Prognostic tools for identification of high risk in people with Crohn's disease: systematic review and cost-effectiveness study

Aetiology, pathology and prognosis

Neither the underlying aetiology of CD nor the factors that determine the course and prognosis of the disease are fully understood. Environmental factors (e.g. smoking), genetic predisposition and dysregulation of the immune system are thought to play a role in the development and course of CD. 2,4

Crohn’s disease can affect any segment of the gastrointestinal tract from the mouth to the anus, but the most commonly affected areas are the distal ileum (the last part of the small intestine) and the colon. 5 CD that is primarily located in the colon often has a high symptom burden, whereas disease affecting the ileum can be extensive but is associated with relatively few symptoms. 6 Diseased segments of the gastrointestinal tract are frequently separated by intervening areas of healthy bowel tissue. 2,4 The size of the inflamed area may be limited to a few centimetres or it could affect an extensive part of the bowel. As well as affecting the lining of the gastrointestinal tract, CD may penetrate the wall of the bowel. 2,4

As CD can affect any part of the gastrointestinal tract, to differing extents, the symptoms experienced by people with the disease vary markedly, which can sometimes make recognition and diagnosis difficult. 2,4 Moreover, the symptoms and severity of the disease can change over time. People with CD most commonly present with:2,4,7

• abdominal pain

• diarrhoea (mucus, pus or blood may be mixed with the diarrhoea)

• tiredness and fatigue

• loss of appetite and weight loss

• anaemia.

Crohn’s disease can also lead to signs and symptoms outside the gastrointestinal tract; these are known as extraintestinal manifestations and have been reported to be more common in CD primarily located in the colon. 6,7 Associated conditions typically occur during flare but can also manifest during remission or before the development of any signs of IBD. Conditions that develop as a result of CD include:7

• arthritis (most commonly of the large joints of the arms and legs, including the elbows, wrists, knees and ankles)

• skin problems (most commonly erythema nodosum)

• eye problems (episcleritis, scleritis and uveitis)

• liver problems (e.g. primary biliary cholangitis).

Flares of IBD indicate a return to active disease and, potentially, symptoms for an individual. Several factors have been proposed as triggers for flare, including poor adherence to treatment, certain medications (e.g. antibiotics and non-steroidal anti-inflammatory drugs), infection, smoking and emotional stress. 8,9 As has been noted for other immune-mediated diseases,10 the course of CD varies widely among affected individuals, making it challenging to predict the severity or frequency of flare occurrence.

As CD is not curable, the goal of management of the condition is to induce and maintain remission. Population-based studies investigating long-term prognosis of CD report that within the first year of diagnosis, 50–65% of people achieve remission and 15–25% experience a low level of disease activity. 11–13 However, 10–30% of people with CD have a relapse or an exacerbation of their condition in the first year. Long-term follow-up (i.e. 10–15 years) indicates that 67–73% of people with CD experience a chronic relapsing course and 13–20% have a chronic disease course with continuous activity. By contrast, 10–13% of those with CD achieve remission for several years. Among those with CD in remission after treatment, relapse rates at 1, 2, 5 and 10 years are estimated at 20%, 40%, 67% and 76%, respectively. 14

Those who develop CD that follows a non-severe course might achieve prolonged remission with no treatment. In contrast to a non-severe course of CD, those people characterised as following a severe course are likely to experience more frequent flares and typically require early aggressive treatment strategies, including multiple treatment escalations and augmentation. People with severe forms of CD are at a high risk of complications of disease, including intestinal obstruction, fistulae and perianal disease, as well as progressive disability and the need for surgery. 2,4,7

The prognostic factors associated with a more complicated, severe course of CD include bowel damage, extraintestinal manifestations of disease, larger number of flares, need for glucocorticoids, and resultant hospitalisations. 15 Other risk factors for a severe course of disease include smoking and fistula formation. Factors present at CD diagnosis that are found to be associated with a worse prognosis are young age (< 40 years), the presence of perianal disease and an initial need for glucocorticosteroid treatment. 16 The presence of known risk factors for flare and for complications in CD could influence the treating clinician’s management of the condition, but consensus on using risk factors to determine the prognosis of disease is yet to be achieved and treatment can vary.

Epidemiology

Crohn’s disease can appear at any age, but it is most often diagnosed in adolescents and adults between the ages of 20 and 30 years, with a second, albeit smaller, peak in diagnosis between the ages of 60 and 80 years. 17 In the UK, it is estimated that CD affects 1 in every 650 people7 and that at least 115,000 people have the condition. 4 The incidence and prevalence of CD have been rising since the mid-1970s, with the highest rates observed in northern Europe and North America. 18 The incidence of CD in the UK is reported to be about 8 per 100,000 people per year,19,20 with an age- and sex-adjusted point prevalence of 144.8 per 100,000 people. 20

Impact of Crohn’s disease

Affecting men and women equally, CD is a debilitating disease that has a marked impact on physical and emotional health, as well as quality of life. Additionally, CD is associated with a high economic burden as a result of disability, loss of work productivity, surgery and hospitalisation. 21 A UK study22 published in 2015 estimated the annual cost of care for a person with CD to be £6156 (£1800 for those in remission, compared with £10,513 for those experiencing relapse), translating to a total UK annual cost of ≈ £700M. Five years after onset, 15–20% of people are affected by their disease to some degree, and between 50% and 80% of people with CD will eventually need surgery as a result of, for example, the development of strictures, perforation of the bowel or failure of drug therapy. 23

Current diagnostic and treatment pathways

Management of Crohn’s disease

The goal of treatment in CD is initially to control or reduce symptoms to induce remission. 31 Once symptoms are under control, maintenance treatment might be given to prolong remission and minimise the risk of relapse. Globally, two pharmacological treatment algorithms are followed in the management of active CD – the ‘step-up’ (SU) and ‘top-down’ (TD) approaches (Figure 1) – both of which involve several tiers of medication and, as the names suggest, are the inverse of each other. 32 Additionally, surgery might be necessary at any stage of the disease but can be considered as an alternative to medical treatment in some people, particularly those in whom the disease is limited to the distal ileum. 31

FIGURE 1.

‘Step-up’ vs. ‘top-down’ treatment algorithms for CD. 5-ASA, 5-aminosalicylate. Note that in the treatment hierarchy shown in the pyramid, the most potent drug therapies are placed at the top.

Currently, National Institute for Health and Care Excellence (NICE) guideline 12931 recommends a SU approach for the medical management of CD. The SU algorithm (see ‘Step-up’ approach) involves starting treatment with the least aggressive medical option available and escalating therapy in reactive stepwise stages in response to recurrent flares or persistently active disease. An alternative treatment path involves an ‘accelerated SU’ plan in which patients who are considered to have more severe disease or who have clinical markers of poor outcome advance rapidly up the treatment ladder, receiving earlier aggressive therapy than those with non-severe disease. The Evidence Assessment Group’s (EAG’s) clinical experts advised that, for those people judged to be at risk of a more severe clinical course (e.g. extensive small bowel disease, perianal disease or upper gastrointestinal disease), most clinicians would prefer to take an ‘accelerated SU’ approach rather than follow the slower, conventional SU algorithm.

The TD approach (see ‘Top-down’ approach) was not recommended by NICE at the time of writing. 31 The strategy involves treatment earlier in the pathway with biological therapies, which are more clinically effective but are also potentially associated with a greater risk of adverse effects (e.g. increased rate of infection and malignancy). 33 The early use of biological therapies in a TD approach is thought to modify the course of CD, to increase the possibility of mucosal healing (preventing structural damage of the bowel), and to be more effective than the SU approach at inducing and prolonging remission;32 the goal of achieving mucosal healing during treatment is gaining acceptance but is not yet part of standard care in the UK.

Another challenge in the management of CD is the timing of treatment de-escalation, which can be defined as either decreasing the dose of a drug or completely ceasing therapy. De-escalation of therapy in both the SU and the TD strategies is typically considered when a person achieves deep remission, which comprises clinical and biological remission. De-escalation is proposed for those at highest risk of potential complications of treatment, such as infection or malignancy, or for those at lowest risk of relapse after the cessation of treatment. De-escalation might not be appropriate for all those achieving deep remission. Factors that need to be accounted for when considering de-escalation of therapy include age, sex, treatments given and severity of CD. 34 A systematic review34 evaluating de-escalating anti-tumour necrosis factor (TNF) or immunomodulator (IM) therapy in people with CD who were in deep remission for at least 6 months found that de-escalating medical therapy in this cohort was appropriate for a small proportion of carefully selected people only, predominantly the elderly and those with non-severe disease.

Neither the SU nor the TD approach is suitable for all people with CD. Considering the risk–benefit profile of the TD approach, some clinicians could be reticent to expose those with mild activity of CD at the time of assessment or those thought to be at low risk of experiencing a relapse to the unnecessary risk of an adverse effect. Conversely, those assessed as at risk of experiencing a severe course of disease are also at risk of undertreatment if the conventional SU approach is followed, with consequent prolongation of symptoms and the inadequate control of disease activity, and the associated long-term risks. Another consideration is cost of treatment; the TD approach is typically more expensive than the SU approach. 33

The ability to easily stratify those with CD by risk of course of disease could help identify the most appropriate treatment strategy for each patient.

‘Step-up’ approach

NICE guideline 12931 advises starting treatment with a glucocorticosteroid [prednisolone, methylprednisolone or intravenous hydrocortisone (for inpatients)] to induce remission in those with a first presentation or a single inflammatory exacerbation of CD in a 12-month period. For those with mild disease who cannot tolerate or who are contraindicated to the recommended glucocorticosteroids, alternative treatments for first presentation or a single inflammatory exacerbation in 12 months are budesonide (another glucocorticosteroid) and 5-aminosalicylate (5-ASA). Additionally, budesonide can be considered for those who have one or more of distal ileal, ileocaecal or right-sided colonic disease. For children or young people for whom there is a concern about growth or adverse effects, NICE advises considering enteral nutrition as an alternative to a conventional glucocorticosteroid. 31

Both budesonide and 5-ASA are less effective than the preferred initial treatment of glucocorticosteroids, but they might be associated with fewer adverse effects; clinical experts advise that, increasingly, 5-ASA is considered to have a limited role in the management of CD. Budesonide should not be considered for those presenting with severe disease activity or exacerbations.

Should remission not be achieved after induction therapy, the next step in the treatment pathway is the addition of an IM (azathioprine, mercaptopurine or methotrexate) to conventional glucocorticosteroid or budesonide, specifically in cases where:31

• a person experiences two or more inflammatory exacerbations in a 12-month period or

• the glucocorticosteroid dose cannot be tapered.

NICE cautions that before offering a patient azathioprine or mercaptopurine, thiopurine methyltransferase activity should be assessed. Azathioprine or mercaptopurine should not be offered when a patient’s thiopurine methyltransferase activity is deficient (very low or absent) and a lower dose of both IMs should be considered if thiopurine methyltransferase activity is below normal but not deficient (according to local laboratory reference values). Alternatively, if it is thought that the patient would be unable to tolerate mercaptopurine or azathioprine, the addition of methotrexate could be considered.

For adults with severe active CD whose disease has not responded to conventional therapy (including IM and/or glucocorticosteroid treatments), or who are intolerant of or have contraindications to conventional treatment, the recommended therapy is escalation to infliximab or adalimumab within their licensed indications; both of these are TNF-alpha inhibitors. 31 Biosimilars of infliximab and adalimumab are available and can be used interchangeably with originator anti-TNFs in clinical practice. Infliximab and adalimumab can be administered alone or in combination with an IM, and the therapies should be given as a planned course until treatment failure (including the need for surgery) or 12 months after the start of treatment, whichever is earlier. Treatment with infliximab or adalimumab could be continued if there is clear evidence of ongoing active disease as determined by clinical symptoms, biological markers and further investigation, including endoscopy, if necessary. However, NICE advises that disease activity should be reassessed at least every 12 months to determine whether continued treatment with infliximab or adalimumab is still clinically appropriate. People whose CD relapses on cessation of treatment with biological therapy should have the option to recommence treatment with infliximab or adalimumab.

For those with moderately to severely active CD in whom treatment with a TNF-alpha inhibitor has failed (i.e. the disease has responded inadequately or lost response to treatment), or who are intolerant to conventional therapies and are contraindicated to anti-TNFs, other biologics, such as vedolizumab (Entyvio^®, Takeda Pharmeceutical Company, Tokyo, Japan) and ustekinumab (STELARA^®, Janssen-Cilag, Beerse, Belgium), are additional treatment options. 31

Once a person affected by CD achieves remission, NICE advises discussing with them, together with their family members or carers, the options for managing their condition, one of which may be no further treatment. 31 For those who choose to proceed with therapy to maintain remission, the available options are:

• azathioprine or mercaptopurine as monotherapy to maintain remission when previously used with glucocorticosteroids (including budesonide) to induce remission and for those who have not previously received these drugs

• methotrexate –

  • for people who required methotrexate to induce remission

  • for people who tried but could not tolerate azathioprine or mercaptopurine for maintenance

  • for people contraindicated to azathioprine or mercaptopurine.

• continued treatment with biological therapy, if appropriate.

‘Top-down’ approach

Although the ‘top-down’ approach is not recommended by NICE, clinicians in specialist centres might choose to offer the strategy to those they consider to have a poor prognosis in terms of outcomes, for example those with complex perianal disease, significant fistulising disease or multiple risk factors. No accepted treatment strategy is available for the TD approach, with disparity in the definition of ‘aggressive’ therapy across studies. TD can involve the early use of biological therapies or of IMs, or a combination of biological therapy and IMs. In two landmark studies35 evaluating the clinical efficacy of early aggressive therapy in those with CD, ‘top-down’ treatment comprised infliximab in combination with azathioprine. However, evidence in support of the effectiveness of the TD approach when it is compared directly with the SU approach is inconsistent,33 with two studies35,36 finding a benefit of early treatment with biologics and one study37 reporting no benefit of early treatment with biologics over the less aggressive strategy. Variation in results across studies could be related to differences in, for example, the definition of ‘early’ intervention and in trial design, outcomes measured, population and trial duration.

Being able to better predict the course of CD would help clinicians to identify those who could benefit most from the early use of aggressive treatments (IMs and biological therapies) and to decide on the most appropriate treatment to manage symptoms. Tools such as the PredictSURE-IBD^TM (PredictImmune Ltd, Cambridge, UK) and IBDX^® (Crohn’s disease Prognosis Test; Glycominds Ltd, Lod, Israel) could potentially help achieve the goal of personalising treatment for those with CD.

IBDX

Glycominds envisages that the IBDX tool can be implemented at three key stages in the management of CD:

• on differential diagnosis of CD from ulcerative colitis

• to assess the risk of developing a more aggressive disease course in those diagnosed with CD who have not yet experienced complications and/or undergone surgery

• to predict the risk of future events in those who have experienced a first CD complication or surgery.

The IBDX tool detects serum levels of specific anti-glycan antibodies, which are a set of serological biomarkers reported to be highly specific to CD with a potential predictive value for severe course of disease. 38 Glycans are saccharides that can be attached to various biological molecules through an enzymatic process called glycosylation. Most glycans are found on the exterior of cell walls and they form the main components of the cell wall surface in many microbes, including fungi, yeast and bacteria. 38

An atypical interaction of environmental, genetic and microbial factors with the immune system is thought to lead to the production of antibodies against intestinal microorganisms in those with CD that results in the gastrointestinal inflammation typical of the condition. 39,40 Examples of microbial antibodies include anti-Saccharomyces cerevisiae antibodies (ASCA; also referred to as gASCA), antibodies against Pseudomonas-associated sequence I2 (anti-I2), and antibodies against the bacterial flagellin cBir1 (anti-cBir1). 41 Anti-glycan antibodies comprise antibodies against ASCA, anti-mannobioside antibodies (AMCA), anti-laminaribioside antibodies (ALCA), anti-chitobioside carbohydrate antibodies (ACCA), anti-laminarin antibody (anti-L) and anti-chitin antibody (anti-C).

Antibodies detected by the IBDX tool include:42

• ACCA

• ALCA

• AMCA

• gASCA

• anti-L

• anti-C.

The IBDX tool is supplied as a set of six biomarker kits (listed above), each of which detects a circulating antibody against the kit-specific antigen in patient serum or plasma by an indirect solid-phase enzyme-linked immunosorbent assay (ELISA). Individual kits contain the relevant antiglycan 96-well microplate (12 × eight-well strips), ELISA reagents, negative control, positive control and calibrators. 43 Each kit can assess up to 90 samples, excluding controls, but the company recommends running samples in duplicate (i.e. a maximum of 45 assays per kit, accounting for controls). The microwell plates, conjugates and controls are specific to each kit, but all other reagents are the same. All kits follow the same procedure (including incubation times), so they can easily be processed at the same time, if desired. On completion of incubation, absorbance of the calibrator, controls and samples can be evaluated spectrophotometrically. Optical density is directly proportional to the amount of bound antibody. Arbitrary units are calculated based on sample optical density and calibrator serum sample optical density. 43 The positivity of each biomarker is assessed based on the cut-off values presented in Table 1.

Those people with CD are considered to be at greater risk for disease complication (stricturing or penetrating) or surgery intervention if they are positive for two or more serological markers. 42 Figure 2 presents a flow chart (adapted from that available in the instructions for the IBDX kit43) summarising how to interpret the complete panel of results from the individual biomarkers.

FIGURE 2.

Overview of interpretation of results from individual IBDX ELISA kits [adapted from the IBDX^® CE MARK kit insert with permission from Glycominds Diagnostics Ltd (2014)43].

Adapted from the IBDX^® CE MARK kit insert with permission from Glycominds Diagnostics Ltd (2014)

The company highlights that anti-glycan antibodies are also detected at the time of diagnosis in people with coeliac disease. However, as noted by the company, initial positivity for various anti-glycan antibodies is lost after people with coeliac disease follow a long-term gluten-free diet. 44 Coeliac disease and IBD can be comorbid, and studies suggest that people with IBD are at an increased risk of coeliac disease. 45 Therefore, the company recommends against using the IBDX kit without exclusion of diagnosis of coeliac disease in those who have not followed a gluten-free diet. The EAG’s clinical experts fed back that, as the symptoms of CD and coeliac disease overlap, most people referred with suspicion of CD are likely to be tested for coeliac disease, which necessitates a blood test. The EAG’s clinical experts commented that the tests for the risk of severe course of CD and for the presence of coeliac disease could be carried out simultaneously.

Potential subgroup analyses

Evidence permitting, the subgroups planned to be investigated were:

• children with a diagnosis of CD compared with adults with a diagnosis of CD

• newly diagnosed CD compared with established diagnosis of CD

• mild activity of disease compared with moderate to severe activity of disease

• presence fistulising or complex perianal disease compared with absence of fistulising or complex perianal disease.

Sensitivity analyses

The planned sensitivity analyses were to include studies deemed to be at high risk of bias that were excluded from the primary analyses. Sensitivity analyses stratified by risk of bias were not conducted, as a lack of sufficient data precluded such analysis.

Results of the systematic literature search

Searches of electronic databases retrieved 6258 records (post deduplication) that were of possible relevance to the review (Figure 3). The initial screening of titles and abstracts led to the identification of 36 publications for review of full texts. Of the 36 articles evaluated, 16 publications, including systematic reviews, were deemed to be relevant to the review. 38,50,66–79 Four records (three full texts38,66,70 and one conference abstract68) provided details for three systematic reviews, the reference lists of which were screened for potentially relevant studies. Additionally, documents supplied by the companies marketing the prognostic tools were reviewed.

FIGURE 3.

The PRISMA flow chart.

Limited evidence is available from the included full-text publications on the prognostic accuracy of PredictSURE-IBD, and no evidence is available on the prognostic accuracy of IBDX, in identifying those at high risk of following a severe course of CD, as determined by measures such as sensitivity and specificity (the prognostic outcomes of interest listed in Table 3). Most of the evidence on the tools’ utility is derived from observational studies that report estimates of the risk of experiencing a clinical outcome associated with an aggressive course of CD, for example need for treatment escalation, development of a complication or surgery. Estimates are presented of an increased risk for those categorised, based on test results, as being at higher risk compared with those determined to be at lower risk of following a severe disease course. No study retrieved reported on the clinical impact of the use of IBDX or PredictSURE-IBD in terms of influencing the treatments given in the management of active CD.

The authors of two studies79,80 were contacted to verify that the kit used in their research was the IBDX tool and not a comparable kit produced by another company. One author confirmed that they had used a kit that was not captured in the scope of this review, and the study was therefore excluded from the review. 80

Summaries of the studies included in the review are presented by prognostic tool evaluated and key characteristics of studies (Table 4). See Report Supplementary Material 3 for a list of full-text publications screened but subsequently excluded (with reasons for exclusion) from the review.

Glycominds

Glycominds provided a list of bibliographic details of the key publications outlining the evidence in support of the IBDX tool. All studies reporting results on the effectiveness of the kit in stratifying those at high risk of following a severe course of CD were retrieved, and subsequently reviewed, by the EAG.

PredictImmune

PredictImmune provided a list of bibliographic details for several publications relating to PredictSURE-IBD, including references describing the research underpinning the development of the signature gene sequence. All studies flagged by the company were retrieved, and subsequently reviewed, by the EAG.

Additionally, in response to queries from the EAG, PredictImmune supplied anonymised individual patient data (IPD) for results from the cohort that provided results for validation of PredictSURE-IBD, together with data for the head-to-head comparison of PredictSURE-IBD with IBDX. The results provided by PredictImmune for this direct comparison are presented and critiqued in Comparison of IBDX and PredictSURE-IBD.

Characteristics of included studies

All studies informing the evidence base on the prognostic accuracy of the IBDX and PredictSURE-IBD biomarker stratification tests were observational in design. Key characteristics of the included studies are summarised in Table 4, with validated data extraction forms for studies available in Report Supplementary Material 5. Twelve publications, describing eight studies, retrieved from electronic searches were included in the assessment of the prognostic accuracy of the tests, with seven of the studies (11 publications) reporting results on the utility of the IBDX kit and one on the utility of PredictSURE-IBD in stratifying those at high-risk of a severe course of CD (see Table 4). Several studies included a mixed population of participants with CD and ulcerative colitis, and reported results separately for those with CD. Most studies included predominantly adults with CD, with one study (three publications) reporting data for an adolescent or a paediatric population. No additional potentially relevant study was identified from hand-searching the bibliographies of three systematic reviews. 38,66,68,70

All included studies assessed outcomes in people reported to have a diagnosis of CD. However, limited reporting was noted across studies relating to the IBDX on stage of diagnosis (newly vs. established) at the time of the test. Baseline characteristics suggest that the samples analysed were provided predominantly by people with established CD (see Report Supplementary Material 5). By contrast, most people enrolled in the study on PredictSURE-IBD had received a recent diagnosis of CD.

Prespecified inclusion criteria for the systematic review presented here required that people have active disease (see Table 3). Although most of the included studies outlined criteria to be met for a diagnosis of CD, only the study evaluating the PredictSURE-IBD tool required people to have active disease to be eligible for enrolment and reported how presence of active disease was determined. 50 In retrospect, given the biomarker targets of the two prognostic tests, the reviewers consider that the criterion of active CD is appropriate for studies assessing PredictSURE-IBD but is not essential for studies reporting on IBDX. As outlined in Chapter 1, Description of the technologies under assessment, the PredictSURE-IBD tool detects a gene sequence associated with CD8+ T-cell exhaustion that arises from an autoimmune response to active disease, and, therefore, it is appropriate to require that people have active CD when blood is taken for analysis; it has been reported that in people with inactive disease after treatment, as determined by endoscopy, the level of CD8+ T-cells increases to a level that is comparable with those observed in healthy controls. 84 By contrast, the IBDX kit detects serum levels of specific anti-glycan antibodies, with specified cut-off values for allocating positive or negative status to each biomarker. Although serum levels of each antibody can change over time, it is purported that status for positivity or negativity for that antibody remains stable throughout the course of disease. 74 Therefore, for IBDX, the reviewers decided to include those studies not specifying a measure of active disease if they met all of the other inclusion criteria and reported an assessment of the six biomarkers included in the IBDX panel.

Analyses presented for evaluation of the six biomarkers forming the IBDX kit typically reported the association of positivity for individual biomarkers, or the positive status for a larger number of biomarkers, with the increased risk of following a severe course of CD, and not the evaluation of all six biomarkers as a collective.

Considering PredictSURE-IBD, the included study described use of the tool in three cohorts, two training cohorts and one validation cohort. 50 Samples from one training cohort (n = 66) were used in biomarker discovery and samples from the second (n = 39) were used in whole blood classifier development. Estimates of prognostic accuracy are available for the validation cohort only. Based on IPD data supplied by the company, the reviewers consider the validation cohort together with the second training cohort (n = 39) to be the most appropriate data set to inform the evidence base on for economic analysis; this is discussed in greater detail in Chapter 4, Development of the health economic model.

Caveats to interpretation of the results for prognostic accuracy of both tests are discussed in Accuracy of prognostic tests.

Quality assessment of included studies

Included studies were assessed for risk of bias and applicability using the QUIPS tool. 59,60 A summary of the results of the assessment of risk of bias and generalisability concerns across studies is presented in Table 5 (see Report Supplementary Material 4 for the full critique of each study).

The QUIPS tool encompasses six domains for the assessment of the validity and bias of studies evaluating prognosis and factors influencing the course of a condition:59,60

• participation

• attrition

• prognostic factor measurement

• confounding measurement and account

• outcome measurement

• analysis and reporting.

Each domain comprises prompting items (between three and seven) for consideration in the overall rating for an item of high, moderate or low risk of bias. 59,60

The IBDX and PredictSURE-IBD tools were designed with the goal of predicting a course of disease based on the levels of biomarkers produced in response to the presence of CD, with stratification to high or low risk of a severe course of the disease determined by the results of laboratory analysis. The extent to which biomarker levels in blood and serum samples change over time in individual people and what factors influence these fluctuations in levels is uncertain. Additionally, as production of the biomarkers assayed is triggered by changes in cellular processes, the effect of physical characteristics that could influence prognosis in CD, for example smoking status and age, on biomarker levels is unclear. Thus, for the studies informing the evidence on prognostic test accuracy reported here, the EAG considers that the importance of the ‘confounding measurement and account’ domain as a determinant of the risk of bias associated with the studies is also unclear. To reflect the ambiguity around the importance of confounding factors, and to capture uncertainty where limited reporting in the publication precluded an assessment of risk for a particular domain, the EAG adapted the QUIPS tool to include an overall assessment of unclear risk.

Around half of the included studies were deemed to have at least one domain with an unclear risk of bias (see Table 5); for conference abstracts, an unclear rating was predominantly associated with the limited reporting of details as a result of space constraints.

Most studies reporting results for the IBDX tool were determined to be at a moderate risk of bias for the population domain as the studies included those with a recent diagnosis and those with an established diagnosis of CD, and, in some studies, those with presence of severe disease at baseline. Data were not analysed separately for the individual subgroups. The population of greatest relevance to the economic evaluation is those with a new diagnosis of CD and who have moderate or severe disease activity. The study assessing the prognostic accuracy of PredictSURE-IBD enrolled those with a recent diagnosis of CD but included any level of disease activity at sample assessment, with the severity of disease activity determined by endoscopy for some people; severity of disease activity at baseline was not available for all those forming the validation cohort.

Most studies were considered to be at a low risk of bias for attrition and for measurement of prognostic factors because all samples taken were analysed with the relevant tool and results were generated as per the company’s individual protocols. Additionally, outcome assessment was deemed to be at a low risk of bias across many studies as the clinicians were masked to the results of the biomarker assessment.

Accuracy of prognostic tests

The EAG notes that limited data were available from the included studies on the prognostic accuracy of the tools in stratifying the risk of a severe course of CD in terms of standard measures of test accuracy, for example sensitivity and specificity. The EAG is unaware of a validated definition for determining whether or not an individual’s CD has followed a severe course, for example a set number of treatment escalations or the development of a complication or a need for surgery. Thus, the EAG considers the criterion required for a true-positive or false-positive result for IBDX and PredictSURE-IBD to be unclear. The EAG considers that it would be challenging to ascertain an accurate estimate of prognostic accuracy of IBDX and PredictSURE-IBD in stratifying a course of CD. Establishing the prognostic accuracy of the tools would require carrying out a prospective study that included a group that received only SU treatment after determination of their risk of course of CD, using clear prespecified criteria for following a severe course. The ongoing PROFILE RCT randomises people to accelerated SU or TD treatment after they are determined to be at high or low risk of following a severe course of CD, and so the two SU groups will provide additional data to inform estimates of prognostic accuracy. 51 Additionally, no study included in the review prospectively followed people whose treatment was determined by results from IBDX and PredictSURE-IBD; the ongoing PROFILE RCT assesses whether or not early treatment with TD strategy affords clinical benefit to those categorised as being at high risk of severe course of CD and should provide data on the clinical impact of using PredictSURE-IBD.

Results for clinical outcomes

The EAG notes that the results presented in this section are on the risk of experiencing an event among those categorised by the tools as being at high or low risk of following a severe course of CD, and are not related to the clinical outcome of treatment decisions based on the stratification of risk using IBDX and PredictSURE-IBD.

Summary of findings for prognostic test accuracy

Methods

A systematic literature review (SLR) was undertaken in July 2019 to identify published economic evaluations of the PredictSURE-IBD and IBDX tools, as well as economic evaluations of treatments for newly diagnosed patients with moderate to severe CD. The searches were also used to identify potential model parameters in case a de novo model was needed. The searches were used to identify resource use and cost data, together with the natural history of CD. Separate searches were carried out for supporting information on utility data.

The following databases were searched for relevant studies:

• Ovid MEDLINE^® Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Daily and Versions^® (via Ovid)

• EMBASE (via Ovid)

• NHS Economic Evaluation Database (NHS EED) (CRD)

• CDSR (via Cochrane)

• CENTRAL (via Cochrane)

• Database of Abstracts of Reviews of Effects (DARE) (CRD)

• Health Technology Assessment (HTA) database (CRD).

Further to the database searches, experts in the field were contacted with a request for details of relevant published and unpublished studies, and reference lists of key identified studies were also reviewed for any potentially relevant studies.

The search strategy for existing economic evaluations of prognostic tests combined terms capturing the tests of interest (PredictSURE-IBD and IBDX) and the target population (adults who have been newly diagnosed with moderate to severe CD, and who have not been offered biologics under current standard care) with economic and health-care resource use terms (adapted from the Canadian Agency for Drugs and Technologies in Health’s search filter for economic evaluations). 85

The target population considered in the SLR to identify economic evaluations of treatments for CD and health-related quality-of-life evidence (adults with moderate to severe CD) was broader than the population considered in the SLR to identify economic evaluations of prognostic tests to account for the fact that patients’ characteristics change along the treatment pathway. The search strategy for existing economic evaluations of treatments for CD also replaced prognostic tool terms with terms related to corticosteroid, IM and biologic treatments. The search strategy for health-related quality-of-life data was not restricted by prognostic tools or treatments, and it combined terms capturing the target population with health-related quality-of-life terms (adapted from Arber et al. 86).

Limits were applied to searches to remove animal studies, letters, editorials, comments or case studies. Only conference abstracts published within the last 2 years were considered for inclusion; it was assumed that any high-quality studies reported in abstract form before that date would have been published in a peer-reviewed journal. Searches were also restricted to studies published in the English language; however, no restriction by setting or geographical location was applied to the search strategy. Full details of the search strategies are presented in Report Supplementary Material 2.

The titles and abstracts of the papers identified through the searches were independently assessed for inclusion by two reviewers using predefined eligibility criteria. The inclusion and exclusion criteria for each of the three reviews are outlined in Box 1. The methodological quality of the full economic evaluations identified in the review was assessed using the Drummond checklist. 87

Economic evaluations of prognostic tests

The SLR identified a total of 115 papers after deduplication and, based on titles and abstracts, a total of three papers were identified as potentially relevant and were obtained for full-text review. Of the three papers identified for full-text review, none was considered relevant for inclusion. Reasons for exclusion are provided in Report Supplementary Material 3. The results of the process to identify evidence are summarised in Figure 4.

FIGURE 4.

The PRISMA diagram of SLR to identify economic evaluations of prognostic tests. DARE, Database of Abstracts of Reviews of Effects; HTA, Health Technology Assessment; NHS EED, NHS Economic Evaluation Database.

Economic evaluations of treatments for Crohn’s disease

The SLR identified a total of 2403 papers after deduplication and, based on titles and abstracts, a total of 80 papers were identified as potentially relevant and were obtained for full-text review. Of the 80 papers identified for full-text review, 32 were considered relevant for inclusion. Of those 32, one Italian study88 specifically compared the cost-effectiveness of the TD and SU approaches. Nice guideline 129 compared nine induction treatment sequences, in a UK setting, composed of four treatment lines. 31

The remaining studies compared individual treatment steps. Given the large number of such studies, data extractions were restricted to UK studies plus the Italian study that compared the TD with the SU approach. Reasons why papers were excluded are provided in Report Supplementary Material 3. The results of the process to identify evidence are summarised in Figure 5.

FIGURE 5.

The PRISMA diagram of SLR to identify economic evaluations of treatments for CD. DARE, Database of Abstracts of Reviews of Effects; HTA, Health Technology Assessment; NHS EED, NHS Economic Evaluation Database.

The type of economic evaluation included in each of the 11 extracted studies was a cost–utility analysis, where the incremental cost-effectiveness ratio (ICER) was expressed as the cost per quality-adjusted life-year (QALY) gained. Of the 11 extracted studies, five were related to NICE guidance, including three NICE technology appraisals (TAs),89–91 one NICE clinical guideline31 and one NICE diagnostics guidance. 92 For NICE guideline 129,31 two economic evaluations were developed, one on treatment sequences for the induction of remission and a second on treatments for the maintenance of remission.

The most frequent type of decision-analytic model used to estimate cost-effectiveness was a Markov model. Three papers also included a decision tree followed by a Markov model to disaggregate the short- and long-term effects. 90,91,93 The time horizons in these analyses ranged from 1 to 60 years (lifetime), while the cycle lengths ranged from 2 weeks to 2 months. Decision trees without any Markov component were used to estimate cost-effectiveness over shorter time horizons (30 weeks and 1 year) in the two remaining analyses. 31,94 A summary of the 11 extracted studies is provided in Table 7 and detailed data extractions can be found in Report Supplementary Material 5; see Report Supplementary Material 4 for the quality assessment of the studies.

TABLE 7 - Summary of the 11 included economic evaluations

ITT, intention to treat.

Economic evaluations reported in the full guideline.

Study (first author and year)	Population	Interventions/comparators	Model type (cycle length)	Time horizon
Marchetti 201388	Newly diagnosed luminal moderate to severe CD	Top-down: first step infliximab plus azathioprine, second step additional infliximab plus azathioprine, third step methylprednisolone plus azathioprine Step-up: first step methylprednisolone, second step methylprednisolone plus azathioprine, third step infliximab plus azathioprine	Markov model (1 month)	5 years
Dretzke 201189 (TA187)	Moderate CD that is refractory to conventional treatment Severe CD that is refractory to conventional treatment	Infliximab induction infusions Infliximab maintenance infusions Adalimumab induction infusions Adalimumab maintenance infusions Conventional treatment (without TNF-α inhibitors, including treatment with aminosalicylates, methotrexate, corticosteroids, azathioprine, metronidazole or surgical intervention)	Markov model (4 weeks)	1 year
Hodgson 201891 (TA456)	Adults with moderate to severe CD in two subpopulations: Anti-TNF-α failure Conventional care failure	Ustekinumab compared with conventional care and vedolizumab for anti-TNF-α failure Ustekinumab was compared with conventional care and adalimumab for conventional care failure	Decision tree followed by Markov model (2 weeks)	1 year
Rafia 201690 (TA352)	Moderate to severe active disease after failure of initial therapy in three subpopulations: The mixed ITT population, which comprised patients who had previously received anti-TNF-α therapy and those who were anti-TNF-α naive Patients who were anti-TNF-α naive only Patients who had previously received anti-TNF-α therapy only	Vedolizumab induction and maintenance infusion Conventional nonbiologic therapies (a combination of 5-ASAs, IMs and corticosteroids)	Decision tree followed by Markov model (8 weeks)	10 years
Mayberry 201395 (NG129)a	Acute exacerbation of CD Active CD in medically induced remission	Nine treatment strategies with four treatment lines for acute exacerbations of CD No treatment, azathioprine, mesalazine, olsalazine, budesonide and glucocorticosteroids compared for active CD in medically induced remission	Decision tree Markov model (2 months)	30 weeks 2 years
Freeman 201696 (DG22)	Moderate to severe active CD treated with infliximab or adalimumab in two subpopulations: Patients responding to treatment Patients who had lost response to treatment	Monitoring of serum anti-TNF-α compared No testing	Markov model (4 weeks)	10 years
Saito 201394	Moderate to severe CD refractory to conventional therapies and naive to biologic therapy	Infliximab induction and maintenance infusions plus azathioprine Infliximab monotherapy	Decision tree	1 year
Bodger 200993	Moderate to severe active CD	Infliximab infusions for induction of remission followed by maintenance treatment Adalimumab injection for induction of remission followed by maintenance treatment Conventional treatment (5-ASAs, immunosuppressive agents, corticosteroids, antibiotics, symptomatic therapies, topical therapies and surgery)	Decision tree followed by Markov model (8 weeks)	Lifetime (60 years)
Loftus 200997	Severe active CD Moderate to severe active CD	Adalimumab induction and maintenance therapy injection Conventional non-biological therapeutics (5-ASA, antibiotics, immunosuppressants and corticosteroids)	No decision-analytic model, costs and benefits were attached to estimated rates of hospitalisation	1 year
Lindsay 200898	Moderate to severe active luminal disease Fistulising CD	Infliximab initial infusions and maintenance treatment Conventional treatment, comprising IMs and/or corticosteroids	Markov model (luminal active CD, 2- to 4-week cycles until week 14 and then 8-weekly; fistulising active CD, one 14-week cycle and one 16-week cycle and then 24-weekly)	5 years
Clark 200399	Chronic active disease resistant to conventional treatment Fistulising CD resistant to conventional treatment	Infliximab as single and episodic infusions Placebo	Markov model (2 months)	Lifetime (40 years)

Health-related quality-of-life evidence

The SLR identified a total of 2221 papers after deduplication and, based on titles and abstracts, a total of 137 papers were identified as potentially relevant and were obtained for full-text review. Of the 137 papers identified for full-text review, 37 were considered relevant for inclusion and 11 of those reported EuroQol-5 Dimensions (EQ-5D) data. The remaining papers considered generic measures, including the SF-36 (Short Form Questionnaire-36 items), the SF-12 (Short Form Questionnaire-12 items), the Psychological General Well-Being Index, the Cleveland Global Quality of Life and the EQ-5D visual analogue scale, and disease-specific measures, including the CDAI and the Inflammatory Bowel Disease Questionnaire. Owing to the large number of relevant papers, the availability of EQ-5D data in these papers and NICE’s preference for EQ-5D data, the EAG decided to restrict the data extraction to primary sources of EQ-5D data. Reasons for exclusion of the ordered papers are provided in Report Supplementary Material 2. The results of the process to identify evidence are summarised in Figure 6.

FIGURE 6.

The PRISMA diagram of SLR to identify health-related quality-of-life evidence. DARE, Database of Abstracts of Reviews of Effects; HTA, Health Technology Assessment; NHS EED, NHS Economic Evaluation Database.

Of the 11 studies that reported EQ-5D data, 10 used the EuroQol-5 Dimensions, three-level version (EQ-5D-3L), and one of those 10 also collected EuroQol-5 Dimensions, five-level version (EQ-5D-5L), data. The remaining paper did not specify which version of the EQ-5D was used. EQ-5D-3L responses were converted into utilities using UK population tariffs in four studies, which were undertaken in Italy,100,101 Germany102 and Hungary. 103 However, each of those four studies used different sources of UK population tariffs to value EQ-5D-3L responses. The sources included Dolan et al. 104 for Benedini et al. ,100 Badia et al. 105 for Mozzi et al. ,101 Dolan et al. 106 for Stark et al. 102 and Dolan107 for Rencz et al. 103 See Report Supplementary Material 5 for full data extractions and see Report Supplementary Material 4 for quality assessment of the studies.

Six of the 11 studies that reported EQ-5D data were undertaken in Spain, and four of those used Spanish population tariffs developed by Badia et al. 108 (for Casellas et al. ,109 Casellas et al. 110 and Huaman et al. 111) or by Rue and Badia112 (for Casellas et al. 113) to convert EQ-5D-3L responses into utilities. The other two studies undertaken in Spain114,115 did not report the sources used to value EQ-5D-3L responses. Finally, one study undertaken in Poland116 valued EQ-5D-3L responses using a Polish population tariff developed by Golicki et al. 117 As for the study that collected EQ-5D-5L responses in Hungary,103 English tariffs developed by Devlin et al. 118 were employed.

PredictImmune’s economic model

During the diagnostic assessment review subgroup meeting, the EAG became aware of the existence of an economic model built by PredictImmune to assess the cost-effectiveness of PredictSURE-IBD. As a result of a request from the EAG, the company supplied the economic model.

Population

The population included in the economic analysis is adults (aged ≥ 16 years) who have been newly diagnosed with moderate to severe CD and who have not been offered biologics under current standard care. The population in the economic model is largely based on the Biasci et al. 50 population. The paper included a training (n = 38 CD patients) cohort and a validation (n = 66 CD patients) cohort; nonetheless, the published paper did not provide sufficient detail on the treatments received by the validation or training cohorts. Therefore, the EAG asked PredictImmune to provide additional treatment data for the study cohort in Biasci et al. ,50 and in response, the company provided the available individual patient data (IPD).

The IPD included 88 patients with newly diagnosed CD and a classification of high- or low-risk disease. However, the EAG had to remove patients from the IPD (as explained in detail in Time to treatment escalation in high- and low-risk patients); therefore, the final population in the model was reduced to 40 patients (23 high-risk patients and 17 low-risk patients). The average age in the EAG-modelled population was 35 years; 65% of patients were non-smokers, with 25% being smokers and 8% being ex-smokers (smoking status was missing for 2%). Thirty-three per cent of patients were male and 55% were female (12% of patients had no information on sex collected). The study did not collect data on patients’ weight, so the EAG assumed a mean weight of 71.4 kg in the model based on results provided in TA456. 119

Intervention and comparator

As per the final protocol, the interventions of interest are the IBDX and PredictSURE-IBD tests. Nonetheless, the base-case economic model included the PredictSURE-IBD test only, while a scenario analysis was undertaken to compare the IBDX™ test against standard care. Although the EAG considers that there are no robust prognostic accuracy data for either test, the development of the model was based mainly on the IPD provided by PredictImmune pertaining to the use of PredictSURE-IBD.

The comparator included in the analysis is standard care. As no test or algorithm is available in the NHS to determine the long-term course of disease or an individual’s risk of developing a severe course of disease, the estimation of prognosis is based on clinical judgement of presenting signs and symptoms, together with the potential risk factors for developing a severe course of disease (more details are provided in Chapter 2, Search strategies).

For the purpose of the economic model, the EAG assumed that the PredictSURE-IBD test (and the IBDX in the scenario analysis) ultimately categorises patients into high- and low-risk disease categories, so that treatment sequences can be allocated accordingly. The treatment sequences included in the economic model were based on clinical expert opinion provided to the EAG and are intended to describe standard care in the NHS for the SU arm and the accelerated treatment pathway of the TD arm, which is not currently recommended in the NHS. The clinical experts added that < 10% of CD patients receive TD therapy in the NHS; thus, the EAG assumed that patients in the standard care arm of the model can receive SU therapy only. The TD treatment approach is assumed to be received only by high-risk patients who have been tested with either PredictSURE-IBD or IBDX.

The two treatment strategies include an induction treatment with prednisolone for 100% of patients in the model. The difference in treatment strategies thereafter is based solely on the fact that the SU strategy includes an additional treatment step with IMs at the beginning of the sequence. The modelled treatment steps include four bundles of different types of therapy: IMs, anti-TNF biologics, second-line biologics and third-line biologics. Clinical expert opinion was used to derive the distribution of treatments in each treatment bundle. The bundles were defined as follows:

1. IM bundle – 80% azathioprine; 10% mercaptopurine; and 10% methotrexate

2. anti-TNF bundle – 40% infliximab, 60% adalimumab and 30% of all patients get the IM bundle

3. second-line biologic bundle – 50% vedolizumab, 50% ustekinumab and 20% of all patients get the IM bundle

4. third-line biologic bundle – 50% vedolizumab, 50% ustekinumab and 20% of all patients get the IM bundle (patients receiving vedolizumab as second-line treatment are assumed to receive ustekinumab as third-line treatment and vice versa).

The order of treatments received in the TD and SU strategies is described in Figure 7.

FIGURE 7.

The TD and SU treatment strategies.

The clinical experts advising the EAG stated that although all newly diagnosed CD patients (in the TD and SU strategies) start treatment with corticosteroids, patients with moderate to severe CD are extremely unlikely to respond to treatment with corticosteroids alone. Therefore, as a model simplification, the EAG did not include this step in the model, as the results would have been the same in both strategies, given that 100% of patients in the high-risk group (in both the TD and the SU arms) would receive initial induction treatment with corticosteroids and move on to the next treatment step. The EAG appreciates that this may result in a minor discrepancy in the costs associated with the two pathways, that is, SU patients may receive a full course of corticosteroids and TD patients are likely to receive a partial course of corticosteroids only. However, given the lack of robust data around the different lengths of treatment with corticosteroids, and the low cost of corticosteroids, the authors consider that this assumption would have a minimal impact on the results. Regarding the potential risk of additional complications associated with the SU strategy, given the delay in initiating treatment with biologics, the EAG notes that Hoekman et al. 120 concluded that in the long term (10-year follow-up) no difference was found in complications, such as new fistulas or surgery, between the TD and SU arms. Furthermore, although not based on comparative evidence, the Biasci et al. data showed only very few events that required surgery, and no patients underwent more than one surgery during their follow-up period while receiving a SU strategy.

Model structure

The EAG adopted a hybrid modelling approach, whereby a decision tree was developed to allocate patients to a response category after initial induction therapy in either the TD or the SU treatment arm. The decision tree is followed by a cohort model, in which state membership was estimated through a series of different Markov health states.

Patients enter the decision tree model (Figure 8) after they are allocated to the test arm (with either PredictSURE-IBD in the base case or IBDX in the scenario analysis) or to the no test arm (standard care). In the test arm, patients are categorised as being at high- or low-risk of following a complicated course of disease, according to test results, whereas those in the no test arm are designated as at high- or low-risk of following a complicated course of disease based on clinical judgement alone. Given that patients in the standard care arm of the model can receive the SU treatment approach only and that the TD treatment approach is assumed to be received by high-risk patients only, the economic model ultimately assesses the cost-effectiveness of TD therapy compared with SU therapy in high-risk patients. The EAG did not identify any direct evidence on the latter. There is, however, an ongoing study (PROFILE51) that will provide data on the relative effectiveness of these treatment strategies in high-risk patients. The EAG considers that this study should also be able to inform the costs and health consequences of ‘misdiagnosing’ patients as high or low risk. The EAG has undertaken a scenario analysis to account for the cost-effectiveness of misdiagnosed cases. The analysis is described in more detail in Chapter 5.

FIGURE 8.

Model for induction treatment.

After being allocated to either the TD or the SU treatment strategy, patients are allocated to induction therapy, at the end of which they are classified as responders (an improvement in CDAI score of > 70) or non-responders (deterioration, no change or an improvement in CDAI score of < 70). Duration of induction therapy differs by class of treatment (i.e. IM, anti-TNF and second-line biologic). If patients respond to induction therapy, they move to the maintenance cohort model (Figure 9), whereas non-responders escalate to the next step in their allocated treatment strategy.

FIGURE 9.

Cohort model for TD and SU maintenance steps. Blue shading indicates the absorbing state of the model.

Responders to their first induction therapy enter the maintenance cohort model in the remission (CDAI score of < 150), mild (CDAI score of 150–220) or moderate to severe (CDAI score of 221–600) health states. Patients can then move between these states during maintenance therapy, reflecting the different levels of response to maintenance therapy. The probability of patients transitioning between these states is also dependent on the treatment class received.

Non-responders to induction therapy escalate to induction in the next step of their treatment strategy, to which they can become responders or non-responders. Patients receiving their second induction therapy are assessed for response and escalation to the next treatment step, similar to patients receiving their first induction therapy (portrayed by the loop in Figure 8).

Patients in the mild and moderate to severe states are at risk of escalating to the next treatment step, and death is the absorbing state in the model.

Escalation to the next treatment step occurs, therefore, for one of two reasons in the model: lack of response to induction therapy or relapse while on maintenance therapy. The former is a default assumption in the model, as 100% of patients who do not respond to induction therapy move to the next step in their treatment strategy. The latter is not estimated explicitly in the economic model, but instead it is assumed that time to treatment escalation (TTE) (taken from Biasci et al. 50) reflects a relapse while on maintenance treatment. This issue is further discussed in Time to treatment escalation in high- and low-risk patients.

The EAG had to estimate surgical events as a standalone outcome in the model. This modelling simplification means that patients do not explicitly leave their health state in a specific cycle to move to the surgery state. Instead, in every model cycle, a proportion of surgeries is estimated, and the associated costs and impact on patients’ quality of life are calculated (this is further discussed in Effectiveness of top-down compared with step-up treatment strategy on time to treatment escalation). Patients who receive surgery in the model have an increased probability of dying which is associated with the procedure.

The economic assessment is taken from the perspective of the NHS and Personal Social Services, and both costs and benefits are discounted at 3.5% per annum. Cycle length in the model is 2 weeks, and the time horizon of the model is 65 years (when modelled patients would be 100 years old).

Clinical input parameters

As mentioned in Model structure, the economic model is ultimately assessing the cost-effectiveness of TD therapy compared with SU therapy for high-risk patients. However, the EAG did not identify any direct evidence on the latter; thus, the clinical data informing the economic analysis had to be derived from multiple sources. This approach is not ideal and creates a patchwork network of evidence, introducing uncertainty to the economic results. The EAG anticipates that this problem will be (at least partially) overcome when the results from the PROFILE trial are available to populate the economic model. The EAG considers that the PROFILE study should also be able to inform the costs and health consequences of ‘misdiagnosing’ patients as high and low risk, thereby allowing an estimation of the cost-effectiveness of undertreating or overtreating CD patients in the NHS.

The EAG notes that the clinical input parameters in the base-case economic model for PredictSURE-IBD and in the scenario analysis for IBDX are the same. The only difference in the cost-effectiveness analyses of the two diagnostic tests is the cost of the test.

The EAG found two main sources of evidence that could be used to model TTE and time to surgery (TTS). Nevertheless, each source could only partially inform the TTE and TTS analyses in the economic model. Whereas the Biasci et al. 50 paper could inform TTE and TTS according to high and low risk of CD complications (for the SU strategy), the D’Haens et al. 35 paper (and its 10-year follow-up study120) could inform TTE and TTS according to TD and SU treatments (for a population with a mixed risk of disease complications).

The Biasci et al. 50 study enrolled patients with active CD who were not receiving concomitant corticosteroids, IMs or biological therapy. Forty patients received treatment with a corticosteroid, followed by an IM (of whom 50% escalated to treatment with an anti-TNF). This treatment strategy was considered to be a good representation of the first three steps in the SU pathway described by the EAG’s clinical experts. Biasci et al. ’s50 included TTE outcomes, however, differentiated outcomes not by treatment strategy, but by risk of severe disease course. Therefore, the data provided in the study could only potentially inform the difference in TTE and TTS for high- compared with low-risk patients receiving SU.

The D’Haens et al. 35 study evaluated the clinical efficacy of early immunosuppression compared with conventional therapy. The study was a 2-year open-label randomised trial at 18 centres in Belgium, the Netherlands and Germany, and randomly assigned 133 patients to either early combined immunosuppression or conventional treatment. The study collected outcome data on time to relapse for 62 patients: 20 patients who received conventional therapy and 42 patients assigned to combined immunosuppression, who received three infusions of infliximab (5 mg/kg of body weight) at weeks 0, 2 and 6, with azathioprine. Additional treatment was given with infliximab and, if necessary, corticosteroids to control disease activity.

Patients assigned to conventional management received corticosteroids followed, in sequence, by azathioprine and infliximab, if needed. If patients responded to treatment with corticosteroids, treatment tapering was initiated. If patients’ symptoms worsened during the course of corticosteroid tapering and did not respond to an increase in treatment dose, treatment with azathioprine was initiated (2–2.5 mg/kg per day). Patients who relapsed after withdrawal of corticosteroids were given a second course of corticosteroids in combination with azathioprine. Any patient who remained symptomatic after 16 weeks of azathioprine treatment received an induction course of infliximab (5 mg/kg body weight at weeks 0, 2 and 6) and continued antimetabolite treatment.

Therefore, although the study forms a reasonable evidence base for measuring the relative effectiveness of anti-TNF versus corticosteroid followed by IM and anti-TNF, it does not differentiate outcomes by risk of severe disease course, only by treatment received. 35 Furthermore, the treatment sequences included in the D’Haens et al. 35 trial only partially reflect the TD and the SU strategies as described by the clinical experts advising the EAG; the TD and SU strategies in the UK include an initial induction with steroid treatment. In the UK, these clinical strategies are differentiated after steroid treatment only, whereby TD patients are given treatment with an anti-TNF and SU patients are given an IM treatment. As the TTE data taken from D’Haens et al. 35 were based on time to relapse, the EAG assumed that relapse meant failure on first treatment in both strategies in the study and, therefore, time to relapse data were based on the comparison of anti-TNF with corticosteroids. Furthermore, D’Haens et al. 35 included a mix of high- and low-risk patients. This means that low-risk patients were overtreated with first-line anti-TNF. The study concluded that TD patients took a longer time to relapse than SU patients.

The Hoekman et al. 120 study was a retrospective review of medical records of patients included in the D’Haens et al. 35 trial, which collected data on hospitalisation, flares, surgery, clinical activity and other outcomes for a median follow-up of 10 years. The study concluded that, in the long term, no difference was found in clinical remission rate, endoscopic remission, hospitalisation, surgery or new fistulas. During the follow-up period, the proportion of patients who received an IM was similar across arms (88% SU and 86% TD; p = 0.76), while the use of anti-TNF was higher in the SU arm than in the TD arm (73% vs. 54%; p = 0.04). However, the authors explained that the lower use of anti-TNF agents observed during long-term follow-up in TD-treated patients was not directly relevant to current clinical practice because it was related to the previous practice of episodic anti-TNF treatment with no anti-TNF maintenance.

Given that the EAG did not find any sources of evidence combining CD outcomes differentiated by risk of disease and by treatment received, the EAG had to choose between Biasci et al. ,50 which differentiated outcomes by patients’ risk of severe disease course, and D’Haens et al. 35 (and Hoekman et al. 120), which differentiated outcomes by type of treatment strategy received (a proxy for TD vs. SU) to form the baseline treatment measure in the model. The EAG chose Biasci et al. 50 because it considered that estimating a relative treatment effect of TD compared with SU (from D’Haens et al. 35 for TTE and from Hoekman et al. 120 for TTS) and applying it to a different population was based on a less flawed assumption than estimating the relative risk of disease to be applied in a different group of patients. Furthermore, given that the purpose of the diagnostic tests is to categorise patients into high- and low-risk disease, the EAG’s preference was to prioritise robust evidence for this component of the model. Additionally, the D’Haens et al. 35 data did not cover the sequences of treatments included in the TD or the SU approaches as per clinical practice in the UK.

The EAG discusses the TTE and TTS data analysis undertaken using Biasci et al. ,50 D’Haens et al. 35 and Hoekman et al. 120 in the next subsections of the report. However, the EAG notes that the caveat of the results of the theoretical economic analysis is the lack of robust evidence available for the relative clinical effectiveness of TD compared with SU strategies for the population defined in the scope.

Time to treatment escalation in high- and low-risk patients

Of the 105 patients included in the Biasci et al. 50 IPD provided to the EAG, 88 were newly diagnosed with CD (Figure 10). Of these 88 patients, 75 received initial treatment with corticosteroids. The EAG also removed 35 patients from the analysis who never received a subsequent IM after corticosteroids (leaving 40 patients for the TTE analysis; see Figure 10). The EAG did not model time to escalation from corticosteroid treatment to IM (SU) or to anti-TNF (TD). This decision was based on the fact that the economic analysis is driven by the impact of giving high-risk patients TD therapy compared with SU therapy; therefore, considering that 100% of patients in the high-risk group would receive initial treatment with corticosteroids, the impact of treatment would cancel out across the TD high-risk and the SU high-risk arms, as the treatment effect from D’Haens et al. 35 was applied for IM compared with anti-TNF (and subsequent treatment steps) in the model only.

FIGURE 10.

Selection of patients from Biasci et al. 50 for time to escalation analysis. Blue shading indicates the number of patients remaining for TTE analysis.

The TTE data from Biasci et al. 50 were used to estimate time to next treatment step in all SU arms of the economic model (the test and no-test arms of the model and the high- and low-risk arms of the no-test model). To extrapolate TTE data to the model time horizon, the EAG analysed the IPD data to create TTE data for time to first escalation.

The final data set comprised 23 high-risk and 17 low-risk patients, with 16 escalation events observed in the high-risk group and four escalation events observed in the low-risk group). The EAG censored patients who did not have an escalation event. Time to treatment escalation was statistically significantly different between the high- and low-risk arms (p = 0.02). Overall, the EAG notes that both the number of patients and events in the analysis are very small and, therefore, the results of the EAG’s analysis need to be interpreted with extreme caution.

The EAG had to make some assumptions in its base-case analysis to use the Biasci et al. 50 data. These consisted of the following:

1. Treatment escalations in the model correspond to patients’ relapse while on current treatment (or a flare).

2. Patients have the same baseline probability of escalating to the next step in the SU treatment strategy (which is estimated from time to first escalation in Biasci et al. 50) regardless of the number of previous escalations.

The EAG acknowledges that these assumptions are a simplification of clinical reality, where time to escalation is likely to depended on the number of previous treatments. Nonetheless, given that patients in remission are assumed not to escalate treatment while they are on maintenance therapy, and given that the probability of remission changes according to treatment step, the total number of patients escalating treatment differs by treatment step, across all treatments.

Furthermore, as mentioned in this section, the EAG did not find any sources of evidence containing complete treatment sequences (for the TD and SU strategies), which would have allowed the estimation of TTE by treatment step and response status.

The EAG considered the possibility of splitting the Biasci et al. 50 TTE data by first escalation and second (or more) escalations. However, the number of events in the second (or more) escalations data set was too small (three events overall) and, therefore, the Kaplan–Meier data were deemed unreliable. The EAG also considered the possibility of splitting the TTE data according to patients’ initial response to treatment (by using a proxy of time to escalation in Biasci et al. 50). However, the EAG decided against this given the already very small size of the Biasci et al. 50 population for the TTE data available.

The TTE Kaplan–Meier data were fitted with exponential, Weibull, Gompertz, log-logistic, log-normal and generalised gamma models in accordance with guidance in NICE Decision Support Unit Technical Support Document 14. 121 The fit of each parametric model was compared with the observed Kaplan–Meier data, and statistical fit was assessed using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). The fitted curves were also validated by clinical expert opinion. Given the small numbers of patients and events across treatment arms, the curves were initially fit dependently for high- and low-risk patients. However, clinical expert opinion provided to the EAG supported the use of different models for high- and low-risk patients, as the clinical expectation is that all high-risk patients will eventually escalate from IM to anti-TNF but that only 65% of low-risk patients will escalate from IM. Among the best-fitting curves, those that support the clinical predictions are the Gompertz curve for low-risk patients and the log-normal curve for high-risk patients.

The EAG acknowledges that the Decision Support Unit advises against fitting different models to same-study arms unless a strong clinical argument exists. The EAG considers that such a clinical argument is present in this case (as supported by clinical expert opinion) and that the nature of the modelled outcome (TTE for different disease severity course) lends plausibility for the difference in the curves’ shape.

According to the AIC and BIC statistics reported in Appendix 2 (see Tables 28 and 29 for high- and low-risk patients, respectively), the three best-fitting models to the high-risk Kaplan–Meier data from Biasci et al. 50 are log-normal, log-logistic and gamma, while gamma, exponential and Gompertz are the three best-fitting models for the low-risk group. The EAG chose the log-normal (for high-risk patients) and the Gompertz (for low-risk patients) curves.

Effectiveness of top-down compared with step-up treatment strategy on time to treatment escalation

To estimate TTE in the high-risk TD strategy arm of the model, the EAG applied a hazard function, derived from D’Haens et al. ,35 to reflect the treatment effect of TD compared with SU treatment on escalations.

The EAG had to make some assumptions in its base-case analysis to use the D’Haens et al. 35 data. These consisted of the following:

1. Randomisation has resulted in balanced populations of high- and low-risk patients in each treatment group.

2. The relative treatment effect of TD compared with SU in a mixed-risk population is the same as the relative treatment effect of TD compared with SU in a high-risk population.

3. Time to relapse is a proxy measure for time to next treatment escalation.

4. The effectiveness of the treatment strategies in D’Haens et al. 35 is a proxy for the treatment effectiveness of the first step in the TD and SU strategies modelled.

The first regimens in the treatment strategies included in D’Haens et al. 35 (corticosteroid vs. anti-TNF) are likely to overestimate the relative effectiveness of the modelled first step treatment in the TD strategy (anti-TNF) compared with the first step in the SU strategy (IM). Counterbalancing the direction of this bias, the anti-TNF regimen in the study consisted of only three infliximab infusions (at weeks 0, 2 and 6), followed by maintenance monotherapy treatment with azathioprine or methotrexate and additional infliximab infusions in the case of clinical deterioration only. As pointed out by the authors of the Hoekman et al. 120 study, since the D’Haens et al. 35 trial, clinical practice has evolved to continued maintenance treatment with infliximab (in cases of a favourable response to induction treatment), which is consistent with UK clinical practice and NICE guidelines. Therefore, although it is not possible to anticipate the overall magnitude or direction of these biases in the data, they work in opposite directions, and so at least partially alleviate the impact of the overall bias in the analysis.

The EAG digitised the time to relapse Kaplan–Meier data in D’Haens et al. 35 and used the number of patients at risk provided in the study to simulate the pseudo-individual patient data using the Guyot et al. 122 method and the algorithm in the survHE R package [R version 1.0.65; The R Foundation for Statistical Computing, Vienna, Austria; https://CRAN.R-project.org/package=survHE (accessed June 2019)]. Subsequently, the EAG fitted a variety of parametric curves to the Kaplan–Meier data (Figure 11) using the process described in Time to treatment escalation in high- and low-risk patients. The EAG notes that the time to relapse was statistically significantly different between the TD and SU arms (p = 0.04).

FIGURE 11.

Time to relapse estimated by the EAG.

The EAG restricted the modelling of the D’Haens et al. 35 data to dependently fitted survival models only. This was to ensure that the relative effect estimated between the two treatment groups was a scaling factor only. Allowing both the scale and the shape of the curves to vary would have resulted in implausible estimates of a relative effect, particularly in the probabilistic sensitivity analysis (PSA), where samples of the curves could theoretically cross.

Given that no relapse events took place for the first 14 weeks of the analysis, both Kaplan–Meier curves show a plateau from week 0 to week 14 (see Figure 11). This made the curve-fitting exercise challenging as the shape of the fitted curves was heavily influenced by the plateau.

According to the AIC and BIC statistics (see Appendix 2, Table 30), the three best-fitting models to the time to relapse Kaplan–Meier data were log-normal, log-logistic and gamma. Figure 12 shows the fitted curves for TD patients along with the time to relapse Kaplan–Meier data, and Figure 13 shows the equivalent curves for SU patients. The log-normal model provided the second-best fit according to AIC and BIC statistics. Given that the TTE data were fitted with a log-normal model and that the hazard function derived from D’Haens et al. 35 was to be applied to the TTE data, the EAG chose the log-normal curve.

FIGURE 12.

Step-up time-to-relapse curves fitted with Weibull, log-normal and log-logistic.

FIGURE 13.

Top-down time-to-relapse curves fitted with Weibull, log-normal and log-logistic.

Given that none of the three best-fitting curves provided a great visual fit to the Kaplan–Meier data (owing to the plateau observed for the initial 14 weeks), the EAG explored the option of truncating the Kaplan–Meier data at 12 weeks (Figure 14) to fit the survival curves.

FIGURE 14.

Time to relapse (truncated) estimated by the EAG.

According to the AIC and BIC statistics (see Appendix 2, Table 31), the three best-fitting models to the truncated Kaplan–Meier data were gamma, log-normal and Gompertz. The log-normal provided the best fit according to the AIC and BIC statistics. Figure 25 (see Appendix 3) shows the fitted curves for TD patients along with the time to relapse Kaplan–Meier data, and Figure 26 (see Appendix 3) shows the equivalent curves for SU patients.

Although the curves fitted to the truncated data provide a better visual fit, the EAG was wary of eliminating 12 weeks of time to relapse data from the analysis. Therefore, the EAG ran the economic analysis with both sets of log-normal curves (i.e. based on the truncated and the original Kaplan–Meier data) and concluded that the impact on the final ICER was minimal. Thus, the EAG decided to use the non-truncated log-normal curves in the model (Figure 15).

FIGURE 15.

Time-to-relapse curves fitted with log-normal curves.

The EAG used the log-normal fitted curves to estimate a hazard function to apply to the high-risk TD arm of the economic model. The EAG applied the relative hazard function to TTE curves in the first step in the TD strategy (anti-TNF). The TTE associated with the remaining treatment steps in both the TD and SU arms was assumed to be the same as TTE for anti-TNF in the TD arm (see Appendix 4, Figure 27).

The underlying assumption in the EAG’s base-case approach is that high-risk patients who initiate treatment with IMs (SU arm) escalate treatment quicker than high-risk patients who initiate treatment with anti-TNF (supported by the data presented in D’Haens et al. 35); however, once SU patients initiate treatment with anti-TNF (their second treatment step), they ‘catch up’ with patients on the TD treatment strategy.

As some high-risk patients who receive SU treatment respond to IM treatment (see Effectiveness of induction and maintenance therapies), having the additional IM step in the SU strategy is advantageous to patients in the EAG’s base-case analysis as patients still subsequently receive treatment with biologics, which are assumed to have the same benefit as biologics in the TD arm. Given the paucity of data to substantiate any further benefits of subsequent treatment steps in the TD and SU approaches, the EAG considered this to be the most conservative modelling approach.

As mentioned in Intervention and comparator and Time to treatment escalation in high- and low-risk patients, the first treatment step modelled in the TD sequence is anti-TNF, while the first step in the SU strategy is IM treatment. Therefore, there is no modelling of escalation from corticosteroids, nor is there any difference captured across TD and SU arms in time to corticosteroid failure and beginning of first treatment.

The assumption that all patients receive steroids but that only patients in the SU strategy would receive a full course of treatment, rather than being switched to biologics in the TD strategy as soon as the test results become available, was not modelled. Including this step in the model for SU only would add further benefits to the SU strategy as it would allow patients a further chance to respond, as well as reducing the chances of receiving the highly expensive biologic treatments (also considering the very low cost of corticosteroids).

The specialist committee members raised a concern about the potential risk of additional complications associated with the SU strategy given the delay in initiating treatment with biologics. The EAG notes that Hoekman et al. concluded that, in the long term (10-year follow-up), there was no difference found in complications, such as new fistulas or surgery, between the TD and SU arms. Furthermore, although not based on comparative evidence, the Biasci et al. 50 IPD reported very few events that required surgery, and no patients underwent more than one surgery during their follow-up period while receiving a SU strategy.

Therefore, the EAG considers that the specialist committee members’ view that early biologics are better than later biologics may apply to those who do not respond to treatment with IMs only. However, removing this step entirely from the model would mean removing the benefit for those who do respond to IMs. As well as this, highly expensive biologics would be added that are potentially unnecessary for those who respond well to IMs.

Nonetheless, the EAG has varied these assumptions in a range of scenario analyses described in Chapter 5, Scenario analyses. Regarding the measure of treatment effectiveness of TD versus SU in the model, the EAG ran three scenario analyses in the model:

1. High-risk patients on anti-TNF after IMs (second step in SU arm) do not do as well as high-risk patients on first-line anti-TNF (first step in TD arm) and, thus, the former group escalate treatment quicker than the latter. Given that the EAG did not find any data to support this reduction in relative treatment effect, a theoretical estimate of half of the base-case relative hazard was assumed (see Appendix 4, Figure 28). Comparison of TTE curves across treatment strategies for high-risk patients (scenario analysis 1 with SU time to escalation from step 2 estimated with half of the base-case relative hazard; Figure 28);

2. Combining scenario 1 with the base-case approach, the EAG assumed that high-risk patients on TD derive a benefit during the first step of the treatment strategy only (anti-TNF in TD compared with SU patients on IM treatment); however, once patients have moved on to the second step in both strategies there is no relative benefit for TD compared with SU. This scenario differs from the base case as the benefit assumed is the same as that used in scenario 1 (see Appendix 4, Figure 29).

3. Assuming that high-risk patients do not respond to treatment with IMs; that is, 100% of patients who receive SU do not respond to treatment and therefore escalate to anti-TNF after induction with IMs.

The three TTE curves for high-risk patients used in the base case and both scenario analyses are reported in Figure 16. The results of these analyses are reported in Chapter 5.

FIGURE 16.

High-risk curves with TD vs. SU effect.

Effectiveness of top-down compared with step-up treatment strategy on surgery

To estimate TTS in the high-risk TD strategy arm of the model, the EAG applied a hazard function taken from Hoekman et al. 120 The study concluded that TTS was not statistically significantly different across treatment arms. The authors discussed several potential explanations for the lack of statistical differences across study outcomes. These included the reasons already discussed in Effectiveness of top-down compared with step-up treatment strategy on time to treatment escalation regarding the D’Haens et al. 35 trial, in addition to the following:

1. The authors mention the relatively early introduction of IMs or infliximab in the treatment regime for patients receiving conventional management as a potential factor in underestimating the relative effectiveness of early immunosuppressant therapy (at the start of follow-up, 66% of SU patients had received IMs and 15% had received anti-TNF treatment, compared with 82% and 20% of TD patients, respectively).

2. The EAG does not necessarily agree with the point above, as the ‘early’ introduction of anti-TNF or IMs in the conventional treatment arm of the study could have been a reflection of the poor performance of conventional therapy and, thus, the need to escalate to anti-TNF treatment faster.

3. The authors also mention the study’s potential lack of statistical power. Conversely, the authors also argue that observed statistically significant differences between groups merely reflect type I errors due to multiple testing (multiple testing correction was not applied in the study).

4. Finally, the study reports that the treatment received by patients beyond year 2 (the end of the D’Haens et al. 35 trial and the beginning of the follow-up study by Hoekman et al. 120) was at the discretion of the treating physician. Consequently, patients’ outcomes might have be influenced by different treatment strategies at the participating sites. The authors added that patients in both arms of the trial were evenly distributed across the participating hospitals, and, thus, in theory, were equally exposed to the treating physicians’ preferences.

In conclusion, the EAG cannot be sure if the timing of immunosuppression therapy has an impact on TTS events, as the data demonstrate a non-statistically significant effect. However, given that there are also plausible reasons that could explain an underestimation of the effect (or a lack of statistical power to detect it), the EAG has applied the hazard function taken from Hoekman et al. 120 to the TTS in the high-risk TD arm of the model in its base-case analysis. As an exploratory analysis, the EAG has assumed that TTS is the same in the TD and the SU arms for high-risk patients. The results of this scenario analysis are reported in Chapter 6.

The EAG digitised the TTS Kaplan–Meier data in Hoekman et al. 120 The study did not provide numbers at risk (except for the total number of patients entering the study). Therefore, the EAG had to manually reconstruct the numbers at risk by visually analysing the Kaplan–Meier data and estimating when (and how many) events happened over time. This task was simplified by the fact that there were no censored events in the TTS data. Subsequently, the EAG used the number of patients at risk to simulate the pseudo-individual patient data using the Guyot et al. 122 method and the algorithm in the survHER package. The EAG obtained the Kaplan–Meier data (Figure 17) and fitted survival models (dependently, owing to the small number of events across the arms) using the process described in Time to treatment escalation in high- and low-risk patients. The EAG notes that TTS was not statistically significantly different between the TD and SU arms in the EAG analysis (p = 0.2).

FIGURE 17.

Time-to-surgery Kaplan–Meier data estimated from Hoekman et al. 120 by the EAG.

According to the AIC and BIC statistics reported in Appendix 2 (see Table 33), the three best-fitting models are the exponential, log-normal and log-logistic. Figure 30 (see Appendix 7) shows the fitted curves for SU patients along with the time to relapse Kaplan–Meier data, while Figure 31 (see Appendix 7) shows the equivalent curves for TD patients. The EAG chose the exponential model, given that it was the best fitting (Figure 18) and for the reasons discussed in Time to surgery in high- and low-risk patients. The EAG used the fitted curves to estimate a hazard function, which was then applied to the TTS curve in the high-risk TD arm of the economic model (Figure 19).

FIGURE 18.

Time-to-surgery curves fitted with exponential curves.

FIGURE 19.

Time to surgery for high-risk patients with TD vs. SU effect.

Utility values

All utilities were adjusted to account for the age and sex of the modelled population, in accordance with Ara and Brazier. 135

Costs

The following costs are considered in the model:

• diagnostic test costs

• treatment costs

• acute and chronic care costs of CD (including costs of surgery).

All costs considered in the model are valued in 2019 Great British pounds. Where unit costs have been obtained from the published literature before 2019, costs were uplifted using the Office for National Statistics’ Consumer Price Inflation Index for Medical Services (DKC3). 136

Summary of base-case model inputs and assumptions

Table 43 (see Appendix 6) reports the key model inputs used in the EAG’s base-case model and how these were varied in the PSA. Table 17 summarises the key assumptions in the EAG’s economic analysis, together with the rationale for these.

Accounting for the cost-effectiveness of misdiagnosed cases

The test accuracy in the base-case economic model for PredictSURE-IBD and in the scenario analysis included in the DAR for IBDX was the same and assumed to be 100%. This is unlikely to reflect the tests’ actual accuracy in clinical practice; however, no robust diagnostic data were found to inform this in the analysis.

There is, however, an ongoing study (PROFILE) that will provide data on the relative effectiveness of these treatment strategies in high-risk patients. The EAG considers that this study should also be able to inform the costs and health consequences of misdiagnosing patients as high or low risk.

In the absence of real data to inform the costs and consequences of misdiagnosing patients according to their risk of disease severity, the EAG has undertaken a theoretical scenario analysis. The EAG assumed that both diagnostic tools are 75% accurate and, therefore, 25% of CD cases are assumed to be misdiagnosed in the analysis.

The EAG assumed that a proportion of patients who are incorrectly diagnosed as having a low-risk course of CD (i.e. high-risk patients) and receive SU therapy do not respond to IMs and, thus, move to anti-TNF treatment after induction therapy. Conversely, patients who are incorrectly diagnosed as being at high-risk (i.e. low-risk patients) and initiate TD treatment are assumed to enter remission with anti-TNF treatment and do not have the need to escalate treatment any further.

The rationale for the EAG’s assumptions is that low-risk patients (misdiagnosed as high risk) do not need to escalate from anti-TNF to other treatment options (second-line biologics) in the model. Given that these are low-risk patients, the EAG assumed that, after 2 years of anti-TNF treatment, 100% of these patients would be in a treatment-free remission state. Similarly, a proportion of high-risk patients (identified as low-risk patients) do not respond to IMs and so move on to anti-TNF. The proportion of high-risk patients who do not respond to IMs was assumed to be the same as in the base-case model (62%). The EAG acknowledges that these assumptions are a simplification of clinical reality; however, no robust evidence was found to inform this scenario.

Varying the assumptions around the measure of relative treatment effectiveness for time to treatment escalation

As some high-risk patients who receive SU treatment respond to IM treatment, having the additional IM step in the SU strategy is advantageous to patients in the EAG’s base-case analysis, as patients in the SU still subsequently receive treatment with biologics, which are assumed to have the same benefit as biologics in the TD arm. Given the paucity of data to substantiate any further benefits in subsequent treatment steps in the TD approach, the EAG considered this to be the most conservative modelling approach.

Nonetheless, the EAG varied these assumptions in two scenario analyses. The scenarios are explained below and summarised in Appendix 10 (see Table 45):

1. High-risk patients on anti-TNF after IMs (second step in SU arm) do not do as well as high-risk patients on first-line anti-TNF (first step in TD arm) and, thus, the former group escalates treatment quicker than the latter. Given that the EAG did not find any data to support this reduction in relative treatment effect, a theoretical assumption was made and varied:

   1. Anti-TNFs in the SU approach are assumed to be only half as effective as anti-TNFs in the TD approach.

   2. Anti-TNFs in the SU approach are assumed to be as effective as anti-TNFs in the TD approach.

This scenario also assumes that the relative benefit in the anti-TNF step of the TD strategy compared with the anti-TNF step in the SU strategy carries through the next treatment steps. Therefore, patients on second-line biologic treatment in the TD strategy have a relative benefit to second-line biologic treatment in the SU arm (as do patients on third-line biologics). It is also assumed that second- and third-line biologic treatment is as effective as anti-TNF treatment in the TD and SU arms (see Appendix 10, Table 45).

1. High-risk patients on anti-TNF after IMs (second step in SU arm) do not do as well as high-risk patients on first-line anti-TNF (first step in TD arm) and, thus, the former group escalate treatment quicker than the latter group. However, once patients have moved on to second- and third-line biologics, SU patients ‘catch up’ with TD patients and there is no further benefit for TD compared with SU. This scenario also assumes, by default, that second- and third-line biologic treatments are less effective than anti-TNF treatment in the TD arm (see Appendix 10, Table 45).

Assumptions around treatment discontinuation in the model

• The EAG assumed that, after 2 years in remission with any biologic treatment, a proportion of patients experience mucosal healing and, therefore, stop treatment permanently. The EAG used the Marchetti et al. 88 paper to inform this scenario. The study reports that, after 2 years in remission, 76% of patients in the TD strategy experience mucosal healing, whereas 40% of patients in the SU arm experience the same outcome (illustrated in scenario 5.3.2ai).

The EAG also varied the Marchetti et al. 88 assumptions and explored the possibility of the TD and SU therapies having the same impact on the 2-year probability of mucosal healing. Therefore, the EAG assumed that both arms would experience the same probability (76% in scenario 5.2.3aii and 40% in scenario 5.2.3aiii) of mucosal healing.

The EAG notes that Hoekman et al. 120 concluded in their 10-year follow-up study that:

mucosal healing 2 years after the start of treatment was associated with a reduced use of anti-TNF treatment [. . .]. Other outcomes, however, did not differ significantly between patients with and without mucosal healing 2 years after the start of treatment.

Hoekman et al. 120

Furthermore, Hoekman et al. 120 also reported that another study has shown that, 2–4 years after randomisation, mucosal healing at week 104 after randomisation, but not treatment allocation, was associated with stable, corticosteroid-free remission. 143

Therefore, although there is some evidence to support the association of 2-year endoscopic mucosal healing with long-term, corticosteroid-free clinical remission, there does not seem to be any evidence that mucosal healing at 2 years differs according to treatment (TD or SU). Of note is that estimates used in Marchetti et al. 88 were taken from another study,143 which the EAG did not have access to.

The company in TA352 assumed that patients discontinued treatment with biologic agents approximately 1 year after maintenance treatment. The Evidence Review Group in TA352 was concerned that a discontinuation rule may not have been appropriate for patients who are not in remission as the NICE recommendation for infliximab and adalimumab suggests that, ‘specialists should discuss the risks and benefits of continued treatment with patients and consider a trial withdrawal from treatment for all patients who are in stable clinical remission. People who continue treatment with infliximab or adalimumab should have their disease reassessed at least every 12 months to determine whether ongoing treatment is still clinically appropriate. People whose disease relapses after treatment is stopped should have the option to start treatment again’ (© NICE 2015 Vedolizumab for treating moderately to severely active Crohn’s disease after prior treatment. Available from www.nice.org.uk/guidance/ta352 All rights reserved. Subject to Notice of rights. NICE guidance is prepared for the National Health Service in England. All NICE guidance is subject to regular review and may be updated or withdrawn. NICE accepts no responsibility for the use of its content in this product/publication.). The EAG notes that duration of treatment with biologics in clinical practice remains uncertain. The clinical experts advising the EAG reported that treatment with anti-TNF and second-line biologics would be given as long as patients continue to show a response.

For completeness, the EAG ran an additional scenario analysis assuming that 100% of patients in continuous remission for 12 months with maintenance treatment of any biologic (i.e. anti-TNF, second-or third-line biologics) discontinue treatment.

Surgery as a final treatment step in the economic model

The clinical experts advising the EAG explained that once patients exhaust all the biologic treatments available, they receive surgery. Therefore, the EAG ran a scenario analysis in which patients escalating from third-line biologic treatment in the model receive surgery. The EAG assumed that surgery had a temporary ‘curative’ effect of 2 years, where patients experience the costs and utility associated with being in the remission state. After 2 years, it was assumed that patients revert to the moderate to severe state, where they remain for the rest of the model.

To test the sensitivity of the results of the model to assumptions relating to surgery, the EAG ran a scenario analysis excluding surgeries from the model.

Accounting for the cost-effectiveness of misdiagnosed cases and assumptions around treatment discontinuation in the model

The EAG combined a range of scenarios to assess the impact of increasing the effectiveness of the TD strategy while decreasing costs with biologic treatments. Scenarios 5.2.5a, 5.2.5b and 5.2.5c explored changing the effectiveness of the diagnostic tool (and TD) through the assumptions made for the misdiagnosis scenario:

1. The EAG combined the misdiagnosis scenario 5.2.1 with scenario 5.2.3ai, where it was assumed that after 2 years in remission, 76% of patients in the TD strategy experience mucosal healing, while 40% of patients in the SU arm experience the same outcome.

2. The EAG also combined scenario 5.2.1 with scenario 5.2.3aii, where it was assumed that after 2 years in remission, 76% of patients in both the TD and the SU strategies experience mucosal healing.

3. The EAG also combined scenario 5.2.1 with scenario 5.2.3aiii, where it was assumed that after 2 years in remission, 40% of patients in both the TD and the SU strategies experience mucosal healing.

Varying the assumptions around the measure of relative treatment effectiveness for time to treatment escalation and assumptions around treatment discontinuation in the model

As with scenario 5.2.5, the EAG explored the impact of combining scenario 5.2.3 (where costs associated with biologics were decreased) with changing the effectiveness of the diagnostic tool (and TD) through the assumptions made for time to treatment discontinuation in the model. The EAG used scenario 5.2.2aii for all the analyses as this is the scenario that assumes the highest relative effective for TD versus SU in terms of TTE:

• The EAG combined scenario 5.2.2aii with scenario 5.2.3ai, where it was assumed that, after 2 years in remission, 76% of patients in the TD strategy experience mucosal healing, while 40% of patients in the SU arm experience the same outcome.

• The EAG also combined scenario 5.2.2aii with scenario 5.2.3aii, where it was assumed that, after 2 years in remission, 76% of patients in the TD and the SU strategies experience mucosal healing.

• The EAG also combined scenario 5.2.2aii with scenario 5.2.3aiii, where it was assumed that, after 2 years in remission, 40% of patients in the TD and the SU strategies experience mucosal healing.

Varying the proportion of patients who respond to immunomodulators and varying the assumptions around the measure of relative treatment effectiveness for time to treatment escalation

One of the scenario analyses carried out by the EAG assumed that no high-risk patients respond to IMs (and therefore these patients do not derive any benefit from response to this treatment). This scenario was intended to portray an extreme clinical reality in which high-risk patients need treatment with a biologic to achieve a response and the impact of this assumption on the final ICER. The ICER for PredictSURE-IBD compared with SU changed from the EAG’s base case of dominated (against the diagnostic tool) to £71,294. Of note is that the EAG tested the impact of varying the proportion of patients who do not respond to IM treatment in the analysis, and when 92% of patients were assumed not to respond to IM treatment the two strategies (TD and SU) became clinically equivalent.

The EAG combined scenario 5.2.2aii with varying the proportion of high-risk patients who receive SU therapy and do not respond to IMs, thereby increasing the relative effectiveness of TD and decreasing the effectiveness of SU in terms of both TTE and the probability of response and remission in the model.

The EAG tested the assumption that 100% of patients do not respond to IMs and varied this percentage to assess the impact on the final ICERs.

Varying the proportion of patients who respond to immunomodulators; varying the assumptions around the measure of relative treatment effectiveness for time to treatment escalation; and varying treatment discontinuation assumptions

• The EAG combined scenario 5.2.6a with varying the proportion of high-risk patients who receive SU therapy and do not respond to IMs (therefore not deriving any benefit from response to this treatment).

• The EAG combined scenario 5.2.6b with varying the proportion of high-risk patients who receive SU therapy and do not respond to IMs (therefore not deriving any benefit from response to this treatment).

• The EAG combined scenario 5.2.6c with varying the proportion of high-risk patients who receive SU therapy and do not respond to IMs (therefore not deriving any benefit from response to this treatment).

All of the above scenarios increased the relative effectiveness of TD in terms of TTE and decreased the costs associated with biologic treatment (to different amounts). For all scenarios, the EAG tested the assumption that 100% of patients do not respond to IMs and varied this percentage to assess the impact on the final ICERs.

Results

The results of the EAG’s scenario analyses are reported in Table 24. The majority of the scenarios still produced a dominated ICER, showing that the TD strategy (via the use of PredictSURE-IBD in the model) is dominated by SU (via the standard care arm of the model), with additional costs and a QALY loss.

Scenario 5.2.1 produced an ICER of £67,741 per QALY gained, with PredictSURE-IBD being more costly than standard care but generating a QALY gain of 0.11. Even though this scenario assumes lower test accuracy, the assumed consequences of misdiagnosis produced a QALY gain with the diagnostic tool. This is related to the assumption of allocating low-risk patients (misdiagnosed as high-risk) to the anti-TNF state in the model, without need for further escalation. Given that treatment with anti-TNF holds the highest remission rate in the EAG’s analysis, and that 62% of high-risk patients (misdiagnosed as low-risk) in the SU arm were assumed to not derive any benefit from treatment with IMs, the results produced positive incremental QALYs for the diagnostic tool (and, thus, for the TD strategy). The EAG also combined this scenario with reducing the costs associated with TD through reducing the time spent on biologic treatment (as per scenario 5.2.3) and presents the results in scenario 5.2.5.

Scenario 5.2.3ai produced an ICER of £44,103 for standard care compared with PredictSURE-IBD, meaning that the diagnostic tool is less expensive than standard care (by £4621) but also less effective (0.10 QALY loss). This scenario reduced the costs of biologic treatment in the TD arm by assuming that a higher proportion of patients in the TD arm achieve mucosal healing and, thus, stop treatment. Even though these patients were ‘kept’ in the remission state, the QALYs generated using this assumption were not enough to produce a QALY gain compared with the benefit patients derive from initial treatment with IMs in SU. The EAG also notes that scenario 5.2.3ai can also be interpreted as a proxy for a scenario assuming de-escalation from biologic treatment in the TD arm to IMs. This is because the scenario reduced treatment costs (by stopping treatment with biologics), which would be similar to replacing biologic treatment with IMs in the model as a result of the low cost of IM treatment.

The other variations of scenario 5.2.3, where the same proportions of patients were assumed to achieve mucosal healing in the TD and SU arms, produced dominated ICERs against the diagnostic tool (and, thus, TD). The EAG notes that Hoekman et al. 120 did not show a difference in mucosal healing for TD versus SU (although it is not clear if the authors investigated the impact that the strategies had on this outcome). Notwithstanding, the authors reported that the rate of mucosal healing reported in another study143 had shown that 2–4 years after randomisation, treatment allocation was associated with stable, treatment-free remission.

Scenario 5.2.5a resulted in a dominant ICER for PredictSURE-IBD (and TD), with the diagnostic tool associated with lower costs and higher QALYs than standard care (and SU). This scenario combines modelling misdiagnosed cases with reducing the costs associated with TD, and therefore generates additional QALYs for the diagnostic tool at a lower cost, given the assumption that a proportion of patients on TD enter a permanent stage of remission. Given that scenario 5.2.5a assumes a difference in the rate of treatment discontinuation for biologics (whereby TD patients have a higher probability of discontinuing treatment – owing to mucosal healing – than SU patients), this scenario produced the highest cost savings for TD. Scenarios 5.2.5b and 5.2.5c produced higher ICERs as the relative costs associated with treatment with biologics (and the diagnostic tool) increased; however, scenario 5.2.5b resulted in an ICER of £29,932 per QALY gained, close to the upper threshold (£30,000) typically used in the NICE decision-making process.

Scenarios 5.2.6a, 5.2.6b and 5.2.6c explored increasing the effectiveness of TD versus SU with respect to TTE, combined with decreasing the treatment costs with biologics. As demonstrated, all scenarios generate a QALY loss for the diagnostic tool compared with standard care. When it is assumed that a higher proportion of patients in the TD arm achieve mucosal healing (scenario 5.2.3ai) than in the SU arm, the diagnostic tool (and TD) becomes cost saving (–£4621), but less effective (–0.10).

Scenarios 5.2.7 and 5.2.8 explored increasing the effectiveness of TD versus SU with respect to TTE, combined with decreasing the treatment costs with biologics and with varying the assumption around the rate of response to IM treatment in the SU strategy.

Scenario 5.2.7 shows that when the relative TTE benefit in the anti-TNF step of the TD strategy compared with the IM step in the SU strategy carries through all the next treatment steps in the model (scenario 5.2.2aii) and when 100% of SU patients are assumed not to respond to treatment with IMs, the ICER amounts to £60,056 per QALY gained. Therefore, even when 100% of high-risk patients do not respond to IMs, the ICER for the diagnostic tool (and TD) compared with standard care (and SU) is still above the NICE £30,000 threshold.

Scenario 5.2.8a shows that when the relative TTE benefit in the anti-TNF step of the TD strategy compared with the IM step in the SU strategy carries through all the next treatment steps in the model (scenario 5.2.2aii), when a higher proportion of patients in the TD arm achieves mucosal healing (scenario 5.2.3ai), and when 100% of SU patients are assumed to not respond to treatment with IMs, the final ICER becomes dominant for PredictSURE-IBD (and TD), with the diagnostic tool being associated with lower costs and more QALYs than standard care (and SU). The diagnostic tool remains dominant until the assumption around the proportion of high-risk SU patients not responding to IM treatment is decreased from 100% to 79%. Of note is that the EAG’s base-case analysis estimates that 62% of high-risk patients do not respond to initial treatment with IMs.

Scenarios 5.2.8b and 5.2.8c show that when the relative TTE benefit in the anti-TNF step of the TD strategy compared with the IM step in the SU strategy carries through all the next treatment steps in the model (scenario 5.2.2aii), when the same proportion of patients in the TD and SU arms achieves mucosal healing (scenario 5.2.3aii for 76% and 40%, respectively), and when 100% of SU patients are assumed not to respond to treatment with IMs, the final ICERs are £28,192 and £43,286, respectively. Both scenarios generate a QALY gain for the diagnostic tool (and TD) compared with standard care (and SU); however, the additional costs associated with TD are higher in scenario 5.2.8c (40% of patients in remission stop treatment with biologics in both the TD and SU arms) than in scenario 5.2.8b (76% of patients in remission stop treatment with biologics in both the TD and the SU arms).

The EAG has produced plots to demonstrate the impact of reducing the percentage of high-risk patients who do not respond to IMs from 100% to 0%. The plot in Figure 22 shows the changes in the incremental costs and QALYs on the cost-effectiveness plane and demonstrates the ICER changing from dominant at 100% non-response to IMs, moving into the south-west quadrant (less costly and less effective for TD) at 79%, then becoming dominated from below 43%. Figure 23 shows the resulting final ICERs, and the drastic variation in these at 79% non-response, when the incremental QALYs become negative.

FIGURE 22.

Incremental costs and QALYs as percentage of high-risk IM non-responders varies.

FIGURE 23.

Resulting ICERs as the percentage of high-risk IM non-responders varies.

Conclusions

1. Estimating the impact of reducing test accuracy was only possible by combining this with an increase in the relative effectiveness of the TD strategy (to attribute consequences to misdiagnosing patients). However, changing this alone in the model still produced ICERs above the NICE upper cost-effectiveness threshold of £30,000 (scenario 5.2.1). When this assumption was combined with decreasing the costs associated with biologic treatment (by assuming different rates of mucosal healing leading to remission), the ICER ranged from dominant to £47,842 for PredictSURE-IBD (and TD) (scenarios 5.2.5a and 5.2.5c, respectively).

2. By itself, increasing the relative effectiveness of TD on TTE did not have an impact on the dominance of standard care over TD (scenario 5.2.2).

3. Assuming that 40% and 76% of patients in remission after 2 years (and 100% of patients in remission after 1 year) on maintenance treatment with anti-TNF and second- and third-line biologics discontinued treatment in both treatment arms also did not have an impact on the dominance of standard care over TD. Nonetheless, when a higher proportion of patients discontinued treatment with biologics in the TD arm than in the standard care arm, this generated a cost saving for TD, although still with fewer QALYs than for SU (scenario 5.2.3).

4. Excluding surgeries from the model did not have an impact on the dominance of standard care over TD, and neither did assuming that surgery has a curative effect for 2 years (scenario 5.2.4).

5. Combining the increase in the relative effectiveness of TD on TTE with the reduction in the costs of biologic treatment did not have an impact on the dominance of standard care over TD when the same proportion of patients were assumed to discontinue treatment with biologics in the TD and SU arms. When a higher proportion of patients discontinued treatment with biologics in the TD arm than in the standard care arm, this generated a cost saving for TD, but with fewer QALYs than for SU (scenario 5.2.6).

6. Increasing the relative effectiveness of TD on TTE and additionally reducing the effectiveness of SU (through assuming a 0% probability of response to IM treatment for high-risk patients) still generated an ICER above the NICE cost-effectiveness upper threshold of £30,000 (scenario 5.2.7).

7. When the increase in the relative effectiveness of TD on TTE and the additional reduction in the effectiveness of SU are combined with a reduction of time on treatment with biologics, the ICERs for PredictSURE-IBD (and TD) drop below the £30,000 per QALY gained threshold with standard care (and SU), depending on the assumptions made for the proportion of patients who discontinue treatment with biologics. When the proportion of patients discontinuing treatment with biologics is 76% in the TD arm compared with 40% in the SU arm, the final ICER is dominant for PredictSURE-IBD against standard care, as long as the proportion of high-risk patients who do not respond to initial treatment with IMs is 79% (or above).

In conclusion, once the relative effectiveness of TD is artificially increased (through TTE, the probability of response to initial treatment, and the impact that it has on low-risk patients) and is combined with decreased time on biologic treatment, the ICERs for PredictSURE-IBD (and TD) compared with standard care (and SU) fall below £30,000, which is the upper threshold typically used in the decision-making process by NICE. However, the EAG notes that these results need to be interpreted with extreme caution as the assumptions made in these scenarios were designed to test extreme clinical scenarios where TD was assumed to be more effective than SU. Nonetheless, the EAG did not find any evidence to substantiate the benefits modelled in these scenarios, and thus concludes that its base-case analysis showing that TD is dominated by SU remains the most conservative assessment of the relative cost-effectiveness of these treatment strategies.

One-way sensitivity analysis

The EAG conducted a number of deterministic one-way sensitivity analyses around the model inputs, as described in Table 25. Figure 24 ranks the key drivers of the model by their impact on the incremental net monetary benefit of PredictSURE-IBD compared with standard care, based on a willingness-to-pay threshold of £30,000 per QALY. The lower and upper bounds of each parameter input were derived from the lower and upper bounds of the 95% CIs of the distributions specified for the PSA. The inputs with the highest impact on the model results were the response to biologic treatments in both the TD and the SU arms. Details of each of the distributions are given in Appendix 6 (see Table 43).

FIGURE 24.

Tornado plot showing one-way sensitivity analyses for PredictSURE-IBD vs. standard care. INMB, incremental net monetary benefit. Note that the bars in the graph represent the change in INMB and the corresponding ICERs are presented at both ends of the bars. Light blue bars represent the lower bound of the parameter changes and dark blue bars represent the upper bound of the parameter changes.

Prognostic accuracy

Twelve publications50,67,69,71–79 describing eight studies were included in the assessment of the prognostic accuracy of the tests. Seven of the studies67,69,72,75,76,78,79 reported results on the utility of the IBDX kit and one study provided data on PredictSURE-IBD for stratifying those at high risk of a severe course of CD. Limited evidence is available from the included full-text publications on the prognostic accuracy of PredictSURE-IBD, and no evidence is available on prognostic accuracy of IBDX as determined by measures such as sensitivity and specificity. Most evidence on the utility of the two tools is derived from observational studies that report estimates of the risk of experiencing a clinical outcome associated with an aggressive course of CD, for example need for treatment escalation, development of a complication or surgery. No study retrieved reported on the clinical impact of the use of IBDX or PredictSURE-IBD in terms of influencing the treatments given in the management of active CD.

All included studies assessed outcomes in people reported to have a diagnosis of CD. However, limited reporting was noted across studies relating to IBDX on the stage of diagnosis (newly vs. established) at the time of the test. Baseline characteristics suggest that samples analysed were predominantly provided by people who had established CD. By contrast, most people enrolled in the study on PredictSURE-IBD had received a recent diagnosis of CD. Although most of the included studies outlined criteria to be met for a diagnosis of CD, only the study evaluating PredictSURE-IBD required people to have active disease to be eligible for enrolment, and reported how the presence of active disease was determined. Given the biomarker targets of the two prognostic tests, the reviewers consider that a criterion of active CD is appropriate for the inclusion of studies assessing PredictSURE-IBD but is not essential for studies reporting on IBDX.

The use of PredictSURE-IBD was associated with a sensitivity and specificity of 72.7% and 73.2%, respectively, in stratifying by need for multiple treatment escalations within 12 months. A negative predictive value of 90.9% for PredictSURE-IBD of predicting multiple escalations within the first 18 months was also reported. The cut-off point for multiple escalations applied in the determination of sensitivity and specificity was two treatment escalations and comprised any type of treatment, including surgery.

Seven studies67,69,72,75,76,78,79 evaluating the IBDX kit were deemed to be of relevance to the review, all of which were observational in nature: three studies75,76,79 were prospective cohorts and three69,72,78 had a cross-sectional design. Clinical heterogeneity across studies in terms of various characteristics (prior complication vs. no complication, previous IBD-related surgery or no surgery, and unclear whether or not people had active disease at baseline) was noted. Two prospective cohort studies75,76 reported an increased risk of experiencing a complication or of requiring surgery among those testing positive for at least two of the six biomarkers included in the IBDX kit.

Two studies reported an increased risk of experiencing a complication or of requiring surgery for those testing positive for at least two of the six biomarkers included in the IBDX kit. Risks of experiencing a complication by positive biomarker status were reported to be:

• OR of 1.5 (95% CI 1.3 to 1.9; p < 0.001; number unclear) based on positivity for a median of two biomarkers

• HR of 2.5 (95% CI 1.03 to 6.1; p = 0.043; n = 20 with no prior complication or surgery) based on positivity for at least two biomarkers

• HR of 2.6 (95% CI 0.92 to 7.2; p = 0.072; n = 20 with no prior complication or surgery) based on positivity for at least three biomarkers.

Considering surgery, three studies reported on the increased risk of surgery. One study reported a trend towards a larger proportion of people with CD requiring abdominal surgery with increasing number of positive biomarkers (n = 517; p < 0.0001 across the groups). Other estimates of higher risk of requiring surgery were:

• OR of 1.5 (95% CI 1.3 to 1.8; p < 0.001; number unclear) based on positivity for a median of two biomarkers

• HR of 3.6 (95% CI 1.2 to 11.0; p = 0.023; n = 14 with no prior complication or surgery) based on positivity for at least two biomarkers

• HR of 2.8 (95% CI 0.80 to 9.6; p = 0.11; n = 14 with no prior complication or surgery) based on positivity for at least three biomarkers.

The study evaluating PredictSURE-IBD reported that those categorised as at high risk of following a severe course of disease had a statistically significantly higher risk of first treatment escalation compared with those designated as at low risk, with a HR of 2.65 (95% CI 1.32 to 5.34; p = 0.006).

Economic

As no robust evidence was identified on the prognostic accuracy of the biomarker stratification tools, the development of the economic model sets a structural framework for analysing future available data on prognostic accuracy and assesses the costs and consequences of treating high- and low-risk patients with both the TD and the SU strategies.

The EAG found two main sources of evidence that could be used to model TTE and TTS. Nevertheless, each source could only partially inform the TTE and TTS analyses in the model. Therefore, clinical data informing the analysis had to be derived from multiple sources.

One of the key underlying assumptions in the EAG’s base-case analysis is that high-risk patients who initiate treatment with IMs (SU arm) escalate treatment quicker than high-risk patients who initiate treatment with anti-TNF (supported by the data presented in D’Haens et al. 35). However, once SU patients initiate treatment with an anti-TNF (their second treatment step), they ‘catch up’ with patients on the TD treatment strategy. As some high-risk patients who receive SU treatment respond to IM treatment, having the additional IM step in the SU strategy is advantageous to patients in the EAG’s base-case analysis as patients still subsequently receive treatment with biologics, which are assumed to have the same effect as biologics is the TD arm. Given the paucity of data to substantiate any further benefits in subsequent treatment steps in the TD approach compared with the SU approach, the EAG considered this to be the most conservative modelling approach.

The EAG also notes that although, in theory, a TD approach would suggest a ‘de-escalation’ of treatments, the clinical experts advising the EAG consistently reported that IMs would not be given to patients who respond well to biologics (instead, treatment with biologics would be continued until loss of response). The experts also explained that, after loss of response with first- or second-line biologics, patients would not be given IMs but instead would undergo surgery as a last treatment option. Nonetheless, the EAG undertook a scenario analysis (5.2.3ai) to explore the impact of de-escalation in the model.

The long-term follow-up study by Hoekman et al. 120 found no difference between SU and TD in 10-year clinical remission rate, endoscopic remission, hospitalisation, surgery or new fistulas. Furthermore, the study concluded that, in the long term, a TD strategy had not been proven to alter the natural history of CD. However, time to relapse was found to be statistically significantly different between the TD and SU arms in the 2-year analysis of the same data (by D’Haens et al. 35).

Hoekman et al. concluded that their study was the first to compare the long-term outcomes for newly diagnosed CD patients who received combined immunosuppression compared with those for patients who received conventional management. The authors added that early combined immunosuppression may be a preferential strategy, given the associated delay in time to relapse. However, the authors noted that the costs and risks of potentially overtreating patients with a potentially ‘benign’ disease course mean that a TD approach should not be recommended as a universal treatment strategy for all patients with newly diagnosed CD.

The EAG’s cost-effectiveness analyses are consistent with the conclusions from Hoekman et al. 120 The ICERs indicate that standard care (and so SU) dominates the use of both diagnostic tools (and so TD), even when assuming that the tests are 100% accurate. In the base-case results, the incremental analysis of cost-effectiveness demonstrates that the TD strategy (via the use of PredictSURE-IBD in the model) is dominated by SU (via the standard care arm of the model), regardless of whether it is assumed that TTE does or does not reset with every new treatment in the model.

To mitigate some of the concerns raised by the specialist committee members, the EAG conducted a range of analyses to test extreme scenarios around increasing the relative treatment effectiveness of the TD approach while decreasing the relative costs associated with TD. The EAG concluded the following:

1. Estimating the impact of reducing test accuracy was only possible through combining this with an increase in the relative effectiveness of the TD strategy (to attribute consequences to misdiagnosing patients). However, changing this alone in the model still produced ICERs above NICE’s upper cost-effectiveness threshold of £30,000. When this assumption was combined with decreasing the costs associated with biologic treatment (through assuming different rates of mucosal healing leading to remission), the ICER ranged from dominant to £47,842 for PredictSURE-IBD (and TD).

2. By itself, increasing the relative effectiveness of TD on TTE did not have an impact on the dominance of standard care over TD.

3. Assuming that 40% and 76% of patients in remission after 2 years (and 100% of patients in remission after 1 year) on maintenance treatment with anti-TNF, second- and third-line biologics discontinued treatment in both treatment arms also did not have an impact on the dominance of standard care over TD. Nonetheless, when a higher proportion of patients discontinued treatment with biologics in the TD arm compared with the standard care arm, this generated a cost saving for TD, albeit still with fewer QALYs than for SU.

4. Excluding surgeries from the model did not have an impact on the dominance of standard care over TD, and neither did assuming that surgery has a curative effect at 2 years.

5. Combining the increase in the relative effectiveness of TD on TTE with reducing the costs of biologic treatment did not have an impact on the dominance of standard care over TD when the same proportion of patients were assumed to discontinue treatment with biologics in the TD and the SU arms. When a higher proportion of patients discontinued treatment with biologics in the TD arm than in the standard care arm, this generated a cost saving for TD, albeit with fewer QALYs than for SU.

6. Increasing the relative effectiveness of TD on TTE and additionally reducing the effectiveness of SU (through assuming a 0% probability of response to IM treatment from high-risk patients) still generated an ICER above NICE’s upper cost-effectiveness threshold of £30,000.

When the increase in the relative effectiveness of TD on TTE and the additional reduction in the effectiveness of SU are combined with a reduction in time on treatment with biologics, the ICERs for PredictSURE-IBD (and TD) can become cost-effective compared with standard care (and SU), depending on the assumptions made for the proportion of patients who discontinue treatment with biologics. When the proportion of patients discontinuing treatment with biologics is 76% in the TD arm compared with 40% in the SU arm, the final ICER is dominant for PredictSURE-IBD against standard care, as long as the proportion of high-risk patients who do not respond to initial treatment with IMs is 79% (or above).

Clinical

Despite extensive systematic searches of the literature, no robust evidence was identified on the prognostic accuracy of the biomarker stratification tools IBDX and PredictSURE-IBD. In terms of sensitivity and specificity as estimates of prognostic accuracy, the EAG is unaware of a validated definition for determining whether or not a person has followed a severe course of CD, and, thus, considers the criterion for a true positive or false positive using IBDX and PredictSURE-IBD to be unclear. The EAG considers that it would be challenging to ascertain an accurate estimate of prognostic accuracy of the tools in stratifying the course of CD and to do so would require carrying out a prospective study that included a group that received only SU treatment after determination of the risk of a severe course of CD. The ongoing PROFILE RCT randomises people to accelerated SU or TD treatment after determining whether they are at high or low risk of following a severe course of CD, and so will provide additional data to inform estimates of prognostic accuracy. One study50 reporting on the sensitivity and specificity of PredictSURE-IBD was identified. The EAG has reservations about the generalisability of the estimates. To determine sensitivity and specificity, the authors of the study applied a cut-off point of two or more treatment escalations to denote a high risk of severe course of CD, with surgery included as treatment escalation. The EAG considers the choice of two escalations to be arbitrary. Additionally, the EAG’s clinical experts fed back that it would be appropriate to consider escalation to CD-related surgery separately from progression to drug treatment, and also to use the development of a complication of CD (fistula or stenosis) as an alternative marker of sensitivity and specificity.

Studies informing the evidence around the effectiveness of the tools predominantly estimated an increased risk of experiencing a clinical outcome for those designated as at high risk compared with those designated as at low risk of following a severe course of CD. Clinical outcomes that could be considered proxies for predicting prognosis of CD are developing a complication (fistula or stenosis), needing CD-related surgery, and a shorter time to and increased frequency of treatment escalations.

For IBDX, estimates were available for increased risk of developing a complication and for need for surgery for those classified as at high risk of following a severe disease course, but estimates were not available for TTE. Conversely, estimates were available for PredictSURE-IBD for TTE but not for risk of developing a complication or need for surgery. Given the disparity in the clinical outcomes assessed with IBDX and PredictSURE-IBD, the EAG considers that no conclusions can be drawn on the comparative effectiveness of the two tools in stratifying people by risk of a severe course of CD.

Another limitation of the identified evidence base is that no study included in the review prospectively followed people whose treatment was determined by results from IBDX and PredictSURE-IBD. The ongoing PROFILE RCT assesses whether or not early treatment with TD strategy affords clinical benefit to those categorised as being high risk of severe course of CD. However, given that people are first stratified as high or low risk using PredictSURE-IBD and, subsequently, are randomised to SU or TD treatment, the EAG considers that there is potential for the misdiagnosis of people who are truly low risk but are categorised as high risk to go undetected. However, an analysis of those randomised to accelerated SU after determination of being at high or low risk of following a severe course of CD will provide additional data to inform estimates of prognostic accuracy.

Economic

The EAG’s model offers methodological advantages when compared with the PredictSURE-IBD model. The main strength of the economic analysis is that it captures partial response to maintenance treatment (as well as remission, relapse, surgery and post-surgical remission). The analysis also uses time to event data (TTE and TTS) more extensively than previous models. Furthermore, the EAG has conducted a series of scenario analyses exploring structural and parameter uncertainty in the economic model. The EAG also conducted a series of scenarios testing extreme clinical assumptions around the potential benefit of TD compared with SU in order to mitigate the concerns raised by the specialist committee members.

However, clinical data informing the analysis had to be derived from multiple sources. This approach is not ideal and creates a patchwork network of evidence, introducing uncertainty into the economic results. Nonetheless, the EAG anticipates that this problem will be potentially overcome when the results of the PROFILE trial are available to populate the economic model.

The test accuracy in the base-case economic model for PredictSURE-IBD and in the scenario analysis for IBDX is the same and assumed to be 100%. The only difference in the cost-effectiveness analyses of the two diagnostic tests is the cost of the test. This is unlikely to reflect the tests’ actual accuracy in clinical practice; however, no robust diagnostic data were found in the analysis to inform this.

The potential benefits of TD treatment for high-risk patients are dependent on two questions that remain unanswered: (1) do some high-risk patients derive a benefit from receiving IM treatment before moving to biologic treatment? and (2) do SU high-risk patients have the same benefits as TD high-risk patients once they initiate the TD treatment pathway (i.e. treatment with anti-TNF)? In the EAG’s model, the potential disadvantage of waiting to initiate treatment with anti-TNF was based on only the increased risk of surgery in the SU arm; however, the negative impact of surgery in the analysis was not enough to offset the advantages of initial treatment with IMs for SU patients.

Finally, the EAG acknowledges that adverse events, specifically those relating to the long-term use of biologics, and the potential benefits associated with surgery were not included in the economic analysis. However, if adverse events were included in the analysis, given that a higher proportion of patients receive biologic treatment in the TD arm, this would have a negative impact on the outcomes in the TD arm of the model compared with the SU arm. Similarly, although the EAG has not captured the potential benefit of surgery in the economic analysis, it notes that to do so would benefit the SU strategy, as a higher proportion of patients receive surgery in the SU arm than in the TD arm. Therefore, including adverse events and the benefits of surgery in the analysis would not change the conclusions likely to be drawn from the current results.

Effectiveness of induction and maintenance therapies

The reasons for excluding studies identified from TA352125 (vedolizumab) and TA456119 (ustekinumab) from the EAG’s analyses are presented in Table 38. The EAG notes that the key differences between the analyses carried out by the EAG and those presented in TA352 and TA456 are exclusion by the EAG of the study carried out by Targan et al. 160 (single dose of 5 mg of infliximab administered) and inclusion of subgroup data from the anti-TNF-naive subgroup of the study reported by Watanabe et al. 127 (see Table 38). In addition, the EAG notes that studies of ustekinumab were not included in TA352, whereas they were included in both TA456 and the EAG analyses.