Clinical effectiveness and cost-effectiveness of use of therapeutic monitoring of tumour necrosis factor alpha (TNF-a) inhibitors [LISA-TRACKER® enzyme-linked immunosorbent assay (ELISA) kits, TNF-a-Blocker ELISA kits and Promonitor® ELISA kits] versus standard care in patients with Crohn's disease: systematic reviews and economic modelling

Mild to moderate disease

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

Hanauer and Sandborn7

Moderate to severe disease

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

Hanauer and Sandborn7

Severe to fulminant disease

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

Hanauer and Sandborn7

Remission

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

Hanauer and Sandborn. Reprinted by permission of Nature Publishing Group from Macmillan Publishers Ltd: The American Journal of Gastroenterology, Hanauer SB, Sandborn W. Management of Crohn’s disease in adults. Am J Gastroenterol 2001;96:635–43,7 copyright 2001

Induction of remission according to the National Institute for Health and Care Excellence Clinical Guideline number 1522

Usually, at first presentation, patients with active CD are recommended to receive monotherapy treatment with conventional steroid therapy (i.e. glucocorticoids including prednisolone, methylprednisolone or intravenous hydrocortisone), which is aimed at inducing remission as a first-line treatment. Alternatively, treatment with budesonide, 5-ASA or enteral nutrition may be offered for patients who do not choose to take or who are intolerant of glucocorticosteroid therapy.

The addition of an immunosuppressant (azathioprine, mercaptopurine or methotrexate) to a conventional glucocorticosteroid or budesonide is recommended as an add-on therapy for inducing remission in patients who have active CD and who have experienced two or more inflammatory exacerbations in a 12-month period, or in whom glucocorticosteroid doses cannot be tapered. As advised in the current online version of the British National Formulary (BNF)18 or BNF for Children,19 the effects of azathioprine, mercaptopurine and methotrexate, as well as levels of neutropenia (in patients on azathioprine or mercaptopurine), should be monitored. 15

In adults and children aged 6–17 years with severe active CD who fail to respond to the first line of treatment with conventional therapy (e.g. immunosuppressants or corticosteroids), or who are intolerant to or who have contraindications to conventional therapy, anti-TNF-α agents (IFX and ADA) are recommended as treatment options within their licensed indications. The administration of anti-TNF-α agents is recommended until 12 months after the start of treatment or until treatment failure (including the need for surgery), whichever occurs first. Reassessment and monitoring of disease activity (at least every 12 months) is advised to ascertain the clinical appropriateness of ongoing treatment. Usually, treatment is initiated with the less expensive drug (i.e. IFX), considering drug administration costs, dose and product price per dose. The use of anti-TNF-α drugs for the treatment of CD is covered in the 2010 NICE technology appraisal guidance 187 [IFX (review) and ADA for the treatment of CD], which is summarised in NICE guidelines. 6

Surgery should be considered early in the course of the disease for patients whose disease is limited to the distal ileum or for children and young people who have growth impairment despite optimal medical treatment and/or who have refractory disease. 2

Maintenance of remission according to the National Institute for Health and Care Excellence clinical guideline 1522

Patients with CD in remission can be managed with or without maintenance treatment. The options for maintenance (including treatment or no treatment) need to be discussed with patients and parents or carers. The discussion should include risk of relapse and the potential side effects of drug treatments. Patients who decide not to use maintenance treatment should agree follow-up plans (e.g. frequency and duration of visits) and should receive information on markers and symptoms of relapse (e.g. unintended weight loss, abdominal pain, diarrhoea or general ill-health) to ensure that they keep their disease appropriately under review with their health-care professionals.

Patients with CD in remission who choose to receive maintenance therapy may be offered a single drug such as azathioprine or mercaptopurine if remission has been induced using a conventional glucocorticosteroid or budesonide. Methotrexate can be offered if remission was induced by methotrexate or to patients who are not able to tolerate, or who have contraindications to, azathioprine or mercaptopurine. Treatment with 5-ASA can be used to maintain remission after surgery.

If remission has been achieved with anti-TNF-α medication, then maintenance with anti-TNF-α with or without an immunosuppressant can be used. Continuation of treatment with IFX or ADA during remission is advised only if there is evidence of ongoing active disease assessed by clinical symptoms, biological markers and endoscopy, if necessary. The balance between harms and benefits of ongoing treatment should be taken into account. The guideline states that patients who relapse after anti-TNF-α treatment may start it again. 15

National Institute for Health and Care Excellence guidelines

The NICE guideline technology appraisal 1876 describes when IFX or ADA should be used to treat patients with severe active or fistulising CD in the NHS in England and Wales.

Infliximab

The guideline states:

Infliximab has a UK marketing authorisation for the treatment of:

1. severe, active Crohn’s disease in people whose disease has not responded despite a full and adequate course of therapy with a corticosteroid and/or an immunosuppressant, or who are intolerant to or have medical contraindications for such therapies

2. fistulising, active Crohn’s disease in people whose disease has not responded despite a full and adequate course of therapy with conventional treatment (including antibiotics, drainage and immunosuppressive therapy)

3. severe, active Crohn’s disease in people aged 6–17 years whose disease has not responded to conventional therapy, including a corticosteroid, an immunomodulator and primary nutrition therapy, or who are intolerant to or have contraindications for such therapies.

NICE (2010) technology appraisal number 187. Infliximab (Review) and Adalimumab for the Treatment of Crohn’s Disease. London: NICE. Available from www.nice.org.uk/guidance/TA187. 6 Reproduced with permission. This information is accurate at time of publication

Administration of IFX should follow this pattern:

. . . 5-mg/kg intravenous infusion over a 2-hour period followed by another 5-mg/kg infusion 2 weeks after the first. If a person’s disease does not respond after two doses, no additional treatment with infliximab should be given. In people whose disease responds, infliximab regimens include maintenance treatment (another 5-mg/kg infusion at 6 weeks after the initial dose, followed by infusions every 8 weeks) or re-administration, otherwise known as episodic treatment (an infusion of 5-mg/kg if signs and symptoms of the disease recur).

NICE (2010) technology appraisal number 187. Infliximab (Review) and Adalimumab for the Treatment of Crohn’s Disease. London: NICE. Available from www.nice.org.uk/guidance/TA187. 6 Reproduced with permission. This information is accurate at time of publication

For fistulising disease, the first three doses at weeks 0, 2 and 6 are considered as induction therapy and additional IFX therapy should not be given if the first three doses have not induced a response. The patient pathway for people responding to IFX induction therapy and moving onto maintenance therapy is given in Figure 1.

FIGURE 1.

Patient pathway of patients with CD on IFX therapy.

Anti-tumour necrosis factor alpha agents

Crohn’s disease is associated with elevated levels of the immune regulatory protein, TNF-α. The reasons for this elevation in CD are still largely unknown. TNF-α is a small cell signalling protein (cytokine) involved in inflammatory responses primarily by influencing regulation of various effector cells of the immune system. TNF-α has been shown to have a role in several inflammatory diseases including CD, UC, rheumatoid arthritis and ankylosing spondylitis. Anti-TNF-α agents bind to cell surface TNF-α and free TNF-α, and block their activity. Blocking of TNF-α with anti-TNF-α drugs has been shown to be successful for some patients with inflammatory diseases, including CD. Anti-TNF-α agents recommended by NICE for the treatment of CD are IFX and ADA. These monoclonal antibodies are introduced into the human body to bind and block TNF-α. They are classed as monoclonal antibodies because they are derived from genetically engineered immune cells, which are all daughters of a single parent cell, so that in culture they generate and secrete antibodies that are all of identical structure to and affinity for TNF-α. 15

Infliximab

Infliximab is a chimeric (mouse–human) monoclonal antibody. It is said to be chimeric because the genetic code determining its amino acid sequences is partly derived from the mouse genome and partly from the human genome. IFX belongs to the immunoglobulin gamma type 1 (IgG1) group of antibody molecules (Figure 2). It should be borne in mind that IgG1 molecules are globular (not linear as in the diagram) and that they are glycoproteins, which have carbohydrate chains attached (not shown in Figure 2). As IFX is generated from cultured mouse cells, the carbohydrate part of the molecule corresponds to that of mouse rather than human glycoproteins. 15

FIGURE 2.

Diagrammatic representation of the structure of an IgG1 antibody molecule. The molecule has two heavy chains and two light chains; the heavy chains are joined by disulphide bonds (S–S) and each light chain is joined to a heavy chain by S–S bonding. The light and heavy chains have a variable region (different from all other antibodies) at the amino (NH₂) end of the chain; these variable regions are responsible for binding antigen. The rest of the heavy and light chains are identical to other IgG1 antibodies and are called constant regions. The proteolytic enzymes papain and pepsin cut the molecule just above or below the S–S bonds holding the heavy chains together. When the split is below the heavy chain S–S bond, this generates a fragment crystallising (Fc) product and a fragment antigen-binding (Fab) product. When the split is above the heavy chain S–S bond, two antigen-binding fragments are formed [F(ab)₂]. COOH, carboxylic acid; HC, heavy chain; LC, light chain.

Infliximab is composed of human IgG1 heavy-chain constant regions and human kappa light-chain constant regions (together representing 70% of the genetic make-up of the molecule), plus mouse-derived heavy- and light-chain variable regions (30% of the genetic make-up, 4/12 domains), which carry the binding sites with high affinity and specificity to TNF-α (see Figure 2). IFX was the first anti-TNF-α agent that was approved and licensed for treating severe active CD and active fistulising CD in adults and children aged > 6 years. It is administered intravenously over 1–2 hours.

Side effects of IFX include:

• allergic reaction to the infusion (or IFX) apparent by:

  • hives (red, raised, itchy patches of skin) or other skin rashes

  • difficulty swallowing or breathing

  • pains in the chest or muscle, or joint pain, fever or chills

  • swelling of the face or hands

  • headaches or a sore throat.

• serious viral or bacterial infections including tuberculosis, especially in people aged > 65 years

• skin reactions including psoriasis (red scaly patches), rashes, skin lesions, ulcers and hives, and swollen face and lips

• worsening of heart problems

• increased risk of cancer or lymphoma

• liver inflammation.

Many of the side effects are reversible if the drug is stopped. 15

Primary non-response

Patients not achieving at least a 70-point reduction on the CDAI during induction therapy are classed as primary non-responders. Incidence rates of primary non-response vary greatly depending on the clinical outcome measured (response/remission) and on the time point that the assessment of response is undertaken. For example, A Crohn’s disease Clinical trial Evaluating infliximab in a New long-term Treatment regimen (ACCENT) I, a randomised controlled trial (RCT) that investigated the benefit of maintenance IFX therapy in 573 active CD patients, assessed response after a single IFX infusion at 2 weeks. 27 The ACCENT II study was a post hoc analysis to determine the efficacy and safety of IFX therapy in patients with fistulising CD in which patients were assessed at 10 weeks. 28 The Crohn’s trial of the fully Human Antibody adalimumab for Remission Maintenance RCT assessed response to ADA at 4 weeks to evaluate the drug’s efficacy and safety in the maintenance of response and remission in 854 patients with moderate to severe CD. 29 However, Ben-Horin and Chowers30 stated that in clinical practice non-response should not be assessed before 8–12 weeks, as remission might still be induced at this time. It is therefore not surprising that a review of incidence rates found that non-response ranged from 20% to 40% in clinical trials and from 10% to 20% in ‘real-life’ series. 30 In contrast, lack of remission at week 4 in patients with luminal CD was reported to be as high as 67% for IFX and 64% for ADA. 31,32 The true magnitude of the rate of primary non-response is therefore difficult to determine.

Factors associated with non-response are believed to include:30,33,34

• severity of disease

• duration of disease

• smoking

• drug elimination

• drug binding

• anti-drug antibodies

• alternative non-TNF-α-mediated disease pathways

• concomitant treatment with immunosuppressants

• prior failure of other anti-TNF-α.

Loss of response

Patients with an initial response to anti-TNF-α treatment can lose response at any time during induction or maintenance therapy despite intensification of treatment (i.e. increase in dose or decrease in dosing interval). Again, lack of a clear definition, assessment at different time points, different outcome measures and different drug doses mean that reported incidence rates of secondary LOR vary considerably across studies. The true extent of this problem is largely unknown. Gisbert and Panes,35 in their review, reported a range of LOR to IFX of 11–48% (mean 37%) for varying lengths of follow-up, and de Boer et al. 33 reported a range of 21–46% for LOR to ADA. For this reason the incidence of LOR is better expressed as the annual risk of LOR per patient-year (13% for IFX35 and 20.3% for ADA36). LOR to ADA and IFX did not differ significantly in a retrospective study of 375 patients; however, patients treated with ADA required more dose optimisation intervention than patients on IFX. 37

The following factors are believed to prevent LOR:14

• pre-medication with steroids

• concomitant immunosuppressants

• maintenance therapy as opposed to episodic treatment.

Mechanisms of LOR to anti-TNF-α agents are still unclear. The next section describes some of the possible mechanisms in more detail.

Enzyme-linked immunosorbent assays for infliximab and adalimumab

For details of the ELISA kits, refer to Appendix 1. All three specified ELISA methods employ similar principles in which, typically, microtitre plates with 96 wells coated with reagent receive the patient serum samples or various standards and calibrators. Reagents are added with wash steps between additions. The final step involves quantifying the amount of a peroxidase label in the titre well, this amount being proportional to the amount of anti-TNF-α or anti-drug antibody in the patient’s sample or in the calibrator standard. 15

The amount of peroxidase present in the well is quantified using a timed incubation with excess substrates [hydrogen peroxide + 3,3’,5,5’-tetramethylbenzidine (3,3’,5,5’-tetramethylbiphenyl-4,4’-diamine)]. Peroxidase catalyses the following reaction:

The incubation is stopped after an appropriate time by the addition of acid and the accumulated chromogen quantified by measuring optical density with a spectrophotometer.

The reagents used for coating the microtitre plate wells and the reagents used in subsequent steps of the assay procedure differ in detail according to manufacturer. The LISA-TRACKER assays for IFX and for ADA are illustrated in Figure 3.

FIGURE 3.

Diagrammatic representation of the LISA-TRACKER assay for IFX and ADA. Procedural steps C and D are detection steps that function to detect the anti-TNF-α that is bound to the microtitre well surface via TNF-α, ensuring a quantitative relationship between anti-TNF-α and peroxidase. Step E quantifies the amount of peroxidase (and, therefore, anti-TNF-α) in the microtitre well (streptavidin has four very high-affinity binding sites for biotin).

Serum samples from patients may contain soluble TNF-α receptors; these could compete with anti-TNF-α for the immobilised TNF-α on the microtitre well plate and may potentially interfere with the assay. The assay quantifies free anti-TNF-α. Samples may contain anti-TNF-α bound to antibodies to anti-TNF-α, especially in patients who have lost a response to treatment. These anti-TNF-α–antibody complexes will be washed away at the first wash step, leaving only free anti-TNF-α bound to immobilised TNF-α. The amount of anti-TNF-α lost at the wash step is likely to vary between patients and is unknown; the practical implications of this are uncertain. 15

TNF-α-Blocker and Promonitor assays for anti-TNF-α drugs differ from the LISA-TRACKER assay in that the well coat is not TNF-α, but rather a reagent (antibody or antibody fragment) able to bind specifically to the TNF-α binding site of IFX or of ADA that is added to the microtitre well in the patient’s sample (or calibrator). After washing, the second reagent is a peroxidase-labelled antibody able to bind the fragment crystallising (Fc) region of the anti-TNF-α antibody (Figure 4). Thus, fewer steps and a single reagent are used to detect well-bound anti-TNF-α drug. Table 1 summarises the information describing the mechanisms underlying these assays.

FIGURE 4.

Diagrammatic representation of TNF-α-Blocker and Promonitor assays for IFX and ADA. Procedural steps C and D are detection steps that function to detect the anti-TNF-α (e.g. drugs IFX or ADA) that has been bound to the well surface via the coating (diamond). This ensures a quantitative relationship between anti-TNF-α (IFX or ADA) in the sample and peroxidase.

Enzyme-linked immunosorbent assays for anti-drug antibodies

The LISA-TRACKER assays for antibodies to IFX and to ADA are illustrated in Figure 5.

FIGURE 5.

Diagrammatic representation of the LISA-TRACKER assay for antibodies to IFX or to ADA. This is a bridging assay in which antibodies to anti-TNF-α in the patient sample bridge the immobilised anti-TNF-α to the biotinylated anti-TNF-α. Procedural steps C and D are detection steps that function to detect the sample antibodies, ensuring a quantitative relationship between anti-TNF-α antibodies and peroxidase. Step E quantifies the amount of peroxidase (and, therefore, anti-TNF-α antibodies). (Streptavidin has four very high-affinity binding sites for biotin.)

This assay quantitatively estimates only free antibodies to anti-TNF-α. Therefore, anti-drug antibodies bound to the drug are lost at the first wash. The amount of bound anti-drug antibody is likely to vary between patients and is unknown. Whether anti-drug antibodies directed at non-idiotypic regions of the drugs (e.g. glycoprotein moieties, variable non-idiotypic mouse regions of IFX, etc.) are detectable or present in samples appears to be insufficiently investigated to date and is therefore uncertain. However, in vitro tests indicate that about 90% of anti-drug antibodies bind to the TNF-binding region of anti-TNF-α drugs. 51 These, and the other anti-drug antibodies, may hasten clearance of drug from the circulation as well as neutralising its binding capacity.

Tumour necrosis factor alpha and Promonitor assays differ from LISA-TRACKER assays in employing a single reagent for detecting well-bound anti-drug antibodies rather than two (biotinylated IFX or biotinylated ADA, plus avidin-conjugated peroxidase). Table 2 summarises the information describing the mechanisms underlying these assays. 15

Brief overview of identified assay methods

There are no gold standard assays for anti-TNF-α agents or for antibodies to anti-TNF-α agents that might provide a robust basis for comparisons between the performances of different assays. According to US Medical Insurance assessments ‘candidate’ assays have been insufficiently investigated to establish any as a gold standard and, according to Steenholdt,60 the evidence is incomplete on how these different assays may compare in practice. 14,61–64

There appear to be four types of assay which differ fundamentally from each other. These are as follows.

1. ELISAs: solid-phase assays. These are available as commercial kits and several in-house methods are mentioned in the literature. Generally, the ELISAs quantitatively measure only ‘free’ anti-TNF-α and ‘free’ anti-drug antibodies, and it is acknowledged that the level of the unmeasured ‘bound’ anti-TNF-α and of ‘bound’ anti-drug antibody may vary considerably between patients. Thus, for some patient samples there is an unknown and unmeasured amount of anti-TNF-α and of anti-drug antibody present, in addition to the measured ‘free’ levels. In theory, this represents a potential deficiency in ELISAs, although whether or not this is serious in practice is difficult to gauge, especially in the absence of an established gold standard. This deficiency appears to have been one stimulus for the development of methods based on alternative principles. It is possible, however, that the relative convenience and cheapness of ELISAs means that this inability to measure total anti-TNF-α and total anti-drug antibody is supportable in practice.

2. RIAs: liquid-phase assays. These are provided as a total service rather than as purchasable kits. They measure total anti-TNF-α and total anti-drug antibody (probably as long as the anti-drug antibody light chain is λ class). These RIAs use ¹²⁵I-labelled human TNF-α and ¹²⁵I-labelled anti-TNF-α. These are commercially available or may be relatively easily constructed from commercially available materials; however, in the absence of purchasable assay kits, it is unlikely that any hospital laboratory would set up such assays for routine use. In these assays the patient’s sample is mixed with a solution containing a fixed amount of ¹²⁵I-labelled TNF-α or ¹²⁵I-labelled anti-TNF-α further antibody (e.g. rabbit anti-human immunoglobulin λ-chain), which promotes the formation of immune complexes that are pelleted by centrifugation. The ¹²⁵I in the pellet is quantified in a gamma counter. Potential disadvantages include: (1) radiolabelled reagents do not keep indefinitely (¹²⁵I decays with a half-life of 59 days); (2) the laboratory needs to be equipped for handling hazardous (radioactive) materials; (3) some staff training may be necessary; and (4) the laboratory requires a gamma counter (preferably automated for high throughput). These factors obviously have cost implications for setting up RIAs.

3. Cell reporter assays: these assays utilise genetically engineered cells that respond to the presence of anti-TNF-α agents by synthesising light-generating enzymes. The enzymes are allowed to accumulate during an incubation period and are then supplied with appropriate substrates resulting in light emission measured with a luminometer. Samples with anti-TNF-α will lead to light emission and samples with antibodies to anti-TNF-α will quench light emission (for further information see Appendix 2).

4. Mobility shift assays: the mobility shift assay depends on detecting the shift in mobility of fluorescent probes when bound to either anti-TNF-α or anti-drug antibodies (for further information, see Appendix 2).

Timing and use of assays

The anti-TNF-α and anti-drug antibody assays are most frequently administered just before the next administration of the anti-TNF-α agent. This is said to allow measurement of a ‘trough’ level of anti-TNF-α and has been adopted to minimise effects from the presence of anti-TNF-α–anti-drug antibody immune complexes in samples. For patients whose response to therapy has waned, the results of the tests are frequently dichotomised using a cut-off assay result. Thus, on the basis of anti-TNF-α assays patients are classified as having therapeutic levels of anti-TNF-α or subtherapeutic levels, and on the basis of anti-drug antibody assay results they are classified as having clinically significant levels of anti-drug antibodies or insignificant levels. Such classifications yield four categories of patient for whom different explanations of failed response are possible. Algorithms have been developed prescribing treatment pathways and/or further diagnostic tests (e.g. colonoscopy) based on such classification. 15

Search strategies for clinical effectiveness

An iterative procedure was used to develop the initial MEDLINE search, with reference to our own scoping searches and those undertaken by information specialists at NICE. Known articles were consulted and checked for relevant terms. Additional phrases were added to find relevant articles that did not include terms for the test name or type of test (e.g. Baert et al. 43) or population (e.g. Vande Casteele et al. 67) in title, abstract or indexing. This search developed for MEDLINE was adapted as appropriate for other databases and sources. The searches for each source are provided in Appendix 3. Searches for studies for cost and QoL were developed separately.

The search strategy comprised the following main elements:

• searching of electronic bibliographic databases

• contact with experts in the field

• scrutiny of references of included studies

• screening of manufacturers’ and other relevant organisations’ websites for relevant publications.

The following bibliographic databases were searched from inception to the date of searching: MEDLINE; MEDLINE In-Process & Other Non-Indexed Citations; EMBASE; The Cochrane Library (including Cochrane Systematic Reviews, Database of Abstracts of Reviews of Effects, Cochrane Central Register of Controlled Trials, NHS Economic Evaluation Database and HTA databases); Science Citation Index and Conference Proceedings (Web of Science); Index to Theses; Digital Access to Research Theses-Europe; Dissertations & Theses; National Institute for Health Research (NIHR) HTA programme; and PROSPERO (International Prospective Register of Systematic Reviews). Searches were undertaken during the period October to November 2014 (see Appendix 3 for exact search dates).

The following trial and patent databases were also searched: Current Controlled Trials, ClinicalTrials.gov, UK Clinical Research Network Portfolio Database, the World Health Organization’s International Clinical Trials Registry Platform and Espacenet (European Patent Office).

Specific conference proceedings, selected with input from clinical experts and specialist committee members, were also checked from January 2010 to January 2015:

• European Crohn’s and Colitis Organisation

• Digestive Diseases Week (meeting of the American Gastroenterology Association)

• British Society of Gastroenterology

• United European Gastroenterology Week

• American College of Gastroenterology.

The following online resources of various health services research agencies, regulatory bodies, professional societies and manufacturers were consulted via the internet:

• International Network of Agencies for HTA publication – www.inahta.org/

• US Food and Drug Administration medical devices – www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/Databases/default.htm

• European Commission medical devices – http://ec.europa.eu/health/medical-devices/

• Theradiag – www.theradiag.com/en/

• Immundiagnostik AG – www.immundiagnostik.com/en

• Proteomika – www.proteomika.com/

• American College of Gastroenterology – http://gi.org/

• European Crohn’s and Colitis Organisation – www.ecco-ibd.eu

• British Society of Gastroenterology – www.bsg.org.uk

• United European Gastroenterology – www.ueg.eu/

• The American Gastroenterology Association – www.gastro.org.

The reference lists of included studies and relevant review articles were checked. Citation searches of selected included studies were undertaken using Scopus. Identified references were downloaded in EndNote X7 software (Thomson Reuters, CA, USA). Included papers were checked for errata using PubMed.

Inclusion and exclusion of relevant studies

During the initial inclusion/exclusion process we identified three different categories of studies which were of interest for the review:

1. studies comparing the performance of different types of assays (assay type comparison studies addressing objective A)

2. studies reporting an algorithm for the management of patients with drug and/or anti-drug antibody level test results (management studies addressing objectives B and C1)

3. studies reporting the correlation of drug and/or anti-drug antibody levels with a patient’s clinical state (response) prospectively or retrospectively (correlation studies).

The third category was included in addition to the original protocol objectives. The reason behind this was the fact that we expected to find only a limited number of management studies to answer the decision questions. The correlation studies were included for two purposes:

1. to provide an overview of the variation in drug-level thresholds used to predict clinical state (for objective A – review of comparative performance of tests)

2. to pool test outcome data for responders and non-responders as an alternative to single-study data to inform the economic model (objective C2 – analysis of correlation between test results and clinical outcomes).

Assay-type comparison studies were considered in two phases because the work involved in objective A (review of comparative performance of tests) was dependent on the available evidence in objective C1. In the first phase, studies were included if they compared assay performance of two or more different test assays (see Objective A: review of comparative performance of test assays measuring anti-tumour necrosis factor alpha and/or anti-drug antibody levels).

Once the management studies to be included were known, the second phase included comparison studies if they compared two or more types of intervention assay or any of the intervention assays with the assays used in the management studies in order to inform a linked evidence approach.

See below for detailed inclusion and exclusion criteria for the different objectives.

Objective A

For objective A, quality appraisal was completed using a modified Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) checklist. 69 In the patient selection domain three questions were included, using the standard version of the tool:

1. Was a consecutive or random sample of patients enrolled?

2. Was a case–control design avoided?

3. Did the study avoid inappropriate exclusions?

An additional question asking for the range of drug and antibody concentrations was added before a judgement was made about applicability. The applicability question was adapted to:

1. Is there concern that the included patients or range of drug/antibody concentrations do not match the review question?

Regarding the index test, two standard questions were included to assess risk of bias:

1. Was the threshold pre-specified?

2. Were index tests interpreted without knowledge of the reference standard?

One additional question was added:

1. Were the number of failed results and measurement repeats reported?

For the reference standard, the two standard questions were used:

1. Were the reference standard results interpreted without knowledge of the results of the index test?

2. Is the comparison test likely to correctly classify the target condition?

The best reference standard for test accuracy to use would be use of standardised spiked samples. Use of spiked samples as a reference standard would allow the accuracy of tests to be compared with reference to the true drug and antibody levels, and would avoid the biases associated with imperfect reference standards. However, with spiked samples test accuracy may not be reflective of test accuracy in clinical practice.

However, when spiked samples were unavailable, one of the comparator tests used in the management studies and considered for a linked evidence approach was used. If the reference standard was one of the four comparator tests then it was classified as unlikely to correctly classify the target condition. This is because, as a result of the lack of evidence, they are an imperfect reference standard. For both comparator and index tests, judgements regarding applicability considered both the test used and the threshold applied.

For flow and timing, the following standard questions were included:

• Was there an appropriate interval between intervention test and comparison test(s)?

• Did patients receive the same reference standard?

• Were all patients included in the analysis?

An additional question was included:

• Were both intervention test and reference standard conducted on all samples?

This is to measure whether or not some patients or samples did not receive any of the index tests. This is of particular concern if the reason for being omitted may be related to the probability of a positive or negative result.

Objective C1: review of clinical effectiveness of test with algorithm combinations

Randomised controlled trials meeting the inclusion criteria were assessed using the Cochrane risk-of-bias tool. 70 The Downs and Black checklist71 was used to assess the quality of non-RCTs meeting the inclusion criteria. The results of the quality assessment provide an overall description of the quality of the included studies and a transparent method of recommendation for design of future studies. Quality assessment was undertaken by one reviewer and checked by a second reviewer, any disagreements were resolved by a third reviewer through discussion.

Objective A: review of comparative performance of tests

We mapped included studies according to the comparisons they undertook. A narrative was produced to summarise the studies that compared the performance of the intervention assays and assays suitable for a linked evidence approach and considering the concordance between the tests. This was assessed using the following outcomes:

1. concordance between tests (split by positive reference standard results and negative reference standard results, or clinical outcomes when available) for therapeutic drug and detectable anti-drug for all index tests and comparators

2. characteristics of cases in which there was disagreement and agreement between tests

3. Bland–Altman plots to show patterns of correlation.

The specific measures of concordance used were percentage agreement between the tests (split between positive reference standard sample results and negative reference standard sample results, when available) and Cohen’s kappa. Two main secondary outcomes were also collected. First, the characteristics of cases in which there was disagreement and agreement between tests may provide information about the reason for and implications of the discordant results. Second, the shape of the Bland–Altman plots shows whether or not the difference between the two tests is dependent on absolute drug and anti-drug levels. Mean bias and the upper and lower limits of agreement were not particularly informative here, as we are interested in only one cut-off point not the whole range of concentrations. Pearson’s correlation coefficient was not considered in detail, as it can have high values even when clinically meaningful differences are present. 72 When there were sufficient studies, a meta-analysis of Cohen’s kappa was considered. 72

Objective B: description of algorithms prescribing patient management following test outcomes for drug and/or anti-drug antibody levels

Algorithms used in management studies were described narratively and compared with the algorithm adapted from Scott and Lichtenstein66 (for LOR) and to the algorithm adapted from Vande Casteele et al. 73 (for responders). Patients or decisions non-compliant with the stated algorithm were quantified.

Time of testing, sequence of testing (drug and antibodies) and sequence of analysis were also considered.

Objective C1: clinical studies evaluating drug monitoring for the management of Crohn’s disease patients (management studies)

Depending on the available evidence, analyses were stratified according to the type of ELISA or other assay, type of drug (IFX or ADA) and patient group (patients with LOR or responders).

Study, treatment, population and outcome characteristics were summarised and compared qualitatively and, when possible, quantitatively in text, graphically and in evidence tables. Pooling study results by meta-analysis was considered; however, meta-analysis was unsuitable for the data identified and we employed a narrative synthesis using text and tables. A detailed commentary on the major methodological problems and biases affecting the studies was also included, together with a description of how this may have influenced individual study results.

We used a linked evidence approach. 1 Evidence on outcomes reported by studies using other test methods (RIA, liquid-phase mobility shift assay and in-house ELISAs) for patient management was linked to evidence on comparative test performance between our intervention tests and these other methods to allow for estimates of anticipated outcome for our intervention assays.

Time of testing, sequence of testing (drug and antibodies) and sequence of analysis were also considered.

When relevant, Kaplan–Meier plots were available, individual patient data (IPD) were reconstructed using the method of Guyot et al. 74 Parametric models were fitted to reconstructed IPD using Stata version 11 (StataCorp LP, College Station, TX, USA).

Objective C2: studies relating test results to clinical state of patients (correlation studies)

For objective C2 we aimed to:

1. Provide an overview of meta-analyses of studies addressing the relationship between drug and/or anti-drug antibody levels and clinical state of patients with CD by producing a narrative of identified systematic reviews with meta-analyses, presenting the reported meta-analyses results, and undertaking a hierarchical meta-analysis of the data presented in the systematic reviews

2. Pool test accuracy data for prediction of patients’ clinical state (response or lost response). This was done as a potentially useful supplement to management studies for informing the economic model.

Studies that provided dichotomised test results and related these to dichotomised clinical status were identified. In particular, studies were sought that reported on both drug and anti-drug antibody test results for individual patients. Two-by-two data for tests were extracted, together with the type of test employed [e.g. ELISA, RIA, homogeneous mobility shift assay (HMSA)], the anti-TNF-α drug administered, dose regimen, patient inclusion and exclusion criteria, timing of testing, method for establishing clinical status, test cut-off point used and study design when these were reported. The populations of interest were (1) responders and responders who lost response; and (2) patients with LOR who continued with LOR or who regained a response.

Meta-analyses of single-test studies (i.e. measuring anti-TNF-α or anti-drug antibodies) were undertaken (1) to provide a pooled estimate for the probability of returning a specified test result after trough anti-TNF-α testing (useful for estimating reflex strategy test result probabilities); and (2) to provide pooled estimates for the probability of returning a specified test results by single test (i.e. anti-TNF-α or anti-drug antibodies) that can be compared for consistency with the corresponding probabilities from the identified patient-level studies.

RevMan version 5.3 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) was used for analysis of sensitivities and specificities. Meta-analysis was undertaken in Stata version 11 using the metandi package. 75,76 The prevalence of clinical status was meta-analysed using a random-effects model with ‘MetaAnalyst’ software (Tufts University, Medford, MA, USA).

Given the prevalence (P) of the condition tested for, and the joint sensitivity (Sens) and specificity (Spec) values from meta-analysis, the probability of returning a positive test result is:

And the probability of returning a negative test result is:

Aim

To compare the performance of the different index tests (three specified ELISA kits) to one another, and to comparator tests that can be used to perform a linked evidence approach, in order to answer the question:

• Do the index tests agree with each other and with the comparator tests with regard to whether or not therapeutic levels of drug and detectable levels of anti-drug antibodies are present and, therefore, will using the tests lead to the same clinical decisions? Comparator tests here are tests with known links to improving patient outcomes from prospective studies with pre-specified algorithms (management studies).

Rationale

In a typical linked evidence approach, test accuracy studies detect cases of disease, and are linked to studies that show evidence of treatment effectiveness in cases detected. The test accuracy studies in this review produce four results:

1. drug positive and anti-drug antibody positive

2. drug positive and anti-drug antibody negative

3. drug negative and anti-drug antibody positive

4. drug negative and anti-drug antibody negative.

However, no trials were found that used the specified index tests to direct treatments for these four patient groups.

The linked evidence approach we therefore used was to evaluate the evidence showing whether or not any comparator tests (drug and anti-drug antibody tests) used in patients with CD improve outcomes (typically a test–treat type of trial), and to assess the accuracy of the index tests versus these comparator tests. These comparator tests form an imperfect reference standard and a simple calculation of sensitivity and specificity of the ELISA index tests against these tests (e.g. using HMSA or RIA) as reference standards might result in either over- or underestimation because of imperfect reference standard bias.

We took studies which investigated if testing for drug and anti-drug antibodies can improve patient outcomes through choosing the treatment prescribed by an algorithm (i.e. in a full RCT), and linked this to the index tests. This method may work satisfactorily if the RCT is of good quality and if there is good evidence for high concordance between the index and (imperfect) reference or comparator tests. When there are discordant results between tests, we do not know which is correct. Alternatively, for spiked samples, we know which test is correct, but not whether or not this would have any impact on clinical outcomes. Therefore, the main outcome for objective A was evaluation of the concordance between the tests. This approach is appropriate for interpreting and synthesising data when the reference standard is imperfect. 72

Results of assay type comparison studies

The search identified 25 relevant studies (reported in 26 references) that compared two or more assays to measure anti-TNF-α and/or anti-drug antibody levels in patients with CD (Figure 7). Of these, 10 were full texts (reported in 11 references)67,86,99,100,103,116,121–123,126,131 and the remainder were conference abstracts, including one unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication). 81,87,89–91,96,97,109,117,118,125,127,129,130 Of the 25 studies, 11 were not further considered as they compared assays other than the intervention assays or as comparators for the linked evidence approach (see Appendix 8). 81,86,96,99,100,103,109,118,121,125,127 Of the remaining 14 studies (15 references), which undertook relevant comparisons [including one unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]67,87,89–91,97,116,117,122,123,126,129–131 (Figure 8), only five (six references) [including one unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]67,87,117,122,123 reported concordance as numerical data or as Cohen’s kappa (see Figure 14). In addition, Proteomika provided information in the form of a benchmark analysis which is commercial in confidence (and, therefore, redacted from the text).

FIGURE 7.

Summary of all of the tests for which there were comparisons in the identified literature. The index tests are shaded green and comparator tests are shaded blue. The comparisons are denoted as (i) for IFX, (a) for ADA, (i*) for anti-drug antibodies to IFX and (a*) for anti-drug antibodies to ADA. Studies include Bodini 2014;81 Corstjens 2013;86 Daperno 2013;87 Egea-Pujol 2013;89 Eser 2013 in Gastroentrology;90 Eser 2013 in J Crohns Colitis;91 Greathead 2014;96 Hauenstein 2012;97 ^aImaeda 2014;100 Imaeda 2014;99 Kopylov 2012;103 McTigue 2013;109 Nagore 2015 (Dr Daniel Nagore, Progenika Biopharma; 2015, personal communication); Ruiz-Arguello 2013;116 Schatz 2013;117 Semmler 2013;118 Steenholdt 2014;122,123 Steenholdt 2013;121 Ungar 2014;125 Vande Casteele 2012;67 Vande Casteele 2013;126 Vande Casteele 2014;127 Wang 2010;129 Wang 2011;130 and Wang 2012. 131 ADAbs, anti-drug antibodies; AHLC, antihuman lambda chain antibody; AJG, American Journal of Gastroenterology; Gastro, Gastroenterology; JCC, Journal of Crohn’s and Colitis; Ste, Steenholdt; TDM, Therapeutic Drug Monitoring; VC, Vande Casteele.

FIGURE 8.

Comparisons that linked the index tests and comparator tests to each other. The index tests are shaded green and comparator tests shaded blue. The comparisons are denoted (i) for IFX, (a) for ADA, (i*) for anti-drug antibodies to IFX and (a*) for anti-drug antibodies to ADA. Studies include Daperno 2013;87 Egea-Pujol 2013;89 Eser 2013 in Gastroentrology;90 Eser 2013 in J Crohns Colitis;91 Hauenstein 2012;97 Nagore 2015 (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication); Ruiz-Arguello 2013;116 Schatz 2013;117 Steenholdt 2014;122,123 Vande Casteele 2012;67 Vande Casteele 2013;126 Wang 2010;129 Wang 2011;130 and Wang 2012. 131 AJG, American Journal of Gastroenterology; Gastro, Gastroenterology; JCC, Journal of Crohn’s and Colitis; Ste, Steenholdt; VC, Vande Casteele. 131

Four comparator tests were identified from the literature linking use of the test to clinical outcomes: these were the RIA,123 the Leuven in-house ELISA,73 the PROMETHEUS ELISA128 and the PROMETHEUS HMSA. 128 All of the test comparisons identified in the search are detailed in Figure 7. The index tests are shown in light green, and those tests with some literature linking use of the test to clinical outcomes are marked in dark green.

Only those studies that compared performance between the different index tests, or between the index and comparator tests, were considered further, as shown in Figure 8. Four comparators were identified, with evidence linking the use of the test to clinical outcomes, as described in Objective C1: clinical studies evaluating drug monitoring for the management of Crohn’s disease patients (management studies). Briefly, use of the RIA was linked to outcomes in a test–treat trial123 and use of the PROMETHEUS ELISA and HMSA were linked to outcomes in a retrospective observational cohort. 128 The latter trial design is not randomised and is also subject to biases to the extent that we considered a linked evidence approach may be inappropriate. Nonetheless, we included these two tests in this section for comparative purposes. An in-house ELISA from Leuven was linked to outcomes in a RCT. 73

Comparisons between the index tests

Results are presented here for all included studies, as outlined in Figure 8. This includes four full papers (five references),67,122,123,126,131 as outlined in the quality assessment in Quality appraisal, and 10 abstracts [including one unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]. 87,89–91,97,116,117,129,130

Adalimumab

(Confidential information has been removed.)

One unpublished abstract that was provided by Proteomika [unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)] compared Promonitor assays with LISA-TRACKER assays for ADA. In this abstract, 40 samples were used from an unspecified number of patients with IBD and an unspecified number of spiked samples. The spiked samples may be the same as described in data provided to us from the manufacturer. For ADA, drug levels were different between the different assays: 6.0 [standard error of mean (SEM) 0.55] for Promonitor assays and 4.9 (SEM 0.39) for LISA-TRACKER assays. Pearson’s R² was 0.83 and the authors concluded from the spiked samples that LISA-TRACKER assays underestimated ADA levels. In addition, 10% of samples were above the upper limit of quantification for LISA-TRACKER assays, and not for the Promonitor ELISA.

In summary, LISA-TRACKER assays may underestimate ADA drug levels and this underestimation will be greatest at higher absolute drug levels. The impact this would have on performance at a set threshold is unclear.

Antibodies to ADA

The same study that reported relationships between the index tests for ADA also reported some information on antibodies to ADA [unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]. They reported a Cohen’s kappa of 0.8 between Promonitor and LISA-TRACKER assays for antibodies to ADA, but it is unclear how many samples were included in this comparison. (Confidential information has been removed.)

In summary, we have one abstract giving a Cohen’s kappa of 0.8 between LISA-TRACKER assays and Promonitor assays, in tests for antibodies to ADA, but it is not known how many samples, and of which type, were included in this analysis.

Infliximab

(Confidential information has been removed.)

There was one abstract comparing Promonitor to LISA-TRACKER assays for IFX [unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]. In this abstract, 69 samples from an unspecified number of patients with IBD and an unspecified number of spiked samples were used. IFX drug levels were different between the different assays: 2.2 (SEM 0.24) for Promonitor and 3.4 (SEM 0.36) for LISA-TRACKER assays, for which the Pearson R² was 0.98 and the authors concluded from the spiked samples that LISA-TRACKER assays overestimated IFX levels. In addition, 23% of samples were above the upper limit of quantification for LISA-TRACKER assays and not for the Promonitor ELISA.

In one abstract, Daperno et al. 87 compared Immundiagnostik with Promonitor for IFX drug levels. In this study, Daperno et al. 87 enrolled a consecutive series of 66 patients (39 CD and 27 UC) undergoing regular IFX dosing by intravenous therapy. It is unclear if additional samples were included. Bland–Altman plots of IFX drug levels showed mean bias of –1.8 µg/ml, indicating that Immundiagnostik estimates were on average lower than those of Promonitor by 1.8 µg/ml, and upper and lower limits of agreement of –10.8 and 7.1 µg/ml, respectively, indicating that 95% of IFX drug levels measured by Promonitor were between 10.8 µg/ml lower and 7.1 µg/ml higher than the same sample scores using the Immundiagnostik test (Figure 9).

FIGURE 9.

Reconstructed Bland–Altman plot comparing Promonitor IFX kits and Immundiagnostik TNF-α-Blocker ELISA kits. Based on data from Daperno et al. 87

In summary, LISA-TRACKER assays showed most variation in results in comparison to spiked samples, with levels of bias dependent on absolute drug levels, so performance at a set threshold cannot be inferred. (Confidential information has been removed.)

Antibodies to infliximab

(Confidential information has been removed.)

One abstract described a comparison of Promonitor and LISA-TRACKER assays for antibodies to IFX [unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]. Cohen’s kappa was 1.0, indicating complete agreement between the two tests, but it is unclear how many samples were included in this comparison. In the case of anti-drug measurements, 75% of samples had to be retested with LISA-TRACKER assays as the original measurements were above the upper limits of the measurement range.

One abstract, by Daperno et al. ,87 compared Immundiagnostik with Promonitor for measurement of antibodies to IFX. Daperno et al. 87 enrolled a consecutive series of 66 patients (39 CD and 27 UC) undergoing regular IFX dosing by intravenous therapy. The two tests showed identical results in just 6 out of 63 cases included in the analysis. It is unclear what is meant in the abstract by ‘identical’ results, but the low proportion that were identical may provide some indication that the two tests for antibodies to IFX should not be considered equivalent.

In summary, one study found perfect agreement between Promonitor and LISA-TRACKER assays for antibodies to IFX and another study found few (6/63, i.e. < 10%) ‘identical’ results between Immundiagnostik and Promonitor.

Comparisons between index tests and comparator tests

Here we outline the studies linking the index tests to comparator tests, which have associated evidence linking use of the test to changes in clinical outcomes.

LISA-TRACKER assays

There are no studies linking LISA-TRACKER assays to any of the comparator tests for detecting ADA or antibodies to ADA.

There is one study by Vande Casteele et al. 67 linking the LISA-TRACKER assay to the Leuven in-house ELISA for IFX and antibodies to IFX. This same study also compares LISA-TRACKER with the Amsterdam Sanquin in-house ELISA and RIA. These tests from the Amsterdam group are not included as comparators, but form part of the linkage pathway between the other index tests and the comparator, Leuven in-house ELISA, and so their relationship to LISA-TRACKER assays is also included here for interest.

Vande Casteele et al. 67 used 62 plasma samples from departments of gastroenterology and rheumatology. Of these, 36 were clinical samples from patients; the remaining 26 samples consisted of 24 spiked samples: 10 samples spiked with IFX, 10 spiked with antibodies to IFX, one spiked with ADA, three spiked with antibodies to ADA and two healthy control samples. The results for these different types of sample are not fully reported separately, but parts are included. Four samples were removed as they were above the upper limit of quantification for the LISA-TRACKER assay. In detecting IFX, the LISA-TRACKER assay gave positive results for 11 samples that were negative when using either Amsterdam or Sanquin in-house ELISAs. Five of these were false-positive spiked samples that did not contain IFX but did contain antibodies to IFX (two samples) and antibodies to ADA (three samples). The remaining six samples were clinical so the true result is not known, but the authors report high levels of antibodies in these samples. The one sample spiked with ADA was a true negative for IFX for both LISA-TRACKER and Leuven assays, but a false positive for the Amsterdam in-house ELISA. The Bland–Altman plots show that, although the relationship between the Leuven and Amsterdam ELISA appears to be independent of absolute drug levels, as drug levels increase measurements using the LISA-TRACKER assay appear to increase slower than those using either Leuven or Amsterdam assays (Figure 10). This means that the levels of concordance between the LISA-TRACKER assay and the other assays will be dependent upon the particular threshold used. The Bland–Altman plot showed no pattern between the Leuven and the Amsterdam IFX tests, and from visual inspection the bias was near zero, with upper and lower limits of agreement between –10 and 10 mg/ml. The performance in detecting antibodies to IFX is less clear, with discordant results reported but not the type of sample in which these results were found. The Amsterdam in-house RIA detected antibodies to IFX in five samples when they were not detected by either of the other two assays, and both the Amsterdam and the LISA-TRACKER assays tests detected antibodies in three samples, which tested negative using the Leuven in-house ELISA. The thresholds used for drug and anti-drug levels, respectively, were 0.1 mg/l and 10 µg/l for LISA-TRACKER assays, 0.3 mg/l and 1 mg/l for Leuven and 0.002 mg/l and 12 arbitrary units/ml (1 arbitrary unit/ml equals approximately 10 µg/l) for Amsterdam. This higher threshold for the Leuven antibody ELISA may explain why fewer cases were detected.

FIGURE 10.

Reconstructed Bland–Altman plots of IFX levels (mg/l) comparing (a) Amsterdam in-house IFX ELISA and Leuven in-house IFX ELISA; (b) Amsterdam in-house IFX ELISA and LISA-TRACKER assays premium IFX kit; and (c) Leuven in-house IFX ELISA and LISA-TRACKER assays premium IFX kit. Based on data from Vande Casteele et al. 67

In summary, in one study, using a range of clinical and spiked samples, there is some evidence that LISA-TRACKER assays may give false-positive results for IFX in the presence of antibodies to IFX or ADA, whereas the Leuven and Amsterdam in-house ELISAs do not. There is also some evidence that the Amsterdam RIA is most likely to detect antibodies, followed by LISA-TRACKER assays and then the Leuven in-house ELISA, but whether or not these antibodies detected are true positives or false positives is unclear.

Promonitor

One letter, by Ruiz-Arguello et al. ,116 compared the Promonitor ELISA with the Amsterdam Sanquin ELISA for IFX and ADA, and the Amsterdam Sanquin RIA for both anti-drug antibodies. In addition, Vande Casteele et al. 67 then linked the Amsterdam Sanquin ELISA and RIA to the Leuven in-house ELISA, which is one of the comparator tests.

The study comparing the Promonitor assay to the Amsterdam Sanquin tests used 120 spiked samples in total, designed to cover concentrations in the clinically meaningful range, 30 samples over the range 0.001–8 µg/ml for IFX, 30 samples over the range 0.001–5 µg/ml for ADA and 30 samples over the range 1–5000 arbitrary units/ml for both anti-drug antibodies to IFX and ADA. 116 The study defines set cut-off points and describes both assays as having no false-positive results for drug levels. However, no details are given of the sensitivity of each assay at those set cut-off points or concordance between them. The analytical sensitivity, meaning the lowest level at which the drug/antibodies are detectable, was given. Analytical sensitivity of the Amsterdam Sanquin assay was slightly higher than that of Promonitor: 10–30 ng/ml for IFX and 2–20 ng/ml for ADA, respectively. Bland–Altman plots of each assay in comparison with the known spiked concentrations gave a mean bias of −0.467 ng/ml [standard deviation (SD) 1.027 ng/ml] and 0.066 ng/ml (SD 0.196 ng/ml) for IFX and −1.140 ng/ml (SD 2.713 ng/ml) and −0.159 ng/ml (SD 0.488 ng/ml) for the Amsterdam Sanquin and Promonitor tests for ADA, respectively. The plots are not provided, but the authors describe a systematic overestimation of the drug levels by the Amsterdam Sanquin ELISA that increases with increased drug levels, which would explain the greater CIs for the mean bias estimate for the Sanquin ELISA. The authors describe this overestimation as occurring at drug levels of > 2 µg/ml. For antibodies to IFX and ADA, only correlation coefficients and analytical sensitivity were reported. Analytical sensitivity of the Promonitor assay was higher than that of Amsterdam Sanquin RIA, at 4 ng/ml and 20 arbitrary units/ml for IFX and 2 ng/ml and 30 arbitrary units/ml for ADA, respectively. Therefore, there is some evidence using spiked assays that the Amsterdam Sanquin assay may overestimate drug levels at higher concentrations, whereas the Promonitor assay may not. In a letter of response, however, Rispens and van der Kleij136 describe an update to their testing procedure which may have corrected the overestimation.

The second link between the Amsterdam and Leuven tests has been described in detail in LISA-TRACKER assays. However, to recap in brief, for drug levels the Bland–Altman plot showed no pattern; therefore, the relationship between the two tests is not dependent on threshold and from visual inspection the bias was near zero, with upper and lower limits of agreements between –10 and 10 mg/ml. For anti-drug antibodies to IFX, the Amsterdam RIA detected more cases than the Leuven assay, with the actual veracity of these unclear.

In summary, although we have some information linking Promonitor to the Amsterdam Sanquin tests, and further information linking these to the Leuven ELISA, it is not in a format from which we can calculate the concordance between the tests at clinically relevant thresholds.

Immundiagnostik

There are no studies linking Immundiagnostik to any of the comparator tests for ADA or antibodies to ADA. In two abstracts, Eser et al. 90,91 compared the Immundiagnostik ELISA with the PROMETHEUS HMSA, and Schatz et al. 117 compared the Immundiagnostik ELISA with the Amsterdam Sanquin in-house tests (for IFX and antibodies to IFX), which are in turn compared with the Leuven in-house ELISAs by Vande Casteele et al. 67

The two abstracts by Eser et al. 90,91 comparing Immundiagnostik ELISA with the PROMETHEUS HMSA method used samples from 90 patients (66 CD and 24 UC). The authors report that HMSA was able to detect anti-drug antibodies to IFX at the mid-infusion point, whereas ELISA returned inconclusive results because of interference from IFX. However, no numerical data were presented comparing the two methods so few, if any, conclusions can be drawn from the study.

The study by Schatz et al. 117 linking Immundiagnostik ELISA to the Amsterdam in-house tests (ELISA for drug levels, RIA for antibody levels) compared performance of the two tests for IFX and anti-drug antibodies to IFX. They used serum samples from 202 paediatric patients, of whom 125 had been exposed to IFX and 77 were IFX naive. Samples were considered positive for IFX if they were above the limit of detectability, which was < 0.8 µg/ml for the Immundiagnostik ELISA and < 0.002 µg/ml for Amsterdam in-house ELISA. Overall agreement using Cohen’s kappa was 0.792. Considering only the IFX-exposed patients, 25 were below the lower limit of detectability for both tests, leaving 87 who tested positive for both, 11 who tested positive only using the Amsterdam ELISA (measurements ranged from 0.1 to 2.3 µg/ml, so some of these will be below the lower limit of detectability using the Immundiagnostik test) and two whose results were not reported. For anti-drug antibodies to IFX, 88 samples were concordant positive, 27 samples were concordant negative and 10 were detected only by the Amsterdam RIA and not the Immundiagnostik ELISA.

The second link between the Amsterdam and Leuven tests has been described in detail in the previous two sections (i.e. LISA-TRACKER assays and Promonitor). For drug levels, the Bland–Altman plot showed no pattern, so the relationship between the two tests is not dependent on threshold, and from visual inspection the bias was near zero, with upper and lower limits of agreements between –10 and 10 mg/ml; for anti-drug antibodies to IFX the Amsterdam RIA detected more cases than the Leuven assay, with actual veracity of these unclear.

In summary, although there are good data linking Immundiagnostik with the Amsterdam in-house ELISA, with agreement for 114 out of 125 samples for IFX and 115 out of 125 samples for anti-drug antibodies to IFX, the link to the Leuven ELISA is not known in terms of agreement of the two tests at a clinically relevant threshold.

Relationship between different comparator tests

One study, by Steenholdt et al. ,122 compared the performance of the Biomonitor RIA used in the test–treat trial with PROMETHEUS HMSA and PROMETHEUS ELISA for IFX and antibodies to IFX. Vande Casteele et al. 126 compared the Leuven in-house ELISA with PROMETHEUS HMSA. One full study, by Wang et al. ,131 and four further abstracts89,97,129,130 compared the performance of HMSA with PROMETHEUS ELISA for IFX and antibodies to IFX, Egea-Pujol et al. 89 also made the same comparison for ADA and antibodies to ADA. However, three of these abstracts (Hauenstein et al. ,97 Egea-Pujol et al. 89 and Wang et al. 130) did not provide data on concordance, Cohen’s kappa or numbers of false-positive, true-positive, false-negative or true-negative test results, and will not be described further here.

The studies by Wang et al. 129,131 compared the performance of HMSA and of PROMETHEUS ELISA. Wang et al. 129 described 20 patients with IBD who had relapsed from treatment with IFX. ELISA detected IFX in 15 out of 20 patients and anti-drug antibodies to IFX in 15 out of 20 patients. HMSA detected IFX in 15 out of 20 patients and anti-drug antibodies to IFX in 18 out of 20 patients. It is not clear if the samples that tested positive on both tests were from the same 15 patients. The focus of Wang et al. 131 was to validate the performance of HMSA rather than to compare it with ELISA. Of 100 samples from healthy control participants, three were false-positive for antibodies to IFX for HMSA. This was to be expected as the cut-off point was determined from the same samples as mean plus two SDs. Repeat measurements of these three resulted in them being below the cut-off point, presumably regression to the mean. ELISA results for the 100 healthy control participants were not reported. Out of 100 patients with IBD selected as positive for antibodies for IFX on ELISA, five did not test positive on HMSA. The authors attribute this to elevated levels of non-specific binding in the ELISA. As we do not have the equivalent data for ELISA results on samples that tested positive using HMSA, it is difficult to draw any conclusions at all. The only comparative data given constitute a plot of correlation that does not appear to show high correlation. The studies therefore did not provide useful concordance data for evaluation.

Vande Casteele et al. 126 compared the Leuven in-house ELISA to PROMETHEUS HMSA. Although the paper does describe some discordant results, the focus of the paper is on outcomes in patients with differing results rather than on comparisons between the two tests. HMSA appears to perform better at detecting antibodies to IFX in the presence of IFX; however, quantifying this is difficult, as reporting focused on other research questions. It is described in the discussion as the HMSA having detected the median 9 weeks earlier. However, in the absence of IFX and with the HMSA cut-off point for antibodies to IFX set at 7.95 U/ml, the Leuven in-house ELISA detected four more cases with antibodies to IFX. The authors report that PROMETHEUS has since lowered the threshold to 3.13 U/ml.

Steenholdt et al. 122 took 66 frozen patient samples (from patients with CD with LOR to IFX) from the test–treat trial123 and reanalysed them using PROMETHEUS HMSA and PROMETHEUS ELISA. In all of these patients, the results of RIA using the samples before freezing were available and determined the treatment pathway at the time of the study. Threshold for positivity for IFX was unclear, but the lower limit of quantification was clearly defined as ≥ 0.15 µg/ml, ≥ 1 µg/ml and ≥ 1.4 µg/ml for RIA, HMSA and ELISA, respectively. 122 However, in the original paper reporting the RCT,123 although the same lower limit of quantification thresholds were given, additional cut-off points of ≥ 0.5 µg/ml and ≥ 3 µg/ml were given for defining therapeutic drug levels for RIA and HMSA, respectively. However, no additional cut-off point was given for ELISA. Thresholds for anti-drug antibodies to IFX were ≥ 10 arbitrary units/ml, ≥ 3.13 arbitrary units/ml and ≥ 1.69 µg/ml for RIA, HMSA and ELISA, respectively. Using RIA, 54 out of 66 (82%) tested positive for IFX, in comparison with 58 out of 66 (88%) for HMSA and 50 out of 66 (76%) using ELISA. The concordance between the three tests is shown in Figure 11. In eight patients IFX was undetectable using all three tests and in 50 patients IFX was detectable using all three tests. In four patients IFX was detected by RIA and HMSA but not by ELISA, and in a further four patients IFX was detectable only by HMSA.

FIGURE 11.

Concordance between RIA, HMSA and ELISA for detecting IFX in 66 patients with CD with LOR to IFX. In eight patients IFX was undetectable using all three tests.

Using RIA, 18 out of 66 (27%) patients tested positive for anti-drug antibodies to IFX, in comparison with 22 out of 66 (33%) using HMSA and 6 out of 66 (9%) using ELISA. The concordance between the three tests is shown in Figure 12. In 43 patients, anti-drug antibodies to IFX were undetectable using all three tests, and in six patients anti-drug antibodies to IFX were detectable using all three tests. In 11 patients anti-drug antibodies to IFX were detected by RIA and HMSA but not by ELISA, in a further five patients anti-drug antibodies to IFX were detectable only by HMSA and in one patient anti-drug antibodies to IFX were detectable only by RIA.

FIGURE 12.

Concordance between RIA, HMSA and ELISA for detecting anti-drug antibodies to IFX in 66 patients with CD with LOR to IFX. In 43 patients anti-drug antibodies were not detectable in any of the tests.

In summary, for IFX, RIA and HMSA agreed for 62 out of 66 patients, with the remaining four patients testing positive on HMSA but not RIA; HMSA and ELISA agreed in 58 out of 66 patients, with eight patients testing positive on HMSA but not ELISA; and RIA and ELISA agreed on 62 out of 66 patients, with the remaining four patients testing positive on RIA but not ELISA. The Bland–Altman plots comparing HMSA with RIA and ELISA with RIA showed a pattern with increasing drug concentration, meaning that these two comparisons are dependent on the absolute values for thresholds chosen (Figure 13). The relationship between HMSA and ELISA appears to be independent of absolute drug concentrations.

FIGURE 13.

Reconstructed Bland–Altman plots comparing PROMETHEUS ELISA, HMSA and RIA. (a) ELISA vs. HMSA; (b) RIA vs. HMSA; and (c) RIA vs. ELISA. Based on data from Steenholdt et al. 122

In the case of anti-drug antibodies to IFX, RIA and HMSA agreed for 60 out of 66 patients, with five patients testing positive on HMSA and not on RIA, and one testing positive on RIA and not on HMSA. HMSA and ELISA agreed in 50 out of 66 patients, with 16 patients testing positive on HMSA but not on ELISA; RIA and ELISA agreed in 54 out of 66 patients, with the remaining 12 patients testing positive on RIA but not on ELISA.

Therefore, there is an indication that RIA and HMSA detect more patients with IFX than does PROMETHEUS ELISA, and this effect is more pronounced with anti-drug antibodies to IFX. We do not know the true measurements for these patients.

Summary

Figure 14 summarises the studies that quantify the link between the index tests and the comparator assays using concordance data. In comparing the three index tests with one another, we have data from two abstracts [including one unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)]. 87

FIGURE 14.

Results of comparisons which linked the index tests and comparator tests to each other of studies reporting concordance data. The index tests are shaded green and comparator tests are shaded blue. Results are listed as either Cohen’s kappa or concordance levels displayed as a fraction for IFX (i), ADA (a), anti-drug antibodies to IFX (i*) and anti-drug antibodies to ADA (a*). Only studies which provide concordance at set threshold or Cohen’s kappa for the comparisons between tests are included Daperno 2013;87 Nagore 2015 (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication); Schatz 2013;117 Steenholdt 2014;122,123 and Vande Casteele 2012. 67 ADAbs, anti-drug antibodies; AJG, American Journal of Gastroenterology; Ste, Steenholdt; VC, Vande Casteele.

For anti-drug antibodies, one abstract [unpublished abstract provided by Proteomika (Dr Daniel Nagore, Progenika Biopharma, 2015, personal communication)] describes complete agreement between Promonitor and LISA-TRACKER assays for anti-drug antibodies to IFX and a Cohen’s kappa of 0.8 for anti-drug antibodies to ADA, but does not report how many samples were included. It also describes the upper limits of the measurement range for LISA-TRACKER assays as low. Daperno et al. 87 compared Immundiagnostik with Promonitor for anti-drug antibodies to IFX and found that the two tests showed identical results in only 6 out of 63 cases in a consecutive series of 66 patients (39 with CD and 27 with UC), but the definition of ‘identical’ was not given. It is not possible from these data to link the three index tests as part of a linked evidence approach.

For each index test we investigated all links to the comparator tests. There was one link to the Amsterdam Sanquin tests (ELISA for IFX, RIA for anti-drug antibodies to IFX),117 which showed agreement for 114 out of 125 samples for IFX and 115 out of 125 samples for anti-drug antibodies to IFX. However, there was only one study then linking the Amsterdam tests to the Leuven in-house ELISA,67 which reported disagreement in at least 8 out of 62 samples, with a lack of clarity regarding the remainder of results. Similarly, there was only one study linking LISA-TRACKER assays to the Leuven in-house ELISA67 and, although it reported disagreement for IFX in at least 11 out of 58 samples and for anti-drug antibodies to IFX in at least 3 out of 62 samples, the results for the remainder were unclear. However, we found no concordance data linking any of the index tests to any of the comparator tests at a clinically meaningful threshold.

In comparing the comparator tests with each other, there was one study122 that described reanalysing the same samples previously used in a test–treat trial. 123 There was agreement between PROMETHEUS ELISA and Biomonitor RIA for IFX in 62 out of 66 samples and for anti-drug antibodies to IFX in 54 out of 66 samples; agreement between PROMETHEUS ELISA and HMSA in 58 out of 66 samples for IFX and in 49 out of 66 samples for anti-drug antibodies to IFX; and, finally, agreement between RIA and HMSA in 62 out of 66 samples for IFX and in 60 out of 66 samples for anti-drug antibodies to IFX. We found no ongoing link to the index tests.

Overall, there was insufficient evidence linking any of the index tests (LISA-TRACKER, Immundiagnostik or Promonitor) to any of the comparators with links to clinical outcomes (HMSA, RIA, PROMETHEUS ELISA or Leuven in-house ELISA).

Results of threshold analysis studies

The search identified 24 studies48,58,77,82,84,85,93–95,100,101,104,105,107,108,110,112,113,115,119,120,126,132,135 that reported a ROC threshold analysis to determine optimal cut-off levels predictive of clinical response for IFX,58,77,82,85,93,101,104,105,110,112,113,119,120,126 ADA48,84,94,100,107,108,115,132,135 or both. 95 Table 8 summarises the studies in terms of the threshold reported, the diagnostic performance of using the threshold, the clinical marker used for assessment of response, the assay used and the value for the area under the curve (AUC). As the area under a ROC curve in this case quantifies the overall ability of the test to discriminate between those individuals who respond and those who do not respond, an AUC value > 0.5 indicates an informative test whereas an AUC of 0.5 represents an uninformative test without discriminatory power (sensitivity + specificity = 1). When considering the reported thresholds we need to bear in mind that response and LOR are poorly defined and studies using different definitions will measure different outcomes, which will have implications on the reported thresholds.

Aim

To provide a narrative description of algorithms used in studies that report clinical outcomes for patients whose treatment options were directed by a test-informed algorithm; and to compare these with related algorithms identified in the literature during scoping as relevant to the NHS. 66,73

Results

Studies and reviews reporting on test results and clinical status of patients have frequently proposed test-based treatment algorithms, but most of these have never been tested or implemented in patients with CD. 44,56,57,61,64,137–139 Here we describe the test-based algorithms that have actually been implemented in studies with patients with CD and briefly compare these with the most similar ‘precursor’ algorithms identified during scoping as relevant to the NHS. None of the algorithms described in this section was used in conjunction with one of the index tests.

Algorithms from management studies

No management studies of CD or IBD patients treated with ADA were found. Three IFX studies fulfilled inclusion criteria for this objective: (1) a RCT in patients with CD with LOR to IFX;123 (2) the TAXIT RCT73 of IBD patients responding to IFX; and (3) a retrospective observational study of IBD patients responding to IFX. 128

Table 9 summarises the algorithm used by Steenholdt et al. 123 Tests were done using commercially available RIAs. Both drug and anti-drug antibody tests are dichotomised so that concurrent testing classifies each patient into one of the four possible combinations of test results:

1. IFX negative and anti-drug antibody positive

2. IFX negative and anti-drug antibody negative

3. IFX positive and anti-drug antibody negative

4. IFX positive and anti-drug antibody positive.

TABLE 9 - Summary of the concurrent testing-based algorithm used by Steenholdt et al.123

s.c., subcutaneously.

Cut-off point	Detectable anti-IFX antibodies	Undetectable anti-IFX antibodies
Subtherapeutic IFX < 0.5 µg/ml	Group 1	Group 2
	Insufficient IFX bioavailability because of the induced immunogenicity of IFX	Insufficient IFX bioavailability because of non-immune-mediated pharmacokinetics of IFX
	↓	↓
	Change to different TNF-α-inhibitor: ADA 80 mg s.c. at inclusion followed by 40 mg s.c. every other week; dose intensification allowed	Intensify IFX treatment: IFX 5 mg/kg intravenously every 4 weeks
Therapeutic IFX ≥ 0.5 µg/ml	Group 4	Group 3
	Consider: pharmacodynamics non-functional anti-IFX antibodies false-positive test	Pharmacodynamics: inhibition of TNF-α is ineffective because of non-TNF-α-driven disease
	↓	↓
	Repeat IFX and anti-IFX antibody analyses and handle accordingly If unchanged results, then act as group 3	TNF-α-inhibitors not effective, so are discontinued. Review of clinical condition at discretion of the investigator: If relapse of CD, use drug(s) with other target (e.g. conventional immune suppressives, glucocorticoids, and/or other biological agents). Consider surgery if appropriate If no relapse, treat underlying problem

An important feature is the proposal of different causes for secondary treatment failure in groups 1–3; these proposals rest on interpretations of the supposed underlying mechanisms leading to the observed test results. The treatment options are prescriptive for two of the groups (1 and 2), but less so for group 3 (patients with LOR who have therapeutic levels of IFX and lack detectable anti-drug antibodies). Thus, the treatment received for group 3 requires further investigation and reflection by the treating clinician and may or may not include relatively expensive biological agents. As most patients fall into this group, these less prescriptive aspects add to uncertainty about treatment cost of the algorithm-based strategy and whether or not discretion relating to cost might play a part in decision-making for group 3 (an expensive biological may or may not be adopted because of perceived cost implications). Furthermore, treatments for this group may be difficult to replicate between different groups of clinicians who may be subject to different health pressures in relation to costs and/or to differing licensing regulations for biological therapies. Results for group 4 (positive test for both IFX and anti-drug antibody) are reviewed with suspicion and require retesting in case of error.

The TAXIT algorithm for patients responding to IFX is based on the hypothesis that an IFX trough level between 3 and 7 µg/ml is optimum for successful maintenance of clinical response; it proposes a strategy of prospective dose adjustment to achieve this target range; this likely requires trough tests before each infusion.

Figure 15 summarises the TAXIT study algorithm as described in the recently published paper. 73 Patients are categorised into four groups according to their trough IFX level: (1) undetectable, (2) low, (3) optimum (3–7 µg/ml) or (4) high (> 7 µg/ml). Group 1 is reflex tested for anti-IFX antibodies and further divided into two subgroups on the basis of anti-drug antibody test results: (1) in those with high anti-drug antibody levels, IFX therapy is stopped; (2) in those with anti-drug antibodies at a lower level (< 8 µg/ml), the dose of IFX is increased. In patients in group 2 (detectable but low trough levels of IFX; < 3 µg/ml) dosing interval is first reduced and then, if necessary, the dose is increased, in attempt to bring IFX trough concentrations within the ‘optimum’ range (3–7 µg/ml). In group 3, whose trough IFX levels are already in the optimum range, dose adjustment is not necessary. In patients in group 4 (high trough levels of IFX) dose interval is increased, followed, if required, by a dose reduction. In the trial, an ‘optimisation phase’ occurred during which the algorithm was implemented to bring patients into the optimum range, and this preceded randomisation. Only those patients already successfully optimised were randomised; thus, if the hypothesis is correct we would potentially expect poor generalisability with higher rates of successful maintenance in the trial than in a broader spectrum of responders.

FIGURE 15.

The TAXIT study algorithm presented in Vande Casteele et al. 73 Reprinted from Gastroenterology, Vol. 148, Vande Casteele N, Ferrante M, Van Assche G, Ballet V, Compernolle G, Van Steen K, et al, Trough concentrations of IFX guide dosing for patients with inflammatory bowel disease, pp. 1320–9.e3, Copyright 2015 with permission from Elsevier. ATI, antibodies to IFX; TLI, trough level of IFX.

Vaughn et al. 128 describe trough monitoring of IFX as a guide to dose adjustment in a group of retrospectively identified IBD patients. Tests were done with a commercial ELISA for the earliest-identified patients, whereas for those identified later the commercial HMSA method was used. Initially, dose adjustments aimed to bring trough IFX into the detectable range, but later the target range was changed to 5–10 µg/ml. The authors quote a typical dose adjustment for those with an undetectable trough IFX level to be an increase in dose to 7 mg/kg with a 6-week interval to the next infusion, followed by a return to 8-week infusion intervals. The authors state that for trough levels < 5 µg/ml, the dose of IFX was increased by ‘50 or 100 mg’. For patients with a trough level of IFX > 10 µg/ml on two testing occasions, the dose was decreased or, if the patient was already receiving 5 mg/kg (in the full paper this is given as ‘5 mg/ml’ and is an assumed typographical error) the infusion interval was increased. No dose adjustment was made for those in target range. This algorithm would be somewhat difficult to implement without assuming that the authors’ preferred policy is to use HMSA testing with a target range of 5–10 µg/ml and to manipulate dosage and dose intervals at the clinicians’ discretion so as to bring the patient into the target range.

Comparison of algorithms that are clinically relevant to the NHS which were identified during scoping

Scoping for this report identified two algorithms likely to be relevant for the NHS: one for patients with LOR for use with an unspecified anti-TNF-α (based on Scott and Lichtenstein,66 who specified IFX) and one for responders to IFX based on the public domain description of the TAXIT trial. The algorithm proposed by Scott and Lichtenstein66 requires concurrent testing (Figure 16).

FIGURE 16.

The Scott and Lichtenstein66 algorithm for patients with LOR. ATI, antibodies to IFX. Reproduced from Current Treatment Options in Gastroenterology, Therapeutic drug monitoring of anti-TNF therapy in inflammatory bowel disease, vol. 12, 2014, pp. 59–75, Scott FI, Lichtenstein GR66 (© Springer Science + Business Media, LLC 2014), with permission of Springer.

This algorithm is similar to that of Steenholdt et al. 123 (see Table 9) in categorising patients with LOR into four groups on the basis of dichotomised drug and anti-drug antibody test results. Drug trough levels are classified as therapeutic or low rather than therapeutic or subtherapeutic as in Steenholdt et al. 123 The suggested treatments for ‘anti-drug antibody positive | drug trough level low’, ‘anti-drug antibody negative | drug trough level therapeutic’, and ‘anti-drug antibody negative | drug trough level low’ groups are the same as those proposed by Steenholdt et al. ,123 namely treatment with an alternative anti-TNF-α drug, treatment with a non-TNF-α drug with a different mechanism of action and escalation of drug exposure, respectively. For the ‘anti-drug antibody positive | drug trough level therapeutic’ group, Scott and Lichtenstein66 make no recommendations, but Steenholdt et al. 123 recommend redeployment to an appropriate group after repeat testing. This is a ‘generalised’ algorithm and therefore differs in detail from that of Steenholdt et al. 123 in that cut-off levels for drug trough levels are not specified and therapies are less prescriptive.

Scoping identified the algorithm presented by Scott and Lichtenstein (Figure 17), which is similar to that of Steenholdt: low drug levels are termed ‘suboptimal’ and undetectable levels of anti-drug antibodies are termed ‘low or undetectable’. In patients with ‘suboptimal drug’ and ‘low or undetectable’ anti-drug antibodies it is recommended that adherence to treatment should be checked and that an immunosuppressant may be added. This version of the algorithm suggests that, in the case of the patients with ‘therapeutic drug’ levels and ‘low or undetectable anti-drug antibodies’, the ‘diagnosis’ of CD is confirmed.

FIGURE 17.

The precursor algorithm for LOR identified in scoping (based on Scott and Lichtenstein 201466). ADAb, anti-drug antibody.

The scope precursor version of the TAXIT trial algorithm is almost identical to the public domain version shown in Figure 18. The differences are (a) no specific trough drug levels are specified; and (b) the addition of an immunosuppressant is recommended for the group with undetectable trough drug levels and low anti-drug antibody levels.

FIGURE 18.

The precursor algorithm based on the TAXIT trial algorithm for IFX responders. ADAb, anti-drug antibody.

Aim

The aim of this section is to assess the evidence from studies that report the clinical impact of implementing a test-informed treatment algorithm for anti-TNF-α recipients with CD.

Results

After screening 2428 studies, we identified three which matched our inclusion criteria for management studies. All three investigated patients treated with IFX. There were no management studies for patients treated with ADA and none of the three studies used one of the index tests. The three studies that fulfilled our inclusion criteria (i.e. Steenholdt et al. ,123 Vaughn et al. 128 and Vande Casteele et al. 73) address several aspects of the decision questions. Other studies were identified (e.g. Afif et al. ,56 Pariente et al. ,59 Roblin et al. 57 and Paul et al. 58) that investigated the clinical utility of therapeutic drug monitoring of anti-TNF-α to predict response to a change in treatment mainly because of LOR to anti-TNF-α. On the basis of their results, the studies retrospectively suggested a test-informed treatment algorithm for IBD patients, but did not prospectively investigate implementation of a test-informed algorithm strategy for patient outcomes and compare that with a strategy that might be similar to standard care. For completeness, these studies have been summarised and their algorithms detailed in Appendix 9. However, as the treatment change in the studies was not prescribed by a standardised algorithm (retrospective studies) and reflected only one treatment change for all patients regardless of test outcome (prospective studies), the studies did not satisfy our inclusion criteria and the outcomes reported were not useful for the health economic evaluation.

Assessment of the risk of bias in the management studies

The risk-of-bias assessment, using the Cochrane risk-of-bias tool,70 for the two included RCTs73,123 is summarised in Table 10 and Figure 19. Steenholdt et al. 123 described a treatment algorithm in patients with LOR and, more recently, provided updated longer-term data (20 weeks’ follow-up). 124 The quality assessment of the retrospective observational pilot study by Vaughn et al. 128 was assessed using the Downs and Black checklist. 71 Further details on the quality assessment of these three studies are provided in Appendix 10 and Table 10.

FIGURE 19.

Risk-of-bias graph across two included RCTs: reviewers’ judgements about each risk-of-bias item. ITT, intention to treat; PP, per protocol.

Summary of the management studies

Vande Casteele et al.: Trough level Adapted infliXImab Treatment study73

Study design

This RCT73 included 251 patients with IBD (173 with CD; 78 with UC) with stable response to IFX therapy who were randomised (1 : 1) to two different treatment strategies: (1) clinically based dosing (n = 123); or (2) IFX trough concentration-based dosing (n = 128) that targeted an IFX trough level of 3–7 µg/ml. Prior to randomisation, a consecutive cohort of 275 IBD patients (186 with CD; 89 with UC) were screened and subjected to an optimisation phase using an algorithm for dose adjustment to identify patients whose trough IFX levels could be successfully brought to the target range. All randomised patients entered with trough IFX levels within the target range. For patients randomised to the clinically based dosing arm, subsequent IFX dosing was in accordance with clinical symptoms and CRP levels (recorded at each infusion) and followed standard clinical criteria. For those randomised to the trough concentration-based dosing arm, IFX dosing continued in accordance with the algorithm. Patients were followed for 52 weeks post randomisation.

Of the 275 consecutive patients, 12 were excluded because of loss to follow-up, or ineligibility, or because their trough IFX levels were undetectable and antibodies to IFX were detected at > 8 µg/ml (n = 6). The remaining 263 proceeded to optimisation. For 12 patients, the optimisation algorithm failed to bring trough IFX levels into the target range; the remaining 251 were randomised to continued dosing based on trough IFX levels or to the clinically based dosing strategy.

Specified primary and secondary outcomes

The primary outcome was the rate of clinical plus biological remission 1 year after randomisation (clinical remission required a HBI score of ≤ 4 for CD and partial Mayo score of ≤ 2 for UC; biological remission required a CRP concentration of ≤ 5 mg/l). Early terminations were considered failures for the primary end point [criteria for termination included safety and failure of IFX therapy defined as persisting clinical symptoms (HBI score of > 4 or partial Mayo score of > 2) on two consecutive visits (including unscheduled visits) and active inflammation based on increased CRP concentration OR endoscopic activity].

Secondary outcomes included durable remission, relapse, trough IFX levels in target range, anti-drug antibody positivity, European Quality of Life-5 Dimensions (EQ-5D) QoL score and total cost of treatment. In the recently accepted paper, an objective was also to compare cost-effectiveness and safety of trough level-based dosing to clinically based dosing of IFX. 73

Optimisation phase

During optimisation patients were first categorised into one of four categories on the basis of trough IFX levels: (1) trough IFX levels > 7 µg/ml; (2) trough IFX levels in the target range (3–7 µg/ml); (3) trough IFX levels < 3 µg/ml; or (4) trough IFX levels undetectable and anti-drug antibodies < 8 µg/ml (patients with undetectable trough IFX levels but anti-drug antibodies > 8 µg/ml were excluded at screening). In each category, dose was adjusted in accordance with the algorithm shown in Figure 15.

Of 72 patients with trough IFX levels < 3 µg/ml (categories 3 and 4), dose escalation brought 69 to a target trough level of IFX. In total, 115 category 2 patients were in range and were randomised, and in 67 out of 72 category 1 patients dose was de-escalated to achieve the target range. A total of 251 patients were randomised: 128 to trough IFX level-monitored dosing and 123 to clinically based dosing.

Infliximab and antibody measurement

The trough IFX and anti-drug antibody levels were measured using an in-house developed ELISA (Leuven in-house ELISA). The trough IFX level was measured using direct ELISA and anti-drug antibody levels using bridging ELISA. The lowest quantification value for IFX and anti-drug antibody limit for the test was 0.3 and 1.0 µg/ml, respectively.

Patient characteristics

The authors reported baseline characteristics and results according to two phases: optimisation phase and maintenance phase (i.e. post randomisation).

Optimisation phase (n = 263): the mean age of patients was 41.0 years (range 30–48.5 years) and, 77.2% of patients were in remission, mean CRP level was 1.7 mg/l and the mean IFX trough level was 4.6 µg/ml (2.5–7.7 µg/ml). Around 5% of patients were receiving immunosuppressant.

Maintenance phase (n = 251): about 55% of patients were female; most patients (69%) were diagnosed with CD, approximately 30% had previously undergone surgery, median duration of disease was 12.5 years (range 6.3–19.9 years), median duration of disease at first IFX exposure was 5.8 years (range 1.7–13.5 years), median time since first IFX was 4.6 years (range 2.1–7.5 years), 82.5% of patients were in remission (CD, 79.8%; UC, 88.5%) and mean CRP and mean IFX trough concentration were 1.4 mg/l (range 0.6–4.2 mg/l) and 4.9 µg/ml (range 3.9–8.5 µg/ml), respectively.

Results: optimisation phase

The results for patients with CD are summarised in Table 17.

TABLE 17 - Remission rates for patients with CD; comparison of after optimisation vs. before optimisation

OR, odds ratio.

ITT analysis.

PP analysis, numbers of patients estimated from reported percentages.

Patient group	Clinical remission, n/N (%)		Statistic (after vs. before) (95% CI)
	After optimisation	Before optimisation
All patients with CDa	138/173 (79.8)	131/178 (73.6)	RR 1.053 (0.936 to 1.186)
Patients with CD dose escalatedb	38/43 (88.4)	28/43 (65.1)	OR 4.071 (1.324 to 12.524); RR 1.297 (1.008 to 1.669)
Patients with CD dose reducedb	35/51 (69.4)	41/51 (80.4)	OR 0.534 (0.215 to 1.325); RR 0.854 (0.678 to 1.074)

Of 178 patients with CD entering optimisation, 131 were in clinical remission (HBI ≤ 4). After optimisation, 138 out of 173 were in remission (ITT: RR 1.053, 95% CI 0.936 to 1.186).

Of 44 patients with CD in the dose escalation group entering optimisation, 43 achieved target trough IFX levels. Of these, 28 were in clinical remission at entry, rising to 38 in clinical remission after optimisation [reported PP, odds ratio (OR) 4.1, 95% CI 1.3 to 12.5; RR 1.297, 95% CI 1.008 to 1.669; p = 0.020]. These patients also showed a significant decrease in mean CRP concentration at the end of optimisation (from 4.3 mg/l to 3.2 mg/l; p < 0.001). The corresponding results for UC patients did not reach statistical significance (p = 1.0 and p = 0.16, respectively). Among patients with CD who underwent dose reduction during optimisation (PP, n = 51), the proportion in remission decreased from 80.4% to 69.4% (PP: RR 0.854, 95% CI 0.678 to 1.074).

No statistically significant changes in clinical remission or in mean CRP concentration by the end of optimisation were observed for CD or UC patients who achieved target trough IFX levels (p = 0.3 and p = 1.0, respectively, for clinical remission and p = 0.56 and p = 0.86, respectively, for CRP levels).

In the dose escalation group, an average of 2.1 optimisations were required to reach target trough IFX levels, and at the end of optimisation the median infusion interval was 6 weeks (range 4–8 weeks). In the dose reduction group, a mean of 1.4 optimisations was required and the median infusion interval was 8 weeks (range 6–12 weeks).

Results: maintenance phase 52 week primary outcome

Almost 90% of patients completed the maintenance phase. The reasons for not completing were, in the clinically based and concentration-based dosing arms respectively, discontinuation because of active disease (4 and 4), serious adverse event (1 and 1), lost to follow-up (2 and 1), pregnancy (3 and 1), inability to maintain the target trough level (0 and 1) and other reasons (2 and 1).

Similar primary outcome rates were observed in both randomised arms [88/128 (68.8%) in the concentration-based dosing arm and 81/123 (66%) in the clinically based dosing arm (p = 0.686)]. Corresponding results for CD and UC patients separately were 63% versus 55% (p = 0.353) and 88% versus 84% (p = 0.748), respectively.

The results did not change when analysis was restricted to those in remission at the start of maintenance.

Results: maintenance phase secondary outcomes

There was little difference between groups in the probability of maintaining durable remission (26% and 27% in the concentration-based dosing arm and the clinically based dosing arm, respectively; p = 0.88).

A higher proportion of patients in the concentration-based dosing arm than in the clinically based dosing arm (74% vs. 57%) had IFX trough concentrations between 3 and 7 µg/ml (p < 0.001), whereas the risk of patients in the clinically based arm having undetectable trough levels of IFX was significantly greater (RR 3.7, 95% CI 1.7 to 8.0; p < 0.001). None of the patients in the concentration-based dosing arm was positive for anti-drug antibodies, but three patients in the clinically based arm were (p = 0.116).

No deaths occurred in any group, but two patients in the clinically based dosing arm required hospital admission: one because of acute appendicitis and another because of ileostomy complications. There were 12 discontinuations in the clinically based dosing arm and 13 in the concentration-based dosing arm.

More patients in the clinically based arm (n = 21, 17%) than in the concentration-based dosing group (n = 9, 7%) relapsed and needed rescue therapy (RR 2.4, 95% CI 1.2 to 5.1; p = 0.018). Relapse was defined as ‘the need for IFX dose escalation (interval decrease and/or dose increase), the addition of steroids or switch to another anti-TNFα and was based on the physician’s global assessment’. 73 Among those relapsing and requiring rescue therapy, relatively more (9/21, 43%) in the clinically based arm than in the concentration-based arm (2/9, 22%) had trough IFX levels of < 3 µg/ml.

Relapse free-survival time was longer in the concentration-based dosing arm than in the clinically based dosing arm (Figure 20).

FIGURE 20.

Kaplan–Meier analysis of time to relapse during maintenance phase. IPD reconstructed using the method of Guyot et al. 74 Conc:-based, concentration-based strategy; clin:-based, clinically based strategy.

Authors’ conclusions

The authors concluded that optimisation of IFX dose to achieve the target trough levels of 3–7 µg/ml is more efficient and is cost-effective relative to clinically based adjustment. Therefore, the authors recommended using dose-to-target optimisation of IFX to achieve the target trough IFX levels and to re-evaluate the level after 6 months. It should be borne in mind that both arms received target dose optimisation prior to randomisation and, therefore, even the comparator group, which received a clinically guided dosing regimen, had already received a phase of trough level monitoring and dose adjustment.

Vaughn et al.128

Study design and conduct

The aim was to investigate the usefulness of proactive therapeutic concentration monitoring and titration of IFX to a target concentration. 128 This was a retrospective observational pilot study of patients with IBD in clinical remission receiving IFX at tertiary health-care centres; patients were identified from records and classified into those who received proactive drug monitoring and those who did not (control group); patients who did not achieve remission were excluded. For both proactive drug monitoring and control groups, clinical remission was defined as ‘lack of symptoms attributable to underlying IBD based on the treating gastroenterologist’s documentation’. 128

The IFX and antibodies to IFX concentrations were measured initially using solid-phase ELISA (PROMETHEUS Laboratories) and later with the HMSA (PROMETHEUS laboratories). The latter test could detect IFX concentrations as low as 1 µg/ml, compared with 1.4 µg/ml for ELISA. In the proactive drug monitoring group, serum trough IFX levels were used to guide dose modifications to achieve target drug levels. Initially the target was detectable IFX; later the target was changed to an IFX concentration between 5 and 10 µg/ml. Typical changes in dose administration in the proactive drug monitoring group were as follows:

• For patients with undetectable trough drug levels, the dose of IFX infusion was increased to 7.5 mg/kg. The next infusion was given after 6 weeks, and after which IFX was given every 8 weeks.

• For patients with detectable trough drug levels < 5 µg/ml, the dose of IFX was increased by 50 or 100 mg.

• In patients with trough drug levels of > 10 µg/ml on at least two occasions, the dose was reduced. However, in those patients who were already receiving 5 mg/kg IFX, instead of dose modification, the treatment interval was increased.

• In patients who had trough levels in the range 5–10 µg/ml, no changes were made.

(Trough concentration was defined as ‘IFX concentration measured at any time up to 7 days before the next infusion’. 128)

Reactive testing was done in both groups whenever there was LOR or if there was a concern that the patient was experiencing side effects as a result of antibody formation. For patients in the control group with LOR, the dose of IFX was increased at the treating physician’s discretion in accordance with a standard of care guideline (typically to 10 mg/kg, but the dose did not reach > 10 mg/kg every 4 weeks).

Patient populations

There were 48 IBD patients in the proactive drug monitoring group and 78 in the control group. They were followed from the start of maintenance therapy until August 2013 or until their last documented clinical encounter. Proactive drug monitoring was initiated at some time during patients’ maintenance and was adopted as a strategy ‘starting in 2009’. The determining difference between groups was that testing was performed only reactively in the control group, but both reactively and proactively in the proactive drug monitoring group to determine any dose changes judged necessary to reach target trough concentration; furthermore, when the dose was escalated in the control group, IFX exposure was probably doubled (e.g. to 10 mg/kg), but dose escalations in the proactive drug monitoring group were of much smaller magnitude (e.g. by 50–100 mg; for a 70-kg individual this raises the dose from 5 mg/kg to between 5.7 and 6.4 mg/kg). Dose de-escalation (to < 5 mg/kg) occurred only in the proactive drug monitoring group.

Two patients were ‘IBD unclassified’, 90 (69%) were diagnosed as having CD and 34 (29%) were diagnosed as having UC. Almost 70% were male. Median age at IFX initiation was 34.9 years in the proactive drug monitoring arm and 35 years in the control arm [interquartile range (IQR) 26.2–49.7 years]. The median age at diagnosis was 23.5 and 25 years in the proactive drug monitoring and control arms, respectively; 30% of patients had undergone IBD surgery previously (40% of the proactive drug monitoring group, but only 25% of the control group); 10% of patients were current users of tobacco, 25% were former users and 56% had never used tobacco; and 52 patients (41%) received combination therapy (44% and 40% of the proactive drug monitoring and control group, respectively). The median duration of IFX before proactive drug monitoring was 43 weeks (IQR 32–72 weeks).

The main reported outcomes comparing the proactive drug monitoring and control groups were the time remaining on IFX (Kaplan–Meier analysis) and the reasons for stopping IFX. Further details about dose changes and trough levels in the proactive drug monitoring group were also provided.

Outcomes: time remaining on infliximab

Patients identified as belonging to the proactive drug monitoring group remained on IFX treatment longer than those identified as belonging to the control group. At 5 years (260 weeks) the probabilities of remaining on treatment were 86% and 52%, respectively. Figure 21 shows the reconstructed Kaplan–Meier comparison between groups. Beyond 5 years there are very few patients at risk; the median duration of IFX treatment before proactive drug monitoring implementation was reported to be 43 weeks (IQR 32–72 weeks). In multiple Cox regression analysis, the probability of patients remaining on IFX therapy was found to be significantly related only to proactive drug monitoring of IFX.

FIGURE 21.

Kaplan–Meier analysis of time to stopping IFX treatment. IPD reconstructed using the method of Guyot et al. 74 The vertical line represents 5 years. Inset shows proportional hazards test plot. TCM, trough concentration monitoring.

The authors also reported on the subgroup of patients that started maintenance IFX after 1 January 2009. As the implementation of proactive drug monitoring was reported to be from 2009, this subgroup would appear to be the more relevant population. Figure 22 shows the reconstructed Kaplan–Meier comparison between the proactive drug monitoring and control subgroups. The reported hazard ratio was 0.3 (95% CI 0.1 to 0.7; p = 0.003). The reconstructed hazard ratio was similar, at 0.24 (95% CI 0.12 to 0.51); it is possible that the authors stratified their analysis by baseline variables (e.g. monotherapy or combination therapy, previous surgery, etc.). Parametric modelling based on reconstructed IPD is provided in Appendix 11.

FIGURE 22.

Kaplan–Meier analysis of time to stopping IFX treatment, patients starting maintenance January 2009. IPD reconstructed using the method of Guyot et al. 74 TCM, trough concentration monitoring.

Outcomes: reasons for stopping infliximab

The reasons for stopping IFX are summarised in Table 18; the most frequent causes for the control group were recurrence of IBD symptoms and acute infusions reactions. Adverse events and high antibody levels were the main causes for the proactive drug monitoring group.

TABLE 18 - Reasons for stopping IFX therapy

Includes: unable to afford copayment, surgery for adhesive small bowl obstruction, and proactive drug monitoring group trough levels and dose changes colectomy for flat low-grade dysplasia.

Reason for stopping	Group (n)
	Proactive drug monitoring	No proactive drug monitoring
Recurrent IBD symptoms	0	15
Adverse events
Pneumonia	0	1
Drug-induced lupus	1	0
Psoriasis	1	0
High antibody concentration	1	0
Infusion reactions
Acute	0	6
Delayed	1	0
Other unrelated to IFXa	1	2

Outcomes: proactive drug monitoring group trough levels and dose changes

For the proactive drug monitoring group, the authors reported data concerning proactive tests undertaken; all reactive tests were omitted. Of initial proactive tests (n = 48), 37 used ELISA and 11 used HMSA methodology. Of subsequent proactive tests (n = 40), seven used ELISA and 33 used HMSA.

Median trough IFX at initial testing was 5 µg/ml (IQR 2.8–9.9 µg/ml), whereas the median subsequent trough level was 7.6 µg/ml (IQR 4.3–12.3 µg/ml).

Dosing adjustment after the initial proactive test was implemented in 17 patients (35%) as follows: dose escalation in 12 patients (71%), dose reduction in three patients (18%) and termination of therapy in two patients (12%). Dosing adjustment in subsequent proactive tests involved 10 patients (25%); these adjustments were described as eight patients (80%) who received dose escalation and two (20%) a dose decrease.

Following proactive drug monitoring, the median dose increment was 100 mg (range 50–250 mg) and the median duration of IFX therapy was 144 weeks (range 36–685 weeks).

A trough level of IFX ≥ 5 µg/ml (lower end of the later target range) was reportedly achieved in 75% of patients (36/48). Of those in who reached this level of IFX, none developed antibodies to IFX or an immune reaction. In one patient, IFX was stopped after colectomy, which was undertaken for flat low-grade dysplasia.

Authors’ conclusions

The authors concluded that proactive trough concentration monitoring of IFX frequently identified patients with low or undetectable trough concentrations and resulted in a greater probability of remaining on IFX. In this study the treatment algorithm was ill-defined and test methods were adjusted during the study. The retrospective observational design means that the selection of patient groups was at risk of bias.

Summary of major findings from three management studies

Three disparate studies were identified that implemented a test-informed algorithm and reported clinical outcomes. One examined patients with CD only, and the other two examined both CD and UC patients. Two were RCTs and the other was a retrospective observational study. None employed designated intervention tests (LISA-TRACKER assays, TNF-α-Blocker or Promonitor ELISA).

The Steenholdt et al. RCT123 used concurrent RIA testing prior to implementation of a treatment algorithm for patients with CD with LOR to IFX; the comparator group received IFX intensification. At 12 weeks after randomisation there was no clinical benefit from the test algorithm strategy relative to dose intensification. For 64% of patients in the algorithm arm (those with therapeutic IFX but no anti-drug antibodies) the algorithm recommended cessation of IFX therapy. That cessation of IFX therapy was not associated with reduced disease control suggests IFX may not be useful for most patients with CD with LOR; however, the criteria for LOR may have been imprecise so that patients appeared to regain response 12 weeks later. Weaknesses of the study include short duration, small size, a large number of withdrawals, the fact that many participants did not receive algorithm-prescribed treatments, and unclear or high risk of bias in several risk-of-bias domains. The authors reported cost savings for the test algorithm group relative to the dose escalation group that are probably attributable to less use of IFX in the algorithm arm. Further studies are required to test the reliability of findings.

The TAXIT RCT (Vande Casteele et al. 73) used a test algorithm for IFX responders to optimise the trough IFX level to a set target range. Tests employed in-house ELISAs. Trough optimisation with dose adjustments did not change the proportion of patients with CD in clinical remission (RR for after vs. before optimisation: 1.053, 95% CI 0.936 to 1.186). After trough level optimisation, patients were randomised to continued trough test monitoring or to clinical monitoring. For the primary outcome (rate of clinical plus biological remission at 52 weeks) there was no difference between groups for patients with CD (54.9% vs. 62.6%; p = 0.353). Time to relapse for CD plus UC patients was superior with test monitoring than with clinical monitoring (p = 0.018). Total cost post randomisation (CD plus UC patients) was slightly lower with test monitoring (€20,723 vs. €21,023). The risk of bias was low for most risk-of-bias domains.

The retrospective observational study of Vaughn et al. 128 compared proactive trough concentration monitoring with dosing based on clinical judgement (test results did not influence treatment). Among clinically managed patients, dose escalations were likely to involve a doubling of IFX exposure, whereas in the trough monitoring group some dose changes were dose reductions, and dose escalations were considerably more moderate than in the clinically managed group. The authors’ major finding was that relative to clinical monitoring, trough monitoring was associated with far superior retention on IFX treatment (hazard ratio 0.3, 95% CI 0.1 to 0.7; p = 0.003). The observational design of the study and the retrospective identification of participants based on medical records mean that the study was at considerable risk of bias.

Limitations of the review of management studies

The most important limitation of this review was that the relevant management studies did not directly address our research questions; a further difficulty was the very limited supply of studies. Three management studies73,123,128 were found in which patients treated with IFX were investigated, but no corresponding studies were found for ADA. The timing of testing specified in the research questions did not correspond with that used in any of the studies. Furthermore, two of the three available studies investigated a mixture of CD and UC patients, and only one study (Steenholdt et al. 123) reported the impact of treatment algorithm in clinical outcomes for patients exclusively diagnosed with CD. In this study there were few patients and follow-up was short; therefore, the power to detect differences in clinical outcome between randomised groups over a clinically meaningful period was limited. However, this study provided some evidence that, at least in the short term, a dose escalation strategy for LOR to IFX may be more costly than the alternative strategy proposed in the authors’ treatment algorithm. Further investigation is required to establish that cost savings are not associated with deterioration in disease control. The TAXIT RCT investigated IBD patients with stable response to IFX. 73 Patients in both randomised groups were optimised to a target trough level of IFX; therefore, a comparator group in which from outset ‘treatment decisions made on clinical judgement without measuring levels of TNF[-]α inhibitor and antibodies to TNF[-]α inhibitors’73 did not exist. The pilot study of Vaughn et al. 128 was a retrospective observational study and, therefore, findings should be viewed with considerable caution. This review of management studies clearly highlights gaps in the evidence and indicates that further studies are needed.

Evidence taken forward to the economic evaluation

Data from the three management studies73,123,128 have been taken forward for economic evaluation. The two RCTs, one for responders and the other for IFX recipients with LOR, have informed model structure and provided information for the base-case economic analysis. The study by Vaughn et al. ,128 which reports substantial clinical advantage for a test algorithm strategy in terms of retention in IFX treatment, has been used in economic evaluation sensitivity analysis.

Search results

The search identified three systematic reviews with meta-analytic pooling of results from multiple studies32,63,111,114 and 31 primary studies38,40,47,52,59,77–85,88,92,94,98–100,102,103,106,108,110,115,120,123,126,133,134 that reported the relationship between test outcomes and clinical status of patients in sufficient detail to allow us to extract 2 × 2 data of diagnostic performance when using a drug and/or anti-drug antibody test to diagnose/predict response or LOR. The systematic reviews are summarised in Published meta-analyses of studies relating test results to clinical state of patients and the primary studies are analysed in Analysis of correlation studies of anti-tumour necrosis factor alpha/anti-drug antibodies level and response.

Published meta-analyses of studies relating test results to clinical state of patients

Results

The literature search yielded several reviews that addressed the relationship between test results and the clinical state of patients with IBD. 36,44,49,61–63,111,114,139 Of these, four were systematic reviews that meta-analysed pooled results from multiple studies. 49,63,111,114 One meta-analysis encompassed several inflammatory conditions in addition to IBD and is not considered further here. 49 The three remaining meta-analyses considered anti-drug antibodies, and one also examined drug trough level tests. 114 Although many of the primary studies included in the meta-analyses presented data in terms of diagnostic or predictive tests (e.g. sensitivity, specificity and other test accuracy measures), the meta-analyses addressed the risk of a particular test result (e.g. negative) in patients with a particular clinical state (e.g. LOR and calculated a RR of a negative test result in LOR relative to state no LOR or, conversely, RR of LOR in patients with negative test relative to those with a positive test). Viewing the tests as diagnostic/predictive permits hierarchical (bivariate) meta-analysis that incorporates covariance between sensitivity and specificity estimates. The RR statistic does not formally allow for covariance between estimated associations. Each of the meta-analyses is considered in turn.

Nanda et al.111

The authors estimated the pooled RR of LOR to IFX in patients with a positive test for anti-drug antibodies relative to those with a negative test for anti-drug antibodies (a greater risk of LOR in patients who were antibody positive than in patients who were antibody negative generates a RR of > 1.0). Eleven studies were included,40,47,55,59,79,83,92,99,103,120,142 one with only UC patients, three studies with mixed IBD populations (one of which reported results separately by UC and CD) and seven studies of CD patients. The comparative numbers of events and patients were reported. The pooled estimate (Figure 23a; RR 3.16, 95% CI 2.00 to 4.98; I² = 70.1%) indicated that the risk of LOR was about threefold higher in those with a positive anti-drug antibodies test than in those with a negative test.

FIGURE 23.

Meta-analysis of data based on that from Nanda et al. 111 (a) Random-effects meta-analysis of the RR of LOR to IFX (anti-drug antibodies positive vs. anti-drug antibodies negative); (b) sensitivity and specificity forest plot of the included studies; (c) summary ROC bivariate meta-analysis of sensitivity–specificity pairs (hollow symbols) with pooled point estimate square solid symbol; and (d) summary ROC bivariate meta-analysis of sensitivity–specificity pairs with pooled point estimate square solid symbol, excluding an influential outlier study. ATI, antibodies to IFX; FN, false negative; FP, false positive; HSROC, hierarchical summary receiver operating characteristic; TN, true negative; TP, true positive.

When viewed as a predictive/diagnostic test143 the same data can be analysed to estimate the sensitivity and specificity, and meta-analysed to generate a pooled joint sensitivity–specificity value (and other test accuracy parameters). 144 In this case, a positive test for anti-drug antibodies is viewed as predictive/diagnostic of LOR. Figure 23b indicates marked heterogeneity among the studies and the trade-off between sensitivity and specificity in the different studies. Figure 23c and d summarises the summary receiver operating characteristic (sROC) meta-analysis results. The meta-analysis test accuracy results are summarised in Table 19. The large RCT-based study by Hanauer et al. 40 was identified as both influential and an outlier; including or excluding this study made little difference to the summary test accuracy estimates, but substantially decreased the 95% CI around the prediction region in sROC space (see Figure 23c). This study differed from the others in having the lowest ratio of positive to negative test results, probably resulting from the number of tests classified as inconclusive.

TABLE 19 - Test accuracy parameters generated by hierarchical MA144

Studies included	Estimate	Sensitivity	Specificity	Diagnostic OR	Likelihood ratio positive	Likelihood ratio negative
Excludes outlier	Point estimate	0.72	0.79	9.87	3.49	0.35
	95% CI	0.64 to 0.78	0.64 to 0.89	4.07 to 23.92	1.85 to 6.60	0.26 to 0.48
All studies	Point estimate	0.70	0.81	9.81	3.63	0.37
	95% CI	0.55 to 0.82	0.67 to 0.89	4.09 to 23.54	2.04 to 6.45	0.24 to 0.58

The implication of these test accuracy results was explored in terms of predictive values as suggested in the Cochrane Handbook for Systematic Reviews of Interventions. 141 As predictive values are influenced by prevalence of the target condition, we determined a pooled random-effects estimate of prevalence (LOR) among the studies (34.7%, 95% CI 25.1% to 44.4%). The point estimates for positive predictive values (PPVs) and negative predictive values (NPVs) at this prevalence were 65% and 84%, respectively. The influence of prevalence on these values is illustrated in Figure 24 across the range of prevalence of the included studies and the 95% CI around the pooled prevalence.

FIGURE 24.

Positive predictive values and NPVs, according to prevalence of LOR at the sROC, model estimates of sensitivity and specificity, as prevalence increases PPV increases and NPV decreases. Data points are PPV and NPV at sROC sensitivity and specificity and pooled prevalence. Dashed vertical lines are pooled prevalence and 95% CI. Thick curves are PPV and NPV for hierarchical model sensitivity and specificity at the pooled prevalence and 95% CI.

The meta-analysis results indicate that the anti-drug antibody test has only moderate accuracy performance in predicting/detecting LOR to IFX.

Lee et al.63

The authors estimated the pooled RR of remission in patients with a positive test for anti-drug antibodies to IFX relative to those with a negative test for anti-drug antibodies (a RR of < 1.0 indicates that anti-drug antibodies are associated with lower risk of remission, consistent with the hypothesis that anti-drug antibodies reduce response to IFX therapy) based on nine studies. 38,40,47,55,83,92,142,145,146 Comparative numbers of events and patients were reported. The fixed- and random-effects RRs are 0.90 (95% CI 0.79 to 1.02) and 0.96 (95% CI 0.77 to 1.19), respectively. Statistical heterogeneity unexplained by chance was 37% (I²-statistic). When the presence of antibodies to IFX is considered as predictor of, or diagnostic of, a lack of remission, then meta-analysis yielded low joint sensitivity specificity values of 0.42 and 0.69, respectively (Figure 25).

FIGURE 25.

Meta-analysis of data based on that from Lee et al. 63 (a) Fixed-effects meta-analysis of the RR of remission (presence of antibodies to IFX vs. absence of antibodies to IFX); and (b) sROC bivariate meta-analysis of sensitivity specificity pairs (hollow symbols) with pooled point estimate square solid symbol. ATI, antibodies to IFX;HSROC, hierarchical summary receiver operating characteristic; M–H, Mantel–Haenszel.

The results indicate that presence of anti-drug antibodies does not strongly increase the risk of lack of remission and that a positive test for the presence of anti-drug antibodies has poor discriminatory power for predicting/diagnosing a lack of remission.

Lee et al. 63 also reported a meta-analysis examining the association between the development of anti-drug antibodies and the use of immunosuppressant therapies. Eleven studies were included38,40,43,45,53–55,142,146–148 and the authors generated a fixed-effects RR (antibodies present with suppressants vs. antibodies present with no suppressants) of 0.50 (95% CI 0.42 to 0.59; I² = 43.4%), indicating a 50% reduction in risk of developing anti-drug antibodies when suppressants are administered. The results of fixed- and random-effects meta-analyses are illustrated in Figure 26.

FIGURE 26.

The RR of anti-drug antibodies with immunosuppressants vs. without suppressants. Data based on Lee et al. 63 ATI, antibodies to IFX; D + L, DerSimonian–Laird; M–H, Mantel–Haenszel.

Paul et al.114

The authors estimated the pooled ratio for the odds of a lack of response in association with a negative test for ADA (i.e. subtherapeutic) compared with the odds of a lack of response in those with a positive test for ADA (an OR of > 1.0 indicates that subtherapeutic ADA levels are associated with lack of clinical response). The comparative numbers of events and patients were not reported and it was difficult to verify the included data from the references provided. The pooled OR differed between three studies of adults with CD (pooled OR 7.5, 95% CI 3.58 to 13.90) and two studies of children with CD (pooled OR 1.59, 95% CI 1.00 to 2.54). The overall pooled OR was 2.60 (95% CI 1.79 to 3.77).

The reported ORs are equivalent to the diagnostic odds ratio (DOR) of a diagnostic test in which subtherapeutic drug levels are the test for lack of clinical response. As such, they are modest relative to the DOR value of 9.6 (odds pooled sensitivity/odds pooled specificity) for studies included in the Nanda et al. 111 review using the test for antibodies to IFX as predictor of lack of response.

The authors also estimated the pooled ratio for the odds of lack of response in association with a negative test for anti-drug antibodies to ADA compared with the odds of lack of response in those with a positive test for anti-drug antibodies (an OR of > 1.0 indicates that the presence of anti-drug antibodies is associated with lack of clinical response). The reported OR of 10.15 (95% CI 3.90 to 26.40) is equivalent to the DOR for presence of antibodies (to ADA) as a predictor for the lack of clinical response. This value is similar to the DOR for antibodies to IFX as a predictor for the lack of response [using the pooled sensitivity specificity pair (0.72 and 0.79) derived from the studies in the Nanda et al. 111 review, which provides a DOR of 9.67 [(0.72/0.28)/(0.21/0.79)].

Analysis of correlation studies of anti-tumour necrosis factor alpha/anti-drug antibodies level and response

Search strategy

A comprehensive search of the literature for published economic evaluations (including any existing models), cost studies and QoL (utility) studies was performed. The systematic search included searching the following electronic databases during December 2014 (from 12 to 17 December 2014):

• MEDLINE (via Ovid) (1946 to Week 3 November 2014)

• MEDLINE In-Process Citations and Daily Update (via Ovid) (11 December 2014)

• EMBASE (via Ovid) (1947 to 15 December 2014)

• NHS Economic Evaluation Database (The Cochrane Library)

• Science Citation Index (Web of Knowledge) (1970–present)

• Cost-effectiveness Analysis Registry

• EconPapers (Research Papers in Economics)

• School of Health and Related Research Health Utilities Database.

The search included terms for CD, anti-TNF-α drugs and the different assay kits, combined with economic and QoL terms. The search was limited to studies published in the English language. The search strategy developed was based on the clinical effectiveness review, with input from a health economist. Details of the full search strategies are provided in Appendix 3.

Inclusion criteria

Only studies meeting the following inclusion criteria were included in the review:

• study type – fully published economic evaluations (including economic models)

• population – people with CD

• intervention – anti-TNF-α drugs (ADA and IFX) and antibody drug testing (LISA-TRACKER ELISA kits, TNF-α-Blocker ELISA kits and Promonitor ELISA kits) for any dosage or treatment regimen

• comparator – standard care treatment: anti-TNF-α drugs (ADA and IFX) for any dosage or treatment regimen

• outcomes – cost-effectiveness or cost–utility studies reporting outcomes as clinical effectiveness measures or utility measures [utility, EQ-5D, Short Form questionnaire-6 Dimensions score or quality-adjusted life-years (QALYs)].

Exclusion criteria

Studies meeting the following exclusion criteria were excluded from the review:

• non-English-language publications

• studies in the health areas where these anti-TNF-α drugs have also been used, such as UC, rheumatoid arthritis, psoriasis and tuberculosis.

Assessment of eligibility and data extraction

All retrieved records (citations and abstracts) were collected in a specialist database (EndNote) and duplicate records were identified and removed. Two reviewers independently reviewed titles and abstracts to identify potentially relevant papers for inclusion. Any discrepancies were resolved by discussion. See Appendix 13 for the table of full-text studies excluded with reasons.

Data extraction was carried out in two stages by one reviewer using standardised data extraction sheets (see Appendix 14) and was then checked by a second reviewer. Stage 1 considered all eligible studies (fully published economic evaluations including any economic models) and stage 2 considered studies assessed for usefulness for populating the economic model. Data extracted during stage 1 included the following:

• study details – author names, source of publication, language and publication type

• baseline characteristics – population, intervention, comparators, outcomes and type of economic evaluation

• methods – target population and subgroups, setting and location, study perspective, time horizon, discount rate, measurement of effectiveness, measurement and valuation preference-based outcomes, resource use and costs, currency, price date and conversion, model type, assumptions and analytical methods

• results – study parameters, incremental costs and outcomes and characterising uncertainty

• discussion – study findings, limitations, generalisability and conclusions

• other – sources of funding, conflicts of interest and comments.

Quality assessment

The quality of full economic evaluation studies that were identified was assessed using the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist (see Appendix 15) by one reviewer and cross-checked by a second reviewer. The CHEERS checklist comprises six dimensions: title and abstract, introduction, methods, results, discussion and other. Under these dimensions, a series of questions check whether or not the criteria have been clearly reported. Any studies containing an economic model were further assessed using the framework for the quality assessment of decision-analytic modelling by Philips et al. 149 (see Appendix 15). The Philips checklist contains two main dimensions: structure of the model and data used to parameterise the model. Under these dimensions several questions assess whether or not the criteria has been clearly reported.

Data synthesis

Information extracted from the included studies were summarised and tabulated. Findings from individual studies were compared narratively.

Summary of the Health Technology Assessment report by Dretzke et al.5

The main aim of this HTA report was to assess the cost-effectiveness of anti-TNF-α drugs in the management of adult patients with moderate to severe CD in the UK NHS. The authors described induction therapy as the use of anti-TNF-α therapy with the aim of achieving remission (a repeated reinduction treatment was considered, rather than a one-off induction therapy) and maintenance therapy as the use of anti-TNF-α therapy to maintain remission in patients who have responded (and continue to respond) to anti-TNF-α therapy when in relapse. Response by the authors was defined as remission within 8 weeks.

The authors developed a Markov model from a NHS and Personal Social Services (PSS) perspective to estimate the incremental cost per QALY gained for both ADA and IFX (anti-TNF-α therapy) compared with standard care. Mortality was not included in the model, as the authors found no difference in the mortality rates that were reported in the clinical trials reviewed and, therefore, felt that a lifetime horizon would not improve the precision of the cost-effectiveness estimate. Instead, the time horizon for the model was 1 year and the cycle duration was 4 weeks. The model for both induction and maintenance therapy started with a cohort of patients in the standard care refractory relapse health state. The model had four main health states and at any time, and on any given treatment, a patient was in remission, in relapse, undergoing surgery or in post-surgery remission.

Transition probabilities for the standard care health states were based on Silverstein et al. 150 Transition probabilities for both the induction and maintenance model were assigned a treatment effect by using relapse to remission probabilities from RCT evidence; however, for the maintenance model there was a lower remission to relapse rate.

The majority of utility values for the model were based on the study by Gregor et al. ,20 which used the time-trade off measure to estimate the health-related QoL in CD. A utility value for surgery was not available in the published literature; therefore, it was assumed that the average utility value for surgery would be equivalent to EQ-5D health state 22222, with a utility weight of 0.516.

The direct costs to the NHS were the sum of the anti-TNF-α costs and type-specific health-state costs. The costs of anti-TNF-α therapy, both induction and maintenance, were derived from the BNF (2007/8),151,152 and administration costs were also included for IFX. Type-specific health-state costs included costs for surgery, which were modelled as the cost of inpatient IBD interventions, and post-surgery remission costs, which were based on outpatient surgical gastrointestinal follow-up. Moderate and severe relapse costs were modelled as the cost of IBD outpatient major and intermediate interventions. Relapse costs were based on a gastrointestinal admission to hospital. Remission costs were modelled using literature. The majority of health-state unit costs were obtained from the NHS reference cost database (NHS Reference Costs 2005 to 2006153).

Incremental cost-effectiveness ratios (ICERs) and cost-effectiveness acceptability curves were presented. One-way sensitivity analyses and probabilistic sensitivity analyses (PSAs) using 10,000 simulations were conducted to characterise uncertainty in the model.

For induction therapy for severe CD, both ADA and IFX dominated standard care (i.e. they were cheaper and more effective). For maintenance therapy for severe CD, neither drug was cost-effective (well above NICE thresholds). For moderate CD, for maintenance therapy for both drugs and induction therapy for IFX, these were not cost-effective (well above NICE thresholds); however, for induction therapy, ADA dominated standard care.

Sensitivity analysis showed that, in patients with severe disease, IFX induction treatment was cost-effective relative to maintenance treatment and standard care in > 99% of cases at all points up to £100,000 per QALY. Likewise, ADA induction treatment was found to be cost-effective relative to maintenance treatment and standard care for thresholds up to £100,000 per QALY.

The key limitations of this model was a short time frame (1-year time horizon); the exclusion of death from the model; no randomised controlled data available for maintenance therapy; and the use of Silverstein et al. 150 data for transition probabilities, which inherently had their own problem (i.e. surgery rates were higher and relapse rates much lower than in routine practice).

Search results for objective D

The literature search identified 2466 records through electronic database searches and other sources. After removing duplicates, 1527 records were screened for inclusion. On the basis of a title and abstract sift only, 1518 records were excluded. The remaining nine records were subjected to full-text screening. A further five articles5,26,154–156 were excluded at the full-text stage, as these studies did not use assay kits to measure levels of TNF-α inhibitors and anti-drug antibodies. The literature search identified four studies73,123,124,157 of the cost-effectiveness of different assay kits for measuring levels of TNF-α inhibitors and of anti-drug antibodies (Figure 27).

FIGURE 27.

The PRISMA flow diagram of cost-effectiveness studies.

Overview of included studies

The literature search identified four studies73,123,124,157 that met our inclusion criteria (studies looking at the cost-effectiveness of different assay kits for measuring levels of TNF-α inhibitors and of anti-drug antibodies) and were reviewed. In the following sections we present an overview of the included studies by population (responders and those showing LOR) of interest.

Comparison of the included studies

All four studies included in this review have been summarised in Table 27. Three studies were based on RCTs73,123,124 and only one study157 presented an economic model. Of the RCTs, two123,124 were conducted in Denmark and one73 was conducted in Belgium. All four studies73,123,124,157 conducted cost-effectiveness analyses: Vande Casteele et al. 73 compared concentration-based with clinician-based dosing; Steenholdt et al. 123,124 compared IFX treatment failure using a treatment algorithm compared with IFX dose increasing; and Velayos et al. 157 compared a testing-based strategy with an empiric dose escalation strategy. All studies73,123,124,157 clearly stated the type of assay used to analyse serum levels and antibodies to anti-TNF-αs. Two studies123,124 used RIA in the base case, one study73 used an assay developed in house and the remaining study157 used a PROMETHEUS ELISA.

The patient populations for three studies73,123,124 included eligible patients with moderate to severe CD, whereas the study by Vande Casteele et al. 73 included patients with UC. The study perspective was not reported in two studies,123,124 whereas the other two studies73,157 conducted the analysis from a third-party payer perspective. The time horizon varied from 12 weeks to 1 year. Steenholdt et al. 123 based their analysis on a 12-week horizon, whereas the other three studies73,124,157 used a 1-year time horizon to estimate the cost-effectiveness of the different strategies.

In two studies,73,157 outcomes were reported as cost per QALYs gained. Vande Casteele et al. 73 used the EQ-5D measure to estimate QALYs, whereas Velayos et al. 157 did not explicitly report how the QALYs were estimated, except to say that they were obtained from a secondary source. 20 The two studies by Steenholdt et al. 123,124 reported outcomes in terms of cost per ITT and cost PP population.

Three studies123,124,157 provided quite a comprehensive breakdown of resource use and costs, whereas the study by Vande Casteele et al. 73 did not elaborate on resource use, apart from the drug costs. Three studies73,123,124 reported costs in 2012 prices, whereas Velayos et al. 157 did not report the price year explicitly; however, we assumed that costs are most likely to be in 2012 prices, as the study was published in 2013.

No studies conducted discounting for either costs or benefits as the time horizon for these studies was ≤ 1 year.

The results and conclusions reported differed between studies, Vande Casteele et al. 73 demonstrated that concentration-based dosing was slightly less effective and less costly than clinically based dosing, but overall differences were small. Steenholdt et al. 123 showed that the intervention based on the algorithm achieved similar clinical and life quality outcomes to dose intensification, but at a lower cost at 12 weeks. These results were maintained at both 20 weeks and 1 year. 124 Velayos et al. 157 showed that the testing strategy was cheaper and more effective than the empiric strategy.

All four studies73,123,124,157 conducted sensitivity analyses to deal with uncertainty around key parameters. The sensitivity analyses ranged from the most simplistic one-way sensitivity analyses123,124 to the more sophisticated probabilistic analyses. 157

Quality assessment

We present, in Appendix 15, a summary of the reporting quality of the studies included in the current review against the CHEERS checklist. 158 Using a 25-point CHEERS checklist, one article73 did not identify the study as an economic evaluation in the title. All studies provided background information to the study and clearly outlined the objectives of the study. Two studies73,157 reported the viewpoint of the economic analysis. All studies described the comparators fully and reported the time horizon. However, because of the short time horizon, no studies conducted discounting of costs and benefits. In addition, the choice of health outcomes was well reported by all four studies;73,123,124,157 however, only one study73 reported how these health states were valued. Resource use and costs were well reported in three studies123,124,157 apart from that by Vande Casteele et al. ,73 who described only the drug costs. The majority of the studies73,123,124 conducted an economic analysis alongside a RCT, whereas one study157 developed an economic model. In terms of analytical methods, study parameters, incremental costs and outcomes and uncertainty were well reported by all four studies. Limitations were provided by all four studies and generalisability was only partially reported by three studies. 123,124,157

From the studies identified, one157 conducted a model-based economic analysis to determine whether or not a testing-based strategy was more cost-effective than an empiric dose escalation strategy. We present, in Appendix 15, a summary of the reporting quality of this study against Philips’s checklist. 149 In general, Velayos et al. 157 conformed to best practice for reporting model-based economic evaluations in terms of clearly stating the decision problem, adequately outlining the objectives, clearly stating the viewpoint of the analysis and describing the model structure, which represented the clinical pathway that patients with CD may follow. Time horizon and cycle length were stated and justified. In terms of the data required to populate the model, Velayos et al. 157 adequately provided references, but they were unclear on the choices made between data sources and the quality of information used in the model. In addition, it was unclear whether or not any expert opinion had been used when choosing baseline information for the model. The other limitations identified were the lack of explanation of pre-model analysis (e.g. calculation of transition probabilities, and methods and assumptions used to extrapolate short-term results into final outcomes) and the omission of half-cycle correction.

Concurrent testing

In the concurrent testing strategy, patients undergo tests for serum anti-TNF-α levels and antibodies to anti-TNF-α simultaneously, and once the test results are available follow the proposed algorithm. Patients are classified, on the basis of their test results, into one of four groups: drug absent and antibodies present, drug absent and antibodies absent, drug present and antibodies present, or drug present and antibodies absent. Alternatively, patients may be categorised according to levels of drug regardless of antibody levels (e.g. as in the TAXIT trial73). Details of test results and proposed algorithms from Steenholdt et al. 123 for patients with LOR and from Vande Casteele et al. 73 for responders are presented in Chapter 3, Objective B: description of algorithms prescribing patient management following test outcomes for drug and/or anti-drug antibody levels.

Reflex testing

In the reflex testing strategy, patients would receive a test to analyse serum anti-TNF-α levels. As a result of testing, two test outcomes are likely: drug absent or drug present. Based on the drug result, patients would undergo further testing for the presence or absence of antibodies. In this section we outline the health states and the pathways for patients undergoing reflex testing for both responder and LOR models. No study was identified that tested an algorithm for reflex testing. The algorithm followed in the model was therefore based on that of the TAXIT73 trial for responders and the Steenholdt et al. 123 algorithm for patients with LOR using concurrent testing. Further details of test results and proposed algorithms are presented in Chapter 3, Objective B: description of algorithms prescribing patient management following test outcomes for drug and/or anti-drug antibody levels.

Proportions

The proportions of patients required to populate various model decision tree branches were obtained from secondary sources [e.g. management studies described in Chapter 3, Objective C1: clinical studies evaluating drug monitoring for the management of Crohn’s disease patients (management studies)] and, when such data were lacking, from clinical input. Proportions that were estimated included partitioning of patients by presence or absence of IFX and of antibodies to IFX in responders and in those with LOR; partitioning of responders according to defined IFX trough levels; and partitioning by treatment options following surgery.

Table 29 summarises the partitioning of IFX responders based on the study of Imaeda et al. ,99 discussed in Chapter 3, Analysis of correlation studies of tumour necrosis factor alpha/anti-drug antibodies level and response, that used concurrent monitoring for the absence or presence of IFX and antibodies to IFX.

The proportions of IFX responders with various trough levels of IFX were based on Vande Casteele et al. 73 (discussed in Chapter 3, Vande Casteele et al.: Trough level Adapted infliXImab Treatment study73). These authors screened a cohort of patients with IBD who were receiving maintenance IFX treatment, and further categorised patients by drug concentration based on test result. Drug levels < 3 µg/ml were considered below the target range, levels between 3 and 7 µg/ml were considered within range and those > 7 µg/ml were above target range. Table 30 shows the proportions of responders with different trough drug levels and the proportions of responders derived from this study.

The partitioning of IFX patients with LOR according to concurrent test monitoring of IFX and antibodies to IFX was based on information obtained from Steenholdt et al. 123 Table 31 summarises these proportions.

Patients who have undergone surgery may receive post-operative treatment to maintain remission. These options include an anti-TNF-α, an immunosuppressant, a combination of an anti-TNF-α and an immunosuppressant or no treatment. Table 32 shows these proportions based on the study of van der Have et al. 160

Table 33 summarises the proportions of IFX responders based on the study by Imaeda et al. ,99 in which reflex testing was used to test for the absence or presence of IFX. In patients in whom IFX was present, we used the proportions according to IFX trough levels based on the Vande Casteele et al. 73 study, as shown in Table 30.

The partitioning of IFX patients with LOR according to reflex test monitoring of IFX was based on information obtained from Steenholdt et al. 123 Table 34 summarises these proportions.

Time-to-event transition probabilities

Table 35 summarises the transition probabilities for time-to-event outcomes used in the models.

Sensitivity analysis

In addition to our base-case analysis, we have undertaken a number of sensitivity analyses. These analyses are summarised below:

1. undertake concurrent testing and reflex testing every 12 months in the responder and LOR models

2. estimate the mean costs and effects associated with each strategy using a 1-year time horizon with 4-week cycle lengths

3. in the responders model – three possible modes of one-off testing:

   1. one-off testing at 3 months followed by yearly retesting

   2. one-off testing at 3 months and one retest for those who regained response

   3. one-off testing at 3 months and no retesting for responders/regained response.

4. in the LOR model – 3-monthly testing for patients with LOR; no testing for patients who have regained response

5. no regain of response following best supportive care (responders)

6. no regain of response following best supportive care (LOR).

Probabilistic sensitivity analyses

Probabilistic sensitivity analyses were undertaken to determine the joint uncertainty in key model input parameters of test results and expected QALYs. The PSA was undertaken based on the outcome of cost per QALY only. In PSA, each model parameter is assigned a distribution reflecting the amount and pattern of its variation, and cost-effectiveness results are calculated by simultaneously selecting random values from each distribution. The distributions used in the PSA are presented in Table 36. We have calculated probabilities that each strategy is the most cost-effective, at a willingness to pay of £20,000/QALY.

Search strategy

Search strategy