeding and weight loss. Clinical signs include pallor, cachexia, abdominal mass or tenderness, perianal fissures, fistulas or abscesses. Systemic symptoms include malaise, anorexia or fever. 8–10 Extraintestinal symptoms related to intestinal inflammation include spondyloarthritis, cutaneous manifestations or ocular inflammation. 10 In children, growth failure may be the primary manifestation of CD. 11

CD can be defined using the Vienna classification (see Disease classification), i.e. by location, disease behaviour (inflammatory, stricturing, penetrating) and age at diagnosis. 12 Stricturing disease refers to the narrowing of the bowel which can lead to bowel obstruction, while penetrating (or fistulising) disease refers to the creation of abnormal passageways (fistulas) between the bowel and other structures such as the skin. Inflammatory disease (non-stricturing, non-penetrating) causes inflammation without any strictures or fistulas.

Approximately 40%–50% of patients present with ileocolonic disease at the time of diagnosis, approximately 30% have isolated small bowel disease and approximately another 30% have pure colonic disease. It is estimated that only 10%–15% of patients have a change in disease localisation in the 10 years after diagnosis. 13 Disease behaviour at diagnosis is inflammatory (non-stricturing and non-penetrating) in 70% of patients, stricturing in 17% and penetrating (fistulas, abscesses or both) in 13% of patients. 14

Where the ileum and colon are affected, this is usually complicated by intestinal obstruction, inflammatory mass or abscess. Where disease is limited to the colon, patients commonly present with rectal bleeding, perianal complications and extraintestinal complications involving the skin or joints. Gastric and duodenal manifestations include nausea and vomiting, epigastric pain or gastric outlet obstruction. 15

Common complications are strictures, fistulas and perianal disease. Fistulas can develop between loops of bowel adjacent to the bladder, vagina or skin. Perianal disease comprises fissures, fistulas and abscesses, and perianal manifestations may precede the onset of bowel symptoms. 8,15 Symptomatic perianal disease requiring therapy occurs in around 35% of CD patients. 16 CD may also be complicated by sequelae related to malabsorption such as anaemia or metabolic bone disease. 15 Rare complications include acute dilatation, perforation and massive haemorrhage, especially when the disease affects the colon.

CD is characterised by recurring flares alternating with periods of remission. Most patients take medication for a large period of their life because if they stop they might experience a disease flare, but some drugs are tapered off during periods of remission, then if a patient experiences a flare he or she then returns to therapy. 13

Aetiology

The aetiology of CD remains unknown. It is generally accepted that the disease is a response to environmental triggers (infection, drugs or other agents) in genetically susceptible individuals. 9 Smoking has been shown to be a risk factor in CD, with suggestions that smokers are more than twice as likely to develop the disease. 17 Areas under investigation to identify pathogenic mechanisms include epidemiology (e.g. diet, drugs, water supply), the gut–environmental interface (e.g. work on luminal bacteria), the inflammatory process (e.g. cell signalling pathways) and genetics (e.g. studies on gene expression). 9 Exacerbating factors include intercurrent infections, smoking and the use of non-steroidal anti-inflammatory drugs, while the issue of stress initiating or exacerbating CD remains controversial. 15

Diagnosis

Recent efforts have focused on discovery of biomarkers that may eventually lead to the development of specific diagnostic tests for CD,18 but as yet no definitive diagnostic test exists for CD. Overlapping features with other IBDs, a potentially insidious onset, and the heterogeneity of manifestations and/or presentation without gastrointestinal symptoms can make diagnosis difficult. 15 Diseases with symptoms in common with CD include infectious diarrhoea, small bowel lymphoma, ulcerative colitis, appendicitis, coeliac disease and irritable bowel syndrome. A detailed clinical history, a physical examination, laboratory tests and endoscopic evaluation are necessary to make an accurate diagnosis. A diagnosis of IBD should be contemplated in patients presenting with chronic (bloody or non-bloody) diarrhoea, particularly nocturnal diarrhoea and/or weight loss, abdominal pain, fever or extraintestinal manifestations. Family history of the disease should be considered. Signs of volume depletion, ulceration of the oral mucosa, perianal lesions or abdominal tenderness may be observed on physical examination. Laboratory tests should rule out infection and look for markers of IBD such as low serum albumin level or vitamin B₁₂ deficiency. Imaging studies of the bowel may be helpful; abdominal radiography may reveal mucosal oedema or dilated loops of small bowel or colon consistent with either inflammation or obstruction. On endoscopy, CD is characterised by deep, linear ulcerations that can occur as segmental areas of mucosal involvement separated by areas of normal intervening mucosa (‘skip lesions’). Biopsy findings usually demonstrate transmural inflammation. 19

CD may be unsuspected and incorrectly diagnosed in the elderly, with as many as 60% of patients being misdiagnosed initially compared with a misdiagnosis rate of only 15% in younger people. The delay in diagnosis has been calculated as 6.4 years after onset of symptoms in older patients compared with 2.4 years in younger individuals. 20

Disease classification

CD is a heterogeneous condition with a variety of clinical manifestations and presentations. The ‘Vienna classification’ introduced a schema to categorise the disease according to three important elements: age at diagnosis (A), location of the disease (L) and disease behaviour (B). The Vienna classification, summarised in Table 1, was revised in 2005 and this modified version, termed the Montreal classification,21 expanded the number of categories within each of the three elements as shown in Table 1.

The categorisation of CD using these classification systems has allowed descriptions of the progression/natural history of the disease and raised the possibility of identifying genes or environmental factors (including treatments) that may be associated with particular features of the disease or with the rapidity of progression from one category to another. For example, using the Vienna classification and retrospective analysis of case notes for a cohort of 290 patients with up to 25 years of follow-up, Louis et al. 22 found that location of disease was relatively stable with only 15.9% of patients exhibiting a change over a decade while disease behaviour was more labile, changing for 45.9% of patients in a decade. Similarly Cosnes et al. 23 reported that over 20 years most patients progressed to penetrating or stricturing disease, that initial location of disease was a determinant of disease behaviour, and that year-by-year disease activity was poorly influenced by previous behaviour of the CD.

Natural history

The disease location of CD is fairly stable; however, the behaviour of the disease can vary substantially during its course. The disease changes from non-stricturing to either stricturing (in 27% of patients) or penetrating disease (in 29% of patients). 14 After the first year of diagnosis, 10%–30% of patients have an exacerbation, 15%–25% have low activity and 55%–65% are in remission; 13%–20% have a chronic active course of disease activity, 67%–73% have a chronic intermittent course and only 10%–13% remain in remission for several years. 14 Most patients with CD will require surgery within 20 years. 14 The lifetime risk for developing fistulas has been reported to be between 20% and 40%. Perianal fistulas are most common, followed by entero-enteric, with many patients developing a fistula at or before diagnosis of CD. 24 CD is associated with an increased risk of colonic carcinoma and the overall mortality is slightly higher than that of the overall population. 9

A Danish study25 of an inception cohort of 373 CD patients found the following disease activity distributions: 80% of patients had high activity at diagnosis, decreasing to an almost stable value of 30% in the following 25 years; a constant 15% of patients overall had low activity; and around 55% could expect to be in remission each year. Individual patients however changed from year to year between relapse and remission. The study further found that over a 10-year period 20%–30% of patients could expect to go into remission each year. There was a slight indication of the disease ‘burning out’, as late in the disease course (more than 15 years post diagnosis) slightly more patients (29%) changed from activity to remission than the 14% who changed from remission to activity. A separate analysis of 171 patients followed for at least 7 years after diagnosis found that, between years 3 and 7, 25% of patients had active disease every year, 22% were in remission and 53% changed between years in remission and years with relapse. 25 This disease course was independent of initial treatment, age, sex, localisation and symptoms at diagnosis or time from onset to diagnosis. With regard to hospital admissions, 83% were admitted during the year of diagnosis; this decreased during the following 5 years to a constant 20% each year.

A US modelling study26 examined a retrospective cohort and estimated a future life expectancy of 46.4 years for a representative CD patient aged 28.1 years at the time of diagnosis. The projected clinical course consisted of 11.1 years in remission (with no medication), 0.51 years of requiring surgery, 18.9 years in post-surgery remission (no medication), 12.7 years of receiving aminosalicylate or a similar medication, and disease severe enough to require corticosteroids or immunosuppressives lasted 3.2 years. This was based on a sample of 174 patients and on treatment practices used between 1970 and 1993, which may have changed over the course of the study.

A Norwegian study12 which followed up 221 CD patients prospectively for 5 years found that during the observation period 28% had undergone surgery. At the time of the 5-year visit 54% used sulfasalazine and 5-aminosalicylic acid, 25% used oral glucocorticosteroids and 13% used azathioprine. There were 16% who had symptoms that interfered with everyday activities and 72% had taken oral glucocorticosteroids at some point during the 5 years.

These cohort studies and the models based on them indicate that the clinical course estimates will vary depending on a variety of characteristics of the patients within the cohort.

Incidence and prevalence

CD can occur at any age, but manifests itself mainly during late adolescence or early adulthood. Peak onset is between 15 and 30 years of age. 27,28 The incidence in younger years is higher in women than in men. 27,29 There is some inconsistency regarding differences in prevalence between women and men overall, with some studies finding a higher prevalence in women, and some finding no difference. 29 There is an increased prevalence among first- and second-degree relatives, suggesting the involvement of genetic factors. 28 CD may also present later in life (sixth and seventh decades) when there tends to be more colonic involvement and disease manifestations may be less severe. 20

The extent of CD varies across the world and is most common in developed countries, with the UK having one of the highest rates. It was previously thought that IBD occurred less frequently among ethnic minorities. However, studies of migrant populations have shown that ethnic and racial differences are more likely to be attributable to lifestyle and environmental influences than true genetic differences. Similar rates of IBD have been found in African-Caribbean and white children and adults in the UK. 17 No association between CD and social class was found in a UK prevalence study; it has been suggested that this is attributable to exposure to risk factors becoming more similar across social classes. 29

In regions with a high prevalence of CD, the incidence increased between the 1950s and 1980s, and stabilised after that, which can be explained by an increased availability of gastroenterology units and increased awareness of the disease. 28,30 Some studies suggest that there is still an upward trend, which may be due to continued variations in environmental risk factors. 29 Increases in less developed countries have recently been noted, and it has been suggested that this is a result of changes in lifestyle (e.g. more exposure to smoking, changes in diet). 28

Table 2 shows the incidence and prevalence of CD in the UK, taken from studies published from 2000 onwards. The incidence ranges from 3.8 to 10 per 100,000 per year and the prevalence ranges from 50 to 375 per 100,000. For children, the British Paediatric Surveillance Unit found an estimated incidence of 5.3 per 100,000 per year. 11 Differences in incidence and prevalence estimates may result from the way data are gathered, changes in disease awareness and diagnosis over time, or changes in disease risk factors. There is no national CD database that could be used to determine numbers of CD patients.

Impact of health problem

Measurement of disease severity in adults

Working definitions of disease severity have been developed by the Practice Parameters Committee of the American College of Gastroenterology,10 and are:

• Mild–moderate disease:

  Mild-moderate Crohn’s disease applies to ambulatory patients able to tolerate oral alimentation without manifestations of dehydration, toxicity (high fevers, rigors, prostration), abdominal tenderness, painful mass, obstruction, or > 10% weight loss.

• Moderate–severe disease:

  Moderate–severe disease applies to patients who have failed to respond to treatment for mild–moderate disease or those with more prominent symptoms of fever, significant weight loss, abdominal pain or tenderness, intermittent nausea or vomiting (without obstructive findings), or significant anaemia.

• Severe–fulminant disease:

  Severe–fulminant disease refers to patients with persisting symptoms despite the introduction of steroids as outpatients, or individuals presenting with high fever, persistent vomiting, evidence of intestinal obstruction, rebound tenderness, cachexia, or evidence of an abscess.

• Remission:

  Remission refers to patients who are asymptomatic or without inflammatory sequelae and includes patients who have responded to acute medical intervention or have undergone surgical resection without gross evidence of residual disease. Patients requiring steroids to maintain well-being are considered to be ‘steroid dependent’ and are usually not considered to be ‘in remission’.

The severity of CD is difficult to assess, and a global measure encompassing clinical, endoscopic, biochemical and pathological features is not available. 38 The most widely used disease activity measures include the Crohn’s Disease Activity Index (CDAI), the Harvey–Bradshaw Index (HBI) or Simple Index, a simplified version of the CDAI, and the Perianal Disease Activity Index (PDAI). A commonly used HRQoL measure is the Inflammatory Bowel Disease Questionnaire (IBDQ). Other measures include the Crohn’s Disease Endoscopic Index of Severity (CDEIS).

The CDAI was developed in the 1970s as there was a need for a single index to assess disease severity. Variables measured include number of liquid stools, abdominal pain, general well-being, extraintestinal complications, use of antidiarrhoeal drugs, abdominal mass, haematocrit and body weight. Scores range from 0 to approximately 600, with higher scores corresponding to more severe disease (see Appendix 1 for full description of the index and the scoring system used). Values of below 150 are suggestive of quiescent disease (remission) and values above 450 are associated with very severe disease. 39 Severe disease is thought to be above 300. Some investigators, however, have arbitrarily labelled CDAI scores of 150–219 as mildly active disease and scores of 220–450 as moderately active disease. 38

The CDAI has been criticised for having limitations. It does not cover aspects of QoL, such as psychological, social, sexual and occupational functioning. A patient with a low CDAI score may still be severely limited by the disease in those areas. 40 Substantial variability exists when different observers review the same case histories and calculate the CDAI score, although this can be reduced after discussion and education about the terminology. The calculation is based in part on a daily diary kept by the patient for 7 days before the evaluation. In practice, some investigators and study co-ordinators assist the patient to retrospectively complete the diary at the time of an evaluation visit; there is no information on the prevalence of this practice. The CDAI score may be low in patients whose primary symptom is drainage of enterocutaneous fistulas, presumably because the presence of an actively draining fistula contributes only 20 points to the score. The CDAI is therefore not an appropriate instrument for assessing the activity of draining abdominal or perianal enterocutaneous fistulas. The CDAI has been criticised for giving too much weight to ‘general well-being’ and ‘intensity of abdominal pain’, as these are relatively subjective items. However, these aspects of disease are important to patients. 41

Clinical studies have variously defined a clinical response as a decrease in CDAI of 50, 60, 70 or 100 points. The US Food and Drug Administration (FDA) and the European Medicines Agency (EMEA) suggested in 2000 that a clinically meaningful decrease in the CDAI score is a decrease of 100 points. 41

The HBI is a modified/simplified version of the adult CDAI. It uses a single day’s reading for diary entries and excludes three variables (body weight, haematocrit and use of drugs for diarrhoea). Code values are added together rather than summing the products of code values and coefficients (see Appendix 1). Scores range from 0 to 20, with higher scores corresponding to worse disease. The CDAI can be predicted reasonably well from the HBI. 42 Other instruments derived from the CDAI are: the Cape Town Index, which includes parameters on subjective symptoms, physician clinical findings and laboratory data; the three-variable version of the CDAI used for survey research; and the van Hees Index, which includes laboratory parameters, sex (male or female) and seven clinical features and excludes subjective, patient-related items such as well-being and pain. 40

The PDAI was developed to account for the morbidity and impairment of QoL of patients with perianal disease, and to evaluate the effectiveness of perianal disease treatment. Variables include discharge, pain/restriction of activities, restriction of sexual activity, type of perianal disease (including number of fistulas) and degree of induration. Scores range from 0 to 20. 16

The reliance on traditional disease activity measures (such as the CDAI) to measure treatment effectiveness fails to take into account the impaired QoL experienced by CD patients. The IBDQ is an HRQoL measure. It is a 32-item questionnaire and evaluates general activities of daily living, intestinal function, social performance, personal interactions and emotional status. Four-dimensional scores cluster items under bowel function, emotional function, systemic function and social function. Scores range from 32 to 224. 43

The CDEIS was developed to take into account endoscopic data, such as lesion severity, when assessing severity of the disease. Variables include the presence or absence of deep or superficial ulceration in various segments of the intestinal tract, the surface involved (in cm), the surface ulcerated (in cm) and presence of ulcerated stenosis. Scores range from 0 to 30. 44

Measurement of disease severity in children

The paediatric CDAI is a multi-item measure of severity that includes linear growth and places less emphasis on subjectively reported symptoms and more on laboratory parameters of intestinal inflammation than the adult CDAI. It includes 11 variables: weight, height, abdominal mass, perirectal disease, extraintestinal manifestation, haematocrit, erythrocyte sedimentation rate, albumin, abdominal pain, number of liquid stools and general well-being. Scores range from 0 to 100: ≤ 10 indicates inactive disease, 11–30 mild disease and > 30 moderate-to-severe disease. 45,46

Adverse events with anti-TNF treatment

A number of adverse events (AEs) have been associated with anti-TNF therapy and have been reported for infliximab and adalimumab. As the immune response is suppressed, infections may be more likely to occur. These include tuberculosis, other bacterial infections including sepsis and pneumonia, fungal infections and opportunistic infections such as pneumocystosis or cytomegalovirus infection. Cases of reactivation of hepatitis B infection have been observed, as have rare cases of jaundice and hepatitis, optic neuritis and onset or exacerbation of demyelinating disorders including multiple sclerosis. A deficiency of TNF may result in the initiation of an autoimmune process, and the occurrence of lupus-like syndrome has been observed. There is the possibility of an increased risk of lymphoma or other malignancies, worsening of heart failure or of AEs of the haematological system (e.g. cytopenias). Infliximab has been associated with acute, infusion-related reactions (including anaphylactic shock) and delayed hypersensitivity reactions. Injection site reactions are common with adalimumab. Common AEs for both infliximab and adalimumab are upper respiratory infections (such as sinus infections), headache, rash, nausea and stomach pains. The development of anti-TNF antibodies may be associated with a decrease in efficacy and predispose the patient to an additional risk of recurrent delayed or acute allergic reactions. 52–56

This report will consider the following patient groups (where information is available): adults with moderate-to-severe active CD intolerant or resistant to conventional treatment; children with moderate-to-severe active CD intolerant or resistant to conventional treatment; and adults with fistulising CD intolerant or resistant to conventional treatment. Where possible, patients with severe (rather than moderate-to-severe) CD will be considered as this is in line with the licence indication.

Degree of diffusion

There is no up-to-date evidence available on the degree of diffusion of adalimumab and infliximab for CD treatment in the UK. The only evidence that is available from routinely collected data is for the total number of adalimumab and infliximab prescriptions for all conditions.

Interventions

Adalimumab and infliximab are drugs for use in patients with severe active CD or fistulising active CD (infliximab) who have not responded to conventional treatment or who have experienced toxicity from these treatments. There has been a distinction made between induction treatment and maintenance treatment, but it is unclear where the boundary lies between these for the interventional drugs. Similarly there has been a distinction between ‘episodic’ treatment, i.e. treatment when a disease flare starts (or at a clinician’s discretion), and maintenance treatment, where patients are treated at regular (scheduled) intervals with the intention of keeping them in remission, but it is unclear where the boundary lies between these treatment strategies. It would be useful to know the most effective dosing regimen for each of the drugs.

Comparators

Conventional treatment includes no treatment, dietary intervention, drug treatment with aminosalicylates, methotrexate, corticosteroids (prednisolone, budesonide and hydrocortisone), azathioprine or metronidazole or surgical intervention.

Given that licences for both drugs are for use only when conventional treatment has failed, it is unlikely that randomised controlled trials (RCTs) would compare the drugs to conventional treatment. Instead, the most likely comparator will be no treatment or placebo, but where patients in all trial arms continue to receive elements of conventional therapy. Another relevant comparator may be a different dosing regimen of the same drug.

For comparisons between both drugs under review, head-to-head comparisons within the same trial would be the ideal scenario. It is important to note that, because of earlier licensing, infliximab could be viewed either as the intervention of interest in some of the RCTs or as part of conventional treatment in others. It would also be useful to establish the effectiveness of both drugs compared with non-drug treatments such as surgery or nutrition, particularly in children.

Population and relevant subgroups

Infliximab is licensed for use in adults and children with severe active CD or in adults with fistulising disease who are intolerant or resistant to treatment. Adalimumab is licensed for severe active CD; current information does not indicate whether this is in adults only.

There is no standard definition for what constitutes severe CD. NICE guidance defines severe as a score of > 300 on the CDAI or 8–9 on the HBI. 1 The group that developed the CDAI defines values of ≤ 150 as quiescent disease and values > 450 as extremely severe disease; no intermediate cut-off point is given for severe disease. 39

The NICE scope for the current appraisal stated that the population of interest consists of patients with ‘moderate-to-severe’ CD. There is no standard definition of what constitutes ‘moderate-to-severe’, but RCTs have described patients with a CDAI of 220–400 as having moderate-to-severe CD. 57 Note that this assessment report is therefore investigating treatments outside their licence indications. The main thrust of the work should be to investigate the clinical effectiveness of treatments in patients with a CDAI score of 300 or more. However, it is unlikely that any RCTs have included only these CD patients. Therefore, the options are:

• To look only at subgroups of patients in RCTs with a CDAI score of ≥ 300. This is unlikely to be a valid comparison unless the RCT stratified patients by being more or less than CDAI 300.

• To widen the inclusion criteria of the assessment report to include RCTs where CD patients had lower CDAI scores. 58

It may be that there is a different effectiveness of the interventions in CD patients with CDAI scores of > 220 compared with > 300.

Most work on measurement of CD has been carried out in adult patients. Where a child has CD, it is unclear how this would be consistently categorised as severe CD or moderate-to-severe CD. Although there is a children’s version of CDAI – Paediatric Crohn’s Disease Activity Index (PCDAI) – it is unclear how well this measure is validated and how it relates to CDAI cut-off points.

It could be important to look at populations of patients who have failed either infliximab or adalimumab therapy to determine if unresponsiveness to a particular drug is a persistent state and whether unresponsiveness to one drug can be linked to similar unresponsiveness to the other. Finally, it is unclear exactly how resistance to treatment is measured or how long a treatment trial would go on for before a patient would be categorised as being resistant or responsive to treatment.

Outcomes

Key factors are the clinical effectiveness of both drugs particularly in terms of enhancing patient QoL, maintenance of remission, delaying disease progression and prolonging survival. More specifically, outcomes could include overall survival, progression-free survival, HRQoL, disease activity (remission, response, relapse, changes in disease activity indices, number of fistulas for fistulising disease), maintenance of response to treatment over time, need for surgery, need for an ostomy, hospitalisation rates, need for steroid treatment, dropout rates from TNF-α treatment and adverse effects of treatment. It is unclear how outcomes such as mucosal healing would impact on clinical outcomes such as QoL.

Where disease severity and effect of treatment is measured by CDAI or PCDAI scores, it is uncertain how large a change in CDAI score constitutes a clinically significant change and whether this would be the same change for more severe CD as for less severe CD.

Trials in patients with fistulising disease will measure fistula closure but it is uncertain whether this is a good measure of effectiveness as abscesses can form if the fistula is no longer patent; so abscess occurrence may be a better outcome measure. Other clinical outcomes could include abscess formation rates and seton use (if reported).

Search strategy

The search strategy was designed to update that undertaken for the previous technology assessment of the clinical effectiveness and cost-effectiveness of infliximab in adults with moderate-to-severe CD5 and to encompass the new anti-TNF therapies identified for review. A search was undertaken to find existing good quality systematic reviews in order to document the evidence base to date. Searches for primary studies were restricted to RCTs. The following sources were searched for relevant primary studies:

• bibliographic databases: Cochrane Library [Cochrane Central Register of Controlled Trials (CENTRAL)] 2007, Issue 2; MEDLINE (Ovid) 2000 to May/June 2007; MEDLINE In-Process & Other Non-Indexed Citations (Ovid) 4 June 2007 and 26 June 2007; EMBASE (Ovid) 2000 to May/June 2007. Searches were based on index and text words that encompass the condition: CD and the interventions: adalimumab, certolizumab pegol, infliximab and natalizumab. [Natalizumab and certolizumab pegol were originally part of this technology appraisal so were included in the searches. They were subsequently dropped from the report after completion of searches (see Protocol modification).] Where it was appropriate, a methodological ‘filter’ was applied to identify RCTs

• EMEA, FDA and other relevant websites

• citations of relevant studies

• contact with experts

• research registries of ongoing trials including National Research Register 2007, Issue 2, Current Controlled Trials and ClinicalTrials.gov

• submissions from industry

• hand search of conference abstracts in 2006 and 2007: British Society of Gastroenterology, Digestive Disease Week, United European Gastroenterology Meeting, European Crohn’s and Colitis Organisation, Federation of Clinical Immunology Societies.

Searches were not limited by language. Full search strategies can be found in Appendix 3.

Inclusion and exclusion criteria

Only studies meeting the following inclusion criteria were included:

• Study design: RCTs (study designs other than RCTs were excluded).

• Population: adults (≥ 18 years) and children (6–17 years) with moderate-to-severe, active CD intolerant or resistant to conventional treatment; adults (≥ 18 years) with fistulising CD resistant to conventional treatment. ‘Moderate-to-severe’ disease includes patients with an average CDAI score of ≥ 220 or those who are described by trial authors as having moderate-to-severe disease.

• Intervention: adalimumab or infliximab (any dosage/treatment regimen).

• Comparator: conventional treatment without TNF-α inhibitors including no treatment, placebo, dietary intervention, drug treatment with aminosalicylates, methotrexate, corticosteroids (prednisolone, budesonide and hydrocortisone), azathioprine, metronidazole or surgical intervention. Adalimumab and infliximab compared with each other. Different dosage or treatment regimens of the same drug.

• Outcomes: at least one of the following: overall survival, progression-free survival, HRQoL, disease activity (remission, response, relapse, changes in disease activity indices, number of fistulas for fistulising disease), need for surgery, hospitalisation rates and adverse effects of treatment.

• Trials that looked at both induction and maintenance of remission were included.

Based on the above inclusion/exclusion criteria, study selection was made independently by two reviewers. Discrepancies were resolved by discussion, with involvement of a third reviewer when necessary. All discrepancies were resolved in this way.

Data extraction strategy

Information on study characteristics, study quality and results for each trial was extracted by one reviewer and checked by a second reviewer. Four reviewers were involved in data extraction. A standardised data extraction form was used, based on the form designed for the previous TAR on infliximab. 5 The data extraction template can be found in Appendix 4. Where necessary the template was adapted to accommodate details relevant to a specific trial. Where required, information was extracted from graphs as follows (see Appendix 5): the graph was scanned into a word document, overlaid with an appropriate template with graph gridlines, and printed and enlarged to A3 size, and information was extracted using the gridline template. To reduce error in this procedure, extracted information was checked by comparing graph readings with any available values in the report text and/or by redrawing the graph using the extracted data and comparing this with the original (see Appendix 5 for examples). Data extraction discrepancies were resolved by discussion, with involvement of a third reviewer when necessary. All discrepancies were resolved in this way.

Quality assessment strategy

Quality assessment was based on the published papers only and note was taken that absence of a quality criterion may be due to lack of reporting rather than actual poor methodological quality. Authors were not contacted for further information. Quality assessment was descriptive, a quality scoring system was not used. The quality criteria assessed were based on guidelines suggested by the Cochrane Collaboration, inviting consideration of threats arising from selection, performance, attrition and detection biases. Individual checklist items were: randomisation, concealment, blinding, comparability of groups, follow-up of trial participants, handling of missing data [intention-to-treat (ITT) analysis], power calculation and selective reporting (see Appendix 4 for checklist). Study quality was assessed by one reviewer and checked by a second. Discrepancies were resolved by discussion, with involvement of a third reviewer when necessary. All discrepancies were resolved in this way.

Handling of manufacturer and other submissions

The main industry submissions (including appendices) were checked for additional relevant trials and additional clinical effectiveness data for included trials. Because editorial constraints meant the results available in published accounts of the trials were necessarily selective, information in the submitted Clinical Study Reports was sourced as required for purposes of balance and completeness. It was not possible to systematically review all such additional information submitted owing to the volume of the submissions [e.g. more than 38,000 pages for the clinical study report of ACCENT (A Crohn’s disease Clinical trial Evaluating inflixmab in a New long-term Treatment regimen) I,2,3 more than 5000 pages for the Clinical Study Report of Targan et al. ,57 both included studies]. No references to specific sections of the clinical study reports were made in the main industry submissions. [Please note that the clinical study reports for the CLASSIC (CLinical assessment of Adalimumab Safety and efficacy Studied as Induction therapy in Crohn’s disease), CHARM (Crohn’s Trial of the Fully Human Antibody Adalimumab for Remission Maintenance) and GAIN (Gauging Adalimumab efficacy in Infliximab Nonresponders) RCTs that were received from the manufacturers of adalimumab started on section 4 and had no page numbers or tables of contents. Also some of the appendices were missing, particularly ones referred to in the text as having all of the raw results in tables. Therefore it is unclear whether some pages are missing from the middle of these reports or not and potentially the most useful appendices were not supplied.] For details on how the submitted economic models were assessed see Chapter 4, Critique of the submission on infliximab by Schering-Plough.

Analysis strategy

The clinical effectiveness section of this report mainly focuses on the results from RCTs and/or RCT trial arms in which the drugs were administered within the limits of their current respective licence indication (see Appendix 6). Results of trials are organised and reported in four categories:

• induction trials in adult populations predominantly or wholly constituted of non-fistulising patients

• maintenance trials in adult populations predominantly or wholly constituted of non-fistulising CD patients

• trials in populations wholly constituted of patients with fistulising CD

• trials in paediatric patients.

Results are reported within these four categories on a trial-by-trial basis except with regard to AEs and side effects which were considered simultaneously across all included trials across both drugs. Most outcome results are presented in forest plots so as to provide an overview of the quantitative spread of effect sizes. These are accompanied with brief narrative commentary. In some instances outcome results are tabulated. Both placebo and intervention rates and both risk difference and risk ratio effect sizes are presented for most outcomes in Tabulation of included studies. The confidence intervals (CIs) quoted were not adjusted for repeated measures.

The clinical heterogeneity of trials, or the existence of only a single trial, precluded pooling of data in meta-analysis. The feasibility of undertaking indirect comparison analysis was considered in-depth in order to assess the relative effectiveness of different drugs because there were no RCTs directly comparing both drugs included in this technology appraisal. However, indirect comparisons were not done because of the variation in placebo effect sizes in the RCTs (induction trials), the lack of similarity in the apparently common comparator (i.e. placebo arm maintenance trials), and the reporting of subgroup results only at follow-up (i.e. variously defined responders only) in many of the RCTs.

Protocol modification

The protocol originally encompassed assessment of the clinical effectiveness and cost-effectiveness of infliximab, adalimumab, certolizumab pegol and natalizumab within their licensed indications for moderate-to-severe CD. At the time of producing the protocol, certolizumab pegol and natalizumab were not licensed for CD, but imminent licensing was anticipated by the commissioners of this report (NICE). After the start of the review process it became clear that neither drug would achieve a licence within the time frame required for this technology assessment and consequently both drugs were dropped from the review. This occurred after completion of the search strategy. As of November 2010 these drugs remain unlicensed for CD.

Quantity of research available

Eleven relevant trials were identified,2,3,45,46,57,58,62–67 some supported by multiple publications. Figure 1 details the trial identification process.

FIGURE 1.

Study identification process.

At the time of writing of this report, 11 hard copies of ordered publications were still outstanding or not available; none of these are likely to contain new trial data (see Appendix 7 for details of publications).

Eleven RCTs were included in total. 2,3,45,46,57,58,62–67 Seven trials meeting the inclusion criteria were identified through the main database searches. 2,3,45,46,58,63,65,67 Two additional studies57,62 from the previous TAR on infliximab were included,5 as were two trials from 2007 which had been published after the search cut-off date. 64,66

Searching through the main industry submissions from both manufacturers did not yield any additional RCTs. The search for conference abstracts yielded no further relevant trials. An abstract of the study by Hommes et al. 61 was identified, which is referred to in Chapter 5, Other relevant factors. This study did not meet the criterion of a population of CD patients who are resistant or intolerant to conventional treatment.

The search for ongoing trials yielded four potentially relevant RCTs, all of adalimumab (see Appendix 8). All were at the recruitment stage (or not yet recruiting) at the time the information was verified by the respective manufacturers. Two were trials (induction and maintenance) of adalimumab in Japanese patients with moderate-to-severe CD. Two multicentre trials of adalimumab were in patients with moderate-to-severe ileocolonic CD and in children with moderate-to-severe CD respectively. Two ongoing trials of infliximab were identified, but did not meet the inclusion criteria as they compared either infliximab with infliximab plus methotrexate or infliximab with infliximab plus azathioprine. No ongoing trials of head-to-head comparisons of adalimumab and infliximab were identified. No preliminary reports of any of these ongoing trials were identified in the manufacturer submissions.

Tabulation of included studies

All of the included RCTs recruited patients having ‘moderate-to-severe CD’ defined according to CDAI scores of between 220 and 450, or 220 and 400; it is therefore likely that they do not reflect the intended licensed population of severe active CD (i.e. CDAI score of more than 300).

The included studies encompassed two trial designs, induction therapy and maintenance therapy, in any of three populations: adults predominantly or wholly non-fistulising, fistulising adults and children. Table 5 gives an overview of the included studies with reference to trial design and recruited patient population.

Of the 11 included RCTs,2,3,45,46,57,58,62–67 nine compared infliximab or adalimumab with placebo. 2,3,57,58,62–67 Two RCTs compared different doses of infliximab only and these were both in children. 45,46 Two RCTs of infliximab were in patients with fistulising disease. 62,65 Both induction and maintenance trials were identified for both drugs. All RCTs were multicentre studies conducted mainly in North America and Europe. No RCTs of head-to-head comparisons of adalimumab and infliximab were identified. No RCTs of adalimumab in children were identified. Based on the information in the published papers, all RCTs were either industry sponsored or in part industry sponsored, had participants from industry involved in study design or manuscript writing, or had one or more authors with industry involvement.

In the induction trials, patients received short-duration anti-TNF or placebo to see if a favourable clinical response was induced. In the maintenance trials, all patients received short-term induction therapy with anti-TNF and then continued with longer term anti-TNF or placebo. In the maintenance trials most published results reported only the follow-up of patients who initially responded to the induction therapy, and results for ‘non-responders’ were generally not provided.

The most widely reported outcomes were based on CDAI scores (see Appendix 1 for details). Although group mean or median CDAI scores were usually recorded at various times of follow-up, the variance of these scores was incompletely reported and trials emphasised binary outcome measures derived by dichotomising CDAI scores. Three such binary measures were used:

• response 70: defined as a reduction of 70 or more in CDAI score relative to baseline

• response 100: defined as a reduction of 100 or more in CDAI score relative to baseline

• remission: defined as a CDAI score of less than 150.

The definitions of the binary measures given above were often qualified by stipulation of additional criteria usually including no requirement for a change in concomitant medication because of worsening clinical condition and no requirement for surgery.

This section describes the results about the effectiveness of the anti-TNF interventions. The results reviewed were taken mainly from publications. When judged necessary for purposes of completeness and balance, information in the unpublished industry trial reports was also sourced.

There are four sections in the clinical effectiveness results: induction treatment in adults (predominantly non-fistulising), maintenance in adults (predominantly non-fistulising), treatments in adult patients exclusively with fistulising CD, and paediatric CD (≤ 18 years old). Within each section infliximab is reported before adalimumab and the earliest trial publication date first. Each of the four sections are organised for each trial as follows:

• description of intervention used in the trial and other unusual points about the trial design

• report of outcomes organised as A, response 70; B, response 100; C, remission; D, other outcomes; and E, other considerations, in the first two sections, primary and secondary outcomes in the last two sections

• quality assessment

• summary for that trial (in box).

Adverse events and side effects are considered simultaneously across all included trials for both drugs at the end of the clinical effectiveness section (see Adverse events), just before the discussion of clinical effectiveness (see Discussion of results and assessment of effectiveness).

Targan et al., 199757 (infliximab)

This RCT had four arms. 57 Patients were randomised to a single i.v. infusion of placebo (n = 25) or of infliximab at 5 mg/kg (n = 27), 10 mg/kg (n = 28) or 20 mg/kg (n = 28). Disease status (remission, response 70 and CDAI score) was monitored at baseline and at weeks 2 and 4 after infusion. The 4-week blinded phase was followed by an open-label phase with a further 12 weeks of follow-up. The primary outcome measure was defined as a response 70 at week 4 with no change in any concomitant medication.

A, response 70

Response 70 at week 4 was the primary outcome. Results for response 70 at weeks 2 and 4 are summarised in Figure 2. For response 70 at week 4 there was a statistically significant difference in favour of the infliximab groups (combined) compared with placebo (p < 0.001). The percentage of placebo patients achieving response 70 was ≤ 16% at both time points and for infliximab groups at week 4, and was between 50% and 81% depending on dose regimen. Point estimates of percentage response were associated with considerable uncertainty. The rate of response 70 at week 4 for the combined infliximab groups was 61% (95% CI 51% to 71%). At week 4 the risk difference (infliximab–placebo) was between 0.34 and 0.65, and risk ratio (infliximab/placebo) was between 3.1 and 5.1 depending on dose. Both risk difference and risk ratio at week 4 reached statistical significance in favour of intervention.

FIGURE 2.

Response 70 rates in Targan et al. 57 At week 4 risk difference p < 0.001, p = 0.0045, p < 0.001, for 5, 10 and 20 mg/kg dose regimens respectively. At week 4 risk ratio p < 0.001, p = 0.022, p < 0.004 for 5, 10 and 20 mg/kg dose regimens respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Table 8 summarises the comparison between different dose regimens for response 70 at week 4. The low-dose regimen (5 mg/kg) appeared more effective than the 10 mg/kg regimen (p = 0.009). The difference between dose regimens for other comparisons did not reach statistical significance.

TABLE 8 - Risk difference between dose regimens in response 70 at week 4 in Targan et al.57

Dose comparison	Risk difference	Lower CI	Upper CI
5 mg/kg vs 10 mg/kg	0.315	0.079	0.551
5 mg/kg vs 20 mg/kg	0.172	–0.058	0.402
10 mg/kg vs 20 mg/kg	–0.143	–0.399	0.114

B, response 100 was not reported

C, remission

Figure 3 summarises remission rates. At 4 weeks, between 25% and 48% of patients in the infliximab groups were in remission, depending on dose, but only one placebo patient achieved remission.

FIGURE 3.

Remission rates in Targan et al. 57 At week 4 risk difference p < 0.001, p = 0.0206 and p = 0.0206, for 5, 10 and 20 mg/kg dose regimens respectively. At week 4 risk ratio p = 0.013, p = 0.076 and p = 0.076 for 5, 10 and 20 mg/kg dose regimens respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

There was a discrepancy between remission rates published in Targan et al. 57 and rates presented in the manufacturer’s submission. The latter for the 5 mg/kg group at week 4 were placebo rate 4% (1/24), infliximab rate 0% (0/24). These remission rates generate a negative risk difference (infliximab–placebo) at week 4 (–0.04). CIs for risk ratios (infliximab/placebo) in the manufacturer’s submission were described as ‘unadjusted’, but were unexpectedly narrow compared with those calculated using standard software packages or using the standard error of ln (risk ratio) given by:69 ([e_i]^–2 + [e_p]^–2 – [T_i]^–2 – [T_p]^–2)^0.6, where e_i, and e_p, are the number of patients with the outcome in the intervention and placebo arms respectively, and T_i and T_p are total number of patients in the intervention and placebo arms respectively. (This discrepancy in CIs applies to CDAI-based binary risk ratios for all trials in the infliximab industry submission.)

Maintenance of initial response to single infusion

At week 4 there were 54/83 (65%) responders (response 70) to infliximab (combined dose groups); by 12 weeks (see E, open-label phase below) there were 34 responders (41%). At week 4, 27/83 (33%) patients given infliximab had gained remission and at 12 weeks 20 patients (24%) were in remission.

D, other outcomes

At week 4 favourable responses to treatment were reported for CDAI scores, for QoL scores (IBDQ), and for C-reactive protein (CRP) levels. The results reported are summarised in Table 9.

TABLE 9 - Mean (standard deviation) values for CDAI, IBDQ and CRP concentrations at baseline and week 4

p < 0.001.

p = 0.003.

p = 0.02.

p = 0.03.

p = 0.001.

Levels of CRP < 8 mg/l are considered normal.

p = 0.004; authors calculated p-values for change from baseline comparing placebo with intervention using analysis of variance with the van der Waerden normal scores blocked according to centre. If the treatment effect was significant, the infliximab treatment groups were compared with the placebo group with linear contrasts.

	Placebo n = 25	5 mg/kg n = 27	10 mg/kg n = 28	20 mg/kg n = 28	All infliximab groups n = 83
Score on CDAI
Baseline	288 ± 54	312 ± 56	318 ± 59	307 ± 50	312 ± 55
4 weeks	211 ± 82	166 ± 76a	226 ± 115b	211 ± 107a	201 ± 103a
Score on IBDQ
Baseline	128 ± 29	122 ± 29	116 ± 23	118 ± 28	118 ±27
4 weeks	133 ± 28	168 ± 36a	146 ± 41c	149 ± 35d	154 ± 38e
CRP (mg/l)f
Baseline	12.8 + 13.9	22.1 + 23.6	23.2 + 34.2	22.4 ± 23.9	22.5 ± 21.4
4 weeks	14.8 ± 18.6	5. 1 ± 9.3g	12.1 ± 18.6	6.9 ± 11.6a	8.3 ± 1.39a

Figure 4 shows the mean difference in IBDQ score (infliximab–placebo) at week 4. Mean difference reached statistical significance only for patients who received the low-dose regimen.

FIGURE 4.

Mean IBDQ scores and mean difference at baseline and week 4 of Targan et al. 57 NS, not significant.

E, other considerations – open-label phase

In the open-label phase of the trial, extending by at least 12 weeks from week 4, non-responder patients at week 4 were eligible for a 10 mg/kg infusion of infliximab. The distribution of this second infusion among the patient groups is summarised in Table 10. Of the original 25 placebo group patients, 19 non-responders received infliximab; 29 non-responder patients who had received a first dose of infliximab received the second dose. Table 10 lists the percentage of the patients (not responsive at week 4) in each group who subsequently achieved response 70 at follow-up weeks 4, 8 and 12 after the second infusion.

TABLE 10 - Numbers of patients receiving second infusion in open-label phase of Targan et al.57

Original randomisation group (n)	Number receiving and not receiving second infusion (%)		Response 70 at times after second infusion (non-responders at week 4 after first infusion)
	Did not receive	Received	Week 4	Week 8	Week 12
Placebo (25)	6 (24)	19 (76)	11/19 (58%)	13/19 (68%)	10/19 (53%)
5 mg/kg group (27)	21 (78)	6 (22)	2/6 (33%)	3/6 (50%)	1/6 (17%)
10 mg/kg group (28)	13 (46)	15 (54)	6/15 (40%)	5/15 (33%)	5/15 (33%)
20 mg/kg group (28)	20 (71)	8 (29)	2/8 (25%)	4/8 (50%)	2/8 (25%)
Combined infliximab groups					8/29 (28%)
All groups (108)	60 (56)	48 (44)	21/48 (44%)	25/48 (52%)	18/48 (38%)

Of patients unresponsive to the first dose of infliximab, 28% (8/29) responded by week 12 following the second dose, compared with 53% (10/19) of patients whose second infusion was their first exposure to active intervention. During this open-label phase there was a lack of a true placebo control group and the results therefore only suggest that some patients poorly responsive to an initial infusion may respond subsequently on receipt of further infusion. Whether a 10 mg/kg second dose represents the most appropriate dose regimen for this second-dose strategy is unknown.

Quality assessment (based on published report)

Randomisation, allocation concealment, and blinding (up to week 4) were all adequate. Baseline characteristics were similar between groups except for CRP levels and for the proportion of patients with ileal involvement. Placebo CRP level {mean 12.8 [standard deviation (SD) 13.9]} was substantially lower than that for the active intervention groups [mean (SD): 22.1 (23.6), 23.2 (34.2) and 22.4 (23.9) for the 5 mg/kg, 10 mg/kg and 20 mg/kg groups respectively]. The potential impact on results of the imbalanced CRP levels is difficult to determine. Follow-up appeared almost complete. The original study protocol did not specify the use of ITT analysis, but the publication stated that patients were analysed according to assignment. A power calculation was conducted; this assumed a 30% response in the placebo group presumably reflecting the authors’ assessment of placebo rates reported in other CD trials. The actual placebo response rate observed was less than half this value (16%) and was low compared with other similar trials. The low placebo rate and imbalance of placebo CRP level may indicate an atypical placebo population possibly stemming from the small sample size of the group (n = 25).

CLASSIC I63 (adalimumab)

In this trial,63 patients (n = 299) were randomised to two subcutaneous injections 2 weeks apart of either placebo (n = 74) or adalimumab at dose regimens of 40 mg then 20 mg (n = 74), at 80 mg then 40 mg (n = 75), or at 160 mg then 80 mg (n = 76). Patients were excluded if they had previously received any anti-TNF treatment. At baseline 11% of patients had fistulas. Outcomes were monitored at weeks 1, 2 and 4 after the first injection. The primary outcome was defined as the proportion of patients in remission at week 4 in the two high-dose adalimumab groups versus the placebo group (tested using chi-squared test).

Targan et al., 1997.57 Summary of effectiveness evidence

A single i.v. infusion of infliximab (5, 10 or 20 mg/kg) was more effective than placebo at delivering a clinical response (a reduction of ≥ 70 points in CDAI score) at week 4 of follow-up (p < 0.005 for risk differences and p < 0.022 for risk ratios). Estimates of the percentage of patients responding to infliximab were associated with considerable uncertainty, and at 4 weeks ranged between 50% and 80% depending on dose. Of the dose regimens used, the lowest appeared to be the most effective, suggesting the possibility that the most appropriate dose could be less than the lowest used in the trial (5 mg/kg). A proportion of patients (∼30%) not responsive at week 4 did respond subsequently when given a second dose of infliximab (10 mg/kg); although it is likely this ‘second-dose’ response required active intervention, this was not properly demonstrated because the trial lacked a true placebo comparator after week 4. The most effective dose regimen for a ‘second-dose’ response was uncertain. After week 4 nearly all trial participants had received active intervention, and inferences about the relation of outcomes to infliximab were obscured. The Targan et al. 57 trial was completed more than a decade ago and no further induction trial of infliximab in this population has been conducted, so the uncertainties described above remain to be addressed.

A, response 70

At week 4 for the less robust measure of a clinical improvement by more than 70 points in CDAI score from baseline (response 70), a statistically significant result was observed for both risk difference and risk ratio for all three dose regimens (results are summarised in Figure 5).

FIGURE 5.

Rates of response 70 in CLASSIC I. 63 At week 4 for risk difference p = 0.029, p = 0.005 and p = 0.004 for 40/20, 80/40 and 160/80 dose regimens, At week 4 for risk ratio p = 0.0357, p = 0.0088 and p = 0.0073 for 40/20, 80/40 and 160/80 dose regimens. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

B, response 100

At week 4 the risk difference for response 100 (intervention–placebo) reached statistical significance only for the highest dose regimen while risk ratio (intervention/placebo) reached statistical significance for the two higher dose regimen groups. The results for response 100 are summarised in Figure 6.

FIGURE 6.

Rates of response 100 in CLASSIC I. 63 At week 4 for risk difference p = 0.279, p = 0.060 and p = 0.0015 for 40/20, 80/40 and 160/80 dose regimens respectively. At week 4 for risk ratio p = 0.284, p = 0.0682 and p = 0.0036 for 40/20, 80/40 and 160/80 dose regimens respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

C, remission rates

Remission rates were the primary outcome in this RCT. For remission rates there was a statistically significant difference in favour of the two high-dose adalimumab regimens relative to placebo for the proportion of patients in remission at (45/151 vs 9/74; p = 0.004). At week 4 the risk difference (intervention–placebo) and risk ratio (intervention/placebo) reached statistical significance only in the highest dose regimen group. Remission rates are summarised in Figure 7.

FIGURE 7.

CLASSIC I remission rates. 63 At week 4 risk difference p = 0.354, p = 0.057 and p = 0.0005 for 40/20, 80/40 and 160/80 dose regimens. At week 4 risk ratio p = 0.359, p = 0.0691 and p = 0.0021 for 40/20, 80/40 and 160/80 dose regimens respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

For each of the three CDAI-based binary outcome measures there was an apparent linear dose response trend with greater effectiveness for higher dose.

D, other outcomes

At week 4, favourable responses to treatment were reported for CDAI scores, for QOL scores (IBDQ), and for CRP levels. The results reported are summarised in Table 11.

TABLE 11 - Mean (SD) values for CDAI, IBDQ and CRP concentrations at baseline and week 4

NR, not reported.

Comparisons vs placebo:

p < 0.01.

p < 0.001.

p < 0.05.

Levels of CRP < 8 mg/l are considered normal.

p = 0.032.

p = 0.0002.

p = 0.0001.

	Placebo (n = 74)	40/20 (n = 74)	80/40 (n = 75)	160/80 (n = 76)
Score on CDAI: mean (SD)
Baseline	296 (60)	299 (57)	301 (61)	295 (52)
4 weeks	240 (NR)	228 (NR)	210 (NR)a	193 (NR)b
Score on IBDQ median (range)
Baseline	131 (52–200)	129 (81–218)	128 (63–200)	127 (37–192)
4 weeks	147 (NR)	147 (NR)	158 (NR)c	158 (NR)c
CRP (mg/l) median (range)d
Baseline	0.9 (0–17.3)	0.9 (0–11.3)	0.9 (0–14.9)	0.7 (0–9.3)
4 weeks	0.8 (0–9.3)	0.3 (0–8.6)e	0.4 (0–34.0)f	0.2 (0–4.6)g

E, other considerations – subgroup analyses

Logistic regression failed to show a relationship between baseline CRP levels or concomitant immunosuppressive therapy and remission rates at week 4 with placebo or adalimumab.

For the small subgroup of patients with fistulas (11%), no significant differences were observed between placebo and intervention with regard to fistula improvement or remission.

Quality assessment (based on published report)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics were reasonably well balanced between groups. There were no losses to follow-up, and withdrawals were limited to 5%. Efficacy estimates appear to have been calculated using ITT analysis, but this was not stated explicitly. A power calculation was conducted; this assumed 20% and 45% remission rates in the placebo and intervention arms respectively (the observed placebo rate in the trial was about 12%). Last observation carried forward was used for analysis of IBDQ scores, but the number of missing data was not stated.

CLASSIC I.63 Summary of effectiveness evidence

Two subcutaneous injections of adalimumab given 2 weeks apart at 40 mg then 20 mg, at 80 mg then 40 mg, or at 160 mg then 80 mg, were more effective than placebo at achieving remission (CDAI score < 150) at week 4 after the first injection (p = 0.004 for the two high-dose regimens combined vs placebo). The percentage of placebo-treated patients gaining remission at week 4 was ∼12% compared with between ∼18% and ∼36% for adalimumab-treated patients depending on dose regimen received. Point estimates of response 70 rates, response 100 rates and remission rates were associated with considerable uncertainty, but for all three outcome measures a trend was evident for higher doses to be more effective. At week 4 of follow-up, risk differences (intervention–placebo) and risk ratios (intervention/placebo) for the highest dose regimen reached statistical significance in favour of adalimumab for all three outcomes. Subgroup analyses failed to identify any baseline characteristics associated with a better response to active intervention relative to placebo.

GAIN64 (adalimumab)

In this trial,64 325 patients were randomised to two subcutaneous injections 2 weeks apart of either placebo (n = 166) or adalimumab at a dose regimen of 160 mg then 80 mg (n = 159). To be included patients had to have been previously exposed to infliximab treatment and found to be intolerant (n = 190), unresponsive (n = 164), or intolerant and unresponsive (n = 40). The primary response was defined as the proportion of patients in remission at week 4 after the first injection.

A, response 70; B, response 100; and C, remission

The primary outcome was remission rates. The remission rate at week 4 was 7% in the placebo group and 21% in the adalimumab group (p < 0.001). This result and those for the secondary outcomes as reported are summarised in Table 12. The CDAI-based binary response outcome measures reported are summarised graphically in Figure 8. At weeks 2 and 4, risk differences (adalimumab–placebo) and risk ratios (adalimumab/placebo) were in favour of the intervention and reached statistical significance.

TABLE 12 - Outcome measures reported in the GAIN trial64

CiC, commercial-in-confidence; NR, not reported.

Comparisons adalimumab vs placebo.

Primary outcome % remission at week 4.

Chi-squared test.

Levels of CRP < 8 mg/l are considered normal.

	Placebo (n = 159)	Adalimumab 160/80 (n = 164)	Difference (95% CI) (adalimumab–placebo)	p a
Remission (rate; %)b
Week 1	4%	6%	2.7% (–2.0 to 7.4)	(CiC information has been removed)
Week 2	6%	21%	14.7% (7.2 to 22)	(CiC information has been removed)
Week 4	7%	21%	14.2% (6.7 to 21.6)	(CiC information has been removed)
Response 70 (rate; %)
Week 1	21%	35%	14.1% (4.5 to 23.7)	0.004
Week 2	33%	52%	19.7% (9.1 to 30.1)	(CiC information has been removed)
Week 4	34%	52%	17.8% (7.3 to 28.4)	(CiC information has been removed)
Response 100 (rate; %)
Week 1	12%	20%	7.4% (–0.5 to 15.4)	(CiC information has been removed)
Week 2	18%	37%	18.4% (8.9 to 27.9)	(CiC information has been removed)
Week 4	25%	38%	13.7% (3.7 to 23.7)	(CiC information has been removed)
CDAI: mean (SD)
Baseline	313 (66)	313 (58)	0
Week 1	287 (NR)	264 (NR)	–23
Week 2	281 (NR)	232 (NR)	–49
Week 4	264 (NR)	226 (NR)	–38
IBDQ score: mean (SD)
Baseline	124 (28)	120 (27)	+ 4
Week 4	139 (NR)	150 (NR)	+ 11	< 0.001
CRP: median (range) mg/ld
Baseline	7.0 (0–235)	9.0 (0–115)	+ 2
Week 4	7.0	5.0	–2
Change from baseline	0	4	4	Significant

FIGURE 8.

Response 70, response 100 and remission rates reported in GAIN. 64 At week 4 risk difference p = 0.001, p = 0.007 and p = 0.0002 for response 70, response 100 and remission respectively. At week 4 risk ratio p = 0.0014, p = 0.009 and p = 0.0006 for response 70, response 100 and remission respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

D, other outcomes

Results for these are also shown in Table 12. Mean CDAI scores reduced from baseline to a greater extent with adalimumab than with placebo (at week 4, p < 0.001 for mean change from baseline). At week 4 the improvements from baseline in IBDQ scores were 30 and 15 for the adalimumab and the placebo groups respectively. CRP levels at week 4 relative to baseline were more normalised in the intervention than in the placebo group. The change from baseline comparing adalimumab with placebo reached statistical significance in favour of adalimumab.

E, other considerations – subgroup analyses

The primary outcome (remission at week 4) was reported for subgroups of patients defined according to: previous response or intolerance to infliximab; receiving or not receiving immunosuppressive agents at baseline; receiving or not receiving corticosteroids at baseline; or having a negative or positive test for antibodies to infliximab. Risk difference was in favour of adalimumab relative to placebo for all subgroups.

A small proportion of patients (14%, n = 45) had draining fistulas or perianal fistulas at baseline. Rates of fistula improvement and remission were similar between placebo and adalimumab groups.

Quality assessment (based on published report)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics were well balanced between groups. There were no losses to follow-up, and withdrawals were limited to 4%. Efficacy estimates appear to have been calculated using ITT analysis for remission and response outcomes. For continuous variables such as IBDQ, last observation was carried forward; the number of missing data for IBDQ was small (eight patients). A power calculation was conducted; this assumed 20% and 35% remission rates in the placebo and intervention arms respectively (the observed rates at week 4 in the trial were 7% and 21% respectively).

GAIN.64 Summary of effectiveness evidence

Two subcutaneous injections of 160 mg and then 80 mg of adalimumab given 2 weeks apart were more effective than injections of placebo at achieving remission (CDAI score < 150) at week 4 after the first injection (p < 0.001). The percentage of placebo-treated patients gaining remission at week 4 was 7% (95% CI 4% to 12%) compared with 21% (95% CI 14% to 27%) for adalimumab-treated patients. At weeks 2 and 4 of follow-up, risk differences (intervention–placebo) and risk ratios (intervention/placebo) reached statistical significance in favour of adalimumab for remission, response 70 and response 100. A statistically significant difference in favour of adalimumab versus placebo was observed for change in IBDQ score at week 4 relative to baseline.

Rutgeerts et al., 199958 (infliximab)

The Rutgeerts et al. 58 trial was an extension of the Targan et al. 57 infliximab induction trial and included 73 of the original 108 patients. Targan et al. 57 consisted of a 4-week comparison between placebo and one dose of infliximab in three arms (5 mg/kg, 10 mg/kg or 20 mg/kg). This was followed after a maximum of 2 weeks by an open-label phase with 12 weeks of follow-up that started with the option of a 10 mg/kg dose of infliximab for week 4 non-responder patients. To be eligible to enrol in Rutgeerts et al. 58 the Targan et al. week 4 responder patients needed to achieve a response 70 at week 8, and the week 4 non-responder patients needed to achieve a response 70 at week 8 after the open-label option of a 10 mg/kg infusion of infliximab. Four weeks after qualifying (week 8 after induction infliximab or 8 weeks after open-label infliximab) the eligible patients were randomised to i.v. infusion of placebo or 10 mg/kg infliximab (designated week 12 of maintenance phase) and a further three infusions at 8-week intervals (a total of four infusions after becoming eligible to participate; administered weeks 12, 20, 28 and 36). Follow-up continued to week 48.

The induction regimen in this study was variable between patients in duration and in exposure to infliximab. In consequence, induction was ill-defined and the distinction between the induction regimen and maintenance regimen was also unclear. The eligible patients could have received any of the following possible infusions of infliximab: no infliximab (placebo), one 5 mg/kg infusion, one 10 mg/kg infusion or one 20 mg/kg infusion; a second infusion of 10 mg/kg could be given (to any patients) at week 4 if there was no response. Four patients received no infliximab (placebo and no second 10 mg/kg dose as a response was achieved). How closely the trial induction phase corresponds to the licence indication is uncertain.

A, response 70

No primary outcome measure was identified. The response 70 results presented (summarised in Figure 9) referred to point prevalence at assessment time points and do not necessarily indicate maintenance of individual patient response. At week 8 > 90% of patients had a response 70 (CDAI reduced by > 70 points relative to baseline in Targan et al. 57). At week 12 (randomisation week) this had diminished to about 75% and by week 48 had further diminished to 33% in the placebo group and 57% in the infliximab group (p = 0.038 for risk difference and p = 0.054 for risk ratio). Point estimates were associated with considerable uncertainty. The authors stated that of patients with response 70 at the last infusion (week 36), 62% of the infliximab group and 37% of the placebo group maintained their response for the 8 weeks to week 44 (p = 0.16).

FIGURE 9.

Response 70 rates in Rutgeerts et al. 58 At weeks 24 and 48 risk difference p = 0.094 and p = 0.038 respectively. At weeks 24 and 48 risk ratio p = 0.108 and p = 0.054 respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

B, response 100

This outcome was not reported.

C, remission

The point prevalence of remission at different follow-up weeks was reported (results are summarised in Figure 10). Point estimates were associated with considerable uncertainty. At randomisation (week 12) ∼38% of patients were in remission in the infliximab group; this increased to ∼ 60% during weeks 16–40. The corresponding values for the placebo group were ∼ 44% (week 12) and 35% (weeks 16–40). Risk difference (infliximab–placebo) and risk ratio (infliximab/placebo) just reached statistical significance (p < 0.05) at most time points for weeks 16–40.

FIGURE 10.

Remission rates in Rutgeerts et al. 58 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

D, other outcomes

Time to loss of response for patients achieving a response at ‘any time’ during follow-up after randomisation was reported. The criteria for loss of response were not explicit. Over 48 weeks it is possible for a patient to enter a response state on several occasions. The publication did not make clear which occasion(s) were used in the analysis, or how and if double counting was avoided. The log rank test for difference between placebo and infliximab groups just failed to reach statistical significance (p = 0.057).

Median CDAI score, median IBDQ score and median CRP concentrations were reported, but range of values and statistical analyses for these outcomes were not presented. The results were in favour of infliximab relative to placebo with greater reduction in CDAI scores, larger increases in IBDQ scores and more ‘normalisation’ of CRP concentrations. The results published are summarised in Figure 11.

FIGURE 11.

Median CDAI, IBDQ and CRP levels reported in Rutgeerts et al. 58 Data taken from published graphs and redrawn. Where necessary the authors carried last observation forward.

Quality assessment (based on published report)

Randomisation, allocation concealment and blinding were adequate. Baseline values for those characteristics reported were evenly balanced, but values for CRP, which was not balanced in the original Targan et al. trial,57 were unclear. Analysis of response 70 and remission rates was by ITT; the results presented were point prevalence values at various follow-up times, and they therefore represent maintenance of response at the group level only and not maintenance by individual patients. For continuous outcomes, last observation was carried forward where necessary, but the number of missing data was not reported. No primary outcome was identified and no power calculation was described; the combined trials appear to have been powered only for the induction analysis of Targan et al. 57 (at week 4 of that study). The maintenance part of the study was probably underpowered. About 33% of patients withdrew.

Rutgeerts et al., 1999.58 Summary of effectiveness evidence

The study recruited patients from among responders (CDAI score reduced by 70 points) following on from the Targan et al. trial57 and the resulting induction phase varied between patients in both duration and dose regimen. Subsequent maintenance treatment with infliximab (four infusions of 10 mg/kg at 8-week intervals) generated a greater proportion of patients with a response 70 and with remission than did treatment with placebo. Point prevalence estimates for these outcomes were associated with considerable uncertainty. The trial left unanswered how well a clinical response is sustained at the individual patient level.

ACCENT I2,3 (infliximab)

This was a free-standing maintenance trial (i.e. newly started). 2,3 There were 580 eligible patients (CDAI range 220–400), of whom 573 received a single induction infusion of 5 mg/kg infliximab. Two weeks later patients were randomised to placebo, to 5 mg/kg infliximab at weeks 2 and 6 and then every 8 weeks to week 54, or to 5 mg/kg at weeks 2 and 6 and then 10 mg/kg infliximab every 8 weeks to week 54 (these groups are here termed 5 mg/kg and 10 mg/kg groups respectively). At week 2 (randomisation week) patients were classified as responders (335/573, 58.5%) or non-responders (238/573, 41.5%) depending on whether they achieved a response 70 (a reduction of > 70 points in CDAI score at week 2 relative to baseline). At week 14 patients who initially responded but then worsened were eligible to cross over to treatment with increased dosage of infliximab; this crossover treatment for the placebo group was termed ‘episodic treatment’. The results for responders were published in 2002 (Hanauer et al. 3) and patients who crossed over to increased dosage after week 14 for most of these analyses were considered as treatment failures.

Effectiveness results published for responders only in 2002 (Hanauer et al. 3) are reviewed below, and results for all patients, irrespective of responder status at week 2 and published in 2004 (Rutgeerts et al. 2), are considered in the following section.

ACCENT I: results for responders

Of the 335 responders (58.5% of those who had received an induction dose of 5 mg/kg infliximab), 110 were randomised to placebo, 113 to the 5 mg/kg infliximab group and 112 to the 10 mg/kg infliximab group.

A, response 70

The published results for responders3 included graphical presentation of point prevalence of response 70 at weeks 30 and 54. These results are summarised in Table 15. A statistically significant difference in rates in favour of infliximab versus placebo was reported for both infliximab groups at weeks 30 and 54. The manufacturer’s submission provided point prevalence rates for response 70 for all assessment visit weeks from 2 to 54. These results are summarised in Figure 12.

TABLE 15 - Published response 70a rates for responders at weeks 30 and 54 in ACCENT I2,3

NA, not applicable.

Response 70 defined as reduction of ≥ 70 CDAI points from baseline and no requirement for medication change or for surgery.

Data read from published graph in Hanauer et al. 3

Intervention vs placebo.

Dose regimen (n)	Week 30b		Week 54b
	Response 70 (%)b	p c	Response 70 (%)b	p c
Placebo (110)	27%	NA	16%	NA
5 mg/kg group (113)	51%	0.0002	38%	0.0001
10 mg/kg group (112)	58%	< 0.0001	47%	0.0001

FIGURE 12.

Response 70 rates for responders throughout follow-up in ACCENT I. 2,3 CIC, commercial-in-confidence; LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Point estimates were associated with appreciable uncertainty. Week 2 response rates of ∼90% had diminished in all groups by week 54 to 15% in the placebo group and 38% and 47% in the 5-mgkg and 10 mg/kg infliximab groups respectively. Risk differences (infliximab–placebo) remained fairly constant from week 14 onwards. Risk differences and risk ratios (infliximab/placebo) reached statistical significance in favour of infliximab at all visit times from week 10 to week 54. It is unclear why week 2 response rates were less than 100%; it is possible some patients with a 70-point CDAI reduction from baseline nevertheless required surgery or a change in concomitant medication for worsening of clinical condition. After week 2, decline of response occurred in both placebo and intervention groups, then after week 10 risk differences remained similar [e.g. for the 5 mg/kg arm risk differences (infliximab–placebo) remained similar after week 14 as follows: at weeks 10, 14, 22, 30, 38, 46 and 54 risk differences were 0.14, 0.23, 0.26, 0.24, 0.21, 0.23 and 0.23 respectively]. This suggested that most benefit of infliximab was delivered in the first 10–12 weeks of the trial.

B, response 100

This outcome was not reported.

C, remission rates

Remission was a coprimary outcome. The results published for remission at week 30 and week 54 are summarised in Table 16. For this outcome patients who worsened and crossed over to ‘episodic treatment’ (allowed from week 14 onwards) were counted as treatment failures (i.e. as no longer in remission). The results reported measured the point prevalence of remission for each group at week 30 and did not require maintenance of response from week 2 to week 30 at the patient level. A statistically significant greater proportion of patients were in remission at weeks 30 and 54 in the infliximab groups than in the placebo group. At week 30 the risk differences (infliximab–placebo) were 18% and 25% for the 5 mg/kg and 10 mg/kg groups respectively and the corresponding numbers needed to treat (NNT) (30 weeks) were 5.66 and 4. Note this NNT estimate does not include non-responders who had been administered induction infliximab. The point prevalence of remission had diminished somewhat by week 54.

TABLE 16 - Remissiona rates for responders reported at weeks 30 and 54 in ACCENT I2,3

NA, not applicable; NR, not reported.

Remission defined as a CDAI < 150 and no requirement for change in medication or for surgery.

Calculated from published values.

Intervention vs placebo.

Data read from published graph in Hanaeur et al. 3

Dose regimen (n)	Week 30			Week 54
	Remission: % (95% CIb) (number)	Odds ratio (95% CI) intervention/placebo	p c	Remission (%)d	p c
Placebo (110)	21% (14% to 29%) (23)	NR	NA	14%	NA
5 mg/kg group (113)	39% (30% to 48%) (44)	NR	0.003	28%	0.007
10 mg/kg group (112)	45% (36% to 55%) (50)	NR	0.002	38%	< 0.0001
5 mg/kg and 10 mg/kg groups combined (225)	42% (36% to 48%) (94)	2.7 (1.6 to 4.6)	NR	33%

The unpublished Industry Trial Report for ACCENT I2,3 provided information regarding the maintenance of remission at the individual patient level for weeks 14–54. The percentages were slightly discrepant with those in the published report as indicated in Table 17.

TABLE 17 - Patient level maintenance of remission reported in ACCENT I2,3

(CiC information has been removed.)

	% in remission at all visits from weeks 14 to 54
	Placebo	5 mg/kg group	10 mg/kg group
Published report	11%	25%	33%
Trial report	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)

The manufacturer’s submission and the Industry Trial Report provided commercial-in-confidence (CiC) point prevalence rates for remission for all assessment visits from weeks 2 to 54. These results are summarised in Figure 13.

FIGURE 13.

Remission rates for responders throughout follow-up in ACCENT I. 2,3 At week 30 risk difference p = 0.0027 and p < 0.0001 for 5 mg/kg and 10 mg/kg groups respectively. At week 30 risk ratio p = 0.0047 and p = 0.00025 for 5 mg/kg and 10 mg/kg groups respectively. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Point estimates were associated with appreciable uncertainty (CiC information has been removed). From week 10, remission rates diminished in all groups and risk difference (infliximab–placebo) diminished or remained fairly constant; risk differences and risk ratios (infliximab/placebo) reached statistical significance at all visit times from week 10 onwards. It is evident that loss of remission was continuous after weeks 6–10 of follow-up and that the advantage of intervention over placebo was mostly gained by about weeks 6–10, the phase of the study during which dose frequency was greatest. Thereafter decline of response was about the same for both placebo and intervention groups despite continued infliximab every 8 weeks in the treatment arms; for example, for the 5 mg/kg arm risk differences (infliximab–placebo) remained similar after week 14 as follows: at weeks 10, 14, 22, 30, 38, 46 and 54 risk differences were 0.15, 0.21, 0.20, 0.18, 0.15, 0.15 and 0.15 respectively.

D, other outcomes

The primary outcome in ACCENT I2,3 was identified as time to loss of response. (Note: a protocol amendment added the proportion of responder patients in remission at week 30 as a coprimary outcome, which has been reported above.) Loss of response was defined as a CDAI of ≥ 175, a CDAI increased by ≥ 35% and a CDAI increased by ≥ 70 points relative to the qualifying value for a response on at least two consecutive assessments, or requirement for change in medication or requirement for surgery. Assessments were scheduled at weeks 0, 2, 6, 10, 14 and then every 8 weeks to week 54. With this definition of loss of response, it is possible for an individual responder to no longer qualify as achieving a response 70 status, but counter-intuitively nevertheless to not have lost response. (For example, an individual with a CDAI of 221 at enrolment would qualify as a responder at week 2 with a CDAI score reduced by 71 points to 150. If this patient’s CDAI subsequently rose to 170 he or she would no longer be in a response 70 but would nevertheless not have lost response because the increase in score from week 2 was < 70 points, < 35% of week 2 score and below a score of 175.) For this primary outcome, patients in the active intervention arms had significantly longer time to loss of response than patients given placebo (p = 0.0002, log rank test). The median times to loss of response are summarised in Table 18.

TABLE 18 - Median time to loss of response in responders in ACCENT I2,3

The upper range is given as > 54, as 54 weeks is the longest follow-up time; the upper end of the range is not actually known.

Dose regimen (n)	Median time (weeks) to loss of response	IQR (weeks)
Placebo (110)	19	10 to 45
5 mg/kg group (113)	38	15 to > 54
10 mg/kg group (112)	> 54	21 to > 54
5 mg/kg and 10 mg/kg groups combined (225)	46	17 to > 54

Published effectiveness results for responders included median CDAI scores and median IBDQ scores. These are summarised in Table 19. For missing values of CDAI and IBDQ, the nearest observation was carried forward. CDAI scores and IBDQ scores diminished and increased respectively to a greater extent in the infliximab groups than in the placebo group. The IQRs for median values during follow-up were not reported.

TABLE 19 - Median CDAI and IBDQ scores for responders during follow-up in ACCENT I2,3

NR, not reported; NS, not significant.

Data read from graph in Hanauer et al. 3

Comparison: 5 mg/kg group vs placebo.

Comparison: 10 mg/kg group vs placebo.

Tests for significance were done by analysis of variance.

Week	CDAI: mediana				IBDQ: mediana
	Placebo (n = 110)	5 mg/kg (n = 113)	10 mg/kg (n = 112)	p	Placebo (n = 110)	5 mg/kg (n = 113)	10 mg/kg (n = 112)	p
0	290	305	305	NSb,c	129	128	130	NSb,c
2	157	155	152	0.01b 0.04c	173	169	173	NR
6	159	138	140	< 0.0001b < 0.002c	165	174	161	NR
10	165	131	127	< 0.0001b,c	160	170	169	NSb,c
14	197	145	125	< 0.0001b,c	155	167	172	0.05b 0.0076c
22	217	163	135	< 0.0001b,c	142	164	169	0.013b < 0.0001 c
30	225	172	150	< 0.0001b,c	144	162	167	0.015b 0.001c
38	238	214	140	< 0.0001b,c	137	151	170	0.015b < 0.0001 c
46	235	200	142	< 0.0001b,c	135	144	169	0.06b < 0.0001c
54	238	192	152	< 0.0001b,c	136	150	167	0.015b < 0.0001 c

The manufacturer’s submission provided information about QoL measures (SF-36). The SF-36 scores were reported separately for mental and physical components for weeks 30 and 54 of the trial, and mean improvement from baseline was reported. SDs of values were provided. The results are summarised in Table 20. Change from baseline for SF-36 physical component reached statistical significance in favour of infliximab at both weeks 30 and 54.

TABLE 20 - SF-36 results reported for responders in ACCENT I2,3

The results for infliximab refer to the 5 mg/kg group only. p-values refer to comparison between infliximab and placebo groups.

SF-36 component	Group	SF-36 score			Mean improvement from baseline
		Baseline	Week 30	Week 54	Week 30	Week 54
Physical Component	Infliximab	33.0 ± 8.5	40.4 ± 11.3	39.2 ± 11.9	7.3 ± 10.3	6.1 ± 10.8
	Placebo	33.9 ± 8.8	37.2 ± 11.3	36.5 ± 11.0	3.1 ± 9.5	2.5 ± 9.0
					p = 0.002	p = 0.014
Mental Component	Infliximab	38.8 ± 11.3	43.2 ± 11.4	43.9 ± 12.2	4.6 ± 12.7	5.1 ± 12.8
	Placebo	39.8 ± 11.3	42.8 ± 12.0	42.1 ± 12.0	2.9 ± 11.2	2.0 ± 10.9
					p = 0.348	p = 0.072

Median daily steroid dose was reduced by week 14 in all groups and then remained constant. The reduction in the infliximab groups was greater than that for the placebo group. The odds ratio for discontinuation of steroid use (infliximab/placebo) at week 54 was 4.2 (95% CI 1.5 to 11.5).

E, other considerations – subgroup analysis of remission rate in severe CD patients

The manufacturer’s submission for infliximab provided CiC information about the proportion of responder patients who initially had severe disease (defined as a baseline CDAI score > 300) and who achieved remission status during follow-up. Results presented referred to patients classified as having severe disease who were randomised to the 5 mg/kg infliximab group [n = 63/113 (56%)] and placebo group [n = 48/110 (44%)]. No information was provided regarding patients with severe disease among non-responders. The remission rates in placebo and 5 mg/kg infliximab arms and the risk difference for this subgroup of patients are shown in Figure 14. Remission rates were slightly poorer in this more severe CDAI group than for all responders, but a similar pattern was shown during follow-up, in that most of the advantage from the intervention was achieved with the first three doses (early phase). Thereafter, remission decayed at approximately similar rates in the two arms even though patients in the intervention arm received further doses of infliximab and risk differences decreased from week 14 onwards.

FIGURE 14.

Remission rates, risk difference and risk ratio (severe disease responders ACCENT I2,3). LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

ACCENT I2,3 (responders): quality assessment (based on published report of Hanauer et al.3)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics were only reported for all patients (i.e. for all responders and for all non-responders). It was therefore not possible to judge if baseline characteristics were evenly balanced between the three arms of responders that were analysed for effectiveness outcomes. Similarly the number of patients who withdrew was reported for all enrolled patients and it was not possible to determine how many responders discontinued their randomised treatment. Where necessary, the nearest or last observation was carried forward for continuous outcomes but the number of missing data was not reported. A power calculation was conducted and based on the primary outcome of loss of response. The definition of loss of response was complex and did not correspond to a failure to maintain a response 70 status, and its clinical meaning was difficult to gauge.

ACCENT I: results for all patients (Rutgeerts et al.2) (infliximab)

The results for all 573 patients who received an induction dose in ACCENT I2,3 were presented by Rutgeerts et al. 2 in a paper published 2 years after that, describing results for responders only. Separate results for non-responders have not been published. The 573 patients were 335 responders and 238 non-responders (defined according to whether a 70-point reduction in CDAI score was attained by week 2 after the induction infusion).

Randomisation at week 2 resulted in allocation of 188 patients to the placebo group, 192 to the 5 mg/kg group and 193 to the 10 mg/kg group.

The authors stated ‘the primary objective of the analysis was to examine the difference in efficacy between episodic and scheduled treatment strategies with infliximab under conditions that simulate clinical practice’. For this purpose the patients in the original placebo group were designated as receiving ‘episodic strategy’, and those in the infliximab groups as receiving a ‘5 mg/kg scheduled strategy’ and a ’10 mg/kg scheduled strategy’ respectively. From week 14 onwards patients who had shown a response to infliximab therapy at any time but then worsened were eligible to cross over to ‘active episodic treatment as needed with infliximab 5, 10, and 15 mg/kg for patients originally assigned to episodic, 5 mg/kg scheduled, and 10 mg/kg scheduled treatment strategies respectively’. This description is confusing as it clearly states that active episodic treatment is given in both episodic and scheduled strategies, which renders a comparison of episodic and scheduled strategies problematic. The publication designates the start of episodic treatment to be week 14 (see Appendix 9 for patient flow through the trial).

ACCENT I.2,3 Summary of effectiveness evidence for responders

Of the 573 patients (with baseline CDAI 220–400), 58.5% (335) achieved response 70 2 weeks after a single induction infusion of 5 mg/kg infliximab. These patients were designated ‘responders’. It is unclear if the three trial arms of randomised responders were well balanced at baseline. Of responders, (CiC information has been removed)% were in remission (CDAI < 150) at week 2. This represented (CiC information has been removed)% of the original 573 patients. The proportion of responders with remission had declined by week 30 to 23% (95% CI 14% to 29%) for those who only received placebo after induction and to 39% (95% CI 30.% to 48%) for those who received four infusions of 5 mg/kg infliximab (at weeks 2, 6, 14 and 22) and to 42% (95% CI 36% to 55%) for those who received four infusions consisting of 5 mg/kg at weeks 2 and 6 and 10 mg/kg at weeks 14 and 22. Risk differences (infliximab–placebo) and risk ratios (infliximab/placebo) for remission at week 30 reached statistical significance in favour of infliximab for both infliximab groups. By week 54 the percentage of patients in remission had diminished further in all three groups. Most of the advantage of intervention relative to placebo was achieved by weeks 10–14; thereafter risk differences remained fairly stable. A similar pattern of results was observed for response 70. Published information regarding maintenance of remission at the patient level (as distinct from group level) was meagre. Between weeks 14 and 54, 11% of placebo patients retained remission at all six study visits; the corresponding values were 25% and 33% respectively for 5 mg/kg and 10 mg/kg infliximab groups. Somewhat lower values of (CiC information has been removed)%, (CiC information has been removed)% and (CiC information has been removed)% respectively were quoted in the Industry Trial Report. Results favouring infliximab over placebo were reported for several other outcomes including median CDAI scores and median IBDQ scores. These measures required last or nearest observation carried forward in order to allow for missing data.

The treatment regimens received before week 14 in each of the randomised groups were as follows:

• placebo/‘episodic group’: 5 mg/kg infliximab week 0, placebo weeks 2 and 6

• 5 mg/kg group ‘scheduled strategy’: 5 mg/kg infliximab weeks 0, 2 and 6

• 10 mg/kg group ‘scheduled strategy’: 5 mg/kg infliximab weeks 0, 2 and 6.

Treatment to week 14 was therefore similar for the two infliximab ‘scheduled strategy’ groups and was determined according to randomisation. From week 14, crossover to an increase in infliximab dosage was allowed in all three trial arms for patients whose CD worsened. The criteria for worsening were ‘an increase CDAI of ≥ 70 points from the qualifying score with a total score of at least 175, an increase in CDAI of 35% or more from baseline value, or the introduction of new treatment for active Crohn’s disease’. From week 14 onwards it was possible for patients in different arms to be receiving identical infliximab treatment; for example, a placebo patient might cross over at week 14 to receive 5 mg/kg and this corresponds to treatment received by a 5 mg/kg ‘scheduled strategy’ patient who did not cross over. This complicates the interpretation of any comparisons between groups.

A, response 70

No primary outcome was identified. Analyses were according to randomised group irrespective of crossover after week 14 to different treatment regimen, and comparisons were drawn between the ‘episodic group’ and the two ‘scheduled strategy’ groups. The results for response 70 for all patients in ACCENT I are summarised in Figure 15. By week 14, statistically significant differences in CD status were evident between placebo group and intervention groups (p-values for risk differences and risk ratios are shown in Table 21). Risk differences and risk ratios for comparison between ‘episodic’ and ‘scheduled’ strategies after week 14 were in favour of ‘scheduled strategies’ but failed to reach statistical significance at most time points. Interpretation of these differences is problematic.

FIGURE 15.

Response 70 rates for all patients in ACCENT I. 2,3 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

TABLE 21 - p-values for comparison of response 70 rates at week 14 for all patients in ACCENT I2,3

	Risk difference (placebo–active intervention)		Risk ratio (placebo/active intervention)
	vs 5 mg/kg group	vs 10 mg/kg group	vs 5 mg/kg group	vs 10 mg/kg group
p	0.00725	0.00326	0.00865	0.00418

B, response 100

This outcome was not reported.

C, remission rates

Figure 16 summarises the published results for rates of remission at clinic visits to end of follow-up (week 54). Week 14 remission rates were greater in the two ‘scheduled treatment’ arms (37.5% in the 5 mg/kg group and 43% in the 10 mg/kg group) than in the ‘episodic’ group (25.5%). p-values for week 14 comparisons between placebo and intervention groups are shown in Table 22.

FIGURE 16.

Remission rates for all patients in ACCENT I. 2,3 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

TABLE 22 - p-values for comparison of remission rates at week 14 for all patients in ACCENT I2,3

	Risk difference (placebo–active intervention)		Risk ratio (placebo/active intervention)
	vs 5 mg/kg group	vs 10 mg/kg group	vs 5 mg/kg group	vs 10 mg/kg group
p	0.0113	0.0049	0.0135	0.0063

Treatment regimens up to week 14 were strictly pre-specified and designed to examine effectiveness for maintenance of the induced response. After week 14 treatment regimens became variable (termed ‘episodic’ by the authors). It is clear that by week 14 the CD status of patients in the placebo/‘episodic’ arm had departed from that of patients in the two ‘scheduled strategy’ arms; this means that at baseline (week 14) for the comparison of ‘episodic’ with ‘scheduled strategies’, the groups were imbalanced. Comparisons between ‘episodic’ and ‘scheduled’ strategies after week 14 are not randomised comparisons. For a randomised comparison of the two strategies patients should have been rerandomised at week 14. Such rerandomisation was precisely the study design adopted by Menter et al. 70 when comparing continuous with intermittent treatment strategies with infliximab in psoriasis.

Risk differences and risk ratios for comparison between ‘episodic’ and ‘scheduled’ strategies after week 14 were in favour of ‘scheduled strategies’, but failed to reach statistical significance at nearly all time points. Interpretation of these differences is problematic because, as described above, the comparisons are not between properly randomised groups and because patients in all groups were allowed the option of ‘episodic’ treatment.

D, other outcomes

Median CDAI score and the proportion of patients with IBDQ score > 170 were reported and are summarised in Table 23 and presented graphically in Appendix 10. By week 14, statistically significant differences in CDAI median scores were evident between the placebo group and the intervention groups. Differences were less pronounced after week 14, especially for the placebo versus 5 mg/kg comparison. The percentage of patients with IBDQ score > 170 did not differ significantly between placebo and 5 mg/kg groups, but after week 14 favoured the 10 mg/kg group relative to placebo.

TABLE 23 - CDAI and IBDQ results for all patients in ACCENT I2,3

NS, not significant.

Data read from graph in Rutgeerts et al. 2

Comparison: 5 mg/kg group vs placebo.

Comparison: 10 mg/kg group vs placebo.

There was no adjustment for repeated measures in the comparisons.

Week	CDAI: mediana				% patients with IBDQ score > 170a
	Placebo (n = 188)	5 mg/kg (n = 193)	10 mg/kg (n = 192)	p	Placebo (n = 188)	5 mg/kg (n = 193)	10 mg/kg (n = 192)	p
0	292	303	297	NSb,c	4.8	5.2	8.3	NSb,c
2	197.5	205	195	NSb,c	35.6	32.3	35.8	NSb,c
6	205	180	180	NSb,c	33.5	41.7	38.3	NSb,c
10	187.5	170	167.5	< 0.05b,c	35.1	41.7	39.9	NSb,c
14	225	185	182.5	< 0.05b,c	29.8	38.0	40.9	NSb< 0.05c
22	212.5	185	167.5	< 0.05b,c	29.3	37.0	44.0	NSb < 0.05c
30	212.5	180	177.5	NSb < 0.05c	33.5	39.6	44.0	NSb < 0.05c
38	200	187.5	170	NSb < 0.05c	35.1	34.9	47.7	NSb < 0.05c
46	205	190	175	NSb < 0.05c	33.5	35.4	48.7	NSb < 0.05c
54	205	185	170	NSb < 0.05c	35.1	37.5	46.1	NSb < 0.05c

The manufacturer’s submission provided information regarding CD-related hospitalisation rates and rates for intra-abdominal surgery. These rates and the relative risk for the 5 mg/kg ‘scheduled maintenance’ group relative to the ‘episodic’ are summarised in Table 24. The results for mucosal healing observed for a small subgroup of patients (n = 58) at European study centres who underwent endoscopy examination are also tabulated. The interpretation of the comparisons is problematical for the reasons already described, in particular after week 14. The extent to which avoidance of hospitalisation and abdominal surgery might depend on the administration of active intervention is not measurable because no true control (placebo) group existed after that time.

TABLE 24 - Endoscopy, hospitalisation and abdominal surgery results: all patients in ACCENT I2,3

NA, not applicable.

The hospitalisation rates were presented differently to other rates as number per 100 patients, rather than number per total at risk. We have calculated the number of hospitalisations based on the reported percentage and the known total numbers of patients.

Comparison for placebo vs infliximab.

Values presented in industry submission for hospitalisation were 0.6 (0.5 to 0.7) and for abdominal surgery were 0.3 (0.2 to 0.6).

	Endoscopy		Hospitalisationsa		Abdominal surgery
	Mucosal healing at week 54	p b	n/N (%)	Relative risk (95% CI)	n/N	Relative risk (95% CI)
Placebo	4/22	NA	71/188 (38%)	NA	14/188	NA
5 mg/kg group	8/19	0.093			5/193	0.348c (0.128 to 0.947)
10 mg/kg group	8/17	0.053			6/192
Combined 5- and 10 mg/kg groups	16/36	0.041	86/305 (23%)	0.591b,c (0.455 to 0.768)	11/385	0.373 (0.173 to 0.806)

Quality assessment of ACCENT I2,3 (all patients): (based on published report of Rutgeerts et al.2)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics at week 0 were well balanced. Where necessary, the nearest or last observation was carried forward for continuous outcomes but the number of missing data was not reported. No power calculation was conducted for the analysis of all patients. The number of patients who withdrew was reported except for patients who crossed over to a 15 mg/kg dose regimen from a 10 mg/kg regimen. The proportion of patients who withdrew before the end of the trial was substantial.

Trial design, withdrawals, crossovers and validity of comparisons

It must be questionable whether the ‘episodic’ (placebo) arm did ‘simulate clinical practice’ as stated to be an objective of the study. Patients in this arm of the study received one dose of 5 mg/kg infliximab at week 0, followed by an interim period of > 3 months with no active infliximab therapy before the ‘episodic’ use of infliximab according to worsening disease (for patients ‘who had responded at any time to infliximab therapy’). There is little evidence to support the idea that this resembles clinical practice. The scheduled strategy is difficult to define as it did not follow a prescribed programme of treatment as might be anticipated by the term ‘scheduled strategy’, but encompassed ‘episodic’ treatment in the same manner as the ‘episodic’ arm.

Because of the large numbers of patients who withdrew from treatment and crossed over to dose escalations, the actual treatments received in the three different trial arms are difficult to define. Figure 17 summarises the progression of patients through the trial with respect to withdrawal from treatment and crossover to increased dose of infliximab.

FIGURE 17.

Withdrawals and crossovers in ACCENT I. 2,3 *(CiC information has been removed). R, randomisation. Note: Dropouts (DOs) and week 14 or later crossovers (XOs) allowed ‘as required’.

Over a period of 1 year, about a quarter of patients withdrew from treatment, and of those allocated active intervention at randomisation only about half completed the trial receiving the treatment regimen to which they had been allocated at randomisation.

The authors’ stated primary objective ‘…was to examine the difference in efficacy between episodic and scheduled treatment strategies with infliximab’. 2 They concluded that the scheduled treatment strategy was superior to episodic treatment. Unfortunately, the comparisons were compromised by strong biases introduced as a result of the study design. These biases are explained below:

1. Crossover to increased infliximab was allowed for patients ‘who had responded at any time to infliximab therapy’ and subsequently worsened. In the placebo group (‘episodic strategy’), 78 of 188 patients (41%) were classified at week 2 as non-responders and received no further infliximab to week 14; these patients were unlikely to become responsive and therefore to qualify for crossover to active intervention. In contrast to this group the week 2 non-responders in the ‘scheduled strategy’ arms received additional doses of infliximab (5 mg/kg) at both weeks 2 and 6, boosting their opportunity to ‘respond at any time’ to infliximab. The greater opportunity to respond at any time in the ‘scheduled strategy’ arms represents a strong bias in their favour in any subsequent comparison with the episodic arm. Relative to the scheduled strategy this resulted in a substantial proportion of patients in the episodic arm being denied access to active therapy. This is reflected in the very large difference between arms in their exposure to infliximab stated to be 3 and 5 times greater in the two scheduled strategy arms than in the episodic arm.

2. Episodic treatment was introduced at week 14 of the trial, but by this time the CD status of patients in the placebo ‘episodic’ arm was significantly inferior to that in the scheduled strategy arms in terms of several efficacy measures. This advantage for the scheduled strategy arms is reflected in increases not seen in the placebo group from week 2 in the response 70 rates and at weeks 6 and 10 in the remission rates. The result is a bias in favour of scheduled strategy for any comparison between strategies at times after week 14. Essentially, the compared arms were unbalanced at the start of the compared strategies (week 14).

ACCENT I.2,3 Summary of effectiveness evidence for all patients

Two infusions of 5 mg/kg infliximab at weeks 2 and 6 after a single induction infusion of 5 mg/kg were better than placebo infusions at generating remission and response 70. At week 14, risk differences (infliximab–placebo) and risk ratios (infliximab/placebo) were in favour of infliximab and reached statistical significance (p < 0.02 for remission, p < 0.01 for response 70).

At week 14, ‘episodic’ treatment was introduced and subsequent comparisons were made between the original placebo arm (designated ‘episodic treatment strategy’) and original infliximab arms (termed ‘scheduled treatment strategies’). Because of bias strongly in favour of scheduled strategy groups, the post 14-week comparisons were not valid estimates of the relative effectiveness of strategies. Biases identified arose from: (a) reduced opportunity for crossover to active therapy for patients in the episodic group compared with the scheduled groups; and (b) gross imbalance in disease status at the start of the strategies (week 14).

Difficulties in interpreting post 14-week comparisons between groups were compounded by the very high rate of withdrawal from treatment and the use of ‘episodic’ treatment in all three arms of the trial so that the distinction between episodic and scheduled strategies was obscured except for the fact that the original infliximab groups were allowed larger dosages of active intervention.

CLASSIC II66 (adalimumab)

The CLASSIC II66 trial was an extension of the previously conducted adalimumab induction trial, CLASSIC I,63 which had enrolled 299 patients. To be eligible for CLASSIC II, patients were required to be in remission (CDAI < 150) at week 4 of CLASSIC I and also 4 weeks later (equivalent to week 8 of CLASSIC I and designated week 4 of CLASSIC II). These patients may have received two subcutaneous injections 2 weeks apart of various doses of adalimumab (40 mg then 20 mg, 80 mg then 40 mg, or 160 mg then 80 mg) or two injections of placebo. Fifty-five eligible patients entered CLASSIC II, this means about 12 patients did not retain remission from weeks 4–8 of CLASSIC I or declined to participate. The 55 patients were randomised at week 4 of CLASSIC II to receive placebo (n = 18) or 40 mg of adalimumab every other week (e.o.w.) (n = 19) or 40 mg of adalimumab weekly (n = 18) from weeks 4 to 54. Thus CLASSIC II analysed only strong responders from the CLASSIC I trial.

For the purposes of the ‘primary efficacy analysis’, patients who had continued non-response defined as ‘a decrease in CDAI ≤ 70 points vs. Week-0 value in CLASSIC I’ were considered treatment failures and became eligible for open-label treatment. This means patients in remission at start of CLASSIC II became treatment failures if they ceased to qualify as response 70 responders relative to their baseline CDAI score in CLASSIC I. In addition, patients who flared during CLASSIC II follow-up were also counted as treatment failures and were eligible for open-label treatment. CD flare was defined as an increase of ≥ 70 points above the week 4 CLASSIC II value (which by definition was < 150) AND a CDAI score > 150 (no longer in remission). Thus a patient in remission at week 4 (CLASSIC II) with a CDAI score of 149 would need to move to a CDAI of at least 219 to be classified as having experienced flare. For this patient a score of 218 would not count as a flare but could count as treatment failure if his or her week 0 CLASSIC I CDAI score had been < 288 (for reference the mean baseline CDAI score at week 0 for 299 CLASSIC I patients was 298).

A, response 70 and B, response 100

Response 100 and response 70 rates throughout follow-up were among the secondary outcome measures of efficacy. Results reported for responses 100 and 70 are summarised in Figure 18. The placebo rates were high for these less rigorous measures of effectiveness, and the risk differences (adalimumab–placebo) and risk ratios (adalimumab/placebo) failed to reach statistical significance at most time points.

FIGURE 18.

Response 100 (upper panel) and response 70 (lower panel) rates in CLASSIC II. 66 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

C, remission

The primary outcome was the proportion of patients in remission at week 56 in each arm of the randomised cohort. Remission throughout follow-up was among the secondary outcome measures of efficacy. For the primary outcome, 10 patients (18%) withdrew before week 56 (five from placebo and five from adalimumab). These were counted as remission failures for the primary analysis. Remission rates at week 56 are summarised in Table 25. Remission rates during the trial are summarised in Figure 19.

TABLE 25 - Remission rates at week 56 in CLASSIC II66 (primary outcome)

NA, not applicable.

Adalimumab vs placebo.

60 patients increased regimen from e.o.w. to weekly.

Trial report data; publication states 94.

Dose regimen (n )	Number in remission (%; 95% CI)	p a
Placebo (18)	8 (44%; 25 to 66)	NA
40-mg adalimumab e.o.w. (19)	15 (79%; 57 to 91)	< 0.05
40-mg adalimumab weekly (18)	15 (83%; 61 to 94)	< 0.05
Open-label [(CiC information has been removed)b]	(CiC information has been removed)c [(CiC information has been removed)%]

FIGURE 19.

Remission rates in CLASSIC II. 66 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Point estimates of remission rate during the trial were associated with considerable uncertainty, reflecting the small number of patients in the trial. The fact that rates rose and fell during follow-up indicated the values reported referred to point prevalence. Nearly half of patients in the placebo group were in remission at week 56 despite not receiving active intervention from 2 weeks prior to randomisation onwards. Risk differences (intervention–placebo) and risk ratios (intervention/placebo) were in favour of intervention at all follow-up times and reached statistical significance at several time points.

D, other outcomes

The results published for continuous measures are summarised in Table 26. These measures involved last observation carried forward to allow for missing values. The amount of missing values was not published but was available (CiC) in the unpublished Industry Trial Report. For week 56 changes in favour of adalimumab relative to placebo were reported for mean IBDQ and CDAI scores.

TABLE 26 - IBDQ scores, CDAI scores and CRP concentrations reported in CLASSIC II66

NR, not reported.

Data read from graph except for weeks 24 and 56; last observation carried forward; the number observations at weeks 24 and 56 (CiC information has been removed) for the placebo, e.o.w. and weekly groups respectively.

Week 4 of CLASSIC I.

Comparison e.o.w. adalimumab vs placebo.

Comparison weekly adalimumab vs placebo.

Last observation carried forward number of observations (CiC information has been removed) for placebo, e.o.w. and weekly groups respectively; CLASSIC I baseline mean (95% CI) CDAI scores for these patients (not published) were (CiC information has been removed) for placebo, adalimumab e.o.w. and adalimumab weekly groups respectively.

	Placebo (n = 18)	Adalimumab 40 mg e.o.w. (n = 19)	Adalimumab 40 mg weekly (n = 18)	p
Mean IBDQ scorea
Week 0b	187.5	181	191.5
Week 4	188.5	187	191
Week 8	178	181	187
Week 12	170.5	182.5	189
Week 16	172.5	181	182
Week 20	170.5	177	186.5
Week 24	167.6	176.3	192.2	< 0.005c,d
Week 32	166.5	182	192	< 0.05c < 0.005d
Week 40	167	179	188	< 0.005c,d
Week 48	163.5	178	183.5
Week 56	162.4	178.4	185.6
CDAI: mean change (95% CI) from baseline in CLASSIC Ie
Week 56	–119.6 (–74 to –65.1)	–158 (–202 to –99.8)	–197.7 (–248 to –147)	< 0.005c,d
CRP concentration mg/dl: median (range) [levels of CRP < 0.88 mg/dl are considered normal]
24	0.5 (0 to 1.2)	0.4 (0 to 1.9)	0.1 (0 to 1.6)	NR
56	0.4 (0 to 0.9)	0.3 (0 to 2.8)	0.3 (0 to 1.2)	NR

At the start of CLASSIC II,66 49% of patients were receiving systemic steroids or budesonide; seven of the placebo group, seven of the e.o.w. adalimumab group, and eight of the weekly adalimumab group. Using the last observation carried forward it was reported that by week 56 the number who had discontinued steroids was four in both the placebo and e.o.w. adalimumab groups, and seven in the weekly adalimumab group.

E, other considerations – open-label study

Most patients from CLASSIC I63 who did not qualify for CLASSIC II66 participated in an open-label study in parallel with CLASSIC II. The results reported were not randomised comparisons and are outwith the inclusion criteria for this report.

Quality assessment (based on published report)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics at week 0 were well balanced. The study was powered for the primary outcome (remission at week 4) of CLASSIC I, and no further power calculation was conducted for CLASSIC II. The number of patients who withdrew was reported; 5 of 18 placebo patients withdrew and 5 of 37 patients given adalimumab withdrew. There were 32 patients (58%) who completed to 56 weeks of double-blind follow-up. The last observation was carried forward as necessary for continuous outcomes, but the number of missing data was not reported.

CLASSIC II.66 Summary of effectiveness evidence

The trial population (n = 55) was recruited from responders in the previous CLASSIC I63 adalimumab induction trial (n = 299). Only responders with a strong response (remission for at least a month) were selected; they had received various induction dose regimens.

Maintenance injections of 40 mg of adalimumab administered weekly or e.o.w. generated a statistically significant greater proportion of patients in remission at week 56 than did placebo (frequency of administration not published). About half of the placebo group and 81% of those who received infliximab were in remission at week 56. Point estimates of response rates were associated with considerable uncertainty due to the small size of the trial. There were no statistically significant differences in effectiveness between e.o.w. and weekly adalimumab regimens.

CHARM67 (adalimumab)

This was a free-standing maintenance trial (i.e. newly started). 67 There were 854 enrolled patients (CDAI range 220–400), of whom 130 (15.2%) had fistulas at screening and baseline. An induction regimen consisting of an 80-mg injection of adalimumab at week 0 and a 40-mg injection 2 weeks later was followed by randomisation of 778 patients at week 4 to one of three arms as follows: placebo to week 56 (n = 261), 40 mg adalimumab e.o.w. to week 56 (n = 260) and 40 mg adalimumab weekly to week 56 (n = 257). There were 76 (8.9%) withdrawals prior to randomisation. Assessment visits were planned for weeks 0, 2, 4, 6, 8, 12, 16, 20, 26, 32, 40, 48, 56 and 60.

At week 4 patients were classified as responders or non-responders. Responders had to have a reduction of ≥ 70 CDAI points relative to baseline. Of the 854 patients given the induction regimen, 499 (58%) were categorised as responders and were the focus of the published effectiveness results. This population was different to that followed up in the other adalimumab maintenance trial, CLASSIC II,66 in that the latter were on average better responders, having achieved remission from induction. The numbers of responders randomised to the three trial arms of CHARM were 170 to placebo, 172 to adalimumab e.o.w. and 157 to adalimumab weekly.

The coprimary outcome measures were designated: the percentage of week 4 responders who achieved remission at weeks 26 and 56. Pre-specified secondary outcomes included (1) percentage achieving response 70 and response 100 at weeks 26 and 56; (2) change in IBDQ score from baseline at weeks 26 and 56; (3) percentage achieving clinical remission at weeks 26 and 56 who were able to discontinue corticosteroid use; (4) percentage achieving clinical remission at weeks 26 and 56 who were able to discontinue steroids for ≥ 90 days; (5) percentage of patients with fistula remission (closure of all fistulas that were draining at screening and baseline visits); and (6) median time in clinical remission among randomised responders achieving remission. Post hoc analyses examined subgroup responses and sustainability of response.

At or after week 12, patients with disease flare (an increase of ≥ 70 CDAI points from the score at week 4 and a CDAI score > 220) or sustained non-response (CDAI score not reduced by ≥ 70 points from week 0) were eligible to cross over to 40-mg adalimumab e.o.w. which could be escalated to 40-mg weekly for patients with continued non-response or recurrent flare. For the primary effectiveness outcome (responders), any patients who crossed over were counted as remission failures.

A, response 70 and B, response 100

The published response 70 and response 100 rates at weeks 26 and 56 are summarised in Table 27. Rates reached statistical significance in favour of adalimumab for both dose regimens at both time points.

TABLE 27 - Reported response 100 and response 70 rates in CHARM67

NA, not applicable.

Adalimumab vs placebo; chi-squared test.

Dose regimen (n)	Number with response 100 (%; 95% CI)		p a
	Week 26	Week 56
Placebo (170)	45 (26.5%; 20 to 34)	28 (16.5%; 12 to 23)	NA
40-mg adalimumab e.o.w. (172)	89 (52%; 44 to59)	71 (41%; 34 to 49)	< 0.001
40-mg adalimumab weekly (157)	82 (52%; 44 to 60)	75 (48%; 40 to 56)	< 0.001
	Number with response 70 (%; 95% CI)
	Week 26	Week 56
Placebo (170)	48 (28%; 22 to 35)	30 (18%; 13 to 24)	NA
40-mg adalimumab e.o.w. (172)	93 (54%; 47 to 61)	74 (43%; 36 to 50)	< 0.001
40-mg adalimumab weekly (157)	88 (56%; 48 to 63)	77 (49%; 41 to 57)	< 0.001

The unpublished Industry Trial Report for CHARM provided (CiC) values for response 70 at time points for all assessment visits. These are summarised in Figure 20. (CiC information has been removed.)

FIGURE 20.

Response 70 rates among responders in CHARM. 67 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Similar CiC results were observed for response 100 and are summarised in Figure 21. From week 8 onwards (CiC information has been removed).

FIGURE 21.

Rates of response 100 among responders in CHARM. 67 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

C, remission

The primary outcome was the proportion of patients in remission at weeks 26 and 56. The results are summarised in Table 28. The difference between adalimumab groups and placebo reached statistical significance in favour of adalimumab for both dose regimens.

TABLE 28 - Remission at weeks 26 and 56 in CHARM67

NA, not applicable.

Adalimumab vs placebo; Cochran–Mantel–Haenszel chi-squared test adjusting for previous anti-TNF use.

Dose regimen (n)	Number in remission (%; 95% CI)		p a
	Week 26	Week 56
Placebo (170)	29 (17%; 12 to 23)	20 (12%; 8 to 13)	NA
40-mg adalimumab e.o.w. (172)	68 (40%; 32 to 47)	62 (36%; 29 to 43)	< 0.001
40-mg adalimumab weekly (157)	73 (47%; 39 to 54)	65 (41%; 34 to 49)	< 0.001

The secondary outcomes of remission rates for each follow-up visit to week 56 are summarised in Figure 22. Risk differences (adalimumab–placebo) and risk ratios (adalimumab/placebo) reached statistical significance in favour of adalimumab at all time points after week 6. Rates of remission in the adalimumab e.o.w. arm diminished through follow-up. From weeks 12–16 onwards, risk differences remained stable so that most benefit of the intervention appeared to be delivered in the first quarter of the trial. The rates reported were group point prevalence values and do not reflect maintenance of remission at the patient level. The difference in rates between the two adalimumab regimens at week 56 was not significant (risk difference p = 0.32, risk ratio p = 0.32).

FIGURE 22.

Remission rates reported during follow-up in CHARM. 67 At week 56 risk difference and risk ratio for both regimens of adalimumab versus placebo, p < 0.0001. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Patient level maintenance of remission was published for weeks 26–56. In the adalimumab arms, 81% of patients in remission at week 26 sustained remission to week 56; this represented 114 patients and 27% of all those randomised to adalimumab. For patients randomised to placebo, 48% of those in remission at week 26 sustained remission to week 56. This represented 14 patients and 5% of all those randomised to placebo. The median time in clinical remission that started at any time was 127 days for the placebo group, 378 days for the adalimumab e.o.w. group and > 392 days for the adalimumab weekly group (p = 0.002 and p < 0.001 vs placebo respectively). Over 56 weeks it was possible for a patient to enter a remission state on several occasions. The publication did not make clear which occasion(s) were used in the analysis or how and if double counting was avoided.

D, other outcomes

Published mean CDAI and IBDQ scores are summarised in Table 29. No variance information was provided. After week 12, CDAI and IBDQ scores for patients who crossed over to increased adalimumab doses were included in the calculation of group mean scores although this was not made explicit. Mean CDAI scores decreased and mean IBDQ scores increased, to a greater degree respectively in the adalimumab groups than the placebo group. Given that a true placebo group did not exist after week 14, the results thereafter are difficult to interpret. The last observation was carried forward; the proportion of patients evaluated at week 56 (CiC information has been removed).

TABLE 29 - Group mean CDAI and IBDQ scores reported for responders in CHARM67

NR, not reported.

Data read from graph in Colombel et al. 67. Values after week 12 were calculated including values for patients who crossed over to increased adalimumab.

Week	CDAI:meana				IBDQ mediana
	Placebo (n = 170)	40-mg adalimumab e.o.w. (n = 172)	40-mg adalimumab weekly (n = 157)	p	Placebo (n = 170)	40-mg adalimumab e.o.w. (n = 172)	40-mg adalimumab weekly (n = 157)	p
0	318	317	310	NR	125	128	123	NR
2	215	200	203	NR	NR	NR	NR	NR
4	170	153	162	NR	166.6	174	165	NR
6	178	150	162	NR	NR	NR	NR	NR
8	183	147	155	NR	NR	NR	NR	NR
12	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
16	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
20	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
26	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
32	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
40	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
48	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR
56	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR	(CiC information has been removed)	(CiC information has been removed)	(CiC information has been removed)	NR

From week 8 the responder patients who were receiving steroids at baseline could begin reducing steroid use (presumably at the physician’s discretion). This involved 66 placebo patients, 58 and 74 patients respectively in the adalimumab e.o.w. and adalimumab weekly groups. The percentage of these patients who were in remission at week 26 and who had discontinued steroids was 3% (2/66) in the placebo group, and 34% (20/58) and 30% (22/74) in the adalimumab e.o.w. and weekly groups respectively. Corresponding percentages at week 56 were 6%, 29% and 23% respectively. The percentage who were in remission at week 26 and who were steroid free for at least 90 days was 3% in the placebo group and 19% and 15% in the adalimumab e.o.w. and weekly groups respectively. Corresponding percentages at week 56 were 5%, 29% and 20% respectively.

Hospitalisation rates

Details on hospitalisation rates from the CHARM trial67 were reported in the industry submission, referenced to published abstracts by Wu et al. 71 and by Feagan et al. 72 The latter abstract reports the hospitalisation rates in the placebo arm and the combined adalimumab arms, which were 22.4% and 14.0% respectively. The 56-week actuarial CD-related hospital admission rates for the placebo and for the combined adalimumab arms were 13.9% and 5.9% respectively. A difference in relative risk was apparent at 2 weeks after randomisation, and placebo patients had 4.5 times the risk of hospitalisation at month 3 as adalimumab patients. Wu et al. 71 used a Cox proportional hazard regression model and found that lower CDAI scores were associated with a decreased risk of hospitalisation and CD-related hospitalisation. Simulated 1-year rates indicated that a 70-point reduction on the CDAI throughout the follow-up period reduced all-cause hospitalisation risk by 28.3% and CD-related hospitalisation by 36.5% at year end. Further simulations indicated that remission was associated with a 43.7% decrease in the 1-year risk of all-cause hospitalisation and a 60.3% decrease in CD-related hospitalisation.

E, other considerations – subgroup analyses and crossover issues

Outcomes for patients with draining fistulas are included in the next section.

The manufacturer’s submission to NICE provided weeks 26 and 56 results for placebo and e.o.w. adalimumab group patients who had severe disease at baseline (CDAI > 300). Results for all severe patients and for severe week 4 responders were provided allowing calculation of results for non-responders with severe CD (Table 30). There were 96 severe CD patients in both placebo and e.o.w. adalimumab groups. Rates of remission, response 70 and response 100 are summarised in Figure 23. Remission rates at week 56 in adalimumab and placebo arms were 35% and 10% respectively; higher rates were recorded for the less stringent response 70 and 100 outcomes. These rates were similar to those reported for all week 4 responders (within 5%; listed in Table 30). The rates in the e.o.w. arm for week 4 non-responders with severe CD were about half those for week 4 responders with severe CD.

TABLE 30 - Response rates for severe CD patients in CHARM67

Outcome	Week	All week 4 responders		Severe CD responders		All severe CD patients		Severe CD non-responders
		Placebo (n = 170)	e.o.w. anti-TNF (n = 157)	Placebo (n = 96)	e.o.w. anti-TNF (n = 96)	Placebo (n = 149)	e.o.w. anti-TNF (n = 135)	e.o.w. anti-TNF (n = 39)
Remission	26	17%	40%	14%	36%	11%	30%	15%
	56	11.8%	36%	9%	33%	8%	27%	8%
Response 100	26	26.5%	52%	28%	56%	21%	47%	26%
	56	16.5%	44%	17%	44%	13%	36%	18%
Response 70	26	28%	54%	29%	58%	23%	47%	28%
	56	17.6%	43%	18%	45%	13%	36%	18%

FIGURE 23.

Response and remission rates for severe disease responders in CHARM. 67 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Other post hoc subgroup analyses

Several post hoc analyses explored the effectiveness of adalimumab among subgroups of patients defined according to various criteria including: baseline CRP level > or < 1 mg/ml; concomitant treatment with or without immunosuppressant medication; and previous experience of anti-TNF therapy or no previous experience. No statistically significant subgroup differences in adalimumab effectiveness were observed.

Premature withdrawal from treatment and crossover due to worsening disease

The published information about withdrawal from treatment and crossover to open-label therapy was difficult to disentangle. The Industry Trial Report provided fuller detail. Of 499 responders 29% (144) withdrew prematurely: (CiC information has been removed)% of the placebo group [(CiC information has been removed)], (CiC information has been removed)% of the e.o.w. adalimumab group [(CiC information has been removed)] and (CiC information has been removed)% of the weekly adalimumab group [(CiC information has been removed)]. The Industry Trial Report stated the overall premature discontinuation rate among all patients was (CiC information has been removed)% [(CiC information has been removed)], with 79 of these occurring before randomisation. Among all 788 randomised patients, withdrawals during the randomised phase were (CiC information has been removed) [(CiC information has been removed)%] in placebo group, (CiC information has been removed) [(CiC information has been removed)%] in the adalimumab e.o.w. group and (CiC information has been removed) ((CiC information has been removed)%) in the adalimumab weekly group, giving an overall rate of (CiC information has been removed)% [(CiC information has been removed)/(CiC information has been removed)] slightly (CiC information has been removed) than for responders only. Of patients randomised to adalimumab maintenance therapy, the rate of premature withdrawal was the (CiC information has been removed).

Crossover to open-label treatment after week 12 involved (CiC information has been removed) [(CiC information has been removed)%] of patients randomised to placebo, (CiC information has been removed) [(CiC information has been removed)%] of those randomised to adalimumab e.o.w. and (CiC information has been removed)/(CiC information has been removed) [(CiC information has been removed)] of those randomised to adalimumab weekly. These numbers represented patients experiencing worsening disease by flare or discontinued response. Transfer to open-label for patients in the weekly adalimumab group involved continuation of the same dose regimen (as crossover was described as ‘…switched to open-label treatment with 40-mg adalimumab eow… escalated to 40-mg weekly for those with continued non-response or recurrent flare’. After crossover ‘…continued non-response with open-label 40-mg weekly dosage resulted in withdrawal’;67 however, there was no published information about how long the state of flare or non-response was allowed to continue before withdrawal was implemented. The number of responder patients who crossed over to open-label was not published. The Industry Trial Report allowed calculation of crossovers and withdrawals among all randomised patients; this information is summarised in Figure 24.

FIGURE 24.

Withdrawals from treatment and crossovers for flare or non-response in CHARM. 67 *There was a discrepancy concerning one patient in the values for the adalimumab weekly group. (CiC information has been removed.)

Quality assessment (based on published report)

Randomisation, allocation concealment and blinding were adequate. Baseline characteristics were reported only for all patients, for all responders and for all non-responders, not for each of the trial arms. It was therefore not possible to judge if baseline characteristics were evenly balanced between the three arms of responders (placebo and e.o.w. or weekly adalimumab groups) that were analysed for effectiveness outcomes. The frequency of placebo injections was not documented. Information about patients who withdrew was reported. After week 12, patients with disease flare or non-response were allowed to cross over to the open-label treatment. It was difficult to determine how many responders and how many randomised patients in each group crossed over to open-label treatment. There was no statement defining how long after crossover flare or non-response was allowed to continue before withdrawal was implemented. Where necessary, the nearest or last observation was carried forward for continuous outcomes, but this was not stated explicitly and the number of missing data was not reported. A power calculation was conducted and based on the primary analysis of 4-week responders achieving remission at weeks 26 and 56.

The published text stated ‘…secondary efficacy analyses were conducted for all treated patients, including both randomised responder and randomised non-responder groups (all randomised patients who failed to achieve a clinical response at week 4)’. 67 Although this might be technically correct, in the sense that analyses were conducted, it is misleading because the results of these analyses were not reported, with the single exception of data on healing of fistulas for a subgroup of patients with fistulas at baseline and screening.

CHARM.67 Summary of effectiveness evidence

Seven hundred and seventy-eight patients given induction injections of 80 mg and 40 mg of adalimumab separated by 2 weeks were randomised at week 4 to maintenance therapy with placebo or 40-mg adalimumab e.o.w. or weekly. Only results for responders were published. Responders were defined as patients who at week 4 had a CDAI score reduced by ≥ 70 points from baseline. At weeks 26 and 56 there were significantly more responder patients in remission in the e.o.w. and weekly adalimumab groups than the placebo group, 40% and 47% respectively versus 17% at week 26, and 36% and 41% respectively versus 12% at week 56 (p < 0.001 for adalimumab vs placebo). The risk difference (adalimumab–placebo) for remission reached statistical significance in favour of adalimumab from week 8 onwards and remained stable from about week 12 or 16 to the end of follow-up (week 56), indicating that most of the benefit from active intervention was delivered during the first quarter to third of the trial.

The proportion of responders (response 70) had diminished to < 50% in all groups by week 56. Response 70 rates diminished (CiC information has been removed) delivered during the first part of the trial. Premature withdrawal from randomised treatments (adalimumab and placebo) was (CiC information has been removed)%; withdrawal rate from active intervention (adalimumab) was (CiC information has been removed) responders (CiC information has been removed)% (CiC information has been removed) non-responders [(CiC information has been removed)%]. Amongst the whole trial population randomised to adalimumab maintenance therapy, (CiC information has been removed)% crossed over to open-label treatment due to flare or non-response. The distribution of crossovers between responders and non-responders was unclear.

Present et al., 199962 (infliximab)

Present et al. 62 was a small study that randomised 31 patients to placebo, 31 and 32 patients respectively to 5 mg/kg or 10 mg/kg infliximab infused at weeks 0, 2 and 6. Follow-up extended to at least week 18 with assessment visits every 4 weeks from week 2 onwards.

Primary outcome

The primary outcome was a > 50% reduction in the number of draining fistulas relative to baseline evaluated by physical evaluation and observed over at least two consecutive study visits at any time during the trial. Secondary outcomes included: complete absence of draining fistula observed over at least 4 weeks (i.e. across at least two consecutive study visits) at any time during the study, time to beginning of response and duration of response. Changes in CDAI and PDAI scores were reported for some patients.

The results for the primary outcome (for 50% reduction in draining fistula occurring at any time over at least two consecutive clinic visits and for complete absence of draining fistula over two consecutive clinic visits) are summarised in Figure 25. For both these outcomes, infliximab at both dose regimens was more effective than placebo (p = 0.002 and p = 0.02 for 5- and 10 mg/kg regimens respectively). The point estimates for response rates were associated with substantial uncertainty because of the small group size; for the combined infliximab groups the response rate was 62% (95% CI 50% to 73%) compared with 26% (95% CI 14% to 43%) for the placebo group (p < 0.001). For those with a response, the median time to response was 6 weeks in the placebo group and 2 weeks in the infliximab groups (Table 33).

FIGURE 25.

Rates and risk differences for a 50% reduction and absence of draining fistulas. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

TABLE 33 - Time to onset of primary outcome

NA, not applicable; NR, not reported.

Only patients with primary response included.

Length of time to the beginning of a responsea (days: median, IQR)
	Placebo (n = 8)	5 mg/kg (n = 21)	10 mg/kg (n = 18)	5 or 10 mg/kg (n = 39)
Median (days)	42	14	14	14
IQR	15–72	14–42	14–42	14–42
p vs placebo	NA	NR	NR	NR

Response and complete response

The median duration of response (defined as the maximum period during which the patient experienced a 50% reduction in draining fistulas) was approximately 3 months. For infliximab patients, 29/63 (46%; 95% CI 34% to 58%) experienced complete absence of draining fistulas for at least two consecutive clinic visits compared with 4/31 (13%; 95% CI 5% to 29%) of patients in the placebo group (p < 0.001).

The median CDAI and PDAI scores reported for baseline and weeks 2 and 18 of follow-up are summarised in Table 34. By week 2, statistically significantly better (i.e. lower) scores were found for the infliximab groups than for the placebo group. The statistical significance of the difference between groups had weakened or disappeared by week 18. Not all patients contributed data for the analyses (i.e. this was not an ITT analysis).

TABLE 34 - CDAI and PDAI scores reported in the Present et al. trial62

Anti-TNF vs placebo using analysis of variance procedure.

Last observation carried forward for missing values.

Outcome	Placebo (n = 25)			Infliximab 5 mg/kg Weeks 0, 2 and 6; (n = 27)			Infliximab 5 mg/kg Weeks 0, 2 and 6; (n = 27)
	Week	Median	IQR	Median	IQR	p a	Median	IQR	p a
CDAI scoreb	0	162	126–265	163	99–284	0.71	203	112–254	0.66
	2	171	114–252	108	83–203	0.04	111	89–164	0.06
	18	160	72–206	104	47–177	0.23	123	58–175	0.32
PDAI scoreb	0	9	7–10.5	8	7–10	0.69	10.0	8.0–12.0	0.31
	2	8	6–10	6	3–7	0.02	6.0	4.0–8.0	0.04
	18	7	4–9	4	1–7	0.05	5.0	3.0–8.0	0.14

Quality assessment based on published report62

Randomisation and blinding were adequate and allocation concealment was likely to have been adequate. Baseline characteristics were generally well balanced although there was a greater proportion of patients in the infliximab groups that had undergone previous segmental resections than in the placebo group. Draining fistulas of < 3 months’ duration were excluded from the primary analysis. However, the number or frequency of these fistulas was not reported, and it was unclear if these were also excluded from the secondary outcome of a complete absence of a draining fistula. Total follow-up time for the primary outcome was unclear. No power calculation was performed. Last observation was carried forward for CDAI and PDAI analyses, but the number of missing data was unclear. There were only six premature withdrawals from treatment, four from the placebo group and one patient from each of the infliximab groups.

Present et al., 1999.62 Summary of effectiveness evidence

Patients with one or more draining fistula of more than 3 months’ duration, and an unreported number of fistulas of < 3 months’ duration, were randomised to placebo or 5 mg/kg infliximab or 10 mg/kg infliximab, by i.v. infusion at weeks 0, 2 and 6. More patients in the infliximab groups than in the placebo group achieved the primary outcome defined as: a reduction in the number of 3-month duration draining fistulas present at baseline by at least 50% lasting for at least two consecutive clinic visits. The percentage of patients responding to infliximab was 62% (95% CI 50% to 73%) compared with 26% (95% CI 14% to 43%) for the placebo group (p < 0.002). The median time to response was 2 weeks for infliximab groups and 6 weeks for placebo group. The duration of response was the same for both groups (median about 12 weeks).

More patients in the infliximab groups than in the placebo group achieved the secondary outcome of absence of draining fistula lasting for at least two consecutive clinic visits. The percentage of patients responding to infliximab for this outcome was 46% (95% CI 34% to 58%) compared with 13% (95% CI 5% to 29%) for the placebo group (p < 0.001).

ACCENT II65 (infliximab)

This was a maintenance trial that recruited 306 patients who had one or more fistulas of at least 3 months’ standing. 65 Of the 306 enrolled patients, 282 were assessed for ‘response’ at week 14 after administration of infusions at weeks 0, 2 and 6 of 5 mg/kg infliximab. ‘Responders’ were defined as those patients with at least 50% reduction in draining fistulas relative to baseline, observed at both weeks 10 and 14. Sixty-nine per cent (195) of patients were classified as responders. Both responders and non-responders were randomised to placebo (96 responders, 43 non-responders) or to 5 mg/kg infliximab (99 responders, 44 non-responders), which were administered at weeks 14, 22, 30, 38 and 46. Assessment visits were scheduled at weeks 0, 2, 6, 10, 14, 22, 30, 38, 46 and 54. After week 22, patients losing response could cross over to 5 mg/kg infliximab from placebo and from 5 mg/kg infliximab to 10 mg/kg infliximab. The fistula status outcome measures were:

• loss of response – defined as a recrudescence of draining fistula, a change in therapy, a need for surgery, dropout due to lack of efficacy, or a worsening of luminal disease activity

• response – defined as 50% reduction from baseline in draining fistula observed at consecutive visits 4 or more weeks apart

• complete absence of draining fistula.

Primary outcome

The primary outcome was designated as time to loss of response in responders. The results are summarised in Figure 26.

FIGURE 26.

Time to loss of response by responders in ACCENT II. 65 Data taken from published graph and redrawn.

The median time to loss of response after randomisation was 14 weeks in the placebo group and > 40 weeks in the infliximab group (p < 0.001 by log rank test). In the infliximab group, 42% of responders lost response, and in the placebo group, 62% lost response. The main reasons for loss of response in the primary outcome were: change in treatment (38% of placebo, 25% of infliximab) or recrudescence of fistula (22% placebo, 16% infliximab).

Response and complete response

At 30 weeks, 33% and 64% of the placebo and infliximab groups respectively had a response (50% reduction in draining fistula from baseline for at least two consecutive visits), and at week 54 the corresponding percentages had diminished to 23% and 46% respectively (p = 0.001). The manufacturer’s submission to NICE contained CiC information for additional weeks of follow-up. These are summarised in Figure 27. Prior to randomisation, except at week 2, the rates were about equal as would be expected, given that all responder patients received identical induction therapy up to week 14 and baseline characteristics were well balanced. At week 2, a surprising difference between groups was observed with higher rates for the patients subsequently randomised to infliximab. At week 14, the placebo group did not receive infliximab. After randomisation at week 14, response rates diminished in both groups. From week 22 the risk difference (infliximab–placebo) reached statistical significance in favour of infliximab, after week 30 risk differences diminished indicating that most benefit for maintenance of response from active intervention was delivered between weeks 14 and 30. By week 30 the intervention group had received two more infusions of infliximab than placebo patients.

FIGURE 27.

Rates and risk differences for ≥ 50% reduction of draining fistulas in ACCENT II65 responders. LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

Responders with loss of response during the post-randomisation phase were allowed to cross over after week 22 to an increased dose of infliximab. The renewed response rate in these crossover patients was reported as 25/41 (61%) in the placebo group (crossed over to 5 mg/kg dose) and 12/21 (57%) in the intervention group (crossed over to 10 mg/kg dose). However, Figure 1 of the ACCENT II65 published report shows 50 crossovers from placebo and 28 from 5 mg/kg infliximab.

The published report provided information about the rates of ‘complete response’ among responders. 65 A complete response was defined as a complete absence of draining fistulas. The definition for a response required ≥ 50% reduction in fistulas for at least 4 weeks, a ‘complete response’ differed in that no minimum duration was specified. It was unclear, but likely, that this definition applied only to draining fistulas of at least 3 months’ standing at baseline. The frequency of draining fistulas at baseline that were of less than 3 months’ standing was not reported. The results for a complete response are summarised in Figure 28.

FIGURE 28.

Rates, risk difference and risk ratio for complete response among responders in ACCENT II. 65 LCI, lower confidence interval; RD, risk difference; RR, risk ratio; UCI, upper confidence interval.

At week 2, after only a single dose of infliximab, 66% (128/195) of patients already had a ‘complete response’. Unexpectedly, more patients who were subsequently randomised to infliximab had a complete response than those subsequently randomised to placebo (p = 0.014). By 14 weeks, 66% and 69% of responder patients who were randomised to placebo and infliximab respectively had a complete response. The rate of complete response in responders diminished in both groups after week 14. From week 22 the risk difference (infliximab–placebo) reached statistical significance in favour of infliximab and from week 30 remained stable, indicating that most benefit in maintenance of response from active intervention was delivered between weeks 14 and 30.

The rate for a complete response among all enrolled patients at week 14 was reported to be 48% (147/306); this generated a 75% [147/(99 + 96)] complete response for responders at week 14 which according to Figure 2B in the ACCENT II65 published report corresponded to week 10 rather than to week 14.

Hospitalisations and major surgery

The manufacturer’s submission to NICE presented results for major surgery and for hospitalisation for all patients in ACCENT II65 whether or not they crossed over. For this purpose the placebo arm was termed ‘episodic treatment’ and the infliximab arm ‘scheduled treatment’. A 2.4-fold lower rate was reported for the scheduled treatment group. There were two important differences between these treatments. Firstly, patients in the ‘episodic’ arm experienced a 4-month mandatory withdrawal of active intervention (from weeks 6 to 22) not experienced by patients in scheduled treatment. Secondly, after week 22 the ‘episodic’ group patients were restricted to 5 mg/kg infliximab at episodes of worsening disease, whereas the ‘scheduled treatment’ group were able to receive 10 mg/kg. Restricted access to treatment (weeks 6–22) and restricted dosage represent biases likely to favour the ‘scheduled treatment’ group for any comparisons after week 6. Furthermore the ‘episodic treatment’ procedure was unlikely to reflect how an episodic strategy might be implemented in real world clinical practice, both with respect to the 4-month gap in active intervention and with regard to restriction of dose. Because of bias in the comparisons made and the probable dissimilarity between the trial episodic treatment and likely clinical practice, it was considered here that the hospitalisation rate for the ‘episodic’ treatment and the comparison with the scheduled treatment were very approximate guides.

The considerations described above also apply to the values reported for the percentages of patients requiring major surgery (13% and 2% in ‘episodic’ and scheduled treatment arms respectively).

Other outcomes reported for ACCENT II

The ACCENT II65 published report presented the median decrease from baseline in CDAI score at weeks 30 and 54 for all patients. Improvements in median CDAI were statistically significantly greater for the infliximab group (p = 0.004). Median increases from baseline in IBDQ scores at weeks 30 and 54 were also significantly greater for the infliximab group than for the placebo group. Baseline scores for all patients by group were not provided and baseline balance was therefore uncertain. In the case of missing values, the last observations were carried forward for the CDAI and IBDQ outcomes. The results are summarised in Table 35.

TABLE 35 - Median CDAI and IBDQ changes ACCENT II65

Baseline scores were reported for responders by group and for all non-responders, but not for all patients by group.

Week	Infliximab (n = 139)	Placebo (n = 143)	p infliximab vs placebo
Median decrease in CDAI score from baseline
30	42	16	0.004
54	40	15	0.004
Median increase in IBDQ score from baseline
30	14	4	0.002
54	10	5	0.03

Further results for the ACCENT II65 trial were presented in two separate papers. 75,76 One reported a post hoc analysis of the subgroup of responder patients with rectovaginal fistulas (11 received placebo and 14 received the 5 mg/kg dose regimen of infliximab);75 the other paper performed a post hoc analysis on incidence of abscess development in patients responding to infliximab with closure of fistulas. 76 The first of these papers was too underpowered for firm conclusions to be drawn. In ACCENT II,65 crossover to an increased dose of infliximab was allowed for all randomised groups (including placebo) from week 22 onwards; this resulted in the mean dose of infliximab in the placebo group (quoted as 20 mg/kg) being approximately half that of the intervention groups (quoted as 40 mg/kg). The post hoc analysis for abscess development compared these two groups and reported no statistically significant difference in rates (15% vs 19%; p = 0.526).

Quality assessment (based on published report)

Randomisation and blinding were adequate and allocation concealment likely to be so. Baseline characteristics for responders were well balanced between placebo and infliximab arms; however, for the all-patient comparisons between infliximab and placebo arms (e.g. of change in IBDQ and CDAI scores relative to baseline) it was not possible to ascertain if groups were balanced at baseline. There was a lack of clarity in the methods section so that it was difficult to determine if the sentence ‘…data for patients who crossed over from placebo to infliximab were censored before crossover occurred…’ referred to the survival analysis of loss of response. If it did, the reason for different handling of crossovers in the compared groups is difficult to interpret. The number of patients who withdrew prematurely was unclear except for discontinuation for AEs. No power calculation was undertaken. The last observation was carried forward as necessary for continuous outcomes, but the number of missing data was not reported.

ACCENT II.65 Summary of effectiveness evidence

After induction infusions of 5 mg/kg infliximab at weeks 0, 2 and 6, 64% of enrolled patients were classified as responders. Responders were defined as patients experiencing at both weeks 10 and 14 a ≥ 50% reduction in the number of draining fistulas that were present at baseline of at least 3 months’ standing.

After week 14, the median time to loss of response by responder patients was greater for patients randomised to continued infliximab treatment of 5 mg/kg at 8-week intervals than for those randomised to placebo (p < 0.001). More responder patients randomised to infliximab at week 14 experienced a response (closure of ≥ 50% of draining fistula for at least 4 weeks) than did responder patients randomised to placebo, and from week 22 the risk difference (infliximab–placebo) reached statistical significance in favour of infliximab. After week 14, response rates diminished in both groups. From week 30, risk differences diminished indicating that most benefit from infliximab was delivered between weeks 14 and 30.

Paediatric Crohn’s disease trials

Patients in these trials were ≤ 18 years of age. Two trials, Baldassano et al. 46 and REACH [A randomized, multicenter, open-label study to evaluate the safety and efficacy of anti-TNFα chimeric monoclonal antibody (infliximab, Remicade^®) in pediatric subjects with moderate-to-severe Crohn’s disease] (Hyams et al. 45), looked at the effectiveness of different doses of infliximab in paediatric CD patients. There was no placebo arm in either trial. There were no trials of adalimumab in children. The outcomes measured are shown in Table 36 and the study characteristics are summarised in Table 37.

TABLE 36 - Outcomes measured in trials of paediatric CD

PCDAI, Paediatric Crohn’s Disease Activity Index.

IMPACT is a measure of HRQoL in paediatric IBD.

	% of patients in remission (PCDAI score < 10)	% achieving response	PCDAI score (mean or median)	Additional outcomes
Infliximab
Baldassano et al. 200346	✓	✓ Decrease in CDAI of ≥ 70 points OR ≥ 10 points on PCDAI	✓	Endoscopic lesion severity score (in consenting patients)
REACH45	✓ (or CDAI < 150)	✓ Decrease in PCDAI of ≥ 15 points and total PCDAI score < 30	✓	IMPACT III score; % discontinuing corticosteroids; change in height status (subgroup); clinical response following crossover

TABLE 37 - Main study and population characteristics: paediatric trials

Study Sponsor	Country	Study length/size	Population: severity of CD (baseline CDAI and IBDQ if stated)	Intestinal areas affected	Main concomitant medication, % not on any medication	Previous/concomitant treatment with anti-TNF inhibitors	Intervention and comparator (dosing regimen)
Baldassano et al. , 200346 Supported by Centocor	Multicentre (USA, Europe)	12 weeks n = 21	Moderate-to-severe active disease despite previous treatment, PCDAI ≥ 30 or modified CDAI ≥ 200 Median PCDAI 56, 45 and 41 infliximab groups (no placebo group), median modified CDAI score 455, 317 and 312 (IBDQ not stated)	Mainly ileum and colon, also colon only and gastroduodenal	Mainly aminosalicylates and corticosteroids, also mercaptopurine or azathioprine, antibiotics, few methotrexate % not on medication (if any) not stated	No details	Single intravenous infusion of 1 mg/kg infliximab, 5 mg/kg infliximab or 10 mg/kg infliximab over at least 2 hours at week 0 (No placebo group)
Hyams et al. , 2007 – REACH45 Supported by Centocor	Multicentre (USA, Canada, Europe)	54 weeks n = 103	Moderate-to-severe CD PCDAI > 30 at baseline Mean baseline PCDAI (SD) 42.1 (9.2) and 40.1 (6.8) (infliximab groups, no placebo group) (no other baseline measures)	Mainly colon and/or ileum, also upper tract	Mainly mercaptopurine or azathioprine, also aminosalicylates and corticosteroids, few methotrexate % not on medication (if any) not stated	Exclusion criteria: previously treated with infliximab or other anti-TNF agent	Three i.v. infusions as induction therapy with infliximab 5 mg/kg (weeks 0, 2 and 6) followed by: five infusions of maintenance therapy with infliximab 5 mg/kg administered at weeks 14, 22, 30, 38 and 46 or three infusions with infliximab 5 mg/kg at weeks 18, 30 and 42

Baldassano et al., 200346 (infliximab)

The small trial of Baldassano et al. 46 examined whether a single dose of infliximab induced a response in paediatric patients. Patients were randomised to a 1 mg/kg (n = 6), 5 mg/kg (n = 7) or 10 mg/kg (n = 8) infusion. Patients were followed up to week 12. The primary outcomes were improvements from baseline in PCDAI and modified CDAI score. Other outcomes were the percentage of patients responding and the percentage in remission.

Table 38 shows the median percentage improvement in PCDAI score at various follow-up times relative to baseline. No clear pattern relating to follow-up time or dose regimen was apparent. To what extent improvement in scores resulted from infliximab treatment is impossible to determine because of lack of an appropriate placebo control group.

TABLE 38 - Improvement in PCDAI score in Baldassano et al.46

Week	Median % improvement from baseline in PCDAI score
	Infliximab dose
	1 mg/kg	5 mg/kg	10 mg/kg
1	47	37	35
2	40	65	53
4	27	57	28
8	32	28	64
12	27	13	40

Response and remission results are summarised in Figure 29. All estimates were associated with great uncertainty due to the small number of participants. The proportion of patients in response approached 100% after 1 week in all groups and then tended to decline during follow-up. There was little difference between the groups. How much of the response was intervention dependent cannot be determined because of the lack of an appropriate placebo control group that did not receive infliximab. For remission, no clear pattern relating to dose or length of follow-up was apparent. Again, because of the lack of an inactive control it is impossible to determine the contribution of infliximab to the observed results.

FIGURE 29.

Response and remission rates reported in Baldassano et al. 46 (results as reported, not ITT). Response was defined as at least a 10-point reduction in PCDAI or at least a 70-point reduction in modified CDAI score; remission was defined as a PCDAI score < 10 or a modified CDAI of < 150.

Quality assessment (based on published report)

Randomisation and blinding were adequate and allocation concealment likely to be so. With such small numbers in each group it is not surprising that some baseline characteristics were imbalanced; notably, the 10 mg/kg group consisted almost exclusively of boys and the baseline CDAI score was substantially higher for the 1 mg/kg group than for the 5 mg/kg and 10 mg/kg groups. The number of patients completing the trial was reported to be 90%. No power calculation was done and analyses did not appear to be ITT.

Baldassano et al., 2003.46 Summary of effectiveness evidence

An induction infusion of 1, 5 or 10 mg/kg infliximab improved PDAI scores relative to baseline. Induction increased the proportion of patients in response (40%–100% depending on dose and follow-up time) and in remission (0%–50% depending on dose and follow-up time). The study was underpowered, so these effectiveness estimates were associated with great uncertainty; no clear pattern was evident relating outcomes to dose regimen. The lack of a placebo control group renders interpretation of results problematic.

REACH45 (infliximab)

The REACH trial45 was called an ‘induction and maintenance’ study. Patients received induction doses of 5 mg/kg infliximab at weeks 0, 2 and 6. Responders were defined as those who reduced baseline PCDAI by at least 15 points and had a score of ≤ 30 at week 10. Responders (only) at week 10 were randomised to either five further doses of 5 mg/kg every 8 weeks delivered at weeks 14, 22, 30, 38 and 46 or three further doses delivered every 12 weeks at weeks 18, 30 and 42. Of 112 patients entering the induction phase, 103 were classified as responders and 99 were analysed. The lack of a placebo control group not receiving infliximab means that it is difficult to determine to what extent maintenance of response after induction was attributable to infliximab intervention. No primary outcome was identified. Response and remission results were reported for weeks 30 and 54 and weeks 10, 30 and 54 respectively. These are summarised in Figure 30. The differences between the two dose regimens for both response and remission at weeks 30 and 54 reached statistical significance (p < 0.05) in favour of the more frequent dose regimen.

FIGURE 30.

Post-induction response and remission rates for responders in the REACH trial. 45 Response defined as decrease in PCDAI of ≥ 15 points from baseline and total ≯ 30. Remission defined as a PCDAI ≤ 10 points.

The REACH publication45 also reported changes from baseline (mean and SD) in PCDAI score, IMPACT III score (a QoL measure; scores range from 35 to 175 with higher scores representing better QoL) and daily corticosteroid use. Last observation was carried forward where values were missing. Information was provided for all ‘responders’ (i.e. the two trial arms combined) or separately for the two different treatment groups, at weeks 10, 30 and 54. The results are summarised in Table 39. The ‘all responders results’ do not represent a randomised comparison but rather a ‘before versus after treatment’ comparison for a subgroup of patients (responders) who were selected because they exhibited a favourable response. Given that a ‘no-treatment control group’ was not included in this trial, the analyses do not provide robust quantitative information about the effectiveness of infliximab for paediatric patients, and the favourable changes reported are difficult to interpret as an indeterminate proportion of the effects observed may have been infliximab independent.

TABLE 39 - Changes from baseline in outcome measures reported for the REACH trial45

NA, not applicable; NR, not reported.

p < 0.001 vs baseline.

Randomisation was at week 10.

Group	Week 10	Week 30	Week 54	n
PCDAI: mean (SD) decrease from baseline [improvement]a
Doses every 8 weeks	–33.2	NR	NR	52
Doses every 12 weeks	–29.4	NR	NR	51
Groups combined	–31 (10)	–25.5 (16)	–27 (16)	103
IMPACT III: mean (SD) increase from baseline [improvement]a
Doses every 8 weeks	NR	24.7	26.5	NR
Doses every 12 weeks	NR	18.3	22.5	NR
Groups combined	23.9 (16)	21 (18)	24 (17)	76
Corticosteroid dose (mg/kg): mean (SD) decrease from baseline [improvement]a
Doses every 8 weeks	NR	NR	NR	52
Doses every 12 weeks	NR	NR	NR	51
Groups combined	0.3 (0.4)	0.39 (0.4)	0.3 (0.59)	103
Patients’ corticosteroid use: n discontinued of N users (%)
Doses every 8 weeks	12 of 24 at baseline (50%)	NR	10 of 12 at week 10 (83%)	NA
Doses every 12 weeks	3 of 12 at baseline (25%)	NR	5 of 9 at week 10 (56%)	NA
Groups combined	15 of 36 at baseline (42%)	NR	15 of 21 at week 10 (71%)	NA

After week 10 responder patients were allowed to cross over to increased infliximab for worsening disease state. The increases in infliximab allowed included transfer from infusions every 12 weeks to every 8 weeks and increase in infusion dose from 5 mg/kg to 10 mg/kg. The proportion of patients who crossed over was 40%. The number of responder patients who withdrew prematurely was reported as 22 (21%), but it was unclear if this included withdrawals of patients who had crossed over to increased infliximab.

Quality assessment (based on published report)

Randomisation was adequate and allocation concealment likely to be so. This was an open-label study with no blinding. 45 Baseline characteristics were well balanced except for steroid use. The number of patients withdrawing was reported, but it was not clear if this also included crossover patients who later withdrew. A power calculation was done and analyses were ITT.

REACH.45 Summary of effectiveness evidence

A 10-week induction phase with infusions of 5 mg/kg infliximab at weeks 0, 2 and 6 was followed by randomisation of responders at week 10 to further 5 mg/kg infusions every 8 or 12 weeks. At week 10, 88% of enrolled patients were classified as responders. Response rates for responders diminished to less than 50% by week 54. The difference between dose regimens reached statistical significance in favour of the 8-weekly infliximab dose regimen. About 40% of patients crossed over to increased infliximab because of worsening disease status. About 20% of patients withdrew from treatment prematurely.

Discussion of results and assessment of effectiveness

Summary of effectiveness results

There were no included RCTs with severe CD patients only. They all included moderate-to-severe CD.

1. The general pattern of results is similar for the two drugs.

2. There is a good initial, clinically significant, improvement for the majority of patients when given induction treatment with infliximab or adalimumab. The short duration induction trials demonstrated that the majority of CD patients with moderate-to-severe disease gained clinical benefit from a single i.v. infusion of infliximab (5 or 10 mg/kg) or two subcutaneous injections of adalimumab (80 mg and 40 mg or 160 mg and 80 mg) separated by 2 weeks. Published estimates for the proportion benefiting depended on the measure of clinical response employed and were associated with considerable uncertainty (e.g. 95% lower CI to upper CI ranged from 13% to 66% and 16% to 47% for remission at week 4 from infliximab and adalimumab respectively depending on dose and trial). Obtaining a valid estimate of effectiveness for the two drugs and for their relative efficacy was plagued with difficulties contingent on the small number of trials, their small size, differences between the populations examined, and uncertainties concerning the most appropriate induction regimen to be used and the imprecision of trial results.

3. Although there exists a core of responders of indeterminate size who maintain an anti-TNF dependent response, in general the initial good response is not well maintained with extended treatment. This is evidenced in three ways:

   1. The percentage of patients in response (or with remission) fades away after the first weeks or so of maintenance therapy.

   2. Large numbers of patients drop out of treatment. In ACCENT I,2,3 34% of patients dropped out, some before dose escalation, some after; in the active treatment arms, 25% withdrew in ACCENT I2,3 and (CiC information has been removed)% in CHARM. 67 Among responders in CHARM,67 about (CiC information has been removed)% withdrew from active treatment.

   3. Large numbers of patients required dose escalation and/or transfer to open label (CHARM67) because of worsening disease. In ACCENT I,2,3 68% in the 5 mg/kg arm required dose escalation, as did 49% of those in the higher dose arm. In CHARM67 about 30% in the adalimumab arms crossed over to escalated dose or open-label therapy.

These results indicate that during extended treatment an appreciable proportion of patients decide there is an unsatisfactory balance between the actual benefit of anti-TNF and its perceived benefits. The withdrawal rates in these trials are not similar to other monoclonal antibody interventions and contrast with the > 90% compliance over 52 weeks observed for i.v. weekly infused eculizumab. 88 The high requirement for dose escalation reflects efforts to resuscitate a fading response; the continuing dropout rate after escalation shows that these efforts meet with limited success.

Introduction

This chapter explores the published cost-effectiveness literature on the costs and benefits of TNF-α inhibitors. Within the UK, the licensed anti-TNF treatments are adalimumab and infliximab. The following section goes on to describe the results of a systematic literature review of these treatments for CD.

When assessing the economic impact of CD, costs can be divided into direct costs and indirect costs. Direct costs refer to the costs of an intervention itself and include the value of all resources consumed in the provision of an intervention, including all side effects occurring as a result of treatment, and all future health-care expenditures contingent on either the intervention or side effects. Direct costs include all goods, services and other resources used, both within and outside of the health-care sector. Health-care costs include all medication, diagnostic tests, supplies, staff and medical facility costs. Costs outside health care can include the costs to the patients and to other public agencies. Indirect costs include those resources consumed that are not directly paid for by any party. As the ability of patients to work is related to general health and the time spent in treatment, indirect costs include productivity gains and losses. Indirect costs also include the productivity costs of unpaid carers including family members.

In CD, the perspective taken may have a significant impact on the costs associated with the disease and the overall conclusions drawn from the evidence. Several perspectives can be adopted and the NICE reference case recommends concentrating only on the direct health-care and public social service (PSS) costs. A researcher may also wish to consider a societal perspective that includes direct and indirect costs to all parties as a sensitivity analysis in an economic evaluation.

Methods for reviewing cost-effectiveness

Results of systematic review of existing cost-effectiveness studies

Model structure and inputs

The economic component of the Schering-Plough submission96 took the form of three cost–utility analyses from the perspective of the health service provider. The model structures were informed by the structure of ACCENT I2,3 and included the ‘episodic’ treatment over which concern has been previously expressed (see Glossary). The term ‘infliximab clinical discretion’ (ICD) has been used in place of ‘infliximab episodic treatment’ here.

The three models in the Schering-Plough submission were: a version considering cost-effectiveness of IMT compared with ICD and SC without infliximab among patients with severe active CD (CDAI scores 220–400) in England and Wales (MODEL A); a second version comparing IMT against SC in fistulising CD (MODEL B); and a third model considering paediatric CD patients (MODEL C).

Infliximab maintenance treatment consisted of 5 mg/kg infliximab at week 0, 5 mg/kg infliximab at weeks 2 and 6 and every 8 weeks thereafter. Those receiving ICD received an induction dose of 5 mg/kg infliximab at week 0 and thereafter 5 mg/kg infliximab according to clinical discretion. Those receiving SC received a placebo infusion at weeks 2 and 6 and every 8 weeks thereafter.

MODELS A–C were primarily based on data from two recent trials, ACCENT I2,3 and ACCENT II. 65 Further trial data came from Targan et al. ,57 Present et al. 62 and REACH. 45 ACCENT I2,3 was designed to compare a single 5 mg/kg infusion of infliximab followed by maintenance or a placebo for patients with CD. Participants were recruited from across North America, Europe and Israel. Participants must have had CD for > 3 months and a CDAI score of between 220 and 400. All participants were given 5 mg/kg infliximab at week 0. At week 2, whether participants were responders or not, they were randomly assigned to one of the following three groups:

• Group I: placebo infusion at weeks 2 and 6, and every 8 weeks thereafter to week 46 (n = 188).

• Group II: 5 mg/kg infliximab at weeks 2 and 6, and every 8 weeks thereafter to week 46 (n = 192).

• Group III: 5 mg/kg infliximab at weeks 2 and 6, followed by 10 mg/kg every 8 weeks thereafter to week 46 (n = 193).

Response was defined as a decrease in CDAI score of ≥ 70 points and a minimum 25% reduction in total CDAI score. After week 14, patients who initially responded but experienced exacerbation of symptoms could cross over to 5, 10 or 15 mg of infliximab on an ‘as needed’ or ‘episodic’ basis.

The ACCENT II65 trial compared long-term treatment regimens for patients with fistulising CD. Participants all had CD with single or multiple draining fistulas and were recruited from across North America, Europe and Israel. All participants were given 5-mg infliximab at weeks 0, 2 and 6. At week 14, all patients, regardless of whether they were responders, were randomly assigned to one of the following two groups:

• placebo infusion at weeks 14, 22, 30, 38 and 46, and follow-up at week 54 (n = 99)

• 5 mg infliximab at weeks 14, 22, 30, 38 and 46, and follow-up at week 54 (n = 96).

In this trial, response was defined as a reduction of at least 50% from baseline in the number of draining fistulas at consecutive visits 4 weeks apart. After week 22, patients in the placebo group who experienced a loss of response could crossover to IMT of 5 mg/kg infliximab.

MODEL A used a Markov model to simulate the progression of patients and to calculate the cost per QALY of the infliximab treatment over a 5-year period. For severe active CD, the model states were active, remission, death, non-responding active (patients who failed to respond either initially by week 2 or discontinued treatment in the second week due to loss of response), surgery, post-surgery remission and post-surgery complications. MODEL B (fistulising) replicated this but expanded the active state to: active + fistula closure, active + fistula, remission + fistula closure and remission + fistula. In the severe active model, patients stayed in the active state (on treatment) for the first 2 weeks before movement to other states. In contrast, patients stayed in the active state for the first 14 weeks within the fistulising model, as assessment of these patients in ACCENT II65 occurred at this point. Transition probabilities for the active state were based on ACCENT II65 and Present et al. 62 trial results. The transition probabilities for the ‘on treatment’ health states were estimated from the Targan et al. 57 and ACCENT I2,3 studies; while the transition probabilities for the ‘off treatment’ health states were estimated from the literature. MODEL C (paediatric) mirrored those of the severe active model with transition probabilities based on data from the Targan et al. ,57 ACCENT I2,3 and REACH45 studies.

The probability of surgery and post-surgery states were obtained from a variety of sources (Marshall et al. ,6 Wolters et al. 97 and Jess et al. 98). The authors assumed an equivalent surgery rate (64%) in all three models (severe active, fistulising and paediatric). Post-surgery complications were estimated from Marshall et al. ,6 which showed no significant differences between groups with and without infliximab prior to surgery, so a weighted average was used. Recurrence rates were based on those from Wolters et al. ;97 while the study contained data from nine European countries this did not include the UK, so expert opinion was sought to confirm the similarity of the estimates with the UK.

The methods for the estimation of quantities and unit costs were, in the main, comprehensively described. The cost of hospitalisation and assessments used data taken from Jewell8 (a UK study). This retrospective observational study (n = 205) compared resource use 6 months pre- and 6 months post infliximab. The pre-infliximab figures were used for SC. Data on post-surgery health states (post-surgery remission and post-surgery complications) were not available so resource use was estimated by an expert panel (consisting of UK gastroenterologists) and the average estimates taken. The cost of immunomodulators was excluded from the analysis on the basis that the efficacy of the treatment would not have been affected (the authors did a post hoc analysis of ACCENT I2,3 and II65 trials that indicated that there was no significant difference in infliximab treatment effect with or without immunomodulators). AEs were assumed to be included in the infusion and hospitalisation costs. The cost of infliximab infusions was estimated using an average adult body weight of 60 kg, which the authors stated was based on previous guidance from NICE. A cost of administering infliximab of £96 was suggested by the authors but was incompletely referenced. The closest match found99 suggests a cost of £257.50 per infusion, which was considerably higher than suggested by the manufacturer.

Health-state utilities were based on a number of different data sources. For severe active CD, health-state preferences for pre surgery were taken from a Spanish study [n = 201 CD patient responses to EuroQoL-5 Dimensions (EQ-5D) and converted into utilities using UK tariffs]. 100,101 Surgery and post-surgery preferences were based on data from a secondary care database of patients in Cardiff and The Vale of Glamorgan, Wales. 34 No information was available for complications post surgery, and a utility value of 0.4 was assigned. The authors justified the value given in terms of complications that would lead to significant hospitalisation.

The transitional probabilities were subject to sensitivity analyses with the exception of post-surgery health states because no treatment effect was assumed beyond surgery. One-way sensitivity analyses was conducted on patient weight, time horizon, discount rate, baseline age and infliximab administration cost, and the resultant cost per QALY was reported.

Model results

The results of the analysis are shown in Table 44.

The results in MODEL A were analysed using both a probabilistic sensitivity analysis (which contained flaws) and a univariate sensitivity analysis [patient weight 70 kg (vial sharing) or 80 kg; time horizon at 2 years or at lifetime; discount rate changes; baseline ages of 30 and 60 years vs 45 years; increased administration costs]. In MODEL A the dominance of ICD over SC remained in all univariate analyses except patient weight 80 kg where it remained highly cost-effective. While the industry submission concluded that maintenance was cost-effective against SC, the relevant comparison where ICD is feasible is between maintenance and ICD. Here, the ICER for maintenance treatment over ICD remains above £400,000 in all analyses.

Critique of the industry submissions for infliximab

The design of the cost–utility analysis is in line with what would be expected for this type of submission but the results are limited by a number of factors. While the comparators appear to be justified, the analysis comparing both SC and infliximab maintenance to ICD care is hampered by the definition of ‘episodic care’ used for the ICD comparison. Although episodic care was described as treatment ‘as required’, no further details were given.

The analysis relied heavily, but not exclusively, on the ACCENT I2,3 and II65 trials. The information used in the economic analysis did not compare treatments consistently across those who were potential responders and non-responders to anti-TNF treatment. This is likely to overestimate the effectiveness of IMT in the industry submission. In addition, the costs of those who did not respond do not appear to have been included, giving lower estimated costs. Both scenarios produce a lower cost per QALY.

In line with the other industry submissions, the primary comparison is with SC. The rationale for this comparator is that the majority of patients eligible for biological treatments in England and Wales still receive SC. While the authors cite market research as evidence of this, unfortunately the information cited is not in the public domain. Also, the ACCENT I2,3 and II65 trials were conducted outside the UK, so it is not possible to determine how ‘standard care’ in the trials compares to that in the UK.

Throughout the submission, a CDAI score of 220–400 was used to indicate severe active CD. While there is no formal quantification of the level at which moderate CD becomes severe active, 220 is lower than has been used in a number of other studies, which makes comparison difficult.

Evaluating MODEL A: active CD (220 < CDAI < 400)

Clinical effectiveness of the comparators in the economic model

The economic model compared maintenance (IMT), ICD and SC. As the details of ICD treatment are unclear, it is not possible to satisfactorily verify or interpret the model. In particular, the description of ICD treatment neither guarantees episodic care nor precludes the use of maintenance treatment. Thus, it both is extremely broad in definition and does not guarantee that clinically identical individuals would receive the same treatment. ICD is limited in the degree to which it represents an identifiable treatment strategy.

Much of the model was based on the ACCENT I2,3 trial. This trial distinguished between those who did and did not respond by week 2 on both the placebo maintenance and infliximab maintenance arms (of which the economic analysis considers only the 5 mg/kg dosage). The placebo arm in the ACCENT I2,3 trial included treatment with (1a) placebo maintenance treatment and (1b) ICD at 5 mg/kg for those not responding to 1a. The infliximab arm in the ACCENT I2,3 trial included treatment with (2a) maintenance at 5 mg/kg and (2b) ICD at 10 mg/kg treatment for those not responding to (2a).

Hanauer et al. 3 compared the outcomes for week 2 responders (1a) with treatment (2a), with those crossing to (1b) and (2b) considered to be treatment failures. Rutgeerts et al. 2 attempted to compare outcomes for both week 2 responders and non-responders in their initial and crossover treatment, attempting to compare (1a) plus (1b) with (2a) plus (2b). The economic model attempted to compare (1a) plus (1b) (as ICD) against only (2a) (as infliximab maintenance).

Confidential clinical information in the industry submission suggested that, however constituted, ICD provides very similar clinical outcomes to maintenance therapy. The clinical study report included data on how many patients retained a response to treatment at week 30. Among week 2 responders, 51% of those receiving IMT retained a response, as against (CiC information has been removed) of those receiving SC (placebo) at week 30 (week 54 clinical study report). The ICD arm is based on those failing on placebo treatment receiving infliximab at 5 mg/kg. All those receiving infliximab on ICD would have been considered failures and on this basis, it would have been expected that ICD would have the same effectiveness in retaining a response as placebo treatment.

A clearer comparison is available using the week 30 clinical study report that includes data for week 2 responders and non-responders, those who crossed over, and those who received protocol-prohibited medication changes or surgery. At 30 weeks, (CiC information has been removed) of patients receiving ICD maintained a clinical response to treatment. As it would be expected that non-responders would have poorer outcomes than responders, these results may have underestimated the effectiveness of ICD for week 2 responders. Given this, it would not be surprising for ICD to match maintenance therapy in health outcomes achieved, and any advantage for maintenance over ICD is likely to be very minor. As maintenance therapy patients received infliximab regularly, whereas ICD patients received infliximab according to clinician discretion (at the same dosage), ICD would be less expensive than maintenance. It appears that IMT is very unlikely to be cost-effective against ICD.

Use of trial data in MODEL A

Although the treatment scenarios were well presented, there were some limitations which were primarily associated with the sample characteristics of the studies used. The active CD treatment strategies estimated were based on ACCENT I2,3 data together with data from another smaller study – Targan et al. . 57 The Targan et al. 57 study was used for transitions between week 0 and week 2, with transitions following week 2 estimated using ACCENT I. 2,3 The use of Targan et al. 57 data is questionable as they come from a smaller trial preferred in place of a larger trial which provided less positive results (ACCENT I2,3).

The economic model also appeared to use two different populations in active CD, with both the SC (placebo) and ICD arms using week 2 responders and non-responders and the maintenance arm using week 2 responders only. If week 2 responders do indeed have a better response than week 2 non-responders, then this is likely to bias the comparisons in the economic model. Given the data above, this bias may account for any positive effect found for infliximab maintenance against ICD.

Assessment group revisions to MODEL A

The cost of drug infusions was estimated using an average adult body weight of 60 kg, which the authors state was based on previous guidance from NICE. 1 This is likely to have underestimated the cost per QALY. The Targan et al. 57 trial recorded mean body weights of between 68 and 74 kg (for different treatment groups), while for ACCENT I2,3 these are recorded only in the (confidential) clinical study reports (CiC information has been removed). ACCENT I aside, there were four larger trials (Targan et al. ,57 CLASSIC I,63 GAIN64 and CHARM67) in the clinical effectiveness section in adults (where n > 100) that gave mean weight of included patients – approximately 71.5 kg.

A weight of 60 kg exactly corresponds to the use of three 100-ml vials of infliximab, and the model therefore assumed no wastage. A revised model was constructed by the assessment group. In this model a weight of 70 kg was used, which remains conservative given that no wastage is incorporated into the model. The price of infliximab within the model was also increased to reflect the cost as published in the British National Formulary (£419.73). 102 In the revised model, administration costs were taken to be half of a day case – H26 (Day Case Rheumatology) – in line with a 2006 HTA report,99 which was £293.67 when adjusted to 2006 prices using the Personal Social Services Research Unit (PSSRU) NHS Pay and Prices Index. 103 The cost-effectiveness of ICD, IMT and SC was calculated using a probabilistic sensitivity analysis (1000 iterations) with cost-effectiveness acceptability curves (CEACs) calculated (Figure 33).

FIGURE 33.

Cost-effectiveness acceptability curve for active CD.

The conclusions of the MODEL A as revised by the evidence review group differ from those of the industry MODEL A, a result of these input changes and correction of errors in the industry sensitivity analysis. For threshold levels between £0 and £2466 per QALY, placebo had the highest chance of being cost-effective. For threshold levels between £2466 and approximately £481,000 per QALY, ICD treatment had the highest chance of being cost-effective. Infliximab treatment according to clinical discretion appears to be cost-effective at thresholds of £20,000–30,000 per QALY, although this is contingent on a series of caveats, including the ill-defined nature of the ‘episodic’ intervention itself.

Evaluating MODEL B: fistulising CD

Use of trial data in MODEL B

For fistulising CD, the industry submission stated that evidence from the ACCENT I2,3 trial suggested that maintenance infliximab therapy may bring significant QALY gains related to improved QoL (as opposed to improved life expectancy) in adults with fistulising CD. However, it is not possible to ascertain from the submission or from published papers of the trial whether the sample included adults with fistulising CD. The submission did however report evidence presented in the ACCENT II65 trial (fistulising CD patients) that showed a significantly longer time to loss of response for infliximab maintenance than for placebo maintenance and significantly improved CDAI scores for the infliximab group versus the placebo group.

The fistulising CD strategies were based on responders only. The ACCENT II65 trial showed that 69% of the sample were responders after the induction period. There was, however, no placebo comparison during this period so it is not possible to determine the proportion of patients who would have gone into remission without infliximab. While the proportion of responders was higher than the ACCENT I2,3 trial, the definition of a responder differs and the time at which they were assessed as a responder/non-responder was much later (14 weeks rather than 2 weeks). This highlights the arbitrary nature of the time point chosen to identify responders and the impact it may have had on the results.

The treatment strategy was modelled on the Present et al. 62 trial (0–14 weeks) and the ACCENT II65 trial (14–54 weeks). The Present et al. 62 trial, like the Targan et al. 57 trial, had relatively small numbers in each arm (31–32). The ACCENT II65 scheduled maintenance arm excluded those who switched to 10-mg ‘episodic’ treatment after week 22. Again it is not clear how SC in the UK compares with that in the Present et al. 62 trial (recruitment was in the USA and Europe) and the ACCENT II65 trial.

Assessment group revisions to MODEL B

As with severe active CD, the cost of drug infusions was estimated using an average adult body weight of 60 kg which the authors state was based on previous guidance from NICE. 1 Questions remain about the suitability of this figure given that it is lower than the values found in clinical trials on which the analyses are based. As above, a figure of 70 kg was used for the average weight of patients, which changed both the cost of infliximab and administration costs to a more accurate figure.

The health-state utilities within MODEL B were based on two different sources:

• A Spanish study (n = 201 CD patients) measuring EQ-5D then converted into utilities using UK tariffs which assumed that the preferences measured are concordant with UK patient preferences. 100,101

• A secondary care database of patients in Cardiff and The Vale of Glamorgan, Wales, measuring surgery and post-surgery preferences. This was based on a small sample size and looked at surgery (< 2 months after surgery, n = 17) and remission post surgery (i.e. > 2 months with no recurrence/complication n = 21). 34

Despite specific utility estimates being available for the fistulising CD model, they were not used because they are generally higher than the utilities found in the Spanish study, which was used to provide values in the severe active CD model. The authors stated that the estimates were not in accordance with the NICE reference case93 because they were taken from CD patients and healthy individuals (n = 32 and n = 20 respectively). The utilities allocated assumed that patients with fistula closure had identical utilities to patients without fistulas. For all other fistula states 0.15 was deducted from utility estimates, the authors gave no explanation for this figure but did include the variable in the sensitivity analysis. While the assessment group notes this as an issue, no changes were made to MODEL B on this point.

The model was rerun using the revised weight and correcting the probabilistic sensitivity analysis for calculation errors. The CEAC for maintenance versus SC suggests that maintenance care was more likely to be cost-effective against SC as the threshold value increased. At £20,000 and £30,000 per QALY, infliximab treatment was found to be cost-effective 32.5% and 48.1% of the time respectively. Following the weight adjustment and recalculation of the CEAC, the curve shifted downwards. Now, at £20,000 and £30,000 per QALY infliximab treatment was found to be cost-effective only 6.3% and 19.1% of the time respectively (Figure 34).

FIGURE 34.

Cost-effectiveness acceptability curve for fistulising CD (IMT versus SC).

Discussion of the Schering-Plough submission

The Schering-Plough submission included three submodels considering (1) active CD for the CDAI range covered in the ACCENT I2,3 trial including both moderate and more severe forms of CD, (2) fistulising CD and (3) paediatric CD. The industry submissions contained errors, some of which were resolved by the manufacturer following the consultation period on the draft HTA report. Others could not be corrected, such as the selective use of responders in only the infliximab maintenance arm in the active CD model.

For active CD, the corrected models suggested that infliximab treatment (ICD) could be cost-effective, while maintenance care was unlikely to be cost-effective even at several times the normal threshold values. The lack of detail on what constitutes ICD or ‘episodic’ treatment is unhelpful. To the degree that they can be investigated, the models provided by Schering-Plough mostly meet the standards in the NICE reference case93 (Table 45). There remain issues regarding the selection of studies, the use of data within the selected studies and some inputs used in the modelling. The utility values used in one model rely on a small sample, but are broadly in line with the reference case. 93

Introduction to the evaluation

An economic analysis was conducted for Abbott for their submission4 to NICE by Analysis Group. The submission comprised two economic models – one comparing the cost-effectiveness of adalimumab as a maintenance therapy against SC and one comparing the cost-effectiveness of adalimumab and infliximab as maintenance therapies. This latter model will be relevant only where both adalimumab and infliximab have been first justified as maintenance therapies versus SC. Where one or both maintenance therapies are not cost-effective versus SC, this comparison provides no information to decision-makers.

This evaluation therefore begins by concentrating on the former model assessing the cost-effectiveness of adalimumab as a maintenance therapy. The model considers both fistulising and non-fistulising forms of CD together, and comprises a printed economic submission and accompanying working model in excel. The economic model contained several assumptions that were not fully explained or justified at its initial submission. Abbott took the opportunity in responding to the draft assessment report to clarify some of these issues raised and their model incorporates elements of the health economic critique to their earlier version.

Model inputs and structure

The stated aim of the company’s primary submission was to produce a comparison of lifetime maintenance on adalimumab versus SC. The adalimumab arm of the models was based on data up to week 56 in the CHARM67 trial, which were then extrapolated to also produce a lifetime analysis by assuming that all those responding at week 56 would continue to respond for the remainder of their lives. A regression based on the CLASSIC I63 trial was used to provide SC outcomes for the CHARM67 arm. The company’s comments received regarding the modified model are acknowledged below.

All patients enrolled in the CHARM67 trial had baseline CDAI scores between 220 and 450. In this trial, all patients were given open-label 80 mg at week 0, 40 mg at week 2 and then randomised blind at week 4 to a placebo, adalimumab 40 mg e.o.w. or adalimumab 40 mg weekly. After week 12, those who did not respond to randomised treatment (defined as a drop of less than 70 points in CDAI) were switched to open-label adalimumab 40 mg e.o.w., as were those ‘responders’ who experienced a treatment flare after week 12. Those not responding to open-label adalimumab 40 mg e.o.w. were switched to adalimumab 40 mg weekly. Those not responding to 40 mg weekly were returned to SC.

In the CLASSIC I63 trial, patients had a baseline CDAI between 220 and 450 and had had no previous exposure to any anti-TNF therapy. There were 299 individuals who were randomised to either placebo (n = 74) or adalimumab induction regimens in weeks 0 and 2 of 40 mg and 20 mg (n = 74), 80 mg and 40 mg (n = 75), or 160 mg and 80 mg (n = 76) respectively. All patients were followed for 4 weeks, and the primary end point was the proportion with a CDAI score < 150 in week 4.

The industry submission compared the cost–utility of the 40 mg adalimumab e.o.w. strategy versus SC. As the standard arm of the CHARM67 trial began with adalimumab induction at 80 mg in week 0 and 40 mg in week 2, this did not provide suitable estimates for either the cost or the effectiveness of SC. As the placebo arm in the CLASSIC I63 trial received no adalimumab, the economic submission used these data to predict health states in the SC arm.

The models in the industry submission were based on both the 56 weeks of the CHARM67 trial and an extrapolation to give a lifetime model. The 56-week model included no discounting, the longer model used a discount rate of 3.5% for both costs and benefits. The lifetime model assumed that health remains constant across the group (in terms of the profile of health states) from week 56 to death and assumed a baseline age of 37 years (in line with the average age for CHARM67), with life expectancy of 66 years. There was no mortality between years 37 and 66 due to treatment, CD or other causes.

The model structure was based around four health states defined as remission (CDAI < 150), moderate (150 ≤ CDAI < 300), severe (300 ≤ CDAI < 450) and very severe (CDAI ≥ 450). Patients enrolled in the CHARM67 trial had baseline CDAI scores between 220 and 450, and fell in only the moderate and severe categories at baseline. Utility data were based on these health states. Costs were calculated based both on trial arms (for anti-TNF medication costs) and on the time spent in each of these disease states (for hospitalisation costs and all other costs). Overall costs and QALY benefits for the CHARM67 trial were calculated for the baseline moderate and severe groups (150 ≤ CDAI < 450) and the baseline severe subgroup (300 ≤ CDAI < 450).

Univariate sensitivity analyses were conducted that modified the method of imputing states for those leaving the trial, incorporated indirect costs and made several other changes to the cost assumptions. Using details in the revised industry model for adalimumab versus SC, it was possible to replicate the multivariate sensitivity analysis provided in the revised model.

Estimates of standard care outcomes

The relationship between CDAI-based health states and prognostic factors was estimated using the CLASSIC I63 trial using an ordered probit regression that predicted the chances of an individual falling into each of the four states (remission, moderate, severe and very severe). Variables were included for baseline CDAI, previous anti-TNF exposure, corticosteroid use, fistulising disease, and time and treatment dummy. Health states for the first 4 weeks in the SC arm were found by applying this regression to the clinical factors observed in the CHARM67 40-mg e.o.w. arm. It was assumed that the proportion of people in each health state would remain constant from week 4 onwards. Although patients previously receiving such treatment were excluded from the CLASSIC I63 trial, the previous use of anti-TNF treatment appeared as a predictor in the ordered probit regression. It is unclear how this effect was estimated.

Estimates of adalimumab maintenance outcomes

The adalimumab outcomes were estimated using data derived from the CHARM67 trial data. Within the CHARM67 trial, 778 patients were randomised but 854 patients were enrolled at week 0 in order to achieve this sample. The 76 patients [(CiC information has been removed) with CDAI ≥ 300] who withdrew prior to week 4 did so for a variety of reasons, including AEs (45), lack of efficacy (13) and, in one case, death. (This death was judged not to be related to the use of adalimumab by the CHARM trial investigators. 67)

The revised industry model incorporated the costs of these non-randomised individuals by including them within the adalimumab modelling arm. Given that the 260 individuals receiving adalimumab comprised approximately one-third of the randomised CHARM cohort, it was assumed that one-third of the non-randomised individuals would have been randomised to this arm. The individuals who were not expected to receive a standard adalimumab course were modelled as if they were SC patients but each incurred an additional £974 in medication costs.

Those expected to be able to receive a standard adalimumab course (i.e. those randomised at 4 weeks) were modelled as the 40-mg e.o.w. arm from the CHARM67 trial, based on randomisation at 4 weeks. CHARM67 randomisation was stratified by 4-week response (reduction in CDAI of 70 points from baseline). At 12 weeks, those not responding (reduction in CDAI of 70 points from baseline) could be shifted to open-label treatment and leave the randomised study. Figure 35 shows the comparison between the randomised data at 4 weeks and the observational groups defined at 12 weeks. Of the 260 individuals, only (CiC information has been removed) received their scheduled treatment at week 56.

FIGURE 35.

Clinical data: CHARM67 evidence versus that used in the economic model. (CiC information has been removed.)

Those patients removed from the trial at 12 weeks were referred to in the industry submission as ‘deleted non-responders’ in the diagram shown in Figure 36 (reproduced as non-CiC) from the economic submission. Missing individuals were those who discontinued from the trial for other reasons, including disease flares and protocol violations.

FIGURE 36.

Adalimumab outcomes from the CHARM67 trial.

The economic model included a mixture of those responding and not responding at 4 weeks. The results in the economic model therefore differ from the coprimary clinical end points of the trial, which concerned week 4 randomised responders only.

In the main analysis, the economic model used the last value carried forward (LVCF) to impute missing values. As a sensitivity test, the model included results where the course of the patient’s disease reverted back to the state the patient was in prior to adalimumab therapy (‘simulated placebo’).

Health-state costs and utility estimates

Each health state was linked to an expected number of hospitalisations using a Poisson regression model based on a variety of clinical and background characteristics. This was used to construct patient-level predictions of hospitalisation events per year. The unit cost was estimated using published UK data (Bassi et al. 37) and inflated using PSSRU figures to produce a cost per hospitalisation of £7441 in 2006 pound sterling. The CHARM67 trial did not record CD-related surgeries, and so this hospitalisation factor incorporated the cost of surgery. (Note that the submission was inconsistent whether the hospitalisation figure applied per year or over the 56 weeks of CHARM. 67)

Other non-hospitalisation disease costs (excluding anti-TNF medication) were estimated using Bassi et al. 37 Bassi et al. used a seven-state classification for CD states similar to Silverstein et al. 26 The model assumed that ‘very severe’ corresponded to ‘indicated for surgery’ in Bassi et al. ,37 with ‘severe’, ‘moderate’ and ‘remission’ corresponding to ‘severe, drug-refractory’, ‘mild disease’ and ‘remitted’ states. Estimated non-hospitalisation disease costs are given in Table 46.

TABLE 46 - Health-state based parameters in the industry submission

2006 pound sterling prices.

Health state	CDAI score	Non-hospitalisation disease costsa	Utility
Remission	< 150	8.45	0.859
Moderate	150 ≤ CDAI < 300	23.66	0.795
Severe	300 ≤ CDAI < 450	43.11	0.693
Very severe	≥ 450	78.55	0.433

Table 46 also displays utility estimates for the four health states that were based on a reanalysis of previously published primary standard gamble data (Gregor et al. 92). These estimates were based on 180 consecutive Canadian patients presenting with CD between December 1995 and December 1996.

Adalimumab cost estimates

The cost per 40-mg adalimumab dose was assumed to be £357.50, with one dose necessary every 2 weeks per patient after an initial three-dose induction in the first 4 weeks. No administration costs were included.

Results of the adalimumab industry submission

The results reported here refer to the adalimumab industry submission produced in response to comments in the draft assessment report. The industry submission 56-week model suggested that for baseline moderate and severe patients (150 ≤ CDAI < 450), the incremental cost of adalimumab 40-mg e.o.w. treatment was £2496 for an incremental increase of 0.0823 QALYs (Table 47). The estimated ICER was £30,319. For patients with severe disease (300 ≤ CDAI < 450), the incremental costs and benefits were estimated at £1254 and 0.1045, giving an ICER of £11,998 per QALY.

TABLE 47 - Results from the industry 56-week model

	Moderate and severe			Severe only
	Adalimumab	Standard care	Difference	Adalimumab	Standard care	Difference
QALYs	0.8566	0.7743	0.0823	0.8384	0.7339	0.1045
Drug costs	£6533	£0		£7119	£0
Health-state related costs	£1249	£2049		£1429	£2407
Hospitalisation	£2028	£5265		£2598	£7485
NHS costs	£9810	£7315	£2496	£11,146	£9892	£1254
ICER	£30,319			£11,998

In the original submission, the univariate sensitivity analysis suggested that for patients with severe CD, adalimumab was close to or below £20,000 per QALY for a variety of different assumptions. When considering both moderate and severe CD together, the baseline assumption was close to £30,000 per QALY and typically exceeded this whenever any adverse change was made to the model assumptions. While the industry submission included an induction regimen at 160 mg and 80 mg at weeks 0 and 2, it should be noted that this was not used in the CHARM67 trial and the results will differ if it is associated with higher AEs. This did not change significantly in the second model.

The second industry submission model assumed that parameters for hospital and health-state related costs were distributed according to gamma distributions. Utilities for the remission state were assumed to be distributed according to a beta distribution, with constant ratios between all four utility values (Table 48). When estimating uncertainty in adalimumab outcomes, a Dirichlet distribution was used based on the time spent in each state by the 260 patients within CHARM. 67

TABLE 48 - Parameter distributions for distributions

Type	Health state	Type	A	B
Hospitalisation costs (£)		Gamma	6.25	1190.56
Health-state related costs (£)	Remission	Gamma	6.25	1.35
	Moderate	Gamma	6.25	3.79
	Severe	Gamma	6.25	6.90
	Very severe	Gamma	6.25	12.57
Utilities	Remission	Beta	3.5280	0.5791

In the second industry model (based on 1000 samples), adalimumab treatment was estimated to be cost-effective below £20,000 per QALY in 62.8% of samples, and below £30,000 per QALY in 80.9% of samples. Figure 37 is based on 5000 samples and gives similar results (61% and 79%, respectively).

FIGURE 37.

Cost-effectiveness acceptability curve for severe disease (LVCF method).

Evaluating the industry submission for adalimumab maintenance versus standard care

The inputs to the industry model of adalimumab maintenance were modified to investigate the robustness of the model. The revised model used the ‘simulated placebo’ method of imputing missing values. Those leaving the CHARM67 trial did so because of disease flare or other issues requiring protocol violating treatments, and so their health may have been poorer than an ‘equivalent’ simulated SC outcome (which represented expected health at 4 weeks). The ‘simulated placebo’ assumption is more neutral with respect to the prognosis of those leaving blinded CHARM67 treatment than the LVCF used in the industry model.

Aside from a preference for ‘the simulated placebo’, the major differences between the second industry model and the revised industry model are the assumptions made regarding the use of adalimumab beyond the study period. In both the first and second industry models it was assumed that all those receiving adalimumab at 56 weeks would continue to do so for their entire lives. In the revised model, the rate of withdrawal from adalimumab maintenance post 56 weeks was increased from zero to that of the CHARM 40-mg e.o.w. arm. Outcomes for patients with moderate CD were also inferred as a separate subgroup where this was possible. Unless otherwise indicated, the model description here is kept as in the industry submission.

Results of revised adalimumab model

The industry submission predicted an ICER of £11,998 per QALY at 56 weeks for those with a baseline CDAI at or above 300. With the preferred ‘placebo method’ of imputing missing data, this rose to £30,964 per QALY. The industry submission did not predict an ICER for the moderate subgroup. In the estimates presented here, it was found that treatment was far less cost-effective than for the severe subgroup, and above £100,000 per QALY using the ‘placebo method’ of imputing missing data.

In the 56-week model it appears that treatment for severe patients is likely to approach £30,000 per QALY under more conservative assumptions, but will fall below £20,000 per QALY under optimistic assumptions. For moderate patients, even optimistic assumptions appear to give relatively large ICERs for adalimumab treatment.

The numbers in Table 49 suggest that treatment of those with severe disease (300 ≤ CDAI < 450) will be cost-effective under optimistic (LVCF) assumptions, and marginally over £30,000 per QALY with the preferred and more conservative assumptions. There is less ambiguity surrounding the treatment of those with moderate disease (150 ≤ CDAI < 300). Even under optimistic assumptions, the smaller additional health benefit of 0.0589 comes at a higher incremental cost, leading to an ICER above £60,000 per QALY.

TABLE 49 - Cost-effectiveness of adalimumab in the second industry models: severe and moderate subgroups; and imputation method

	Severe patients only			Moderate patients only
	Adalimumab	Standard care	Difference	Adalimumab	Standard care	Difference
Values imputed using LVCF – optimistic estimate
QALYs	0.8384	0.7339	0.1045	0.8769	0.8180	0.0589
Drug costs	£7119	£0		£6029	£0
Health-state related costs	£1429	£2407		£1046	£1663
Hospitalisation	£2598	£7485		£1465	£2868
NHS costs	£11,146	£9892	£1254	£8540	£4531	£4009
ICER	£11,998			£68,065
Values imputed using simulated placebo – pessimistic estimate
QALYs	0.8225	0.7339	0.0886	0.8605	0.8180	0.0425
Drug	£7119	£0		£6029	£0
Health-state related costs	£1565	£2407		£1205	£1663
Hospitalisation	£3952	£7485		£2099	£2868
NHS costs	£12,636	£9892	£2744	£9333	£4531	£4802
ICER	£30,964			£113,008

Figure 38 shows the CEAC for the treatment of severe disease where the (conservative) simulated placebo method is used. Across 5000 samples, 24.4% of ICERs fell below £20,000 per QALY and 44.1% fell below £30,000 per QALY. These figures compare to 61.2% and 79.1% respectively using LVCF.

FIGURE 38.

Cost-effectiveness acceptability curve for severe disease (placebo method).

For moderate disease, Figure 39 shows the CEAC for the optimistic case (LVCF). Here, only 1.4% of samples fall below £20,000 per QALY and only 7.9% fall below £30,000 per QALY. Under the pessimistic assumption (simulated placebo) these figures fall to 0.04% and 2.0% respectively.

FIGURE 39.

Cost-effectiveness acceptability curve for moderate disease (LVCF method).

Industry submission comparison of adalimumab and infliximab maintenance therapies

The Abbott submission4 also included a cost model comparing adalimumab and infliximab maintenance regimens. The stated aim of the maintenance comparison was to compare adalimumab against infliximab on the basis that infliximab is the alternative most likely to be displaced by the prescription of adalimumab. However, neither adalimumab maintenance nor infliximab maintenance would be the most appropriate comparator in such an analysis. Owing to a lack of trial results comparing these treatments directly, the comparison is secondary in nature.

The infliximab comparator used appears to combine those who were judged to be responders and non-responders on the 5 mg/kg arm of ACCENT I2,3 using the Rutgeerts et al. 2 data including both 5 mg/kg standard dosage and 10 mg/kg ‘as needed’ dosage. The adalimumab comparator used adalimumab maintenance at 80-mg/40-mg induction with 40-mg dosage e.o.w. The adalimumab outcomes were found using a weighted sample from the CHARM67 trial for those with CDAI between 220 and 400 (in line with ACCENT I2,3) and with weights derived so that the same gender distribution, median age and CDAI quartile figures (lower quartile, median, upper quartile) held across the infliximab and modified adalimumab groups. Those with CDAI above 400 were excluded from the analysis for comparability with the ACCENT I2,3 trial. Those who had previously used anti-TNF treatments were not excluded from the adalimumab group, although this was an exclusion criterion in ACCENT I. 2,3 Missing data for both comparators were inferred using LVCF.

The adalimumab arm costs were found by assuming that all those receiving adalimumab and not responding at week 12 would have continued to receive it, which led to higher costs than in the main model. The infliximab arm drug costs were assumed to include an average wastage of 0.5 vials per infusion. Hospitalisation costs were estimated using the Poisson regression for the adalimumab group and observed hospitalisations (plus inferred data from the adalimumab group) in the infliximab group. The model used an excess hospitalisation of 0.098 per infliximab patient per 56 weeks.

As the infliximab data used remission/non-remission rather than the four health states of the main adalimumab model, the health status based costs were estimated using the frequencies reported in Bassi et al. 37 A cost of £38.48 per week was calculated for non-remission costs, and the cost of remitted patients per week was assumed to be £8.45. Overall, the model suggested an excess cost for infliximab patients of £4414 over 56 weeks, of which the majority was due to medication costs (£3526).

While this model also attempted to compare health outcomes, no summary quantitative figures were provided on which to base a cost–consequences analysis, cost-effectiveness or a cost–utility analysis. Using the proportion of patients in remission (partially inferred using LVCF assumptions), it was claimed that adalimumab led to a higher proportion of patients in remission from week 6 onwards. In conjunction with the cost findings, the model claimed that adalimumab maintenance dominated infliximab maintenance.

This model is reported but not analysed in depth. The model compared adalimumab maintenance against infliximab maintenance without comparing either against a standard treatment. As a comparison of non-standard treatments, it fell outside the scope of the assessment. Furthermore, given that both adalimumab and infliximab maintenance appeared to have ICERs far outside the suggested ranges, the results of this model are of little practical relevance to the decision problem faced.

It was also noted that the infliximab regimen modelled here included the 10 mg/kg dosage only. Given the uncertainty relating to which treatments were actually received by patients in ACCENT I,2,3 on what basis these treatments were received, and to what degree the treatments received would be legitimate in a NHS context, it would be difficult to place any confidence in this model.

Introduction

The overall decision problem for this appraisal was ‘What is the cost-effectiveness of anti-TNFs in the management of moderate-to-severe CD in the UK NHS?’. In order to undertake cost-effectiveness analyses to address this decision problem it was necessary to (a) define moderate and severe CD; (b) specify the specific roles for anti-TNFs in the management of CD that were to be evaluated; and (c) specify the patient groups for whom cost-effectiveness would be assessed. The first part of this section addresses these questions. The second part describes the de novo cost-effectiveness model developed to answer the decision problem, and provides base-case results. A series of sensitivity analyses to explore the impact of uncertainty in key model parameters on the estimates of cost-effectiveness were then undertaken. The section finishes with a discussion of the implications of the results of this model.

Disease severity can be expressed in terms of current status or life course. It can be measured using a wide range of indices including frequency of symptoms, severity of symptoms, biochemical activity levels and intensity of treatment required. Available evidence does not provide a strong basis for differentiating CD severity in terms of life course. Munkholm et al. 25 reported that ‘The clinical course of CD differs markedly over time, with ever-relapsing cases, to a quiescent course with remission for several years, interrupted by years with relapse’. No predictive factors have been found for the subsequent course with regard to age, sex, extent of disease at diagnosis and treatment in the year of diagnosis.

The current severity of CD is difficult to assess, and a global measure encompassing clinical, endoscopic, biochemical and pathological features is not available. 39 The most widely used disease activity measures include the CDAI, the HBI or Simple Index, a simplified version of the CDAI, and the PDAI. A commonly used HRQoL measure is the IBDQ. Other measures include the CDEIS.

The CDAI is the measure used most extensively in the anti-TNF clinical trials. It measures current disease severity using a recall period of the previous 7 days. Variables captured in the measure include number of liquid stools, abdominal pain, general well-being, extraintestinal complications, use of antidiarrhoeal drugs, abdominal mass, haematocrit and body weight. Scores range from 0 to 600. Values of < 150 are suggestive of quiescent disease (remission) and values > 450 are associated with very severe disease. 40 Some investigators have arbitrarily labelled CDAI scores of 150–219 as mildly active disease and scores of 220–450 as moderately active disease. 39

Given that the anti-TNF trials use CDAI to measure disease severity and to determine response, the cost-effectiveness analysis used the following definitions of disease severity:

• severe disease: CDAI > 300

• moderate disease: 220 < CDAI ≤ 300.

It should be noted that in line with the decision problem and the use of CDAI in the trials, this definition says nothing about the frequency of relapse. A patient who has been in remission for 5 years and relapses with moderate disease refractory to standard therapy is as much a target for treatment with anti-TNFs as a patient who has had two relapses in the last 12 months, with moderate disease that is refractory to standard therapy.

The scope for this appraisal identified three categories of use of anti-TNFs in the management of CD: induction, episodic and maintenance. There is some uncertainty as to the precise definition of each of these categories.

Maintenance therapy is perhaps the most straightforward to define. It can be described as the chronic use of anti-TNF therapy to maintain remission in patients who have responded (and continued to respond) to anti-TNF therapy when in relapse. In maintenance therapy, the key challenges in arriving at a working definition are:

1. What is the criterion for defining a patient as a responder? Is it the achievement of remission (achieving a CDAI score below a defined threshold) or improvement (a defined minimum fall in their CDAI score)?

2. In those currently receiving treatment, is non-response identified by looking at their response over a period of time, or their response after a certain number of treatments?

Defining ‘episodic’ treatment is less straightforward (see Glossary). Seven different working definitions of episodic treatment were identified in the literature and submissions to this appraisal. In the previous appraisal of anti-TNFs in the management of CD, episodic treatment was defined as giving up to three additional courses of treatment when a patient experienced a disease relapse if that patient initially responded to anti-TNF therapy. The relapse could have occurred once in several years or much more frequently. The key uncertainties in this definition are the same as with maintenance therapy above.

Induction treatment is the use of anti-TNF therapy with the aim to achieve remission. It is not straightforward to draw a distinction between repeated use of anti-TNF as induction on the one hand and episodic therapy as described above. Induction therapy may merely be the initial application of anti-TNF to a patient in relapse which establishes his or her responder status prior to the subsequent provision of ‘episodic’ or maintenance therapy. To consider the cost-effectiveness of induction therapy divorced from its value in informing future decisions on ‘episodic’ or maintenance therapy would be clinically unrealistic and would produce an inaccurate estimate of its cost-effectiveness. As with ‘episodic’ and maintenance treatment, the definition of response and the identification of responders are important issues.

Given the problems with assessing the cost-effectiveness of induction therapy in isolation, one-off induction therapy has not been modelled. Instead, the cost-effectiveness of ‘induction’ as a series of individual induction treatments received when a patient falls into relapse that is non-responsive (or given past treatment expected to be unresponsive) to standard therapy was examined. Thus a repeated reinduction treatment was considered here. Each individual induction treatment was composed of an appropriate dosage regimen based on clinical trial information. Those who did not respond within 8 weeks of induction (i.e. after induction doses at 0, 2 and 6 weeks for infliximab and 0 and 2 weeks for adalimumab) were deemed non-responders and transferred to SC. In maintenance, treatment occurred every 8 weeks for infliximab and every 2 weeks for adalimumab. On relapse, doses increased to match the higher induction regime. For infliximab, treatment occurred at relapse, relapse + 2 weeks and relapse + 6 weeks. For adalimumab, treatment occurred at relapse (at the higher induction dose), then relapse + 2 weeks, relapse + 4 weeks and relapse + 6 weeks at the standard dose in line with maintenance treatment. Remission status following this reinduction dose was assessed at relapse + 8 weeks, and subsequent treatment was determined by whether or not remission was achieved: maintenance treatment continued where remission was achieved by 8 weeks; treatment reverted to SC where remission was not achieved at 8 weeks.

There are a number of alternatives to defining responder status. Within the trials, responders were defined in two distinct ways: (1) patients whose CDAI improved by a pre-specified amount following administration of anti-TNF; and (2) patients who achieved remission following administration of anti-TNF. For the purpose of economic evaluation the first approach to defining a responder was problematic as it said nothing about the relative improvement in health for any given reduction in CDAI score. Both the health gain associated with any given improvement in the CDAI and the value attached to the health gain would depend upon the pre-treatment CDAI. Defining responders using a pre-specified improvement in CDAI does not differentiate between patients for whom treatment controls the disease and patients for whom treatment merely reduces the severity of the symptoms. Thus it was not possible to ascribe a robust utility value for the health of responders defined in this way. By contrast, patients in remission (CDAI < 150) will typically have few CD-specific symptoms and this is a health state for which it would be possible to ascribe a robust utility value. For this reason, response was defined as achieving remission following anti-TNF therapy. This method of defining responders may bias results against treatment in patients with severe disease for whom achieving an absolute reduction in CDAI score may be more difficult. By using an absolute score for response rate it is not possible to take account of all possible health benefits that a patient may gain from a large change in relative response. Yet, as stated above, there is no possible way of determining the health gain associated with a particular relative reduction in CDAI score given the data available. So even when considering the possible bias this approach might introduce in to the model, it is still the only methodologically appropriate way of valuing health states.

The scope for this appraisal identified a number of patient groups for whom the NICE Appraisals Committee would be interested in obtaining specific estimates of the cost-effectiveness of anti-TNF therapy: adults; children; severe CD; moderate CD; fistulising; and non-fistulising.

The randomised placebo-controlled trial evidence for the anti-TNFs did not include paediatric trials, even though infliximab does have a licence for use in the paediatric population. In the absence of estimates of effectiveness that can be used to model the magnitude of effect for anti-TNFs compared with SC, robust modelling of the cost-effectiveness of anti-TNF therapy in a paediatric population was not possible. However, to assist the committee in its deliberations; a scenario analysis using the adult models was presented, where paediatric administration and drug costs were substituted for the adult costs. This was equivalent to assuming that treatment was equally effective in paediatric and adult populations and that all other costs of treatments and utilities gained remained the same.

Separate models are presented for patients with moderate and severe disease, as the value of the health gain associated with remission will be systematically different for these two patient groups, as would be the likely costs of managing relapse.

While the trials of anti-TNFs differentiated between fistulising and non-fistulising disease, it was not possible to identify a long-term usual care cohort study for fistulising patients. In the absence of this evidence, the trial-based evaluations submitted by the manufacturers provided the best available estimates of the likely cost-effectiveness of treatment in the fistulising population. However, it is important to emphasise the difference in the characteristics of the trial populations and the characteristics of the population who are the focus of the decision problem for this appraisal as defined in the NICE scope – i.e. all patients with moderate-to-severe disease that is refractory to standard treatment.

The trial populations were patients in relapse. However, frequency of relapse was not an inclusion criterion for treatment in the decision problem for this appraisal. Patients who relapse more frequently have more capacity to benefit from an effective treatment and therefore, assuming effectiveness is not lower in patients who relapse frequently, treatment would be more cost-effective in those patients than in the population defined for this appraisal.

In summary, de novo cost-effectiveness analyses for adults with moderate-to-severe CD were presented, where response was defined as remission within 8 weeks. The objective of the cost-effectiveness analysis was to estimate the incremental cost per QALY between (a) SC, (b) induction therapy and (c) maintenance therapy for moderate and for severe disease.

Model structure

The NICE methods guide recommends that models of chronic diseases normally adopt a lifetime horizon because exacerbations in chronic diseases usually relate to reduced life expectancy. 93 Analysing a cohort of patients followed from 1993–4 to 2003–4, Wolters et al. 97 reported that there was no significant CD-specific mortality, with the age at diagnosis being the only significant association with mortality. As none of the clinical trials provided evidence of an impact upon mortality, it was reasonable to assume there was no differential mortality rate and, therefore, a lifetime horizon would not add meaningfully to the precision of the cost-effectiveness estimate. The time horizon for the Markov model was 1 year (with sensitivity analyses of 5, 10 and 20 years) and the cycle duration was 4 weeks, i.e. the model had 13 cycles and did not include mortality. All models were constructed and analysed in treeage pro 2008 (TreeAge Software Inc., Williamstown, MA, USA). Both the induction and maintenance model started with a cohort of patients suffering from an SC-refractory relapse.

In the absence of time-dependent transition probabilities for the natural history of the cohort with moderate-to-severe CD at onset, a natural history cohort that reflects the average transition rate over time was used. If the appropriate data had been available, the preferred model structure would have been a time-dependent state transition model, allowing the possibility to capture more accurately the performance of the therapies in the early phase of the disease, where relapse rates are notably higher, as in the short-duration RCT data provided by the manufacturers. In the absence of such data a balance must be struck between the potential underestimate of treatment in the early stages of the disease and the overestimate of the value of maintenance therapy in longer established disease.

Conceptually, the cost-effectiveness model developed was an expanded version of a four-state Markov model. At any time and on any given treatment, a patient was in remission, in relapse, undergoing surgery or in post-surgery remission. For ease of identification, each state is identified by a prefix showing the type of treatment received be it SC, maintenance (MNT – in this chapter only when referring to the Markov model) or induction (IND – in this chapter only when referring to the Markov model). For example, those who received SC therefore occupied states SC remission, SC relapse, SC surgery or SC post-surgery remission within the Markov model. Figure 40 shows the basic structure of the SC model.

FIGURE 40.

Schematic of SC arm of de novo cost-effectiveness model. Incr, incremental; Init, initial; rwd, reward.

The models for induction and maintenance treatments were conceptually similar: each involved one set of states for anti-TNF treatment and another for SC treatments (where necessary). For IND there were three treatment-related states – remission and surgery-related (IND remission, IND surgery, IND post-surgery remission). There were also two relapse states, one for each 4-week period in which response was assessed (IND relapse and IND relapse 2). Those failing to respond to treatment after 8 weeks of continued relapse (i.e. passing through both IND relapse and then IND Relapse 2) transited to the SC states (SC remission, SC relapse, SC surgery, SC post-surgery remission). In order to correctly assign anti-TNF costs, this occurred through a temporary transition state which was identical to SC relapse in its transition probabilities and utility and which only differed in its costs. Maintenance treatment used the same structure as induction.

Transition probabilities for the SC states were based on the Silverstein et al. 26 cohort and the derivation process is detailed in Standard care transition probabilities. Transition probabilities for the IND model assigned a treatment effect by using relapse to remission probabilities from RCT evidence. Transition probabilities for the MNT model assigned the treatment effect by using the this same relapse to remission data and, by inference, a lower remission to relapse rate. Details of the derivation of IND and MNT models are given in Induction transition probabilities and Maintenance transition probabilities below.

Standard care transition probabilities

The population of interest for this appraisal was an inclusive one, rather than the tightly defined populations often found in clinical trials. Therefore, it was important to identify evidence from a population cohort that included patients who were resistant to standard therapy. It was also important the cohort reported data from the time before the advent of biological therapy.

The frequently cited study by Silverstein et al. ,26 as mentioned above, met these criteria. This study reported a 2-monthly transition matrix estimated from 20 years’ follow-up of an inception cohort of 174 patients. Patients were characterised as being in one of eight states: remission, mild, drug-responsive, drug-dependent, drug-refractory, surgery, post-surgery remission and death. Figure 41 shows a schematic of Silverstein et al. ’s26 Markov model.

FIGURE 41.

Schematic of Silverstein et al. ’s26 clinical classification. Incr, incremental; Init, initial; rwd, reward.

As the model did not include death as a state, this had to be removed from the Silverstein et al. 26 matrix when estimating transition probabilities. A relationship existed between the SC remission and SC relapse states and the Silverstein et al. 26 states for mild disease (mild) and active disease (drug-dependent, drug-responsive, drug-refractory). Within Silverstein et al. ,26 patients were effectively managed with non anti-TNF treatments in both the drug-dependent and drug-responsive states; these states were therefore combined to form the SC remission state. In contrast, in the drug-refractory state, the patient was not effectively managed by treatment and was not responding to SC. It was this group who were candidates for anti-TNF treatment and hence this group was used to form the SC relapse state. The process for removing mild and death states from the Silverstein et al. 26 model was slightly more complex and appears below. The four-state matrix: remission, relapse, surgery and post-surgery remission was derived using the following steps.

Begin with the Silverstein et al. 26 transition matrix. Reprinted with permission from Elsevier.

Step 1: Removing death. It was supposed that death from all states was equally likely. The chance of death in each (t0) state was divided by six and this was added to the six non-mild, non-death states.

Step 2: Removing the mild state. A more complex process was used for the mild state. Here, the issue was not now where one entered the state from, but where one would exit to after leaving mild.

1. 2.1. Each of the non-death transition probabilities out of mild were deflated (subtract?) by the total chance of leaving the mild state (0.090). These probabilities for the exit state for mild were: remission, 0.636; drug-responsive, 0.092; drug-dependent, 0.068; drug-refractory, 0.107; surgery, 0.065; and post-surgery recovery, 0.031.

2. 2.2. These exit probabilities were multiplied by the chance of entering mild from each of the other initial health states (t0) and used to distribute the probabilities. Here, the chance of a person in remission remaining in remission increased by 0.070 (the chance of leaving for mild) × 0.636 (the chance of moving from mild to remission).

3. 2.3. The initial Silverstein et al. 26 transition probabilities were increased by the probabilities in (2.1) and (2.2).

Step 3. The drug-responsive state has Markov probabilities summing to 1.00001 due to rounding error in the original paper. These have now been corrected.

Step 4. Steps 1–3 produce a matrix in six states. The states remission, drug-responsive, and drug-dependent were then combined into a single remission state.

1. 4.1. The chance of being in any one of these states was assessed using figures in Silverstein et al. 26 Of the three states, there was an 89.1% chance of being in remission, a 2.1% chance of being in a drug-responsive state, and an 8.8% chance of being in a drug-dependent state.

2. 4.2. The chance of remaining in the (broader) remission state was calculated as the average of the chance of moving to any of the three earlier states from the three earlier states, weighted by the 89.1%, 2.1% and 8.8% from (4.1).

3. 4.3. The chance of moving from this (broader) remission state to a relapse, surgical or post-surgery state was also taken as a similar weighted average.

4. 4.4. The chance of transiting to the (broader) remission state was calculated as the sum of the probabilities of the earlier states comprising the (broader) remission state.