Randomised controlled trial of tumour-necrosis-factor inhibitors against combination intensive therapy with conventional disease modifying anti-rheumatic drugs in established rheumatoid arthritis: The TACIT trial and associated systematic reviews

Key impacts

Rheumatoid arthritis (RA), one of the commonest disabling diseases in the UK, remains a major health-care problem. 1–3 It affects almost 1% of UK adults and is more common in women. There are two peak ages of onset, early adulthood (mainly women) and later life (equal sex distribution). There are internationally accepted classification criteria for RA; from time to time these been revised and modernised. 3–6

Its main impacts are increasing disability and reduced quality of life. 7 Both are substantial and persistent and reflect the combined effects of persisting joint inflammation, progressive joint damage and extra-articular features of RA. 8 Another significant impact of RA is reduced life expectancy, which is mainly due to associated comorbidities such as coronary artery disease. 9 The final major impact of RA is the substantial costs in terms of medical and social care and lost employment. 10

Disease course and outcomes

The primary clinical feature of RA is chronic, usually persistent, inflammatory synovitis, initially mainly affecting the small joints of the hands and feet but subsequently spreading to involve multiple other joints. 7 Without adequate treatment many patients will develop joint damage, classically erosions, but also joint space loss and secondary osteoarthritis. 11 In addition, RA may be associated with extra-articular features, such as nodules and interstitial lung disease,8 and with comorbidities, such as the increased risk of cardiovascular disease and infection. 12

Diagnosis combines clinical features with laboratory tests such as acute phase markers [erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP)],13 rheumatoid factor, anticyclic citrullinated peptide (anti-CCP) antibodies14–16 and imaging [ultrasound, magnetic resonance imaging (MRI) and/or radiography]. 17–19 Definitive differentiation from other forms of inflammatory arthritis is difficult in early arthritis but usually uncontroversial in established disease.

The outcome of RA is highly variable, ranging from mild disease with limited impact on a patient’s life to severe unremitting disease unresponsive to treatment. Some features are known genetically and epidemiologically to be associated with a poorer outcome, including specific human leucocyte antigen genotypes, smoking and the presence of anti-CCP antibodies. 20–22 However, it has proved difficult to develop an outcome predictor at the level of the individual patient, which would be required to develop tailor-made individual treatment regimens.

Disease costs

Rheumatoid arthritis results in high medical and social costs,23 with drug costs a significant part of the economic cost. Conventional drugs are relatively inexpensive whereas newer biological agents are very expensive; over time, drug costs have risen substantially. A second cost component is other medical care. These costs are modest in the short term but rise substantially when surgical treatment or supportive long-term medical treatment is needed for disabling severe RA or for comorbid disease. The final costs are societal costs. These include loss of work, support from family and carers and costs of care within the community. These societal costs usually exceed medical expenses and rise with disease duration and severity.

Historically, in the period before biological treatments were available, the direct and indirect costs were estimated to be in the region of £55–70M per million of the population,24 with a total disease cost of £4B for the UK as a whole. 25 Since the introduction of biological treatments, drug costs have increased substantially. A report by the National Audit Office in 2009 estimated that RA costs the NHS around £560M a year in health-care costs, with the majority of this in the acute sector. 26 This report estimated that the costs to the NHS of biologics for treating RA were around £160M annually. As biologics prescribing for RA has continued to increase, the current costs are likely to be substantially higher but may be balanced by reductions in other medical costs, such as orthopaedic interventions for RA, if high-cost drug treatments improve medical outcomes. The National Audit Office report also estimated that the additional cost to the UK economy of sick leave and work-related disability for RA is £1.8B a year.

Assessments

Assessments in RA mainly look at joint inflammation. Clinical-based assessments include swollen and tender joint counts and global assessment, which estimates overall disease activity and health status. Standard joint counts focus on 28 joints in the hands, upper limbs, and knees. Some experts prefer extended 66 and 68 joint counts; these include the feet. Laboratory measures include the ESR, CRP or both. Patient-based measures span pain, global assessment and disability. 27–29 The Health Assessment Questionnaire (HAQ) measures disability. 30 Other areas, such as fatigue and depression,31,32 are very relevant to patients but are not always formally assessed. Patient-based measures are especially important because they measure an individual’s perspective of the burden of their RA.

A number of combined indices amalgamate individual assessments. A widely used combined index is the Disease Activity Score for 28 Joints (DAS28), which combines the numbers of swollen and tender joints (hands, arms and knees) out of a total of 28, a patient’s global assessment and the ESR to indicate a patient’s current status. 33 As calculating the DAS28 involves a complex mathematical formula, simplified variants have been devised. 34 The Simplified Disease Activity Index uses the number of tender and swollen joints (out of a total of 28), doctors’ and patients’ global assessments and CRP level. The Clinical Disease Activity Index is similar but omits CRP level. The American College of Rheumatology (ACR) improvement criteria, which gauge change in status in clinical trials, include falls in joint counts and several other measures (patients’ and doctors’ global assessments, ESR, pain and HAQ). They record 20% (ACR20), 50% (ACR50) and 70% (ACR70) improvements in five of the seven measures. 35

Juxta-articular erosions characterise progressive, established RA and are usually irreversible. They can be readily identified on radiographic images of the hands and feet. Two typical erosions are sufficient for diagnosis. 36 Extensive damage seen on radiographs suggests that RA is inadequately controlled. Rapid progression of joint damage needs intensive treatment. 37 Several scoring systems are used to quantify damage seen on radiographs in research studies. Although new imaging modalities such as ultrasound and MRI can assess structural changes, they are not yet widely used except in research. 38

Treatment goals

The overall treatment goal is making patients feel better and minimising the impact of RA on their lives. 39 The main immediate treatment goal over the last two decades has been to reduce disease activity. Reducing joint and systemic inflammation is beneficial in itself. Crucially, it is also associated with other benefits including decreased disability, improved quality of life and reduced progression of joint damage. A dominant theme has been to treat patients with active RA; in the main, current treatments mean that few patients now have persisting active disease.

More recently there has been a shift towards making remission the main goal. An ideal treatment would result in the majority of patients achieving remission with no active joint inflammation and no functional deterioration or erosive progression. 40 Although 10–50% of patients with early RA can achieve remission,41 only a small minority of established RA patients achieve sustained remission. An associated difficulty in determining the frequency of remission depends on how it is defined and the intensity of treatment. 42

Relatively cheap, readily available disease-modifying antirheumatic drugs (DMARDs) such as methotrexate have made major inroads into managing active RA. DMARDs were initially given as monotherapies but in recent years there has been greater emphasis on using combinations of two or more DMARDs as this has been shown to be more effective in disease control. 43

Since the mid-1990s a new treatment approach has been developed – the use of targeted biological treatments. They are usually given in combination with methotrexate or other DMARDs. Biologics have revolutionised the treatment of severe RA, for which they appear highly effective. A major limiting factor is their high cost. 44

Reducing disease activity appears a clear-cut well-defined goal. However, the degree of reduction required for a good ultimate outcome is not yet known. Intensive treatment aimed at inducing remission45 appears an inevitable next step. However, it is not clear whether or not this is appropriate for every patient. Furthermore, there remains uncertainty about the appropriate definition of remission in RA. 46

Conventional disease-modifying antirheumatic drugs

Disease-modifying antirheumatic drugs are a diverse range of drugs. 47 They form a single group because they both improve symptoms and also, to a greater or lesser extent, modify the course of the disease. This means that they reduce the progression of erosive joint damage and decrease disability. 48,49

Many drugs have some features of DMARDs but only a few have been accepted into clinical practice. The use of DMARDs varies, with a small number being particularly favoured. The current situation is summarised in Table 1. At present, methotrexate is the dominant DMARD because of its greater efficacy and retention compared with other DMARDs. 50 As the most widely used DMARD, methotrexate is now considered by regulatory agencies as a benchmark against which new agents must be tested. The majority of RA patients treated with DMARDs in most UK specialist units either are currently taking or have previously received methotrexate. Sulfasalazine, leflunomide and hydroxychloroquine (the last largely as part of a combination regime) are the only other DMARDs used to any appreciable extent in the UK. 51

The efficacy of DMARDs involves reduced features of joint inflammation, such as fewer swollen joints and a lower ESR, a reduction in the progression of joint damage, particularly erosive damage, decreased levels of disability and improved quality of life. The harms, or adverse events, related to DMARDs include common problems seen with most DMARDs such as low white cell or platelet counts and unique toxicities with specific DMARDs. There is a reasonable evidence base for their use as monotherapies. 52–58

Steroids

The commonest use of steroids in RA is as adjunctive agents to control disease flares; they may be given intra-articularly, intramuscularly or orally. Because in early disease it has been suggested that steroids exert a disease-modifying effect, they form an initial but temporary component of several early arthritis combination regimens. They are also widely used as part of intensive DMARD combination therapy regimes in patients with uncontrolled established disease. There is a reasonably strong evidence base for their use. 59–61

Disease-modifying antirheumatic drug combinations with and without steroids

Disease-modifying antirheumatic drugs can be used in combination (see Table 1). This approach, initially advocated by McCarty,62 has been examined in many clinical trials. Initial studies evaluated combinations that turned out to have excessive toxicity (gold–hydroxychloroquine)63 or limited efficacy (methotrexate–azathioprine). 64 This toxicity led early reviews to suggest that risk/benefit ratios were unfavourable compared with monotherapy. 65

However, the situation changed when randomised controlled trials (RCTs) of methotrexate–ciclosporin,66 methotrexate–sulfasalazine–hydroxychloroquine67 and methotrexate–sulfasalazine–steroids68 reported improved disease control in active RA with mild or no excess toxicity; similar results were obtained in subsequent combination therapy studies. Combination DMARDs (cDMARDs) may not be required for all RA patients. In the only RCT of mild, early RA patients on stable DMARD monotherapy, they did not add benefit. 69

Overall, from our 2005 systematic review,70 and as suggested by a gradual expansion of its use in routine practice,71 the benefits of combination therapy are now thought to outweigh the risks in patients with active disease not controlled by monotherapy. They are recommended in UK national guidelines72 for early RA patients with active disease to avoid delay in bringing the disease under control, which is known to be associated with a poor outcome.

The RCT evidence for using DMARD combinations is of crucial importance to the Tumour necrosis factor inhibitors Against Combination Intensive Therapy (TACIT) trial. It is summarised in detail for both early and established RA in two systematic reviews (see Chapter 3).

Tumour necrosis factor inhibitors

These agents were developed in the late 1980s to target tumour necrosis factor (TNF)-α, a cytokine of central importance in the pathogenesis of RA, which exerts its effects by binding to type 1 and 2 receptors on immune, inflammatory and endothelial cells in the lymphoid system and joints and in less well-studied systems such as the central nervous system. 73

The proof of principle for inhibiting this cytokine came from an open-label clinical study in which patients with RA received a single infusion of a tumour necrosis factor inhibitor (TNFi). Patients showed a rapid response, including an early fall in CRP level. However, the anti-inflammatory effect lasted only 6−12 weeks and was followed by a return of active disease. 74 As a result, patients were retreated with further infusions; these showed responses of similar magnitude and duration. 75 The scene was set for a major clinical development programme.

There are currently five TNFis available to treat inflammatory arthropathies, summarised in Table 2. All have been shown to be effective in large clinical trials, which have been collated in systematic reviews. 76–80 These TNFis can be subdivided into first-generation agents (comprising etanercept, infliximab and adalimumab) and second-generation agents (comprising certolizumab and golimumab). In RA, all of these agents are licensed for use in routine clinical care; they are also approved by National Institute for Health and Care Excellence (NICE) for use in the NHS although in some cases this has required a financial risk-sharing agreement. 81–83

There is no clear-cut evidence that any one of these agents is superior to any other, and practical issues, including cost, determine which is chosen. There have been network meta-analyses of the efficacy and toxicity of different TNFis and these suggest potential minor differences in efficacy and adverse event risks. 84–86

Infliximab must be given concurrently with methotrexate (or another DMARD in methotrexate-intolerant patients) to prevent the formation of human antichimeric antibodies. 87 The licence for adalimumab also requires concomitant methotrexate unless the patient is intolerant. Although concomitant treatment is not required for etanercept, substantial data suggest that combination treatment is more effective, especially in terms of the effect on bone erosion. Therefore, all three drugs are almost always given with methotrexate or another DMARD. 88

The RCT evidence for using TNFis in combination with methotrexate and other DMARDs is also of crucial importance to the TACIT trial. This evidence is also summarised in detail in the systematic reviews in Chapter 3.

The question of what to do when a TNFi failed was a crucial question, particularly in the early 2000s when other biologics were not available. There is only limited information about the relative merits of switching from one TNFi to another. The only RCT studied golimumab in patients who had failed another TNFi; this showed some benefit from the switch. 89 The relative benefits of switching TNFis in patients who, for one reason or another, have not responded to their first biologic has also been addressed using observational data from registries and similar studies. Again, these studies provided some evidence that switching TNFis can give clinically useful improvements although response rates for second and subsequent TNFis are lower than for first-time use. 90

More recently, several trials evaluating non-TNF-targeted biologics, including abatacept, rituximab and tocilizumab, have provided convincing evidence that non-TNF-targeted biologics are effective in patients who have failed with TNFis, and this is increasingly the preferred approach. 91,92

Other new agents

A number of other biological treatments have been licensed, and in some cases approved by NICE, for treating RA. An early agent, anakinra, which is an interleukin-1 (IL1) receptor protein, is relatively ineffective93 and is not often used for treating RA. It is, however, highly effective in a range of other disorders including acute gout, some forms of juvenile arthritis and some familial periodic fevers. Further anti-IL1 agents are in late-stage development, currently for these indications.

Rituximab targets B cells and is highly effective in active RA. 94 Its mechanism of action is controversial as the presence of rheumatoid factor is not essential for its efficacy. Tocilizumab targets IL-6 and is also highly effective in active RA. 95 The third effective biological treatment, abatacept, targets costimulatory molecules on T lymphocytes. 96 Although some of these other biological treatments are licensed to be used as first-line treatment in methotrexate incomplete responders with RA, network meta-analysis and similar comparative studies84,85 show that these different biologics have comparable efficacy. TNFis are the most widely used treatment in methotrexate incomplete responders. Therefore, comparing cDMARDs with TNFis will provide results of general interest.

Several new non-biological agents such as kinase inhibitors97,98 are being developed. One of these agents, tofacitinib, has been licensed in the USA99 but is not yet approved for use in Europe. Depending on their cost, and relative efficacy and toxicity, such orally active agents may also change the treatment pathways for RA. However, for the present their roles are uncertain.

Non-TNFi biologics and new oral DMARDs are not part of the TACIT trial. Most of their actual or projected use is for patients who have failed to respond to both conventional DMARDs and TNFis. This late-stage treatment pathway for RA, which is complex and less well defined than the earlier management stages for RA, is not the key focus of the TACIT trial. We have therefore not reviewed it in detail.

Supportive and symptomatic treatment

As with all long-term disorders the management of RA requires multiple inputs from a range of health-care professionals from primary and secondary care. Patients need to be fully informed about their condition and be able to access advice; this is one of the key roles of the specialist nurses. Patients need effective treatment for pain, using analgesics and non-steroidal anti-inflammatory drugs,100,101 and their comorbidities, notably ischaemic heart disease, need to be appropriately managed. 102 Finally, they need access to physiotherapists and in some cases occupational therapists and need to be encouraged to take regular exercise. 103,104 The appropriate use of all of these treatments is crucial to ensure a good outcome. However, they are outside the focus of the TACIT trial and so have not been considered in detail.

Treat to target

There is evidence that intensive treatment is important in early RA both to suppress disease activity105–109 and to maintain low disease activity when it has been reduced. Welsing et al. 110 investigated the longitudinal relationship between disease activity and radiological progression in two independent follow-up cohorts. Both showed significant relationships between disease activity and radiological progression, but only in patients seropositive for rheumatoid factor. The results support systematic monitoring to achieve persistent low disease activity. This approach, termed ‘tight control’ or ‘treat to target’, includes several standard procedures such as:

• a predefined treatment protocol to which the treatment of individual patients is adjusted

• being able to assess whether or not the treatment chosen is necessary and effective

• incorporating measures to ensure that patients are not overtreated.

Many groups have reported on aspects of tight control. 111–114 Most used the Disease Activity Score (DAS) or DAS28 to guide treatment or as the primary end point. Overall clinical and radiological outcomes were more favourable in patients receiving tight-control regimens; in particular, remission rates were generally higher with tight control than with conventional therapy. These improved clinical and radiological outcomes did not appear to be at the cost of increased drug toxicity.

Access to high-cost treatments

Different countries have taken divergent approaches to the use of high-cost treatments such as biologics. A range of international groups, specialist societies and regulatory bodies recommends TNFis for patients with active RA who have failed to respond to conventional DMARDs. 115–118 The current UK consensus recommends that TNFis are started only in patients who have a DAS28 of > 5.1115 and who have failed to respond to adequate therapeutic trials of two standard DMARDs including methotrexate. 119 Some UK experts believe that the threshold for active RA should be reduced to a DAS28 of > 3.2 with at least three or more tender joints and three or more swollen joints. 120 There are major national differences in the guidelines followed by rheumatologists for starting biological treatments and considerable diversity in biological treatment use across Europe. 121–124

Economic modelling and biological treatments

Economic modelling conventionally extends beyond conventional RCTs,125,126 bringing together cost and outcome evidence from a range of sources. Several modelling methods are used including simple decision trees, Markov models and individual sampling models. Most economic studies evaluate the impact of biological treatments on quality-adjusted life-years (QALYs). In the absence of direct QALY measures, values may be inferred from other available outcomes. 127 Recent systematic reviews of health economic studies in RA highlight the different conclusions reached based on assessments of the much the same set of published evidence. Schoels et al. 128 identified 21 relevant studies of biological treatments and, based on willingness-to-pay thresholds of US$50,000–100,000 per QALY, they concluded that combinations of TNFis with methotrexate were cost-effective after conventional DMARD failure. The sequential use of TNFis has been a difficult problem to resolve; however, Brennan et al. 129 reported favourable incremental cost-effectiveness ratios (ICERs) for using a second TNFi compared with DMARD treatment. A different perspective was taken in a systematic review by van der Velde et al. 130 They concluded that the economic evidence suggests that biological treatments are not cost-effective compared with DMARDs for RA in adults at a cost-effectiveness threshold of C$50,000 per QALY and that there is mixed evidence of cost-effectiveness in selected populations at a willingness-to-pay threshold of C$100,000 per QALY.

Alternative approaches for accessing high-cost biological treatments in rheumatoid arthritis

Different groups of experts have reached widely differing views on when biological treatments in general and TNFis in particular should be started in RA. Almost all experts recommend usually using them in combination with methotrexate or other DMARDs. There is also universal agreement that they should be reserved for active RA, although there is uncertainty about what constitutes active disease.

Most experts recommend that they are used in patients who have failed to respond fully to methotrexate and who continue to have active RA. Many trials have been undertaken in such patients with positive findings. Most countries in continental Europe and North America have adopted this approach. The UK is more conservative and TNFis are generally used after patients have failed two DMARDs and still have active disease.

There are many alternative ways in which TNFis could be used in RA. One approach, which might be the most effective and cost-effective, is to reserve them until RA patients have tried and failed to respond to intensive DMARD combination treatments. Such an approach would place TNFis slightly lower down the therapeutic cascade; however, this might be of greater overall benefit to the health service by optimising the use of resources without causing patients any problems. This option is specifically explored in the TACIT trial.

Limiting the use of high-cost treatments has always been a component of Western health care. No country allows universal access to all high-cost treatments for all patients who would like to have them. It is equally important to ensure that patients are not denied effective treatments. If intensive cDMARDs are effective in some patients and the use of biological treatments can be changed so that they are given to patients most likely to show substantial benefits, the needs of both patients and health-care funders can be met. TNFis are not universally effective in RA; they have a response failure rate in the region of 20–30% of patients. 131–134 Some patients who would fail TNFis may respond to intensive cDMARD regimens. As the number of treatment options is limited in RA, and as patients rarely move from biological to non-biological treatment, there is a sound logic in exhausting treatment options in an organised sequence. If TNFis were curative such an argument would be misplaced; however, the balance of evidence indicates that DMARDs and biological treatments are both suppressive therapies that do not appear to alter the long-term clinical phenotype.

Systematic reviews of intensive treatments

The two treatment strategies used in the TACIT trial – cDMARDs and TNFis given in combination with methotrexate or another DMARD – have both been studied in RCTs in RA. Systematically reviewing these previous trials is crucial for both designing the TACIT trial and interpreting its findings.

We have previously published three systematic reviews of combination therapy trials: one looked at all combinations at all time points;70 the second compared DMARD combinations and TNFi/methotrexate combinations in early RA;135 and the third specifically examined the toxicity of combination therapies. 136 A number of other groups have also published systematic reviews of both cDMARDs and TNFis. 65,85,137–144 These systematic reviews all show that both DMARD combinations and TNFi/methotrexate combinations are effective in both active early and established RA. Their effectiveness in clinical trials is broadly comparable, although there are insufficient head-to-head clinical trials of these two treatment strategies. Overall, these published systematic reviews provide strong clinical support for undertaking a head-to-head trial such as the TACIT trial.

In early RA the relative clinical effectiveness and cost-effectiveness of conventional DMARD combinations have been considered in detail in guidance from NICE. 72 This guidance recommends cDMARDs, including methotrexate, as first-line treatment for early active RA. Biological treatments were excluded as first-line therapy by NICE because they were not considered cost-effective in these patients. Not all guidance accepts this conclusion; for example, the ACR advises the use of biological treatments (TNFis) combined with methotrexate as one initial therapy for active early RA patients with a poor prognosis118 as an alternative to cDMARDs.

In patients with established RA disease the overall value of intensive DMARD combination regimens compared with TNFi/methotrexate combinations is less certain, particularly as these patients will usually have failed one or more previous DMARDs. The TACIT trial is aimed at these patients because there is genuine uncertainty about the relative merits of cDMARDs in such cases.

Hypothesis

The TACIT trial was designed to test the hypothesis that patients with active RA who meet the NICE criteria for treatment with TNFis will gain equivalent benefit over 12 months at substantially less expense and without increased toxicity from starting treatment with intensive combination therapy with DMARDs.

Testing the hypothesis

The TACIT hypothesis would be rejected if the primary outcome measure – the HAQ – showed substantial clinically important improvements at 12 months in patients randomised to receive TNFis. The TACIT trial was designed to confirm or refute the equivalence of treatment with TNFis and cDMARDs in improving HAQ scores over 12 months.

The TACIT hypothesis would also be either rejected or substantially weakened if economic evaluations at 12 months – including health and social care costs, societal costs, cost-effectiveness and cost–utility – showed disadvantages in patients randomised to receive cDMARDs or if the adverse event profile was substantially worse with cDMARDs.

The TACIT trial also collected a range of secondary outcomes to help evaluate the clinical usefulness of cDMARDs in these patients. These included assessing joint damage, quality of life, disease activity using the DAS28 and retention rates on such DMARD treatment. We considered also collecting other outcomes such as ACR response rates and responses on other indexes such as the Simplified Disease Activity Index. However, we concluded that measuring the same outcome in multiple ways, particularly using methods not followed in the UK where the trial is based to inform routine practice, would be counterproductive.

Systematic reviews

In the last few years more trials have been published about DMARD combinations and TNFi/methotrexate combinations. To ensure that the results of the TACIT trial can be placed into an appropriate context it is essential to provide an updated systematic review.

As there are different issues in early RA and established RA we have undertaken two reviews of these different aspects of RA treatment. As it is crucial to define whether or not there are differences in these two clinical settings in the relative efficacy of cDMARDs and TNFis, we used similar methods in both reviews. Our analytical approach is also comparable.

Consequently, the two systematic reviews assess the efficacy and toxicity of combination treatment with both cDMARDs and TNFis with methotrexate in the two groups of RA patients. The first group was patients with early disease, which is disease of < 3 years’ duration. The second group was patients with established disease who have failed one or more DMARDs. The reviews evaluate treatments in trials that compared (1) cDMARDs with DMARD monotherapy; (2) TNFis plus methotrexate with methotrexate monotherapy; and (3) cDMARDs with TNFis plus methotrexate. The trials that enrolled patients with early RA were analysed separately from the trials that enrolled patients with established RA.

Eligibility criteria

The trial was aimed at patients with RA attending outpatient rheumatology clinics in England who met the current NICE criteria for receiving TNFis. 81

Tumour necrosis factor inhibitors

The three licensed agents available when the trial started – adalimumab, etanercept and infliximab – were allowed at standard doses (British National Formulary148). The choice of TNFi reflected patient preference and local circumstances. Methotrexate was also given to maximise efficacy and (in the case of infliximab) reduce the formation of antichimeric antibodies. Patients intolerant to methotrexate took another DMARD. DAS28 at 3 and 6 months defined responses to therapy.

Patients had their TNFi stopped for one or more of three reasons:

• lack of effect as defined by the NICE criterion,81 that is, a change in DAS28 of < 1.2 at 3 or 6 months

• an adverse event that, in the opinion of the supervising specialist, necessitated treatment withdrawal

• patients could stop therapy for any reason should they wish (reasons to be specified if patient willing).

Patients in whom one TNFi was stopped were able to start another. This option represented current UK practice when the trial started. Patients who failed two TNFis for whatever reason were not able to start a third agent and required alternative treatment such as cDMARDs.

The principles of the treatment algorithm were as follows:

1. start a TNFi of choice on the basis of local circumstances and patient preference

2. assess at 3 months: no change if good response (change in DAS28 ≥ 1.2); change to second TNFi if change in DAS28 is < 1.2

3. assess at 6 months: no change if good response (change in DAS28 ≥ 1.2); change to second TNFi if change in DAS28 is < 1.2; if two biologics already given and DAS28 change is < 1.2, TNFi stopped and patient offered DMARD combination or other therapy.

Combination disease-modifying antirheumatic drugs

Those cDMARDs with proven efficacy over DMARD monotherapy in RCTs were used, including:

• triple therapy with methotrexate (methotrexate–sulfasalazine–hydroxychloroquine)

• other methotrexate combinations (methotrexate–ciclosporin, methotrexate–leflunomide and methotrexate–gold)

• a sulfasalazine combination (sulfasalazine–leflunomide)

• additional monthly steroids [intramuscular Depo-Medrone (120 mg stat) or equivalent] were used if needed.

The DMARD combinations were stopped for three reasons: adverse events, patient-initiated withdrawals (which are identical to those reasons for stopping a TNFi) and lack of effect (change in DAS28 < 1.2), which is similar to that for TNFis but was implemented only at 6 months.

The principles of the treatment algorithm were:

• initially: maximise initial DMARD/optimise administration (e.g. parenteral methotrexate); start second/third DMARD; give intramuscular Depo-Medrone^® (methylprednisolone, Pfizer) (whenever possible)

• second step: maximise dose of second/third DMARD

• third step: change combination (repeated if needed)

• additional option: continue with intramuscular Depo-Medrone monthly short term if RA remains active

• assess monthly and change treatment if change in DAS28 is < 1.2 or DAS28 is > 3.2

• at 6 months start a TNFi if change in DAS28 is < 1.2.

The target doses of different DMARDs used in combinations were as follows:

• methotrexate: 25 mg weekly – preferably by intramuscular injections although could be oral (achieved by 5-mg increments)

• sulfasalazine: 3 g daily (starting at 500 mg daily and increasing by 500-mg increments)

• hydroxychloroquine: 400 mg daily (starting at 200 mg and increasing as one increment)

• ciclosporin: 3.5 mg/kg (starting at 2 mg/kg and increasing incrementally depending on creatinine levels)

• leflunomide: 20 mg/day (staring at 10 mg/day and not increasing if used in combination with methotrexate)

• gold: 50 mg/month (starting with test dose, then 50 mg/week for 20 weeks, then 50 mg/month)

• intramuscular Depo-Medrone: 120 mg/month for 3 months; further courses were given if the RA was still active.

Dose adjustments to all drugs depended on both disease activity and evidence of adverse events. Decisions about changes in treatment were made by the supervising rheumatologist but were reviewed by the principal investigator (DS) or deputy to ensure that the algorithm was followed.

Concomitant therapy

Non-opiate analgesics and non-steroidal anti-inflammatory drugs were used as needed at standard doses. Patients taking methotrexate also received folic acid (5 mg/week) to limit adverse events. Patients taking steroids received bone protection (e.g. alendronate and calcium/vitamin D). Other drugs (e.g. antihypertensives) were used as needed. Patients taking oral prednisolone up to 10 mg at entry stayed on treatment. Intra-articular steroids were used as required.

Safety monitoring

Safety monitoring followed national guidelines with monthly blood counts and liver function tests plus measurement of renal function (creatinine), urinalysis and blood pressure recording for some DMARDs. 149–151 Patients were screened for tuberculosis.

Primary clinical outcome

Patient self-assessed outcomes were used in the TACIT trial. 152 The HAQ was the primary clinical outcome measure. 153 Although it is sometimes termed the HAQ Disability Index, most reports refer to it simply as the HAQ. The HAQ is a self-assessed questionnaire that is completed by patients. It primarily measures disability and is the dominant disability assessment in RA. 154 Scores range from 0 to 3, with higher scores indicating greater disability. It has established reliability and validity and has been used in many published RCTs in RA. HAQ scores were measured initially and at 6 and 12 months.

Secondary clinical outcomes

The European Quality of Life-5 Dimensions (EQ-5D)155 and the Short Form questionnaire-36 items (SF-36)156 were measured initially and at 6 and 12 months. These are also self-assessed questionnaires completed by patients. They measure health-related quality of life and can be used to estimate health utility and have been extensively studied in RA. 157–160

Plain radiographs of the hands (including the wrists) and the feet were taken initially and at 6 and 12 months. These are widely used outcome measures. 161 Digital images of the radiographs were read at the end of the trial by a single observer (DS) experienced in reading radiographs using the Larsen score,162 modified for minor changes. 163 The radiographs were assessed in known date order without knowledge of the treatments that patients had received.

Disease activity scores for 28 joint counts were measured initially and every month throughout the trial. 164 Scores are calculated using tender joint counts and swollen joint counts for 28 joints assessed by trained specialist nurses, the ESR and patients’ global assessments of their disease activity recorded on a 100-mm visual analogue scale (VAS). DAS28 was used to guide treatment based on the predefined treatment targets and to assess responses to treatment.

Adverse effects were recorded each month by patient reporting. Specific events such as hospital admission were also recorded following international guidance. 165

International recommendations

These outcome measures have all been recommended by international bodies including the European Medicines Agency166 and the US Food and Drug Administration. 167

Health economic assessments

An adapted version of the Client Service Receipt Inventory (CSRI) was used168 to measure individual-level resource use. It covered sociodemographics, the use of (all-cause) secondary and community-based health and social care services, time off work because of illness, receipt of social security benefits and medication prescribed in addition to the study treatments. The CSRI has been previously used successfully in trials in arthritis. 169 We also measured take-up rates (without intention to attribute costs) for NHS/social services contributions towards more exceptional resources, such as special mobility equipment, adaptations to the home or transport to health care. The CSRI was administered as a self-complete questionnaire retrospectively for the previous 3-month period at three assessment points: baseline, 6 months after randomisation and 12 months after randomisation. Use of trial medications was recorded separately and prospectively by the clinical/research teams over the entire study period in the form of medication name, dose, frequency and duration of use.

As discussed earlier, health-related quality of life was assessed by self-report questionnaires at baseline, and 6 and 12 months using the EQ-5D and SF-36 for the purpose of estimating QALY gains.

Data collection

Patient details and outcomes, with the exception of radiographic outcomes, were collected using an academic online database system with direct entry of data into the electronic case record form (see www.medscinet.net/tacit). This electronic data capture (EDC) system collected information anonymously using patients’ initials and date of birth as identifiers.

Recruitment

Patients were recruited from rheumatology clinics in England, which were divided into three sectors: London and the south of England, the Midlands and the north of England.

Potentially eligible patients were identified by rheumatologists and clinic nurses at the participating centres. Rheumatologists approached potentially eligible patients and outlined the trial to them. Patients interested in participating were given the patient information sheet and were then contacted by telephone at least 24 hours after receiving the patient information sheet to see whether or not they were interested in participating. If they were, a screening assessment was arranged.

Screening involved making the following checks against NICE guidance to ensure that patients were eligible to receive TNFis:

• ensure that patient has failed adequate treatment with two DMARDs and has a DAS28 > 5.1

• negative screen for tuberculosis including chest radiography (and other local measures such as Mantoux testing where applicable)

• repeat DAS28 assessment 4 weeks after the initial DAS28 assessment to ensure that it remains > 5.1.

The screening assessment pages were collected anonymously using the EDC system. Once complete and eligible, the EDC system automatically assigned consecutive patient numbers to patients in chronological order.

Randomisation

Randomisation numbers were formed of four numbers and prefixed by the region identifier (i.e. 1 for London and the south, 2 for the Midlands and 3 for the north). The allocation sequence for randomisation was generated by the EDC system. Block randomisation was used in blocks of four with allocation balancing. Randomisation was stratified by region. Formally, the trial was designed to test the null hypothesis of a difference of > 0.22. With a (one-sided) testing level of 5%, a sample size of 176 was required to achieve 90% power. To allow for a dropout rate of 5–7%, we planned to recruit 190 patients. The clinicians at each of the trial centres and the trial co-ordinator were unaware of the allocation sequence.

Once a randomisation number was allocated, the EDC system automatically informed the researcher at the individual centre and the trial co-ordinator by e-mail. The trial co-ordinator informed the pharmacy at site of the randomisation. The patient was then informed that they had been recruited to the trial and the baseline assessment was arranged.

Data collected during screening

As part of screening data were collected on:

• number of patients who were potentially eligible to receive TNFis

• reasons why patients chose not to enter the clinical trial (insufficient disease activity, consent not given and other reasons)

• numbers of patients randomised.

Baseline assessment

Delays between screening and baseline were anticipated because of the pragmatic nature of the trial and local practices relating to the supply and delivery of TNFis. It was recognised that patients may therefore require additional treatment between the screening assessment and the baseline assessment, which would usually be an intramuscular steroid injection. The following rules were therefore applied:

• patients were given an appropriate dose of intramuscular steroid if needed

• the baseline assessment was delayed for 1 month after the date of injection.

Eligibility based on a DAS28 of > 5.1 at screening was not required to be maintained at the baseline assessment as this is not a requirement for receiving TNFis in routine practice.

Blinding

The TACIT trial was not blinded and both clinicians and patients knew to which treatment strategy patients had been allocated.

Recruitment and follow-up patterns

Recruitment was recorded by year and region. The numbers of patients enrolled – excluding patients who had been withdrawn from therapy and who were unwilling to continue follow-up – were reported by treatment arm. The numbers of patients who withdrew from therapy, who were lost to follow-up or who died while taking part in the study were also reported by treatment arm.

Baseline comparability

Baseline characteristics were summarised by randomised group. Summary measures for the baseline characteristics of each group have been presented as means and SDs for continuous (approximate) normally distributed variables, medians and interquartile ranges (IQRs) for non-normally distributed variables, and frequencies and percentages for categorical variables.

Intention-to-treat population

Except for enrolled patients who withdrew consent or who were found to be ineligible at the baseline visit and so never received any treatment and for whom no data were therefore available, analyses on an intention-to-treat (ITT) basis reflect the randomisation process. We also carried out two additional analyses on the following populations:

1. a complete-case population: these were observations that subjects completed without missing data or violation of the protocol; this analysis is therefore referred to as a ‘complete-case analysis’ throughout this report

2. a per-protocol population: these were observations that excluded those patients who were found to deviate from the protocol.

The allowed variations to the protocol are shown in Appendix 1. The results of the per-protocol analysis were similar to those of the ITT analysis. Therefore, the results of the ITT and complete-case analyses have been presented in this report.

Imputing missing data

All participants had observations at baseline. However, some subjects had missing data on the outcome variables at 6 months, 12 months or both. The outcome variables that were measured at baseline and 6 and 12 months (HAQ, SF-36, EQ-5D and Larsen score) were imputed under different assumptions from those used for the DAS28 and its components, because DAS28 was measured monthly.

All missing data were imputed regardless of the reasons why it was missing. For the subjects who had missing outcomes, the baseline outcomes and other explanatory covariates (treatment group, sex, age, ethnicity, region and disease duration) were used to impute the missing data, assuming that unobserved measurements were missing at random.

For the subjects who had missing outcomes at 6 months, under the monotone assumption, baseline outcomes and explanatory covariates were used to impute these missing values. Then, for those patients who had missing outcomes at 12 months, baseline and 6-month outcomes with explanatory covariates were used to impute the missing values at 12 months. If the outcome variables were missing at 6 and 12 months then the outcome variables at 6 months were imputed first followed by the outcome variables at 12 months.

The DAS28 and its component were imputed using multivariate sequential imputation using chained equations. First, all missing values were filled in by simple random sampling with replacement from the observed values. The first variable with missing values, say DAS28 at month 1, was regressed on all other variables, DAS28–0, DAS28–2 through to DAS28–12, restricted to individuals with the observed DAS28–1. Missing values for DAS28–1 were replaced by simulated data points drawn from the corresponding posterior predictive distribution of DAS28–1. Then, the next variable with missing values was replaced using the same cycle. 176

The imputation was 20 cycles; at the end of the first cycle one imputed data set was created and the process was then repeated to create 20 imputed data sets. The 20 data sets were combined using Rubin’s rules;177–179 therefore, the estimates and standard errors presented here are the combined ones. As an additional check of the robustness of the analyses performed to the missing at random assumption, we further analysed the individual HAQ scores, EQ-5D scores, Larsen scores and DAS28 and its components) using the linear increments method of Diggle et al. 180 to handle the missingness. As the results obtained using this approach were qualitatively the same as those of the multiple imputation approach adopted, we report only the findings from the standard multiple imputation analyses.

Adjustment for design factors

Randomisation was stratified by region and therefore analyses of outcomes in the univariate or multivariable analyses were adjusted for region.

Outcomes assessed every 6 months

The primary outcome (HAQ score) and three of the secondary outcomes (EQ-5D score, SF-36 summary scores and Larsen score) were measured at baseline and 6 and 12 months. As there were not a significant number of zero values for the HAQ and other outcomes during follow-up, a linear regression model was used to analyse the change in these outcomes at 6 and 12 months. Thus, change was defined as either 12- or 6-month score minus baseline score. The unadjusted univariate analysis (model 1) was adjusted for region to account for design effect. The adjusted multivariable model (model 2) included sex, ethnicity, age, region, disease duration and baseline covariate as explanatory variables. Interactions between treatment and sex were assessed in the adjusted model 2 using the Wald test. The sex-specific interactions were not significant (for all outcomes p > 0.70). The treatment regression coefficient provided an estimate of the mean differences in HAQ, EQ-5D, SF-36 and Larsen scores.

For individual components of the SF-36 we used generalised estimating equations (GEEs) to estimate the effect of treatment including baseline values as a covariate for these outcomes. Working correlation matrices were unstructured, which was not unduly restrictive given that measurements were taken only at three time points. As the data were analysed longitudinally, time was included as a covariate in models 1 and 2. A final model tested specifically for interactions between treatment and sex and treatment and time using the Wald test. The sex-specific interactions were not significant (for all outcomes p > 0.50 in the overall test of all interaction terms). However, the interaction term between time and treatment was of borderline significance for some SF-36 domains (physical functioning, general health perception). We therefore report the period-specific treatment effect for those variables that had significant interaction terms.

Outcomes assessed every month

The DAS28 and its components were measured monthly and were therefore analysed separately. Changes in DAS28 and its components were analysed using GEEs to estimate the effect of treatment including baseline values as a covariate. Working correlation matrices were autoregressive with lag 1. In this analysis interactions between time and treatment and sex and treatment were also assessed and were found to be non-significant. Treatment effects were examined as subanalyses in two periods (1–6 months and 7–12 months). The estimates were presented as mean treatment effects (beta coefficients) with 95% confidence intervals (CIs). The sandwich estimator of error was used with the aim of obtaining robust estimates of precision. Statistical significance was determined at the 5% level using a two-sided test throughout. These analyses were based on the assumption that patients stayed in their original randomised treatment arm and thus ignored subsequent treatment switches.

Exploratory analyses

The patients randomised to start cDMARDs fell into two categories. The first category included those patients who remained on cDMARDs throughout the TACIT trial. The second category included those patients who switched to a TNFi after ≥ 6 months because they had not fully responded to cDMARDs. The outcomes of these two categories of patients have been compared in a series of exploratory analyses, recognising that these are non-randomised in their original treatment arm. These analyses were carried out for all populations (ITT, complete case and per protocol).

Four additional analyses used all observed data only and missing data were not imputed. This approach was taken when patients were divided into discrete response categories. These analyses comprised: (a) changes in Larsen score; (b) the development of new erosions shown by categorical increases in Larsen score; (c) clinical response to treatment indicated by a decrease in DAS28 of ≥ 1.2; and (d) achieving remission indicated by a DAS28 of ≤ 2.6. Analytical approaches used specifically with all observed data were the construction of Kaplan–Meier plots and a comparison of treatments using the log-rank test.

Toxicity

The proportion of serious adverse events was compared across randomised groups using Fisher’s exact test as appropriate.

Software specification

All data management and analyses were carried out using Stata version 12.0 (StataCorp LP, College Station, TX, USA) and the R statistical package (The R Foundation for Statistical Computing, Vienna, Austria181).

Costs

Unit costs were applied to resource use data to calculate cost per participant. Unit cost estimates, their sources and any assumptions made for their estimation are detailed in Appendix 2. Medication unit costs were converted into cost per mg based on the most cost-efficient pack size, choosing maintenance prices over initial treatment prices and generic prices over branded ones to obtain conservative estimates.

Total costs were computed for each participant at each assessment point from two perspectives: health and social care and societal. Health and social care costs included the costs of inpatient services, outpatient services, primary care services, other community-based services, social services, trial medications and other prescribed medications. Two sets of societal costs were calculated, one that included health and social care costs plus participant lost productivity because of absence from work and one that included health and social care costs, participant lost productivity because of absence from work and, additionally, the cost of social security benefit payments received.

For the economic evaluation, costs generated from the 3-month CSRI data were extrapolated (multiplied by 2) to cover the full 6 months before each follow-up point. All costs are reported in pounds sterling at 2010/11 prices. Discounting was not necessary as all costs were related to a 1-year period.

Outcome measures

Cost-effectiveness analyses were based on the primary outcome measure (the HAQ). Cost–utility analyses were based on QALYs derived from both the SF-36 and the EQ-5D. Utility weights appropriate to each measure were attached to the SF-36- and EQ-5D-produced health states at baseline and 6 and 12 months. 182,183 QALY gains between baseline and 6 months and between 6 months and 12 months were then calculated using the total area under the curve approach with linear interpolation between assessment points (and baseline adjustment for comparisons184).

Analyses, missing data and sensitivity analyses

Data were analysed using IMB Statistical Product and Service Solutions (SPSS) Statistics for Windows (version 20; IBM Coporation, Armonk, NY, USA) and Stata (version 11.2). Participants had individual unit costs applied based on the exact medication that they were prescribed, not on which arm they were in. Therefore, appropriate costs were applied regardless of switching during the trial.

Costs and outcomes were compared at 6 and 12 months and are presented as means and SDs. Mean differences and 95% CIs were obtained using non-parametric bootstrap regressions (1000 repetitions) to account for the non-normal distribution commonly found in economic data, with adjustment for region as this was a stratification factor in the randomisation process. Although this was a RCT and participants in all groups were expected to be balanced at baseline, baseline costs and outcomes could be predictors of follow-up costs. To provide more relevant treatment effect estimates,185 regressions to calculate mean differences in costs at follow-up included covariates for baseline cost from the same cost perspective, baseline HAQ score, duration of illness, age, sex, region and ethnicity. Outcome comparisons (for the economic evaluation) at follow-up included covariates for baseline values of the same outcome plus baseline HAQ score, duration of illness, age, sex, region and ethnicity.

Data were entered via an EDC system using the MedSciNet database (MedSciNet AB, Stockholm, Sweden; http://medscinet.com) which was programmed to disallow individual-item non-response on the CSRI service use section. There was thus no item non-response for this part of the CSRI. For lost employment data, if the CSRI indicated that this was positive but the amount was missing, the mean lost employment cost for that arm at that time point (only for those who had lost employment and had valid data) was substituted. For social security benefit data, if the CSRI indicated that this was positive but the amount was missing, unit costs for specified benefits were applied. When receipt of benefits was positive but specific benefits were unspecified, the mean benefit cost for that arm at that time point (only for those who received benefits and had valid data) was substituted. For non-trial medication data, if the medication name was missing but other information (e.g. dose) indicated some use, an average prescription cost (from Department of Health prescription cost analyses; see http://data.gov.uk/dataset/prescription_cost_analysis_england) was assumed. If a medication name was provided but usage quantity was missing, an average prescription cost for that particular medication was assumed.

Analyses were based on available cases for each analysis, that is, they excluded non-responders to the CSRI, HAQ, EQ-5D or SF-36 at each time point if there were any. To explore the potential impact of excluding non-responders we examined the sociodemographic and clinical characteristics of those included in the analyses and those in the full sample. We also carried out an ITT analysis, imputing missing 6- and 12-month total costs and outcomes using the multiple imputation command in Stata (version 11). Imputations of missing 6- and 12-month costs were based on variables expected to predict follow-up costs: baseline HAQ score, duration of illness, age, sex, region, ethnicity, trial arm and equivalent baseline cost (and equivalent cost at 6 months for 12-month imputations). Imputations of HAQ scores at 6 and 12 months were based on baseline HAQ score, duration of illness, age, sex, region, ethnicity and trial arm (and HAQ score at 6 months for 12-month imputations). Imputations of missing QALYs at 6 and 12 months were based on baseline HAQ score, duration of illness, age, sex, region, ethnicity, trial arm and equivalent baseline utility score (and utility score at 6 months for 12-month imputations). Cost and outcome data for the resulting imputed full sample were analysed and presented as per the base (available) case data.

Cost-effectiveness and cost–utility analyses

Accounting for the three cost perspectives and three outcomes, there were nine possible cost–outcome combinations to consider in the economic evaluation. ICERs were calculated for any combination that showed both significantly higher costs and better outcomes in either the intervention group or the control group.

Uncertainty around cost-effectiveness/cost–utility was explored using cost-effectiveness acceptability curves (CEACs) based on the net-benefit approach. This followed the Bayesian approach to cost-effectiveness analysis outlined by Briggs. 186 These curves address some of the problems associated with examining ICERs and show the probability that one intervention is cost-effective compared with another other for a range of values that a decision-maker would be willing to pay for an additional unit of each outcome (i.e. per additional QALY or per additional point improvement in HAQ score). Net benefits for each participant were calculated using the following formula, where λ is the willingness to pay for one additional unit of outcome:

A series of net benefits were calculated for each individual for λ values ranging between £0 and £50,000 per QALY gained and per point improvement on the HAQ. After calculating net benefits for each participant for each value of λ, coefficients of differences in net benefits between the trial arms were obtained through a series of bootstrapped linear regressions (1000 repetitions) of group upon net benefit, which included the baseline values of the same cost category and the same outcome as covariates plus baseline HAQ score, duration of illness, age, sex, region and ethnicity. The resulting coefficients were then examined to calculate for each value of λ the proportion of times that the cDMARDs group had a greater net benefit than the TNFi group. These proportions were then plotted to generate CEACs for all three outcomes from the health and social care perspective at 6 and 12 months.

Search strategies

Ovid MEDLINE and EMBASE were searched from 1946 to 2013. The terms used in the search strategies are found in Appendix 5. This was then limited to English language and clinical trials. The Cochrane Library was also searched using the terms ‘early rheumatoid arthritis and combination therapy’ or ‘early rheumatoid arthritis and anti-TNF’ for the early RA search and ‘rheumatoid arthritis and combination therapy’ or ‘rheumatoid arthritis and anti-TNF’ for the established RA search. The titles and abstracts were then assessed by two reviewers (MM, DS) independently. If there were any doubts regarding the eligibility of a particular study it was discussed between the two reviewers until agreement was reached.

Selection criteria

Assessing the risk of bias

The risk of bias was assessed using criteria recommended by Viswanathan et al. 188

Outcome measures

Statistical analysis

Results were analysed using Review Manager 5.1.6 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark). The random-effects model based on DerSimonian and Laird’s method189 was used to estimate the pooled effect sizes; this gives more equal weighting to studies of different precision than a simple inverse variance weighted approach, so accommodating between-study heterogeneity. For all meta-analyses we performed Cochran’s chi-squared test to assess between-study heterogeneity and quantified I²-statistics. 190,191 We considered a p-value < 0.05 as statistically significant.

In RCTs with two or more combination arms, the treatment arm with the best outcome was used. We also carried out sensitivity analyses using methotrexate monotherapy as the comparator arm. Methotrexate is the most commonly used DMARD monotherapy and is considered the most effective DMARD. Therefore, including this DMARD as a comparator ensures that trials had an effective comparator agent.

Results of the Tumour necrosis factor inhibitors Against Combination Intensive Therapy trial

The results of the TACIT trial are presented in five sections. The first section describes screening, randomisation, patients studied and the treatments that they received. The second section describes the impact of treatment on disability, quality of life and erosive damage; these outcomes were assessed every 6 months and include the HAQ, which was the primary outcome. The third section describes the impact of treatment on disease activity; it focuses on the DAS28 and its individual components, with outcomes assessed every month. The fourth section describes adverse effects encountered during the trial and the final section reports the economic evaluation.

The report presents the results from the ITT and complete-case populations. The results of the per-protocol population were similar to those of the ITT population and these findings are considered only in summary form. However, detailed tables of the complete-case analyses are presented in Appendix 3.

Results of the systematic reviews

The two systematic reviews focus on early RA and established RA and these are reported after the trial results.

Screening and randomisation

Between September 2008 and December 2010 432 patients were screened at 24 rheumatology clinics in England. In total, 218 patients were excluded, 196 because they did not consent to participate and a further 20 because they were not eligible to participate; in addition, for two patients no reasons were recorded. The remaining 214 patients were randomised: 107 to receive cDMARDs and 107 to receive TNFis (Figure 1). The final assessment for the TACIT trial was in December 2011.

FIGURE 1.

Consolidated Standards of Reporting Trials (CONSORT) diagram for the TACIT trial.

After randomisation three patients in the cDMARD group did not receive the intervention; this was because the patients changed their minds about participating in the trial after randomisation but before receiving treatment. There were also six patients in the TNFi group who did not receive the intervention. In three cases this was also because the patients changed their minds about taking part in the trial after randomisation but before receiving treatment. In one patient new information about a previous non-melanotic skin cancer resulted in the supervising rheumatologist considering that the patient should not receive biological treatment; in another patient the supervising rheumatologist changed his opinion about the suitability of the patient for intensive treatment; and the final patient mistakenly self-injected the TNFi, which had been delivered before the formal baseline assessment had been carried out.

Patients studied

Although 214 patients were randomised (107 to receive cDMARDs and 107 to receive TNFis), only 104 patients started cDMARDs and 101 patients started TNFis and the trial report focuses on these 205 treated patients. Their baseline characteristics are summarised in Table 3. Both groups had similar demographic characteristics including region, sex, ethnicity, disease duration, weight and height. Most clinical variables were also similar between groups including DAS28, the individual components of the DAS28, HAQ and EQ-5D scores, and SF-36 domain and summary scores. The only variable to show a baseline difference was Larsen score, with a mean score of 45.1 (SD 41.9) in the cDMARD group and 37.9 (SD 38.8) in the TNFi group.

During the 12 months of the trial 16 out of the 205 treated patients (8%) were lost to follow-up, nine patients in the cDMARD arm and seven in the TNFi arm (see Figure 1). A further 42 out of the 205 patients (20%) discontinued their intervention but remained under follow-up; these comprised 23 patients in the cDMARD arm and 19 patients in the TNFi arm.

In total, 10 out of 104 patients (10%) in the cDMARD arm stopped treatment because of toxicity, one (1%) stopped because of disease progression and 21 (15%) stopped for other reasons, including because a patient decided to stop treatment. In the TNFi arm six out of 101 (6%) patients stopped treatment because of toxicity, four (4%) stopped because of disease progression and 16 (16%) stopped for other reasons, including because a patient decided to stop treatment.

The main analysis evaluated 205 patients in the ITT group (104 in the cDMARD group and 101 in the TNFi group). The complete-case analysis evaluated the 147 completers (72 in the cDMARD group and 75 in the TNFi group).

Drug treatments

Individual treatments in the combination disease-modifying antirheumatic drug group

Most patients received combinations of two or three DMARDs during the course of the trial (Table 4). A minority had four or five DMARDs. As patients were receiving DMARDs when enrolled, combinations were usually given in a step-up approach. The most common combination was methotrexate and leflunomide. Other frequently used combinations included methotrexate and ciclosporin, methotrexate, sulfasalazine and hydroxychloroquine and methotrexate and gold. A variety of other combinations were used occasionally.

TABLE 4 - Individual treatment regimens in the cDMARD group

Three patients were to have switched to a TNFi but never started because they withdrew from the trial.

Therapies	No. of patients
No. of DMARDs combined
One	0
Two	46
Three	48
Four	8
Five	2
Total	104
DMARD combinations
Methotrexate/leflunomide	62
Methotrexate/ciclosporin	17
Methotrexate/sulfasalazine/hydroxychloroquine	13
Methotrexate/gold	10
Other	2
Total	104
Use of steroids
Oral prednisolone	24
Depo-Medrone injections	3
Total	27
Switched to TNFis
Adalimumab	25
Etanercept	14
Infliximab	4
Withdrew before startinga	3
Total	46

In total, 27 patients received glucocorticoids (steroids). In 24 patients these were given as oral prednisolone. Most of these patients receiving steroids (16 cases) had oral prednisolone combined with two DMARDs. An additional three patients received Depo-Medrone injections.

After 6 months, patients who had failed to achieve a decrease in DAS28 of ≥ 1.2 could receive a TNFi. Altogether, 46 out of 104 patients (44%) were recommended to switch to a TNFi. Three patients withdrew from the trial before starting a TNFi and therefore 43 out of 104 patients (41%) actually switched to a TNFi. The majority of these patients received adalimumab (see Table 4). The times at which the new treatments were actually started (as opposed to when they were recommended) are shown in Figure 2. TNFis were started after an average of 9 months (range 7–12 months).

FIGURE 2.

Actual month that TNFis were started in cDMARD patients switching to TNFis.

Changes in Health Assessment Questionnaire scores

Six-month outcomes in the intention-to-treat population

At 6 months the HAQ score decreased by a mean of 0.28 (95% CI 0.18 to 0.38) in patients randomised to cDMARDs and by a mean of 0.35 (95% CI 0.23 to 0.46) in patients randomised to TNFis (see Table 7). This difference was not significant in either the unadjusted or the adjusted model (see Table 8). The overall pattern of change is shown in Figure 3.

FIGURE 3.

(a) Health Assessment Questionnaire; and (b) EQ-5D mean scores (95% CIs) by treatment group in the ITT population.

The effects of patients in the combination disease-modifying antirheumatic drug group switching to tumour necrosis factor inhibitors

In total, 58 of the 104 patients in the cDMARD group remained on cDMARDs and 46 switched to TNFis after 6 months. Over 12 months both sets of patients showed similar changes in HAQ score and there was no evidence of a difference between groups by linear regression analysis (Table 9 and Figure 4). Comparing changes in HAQ scores in both of these groups using general estimating equations provided no evidence that there were any significant differences between the two groups in both an unadjusted and an adjusted model (Table 10).

TABLE 9 - Primary and key secondary outcomes [mean scores and changes in scores (95% CIs)] in the cDMARD group (n = 104) by treatment status

p = 0.81 comparing change at 12 months between groups by linear regression.

p = 0.069 comparing change at 12 months between groups by linear regression.

p = 0.77 comparing change at 12 months between groups by linear regression.

Measure	Stayed on cDMARDs (n = 58)				Changed to TNFis (n = 46)
	Initial	6 months	12 months	Change 0–12 months	Initial	6 months	12 months	Change 0–12 months
HAQ score	1.82 (1.65 to 1.98)	1.42 (1.23 to 1.60)	1.38 (1.17 to 1.60)	0.43a (0.29 to 0.58)	1.77 (1.62 to 1.92)	1.64 (1.46 to 1.82)	1.31 (1.10 to 1.51)	0.46 (0.30 to 0.62)
EQ-5D score	0.35 (0.27 to 0.43)	0.57 (0.50 to 0.64)	0.61 (0.53 to 0.69)	−0.26b (−0.35 to −0.17)	0.44 (0.35 to 0.52)	0.49 (0.40 to 0.57)	0.57 (0.48 to 0.65)	−0.13 (−0.24 to −0.02)
Larsen score	44.7 (33.6 to 55.8)	45.3 (34.1 to 56.5)	45.9 (34.7 to 57.0)	−1.13c (−2.63 to 0.38)	45.5 (33.4 to 57.6)	46.5 (34.3 to 58.7)	46.9 (34.6 to 59.3)	−1.43 (−2.92 to 0.06)

FIGURE 4.

(a) Health Assessment questionnaire; and (b) EQ–5D mean scores (95% CIs) in cDMARD arm patients (n = 104) in the ITT population by treatment status.

TABLE 10 - Effects on HAQ score in cDMARD arm patients (n = 104) in the ITT population by treatment status: adjusted and unadjusted assessments of treatment effect using GEEs

Demographic variables are age, sex, ethnicity, disease duration and region. The reference group is no switch.

Outcome	Model 1 (unadjusted): treatment + region + time		Model 2 (adjusted): treatment + demographics + baseline score + time
	Coefficient (95% CI)	p-value	Coefficient (95% CI)	p-value
Change in HAQ score
12 months	0.14 (−0.03 to 0.31)	0.103	0.12 (−0.05 to 0.29)	0.185

Changes in European Quality of Life-5 Dimensions scores

Changes in Short Form Questionnaire-36 items scores

Changes in Larsen scores

Intention-to-treat population

The initial Larsen scores differed between groups (see Table 7): in the cDMARD group the initial mean score was 45.1 (95% CI 37.0 to 53.2) and in the TNFi group it was 37.9 (95% CI 30.2 to 45.6). The Larsen score was the only clinical variable to show baseline differences and no clinical significance was attached to this difference.

Progression over 12 months was similar between the groups (Figure 5). With cDMARDs the Larsen score increased by 1.26 and with TNFis it increased by 1.37. Progression over 6 months was also similar. These differences between the treatment groups were not statistically significant (see Tables 8 and 12).

FIGURE 5.

Mean Larsen scores (95% CIs) by treatment group in the ITT population.

An exploratory analysis examined individual changes over 12 months using all observed data for both groups (Figure 6); this showed no evidence of a different pattern of progression between the groups.

FIGURE 6.

Individual changes in Larsen score over 12 months using all observed data.

Another exploratory analysis evaluated the development of one (increase in Larsen score of 2–5) or many new erosions (increase in Larsen score of > 5) using all observed data for both groups; this is summarised in Figure 7, showing that there were no differences between the groups. By the end of the trial, 23 out of 91 patients (25%) randomised to receive cDMARDs developed one new erosion and 12 out of 91 patients (13%) developed two or more erosions. In the group randomised to receive TNFis, 19 out of 93 patients (20%) developed one new erosion and 13 out of 93 patients (14%) developed two or more erosions.

FIGURE 7.

Development of new erosions over 12 months using all observed data.

The effects of patients in the combination disease-modifying antirheumatic drug group switching to tumour necrosis factor inhibitors

Over 12 months both sets of patients showed small increases in Larsen scores (see Table 9). In patients remaining on cDMARDs Larsen scores increased by a mean of −1.13 (95% CI −2.63 to 0.38) and in patients switching to TNFis Larsen scores increased by a mean of −1.43 (95% CI −2.92 to 0.06). Comparing these changes in Larsen scores over 12 months by linear regression provided no evidence that the difference was significant (see Table 11). We also examined individual changes over 12 months for both sets of patients using all observed data (Figure 8); this showed no evidence of a different pattern of progression between the two groups.

FIGURE 8.

Individual changes in Larsen score over 12 months in the cDMARD group by treatment status using all observed data.

Changes in Disease Activity Score for 28 Joints scores

Intention-to-treat population

Initial DAS28 were similar in both groups. In the cDMARD group the initial mean DAS28 was 6.21 (95% CI 6.04 to 6.39) and in the TNFi group the initial mean score was 6.30 (95% CI 6.14 to 6.46). The patterns of change are shown in Table 13 and Figure 9.

TABLE 13 - Mean DAS28 and components scores (95% CIs) by treatment group in the ITT population

Month of assessment	cDMARDs (n = 104)					TNFis (n = 101)
	DAS28	Tender joint count	Swollen joint count	ESR (mm/hour)	VAS	DAS28	Tender joint count	Swollen joint count	ESR (mm/hour)	VAS
0	6.21 (6.04 to 6.39)	16.38 (15.02 to 17.75)	10.51 (9.34 to 11.68)	33.14 (28.10 to 38.19)	68.13 (64.32 to 71.95)	6.30 (6.14 to 6.46)	17.48 (16.15 to 18.80)	10.79 (9.47 to 12.11)	30.13 (25.65 to 34.61)	68.18 (64.00 to 72.36)
1	5.32 (5.05 to 5.59)	11.81 (10.07 to 13.56)	6.76 (5.55 to 7.97)	32.44 (27.43 to 37.46)	54.87 (49.90 to 59.84)	4.67 (4.38 to 4.95)	10.69 (8.93 to12.46)	5.76 (4.53 to 6.99)	19.56 (15.78 to 23.34)	46.15 (41.09 to 51.20)
2	5.02 (4.76 to 5.27)	9.85 (8.36 to 11.34)	5.72 (4.72 to 6.72)	30.23 (25.51 to 34.95)	51.96 (47.22 to 56.70)	4.30 (3.98 to 4.61)	7.84 (6.47 to 9.21)	5.02 (3.74 to 6.31)	21.54 (17.27 to 25.81)	43.18 (38.14 to 48.22)
3	4.92 (4.63 to 5.21)	9.97 (8.36 to 11.59)	5.37 (4.26 to 6.47)	30.10 (24.81 to 35.39)	50.94 (45.74 to 56.15)	4.28 (3.95 to 4.60)	7.71 (6.12 to 9.31)	4.43 (3.27 to 5.59)	23.44 (19.08 to 27.80)	43.16 (38.10 to 48.23)
4	4.73 (4.45 to 5.02)	8.40 (7.03 to 9.78)	4.91 (3.85 to 5.97)	31.20 (26.05 to 36.36)	45.38 (39.69 to 51.07)	4.31 (3.97 to 4.64)	8.16 (6.54 to 9.78)	4.07 (2.95 to 5.20)	22.23 (18.31 to 26.15)	45.40 (39.48 to 51.31)
5	4.66 (4.36 to 4.96)	8.51 (6.98 to 10.04)	5.44 (4.33 to 6.55)	29.87 (24.95 to 34.79)	43.62 (37.87 to 49.37)	4.13 (3.81 to 4.44)	7.75 (6.06 to 9.45)	4.38 (3.15 to 5.62)	20.51 (16.96 to 24.06)	40.60 (35.02 to 46.18)
6	4.78 (4.45 to 5.12)	10.16 (8.39 to 11.93)	6.11 (4.74 to 7.49)	29.25 (24.16 to 34.33)	48.25 (42.42 to 54.09)	4.23 (3.89 to 4.58)	8.57 (6.87 to 10.27)	4.66 (3.41 to 5.90)	21.80 (17.64 to 25.96)	40.51 (34.98 to 46.03)
7	4.57 (4.27 to 4.87)	8.34 (6.68 to 10.01)	5.76 (4.53 to 7.00)	28.21 (23.03 to 33.39)	42.51 (36.95 to 48.07)	4.17 (3.81 to 4.53)	8.01 (6.35 to 9.67)	4.64 (3.34 to 5.93)	21.01 (17.18 to 24.85)	40.73 (34.72 to 46.73)
8	4.25 (3.92 to 4.59)	7.42 (5.98 to 8.85)	4.54 (3.43 to 5.64)	24.14 (19.29 to 28.99)	41.30 (35.45 to 47.14)	4.05 (3.69 to 4.41)	6.62 (5.14 to 8.09)	4.23 (3.04 to 5.42)	21.77 (17.57 to 25.97)	42.56 (36.91 to 48.22)
9	4.21 (3.87 to 4.56)	6.93 (5.40 to 8.47)	3.99 (2.98 to 5.00)	25.47 (20.57 to 30.37)	41.56 (35.78 to 47.35)	4.08 (3.73 to 4.42)	6.68 (5.14 to 8.22)	4.30 (3.07 to 5.52)	22.92 (18.62 to 27.21)	40.86 (35.36 to 46.36)
10	4.05 (3.73 to 4.37)	6.17 (4.74 to 7.60)	3.69 (2.78 to 4.61)	25.42 (20.44 to 30.40)	38.33 (33.02 to 43.65)	3.90 (3.56 to 4.24)	6.51 (4.90 to 8.12)	3.42 (2.33 to 4.51)	21.48 (17.34 to 25.62)	39.33 (33.64 to 45.03)
11	4.03 (3.74 to 4.31)	6.67 (5.09 to 8.25)	3.27 (2.31 to 4.24)	23.28 (18.69 to 27.87)	40.64 (34.69 to 46.59)	3.84 (3.48 to 4.20)	6.16 (4.57 to 7.76)	3.50 (2.25 to 4.75)	22.01 (17.73 to 26.28)	37.69 (31.82 to 43.56)
12	4.04 (3.74 to 4.34)	6.32 (4.88 to 7.77)	3.39 (2.63 to 4.14)	25.03 (20.41 to 29.65)	39.21 (33.23 to 45.19)	3.89 (3.53 to 4.24)	6.81 (5.22 to 8.40)	3.20 (2.25 to 4.14)	20.32 (16.04 to 24.59)	43.03 (36.79 to 49.27)

FIGURE 9.

Mean DAS28 (95% CIs) by treatment group in the ITT population.

By 6 months the DAS28 had fallen in the cDMARDs group to 4.78 (95% CI 4.45 to 5.12) and in the TNFis group to 4.23 (95% CI 3.89 to 4.58). By 12 months the DAS28 had fallen further in the cDMARDs group to 4.04 (95% CI 3.74 to 4.34) and in the TNFis group to 3.89 (95% CI 3.53 to 4.24). The initial change in DAS28 was greater in patients randomised to TNFis and there was a significant difference between groups within the first month of treatment. After 1 month the mean DAS28 in the cDMARDs group fell to 5.32 (95% CI 5.05 to 5.59) and the mean score in the TNFis group fell to 4.67 (95% CI 4.38 to 4.95, p = 0.001).

Longitudinal analysis (Table 14) showed that there was a significant difference between treatment groups over the whole 12-month period. Patients randomised to TNFis achieved greater overall reductions in DAS28 than those randomised to cDMARDs in both the unadjusted model (−0.48, 95% CI −0.79 to −0.17; p = 0.002) and the adjusted model (−0.40, 95% CI −0.69 to −0.10; p = 0.009). Comparing the initial and final treatment periods in the adjusted model showed a difference in the pattern of change. In the first 6 months there was a greater reduction in DAS28 in patients randomised to TNFis than in patients randomised to cDMARDs, with a coefficient of −0.63 (95% CI −0.93 to −0.34, p < 0.001). In the second period there was no difference between groups, with a coefficient of −0.19 (95% CI −0.55 to 0.18, p = 0.317).

TABLE 14 - Longitudinal analysis of treatment effects on disease activity: analysis of changes in DAS28 and its components using GEEs in the ITT population

Demographics variables are age, sex, ethnicity, disease duration and region. The reference group is cDMARDs.

Time period	Variable	Model 1 (unadjusted): treatment + region + time		Model 2 (adjusted): treatment + demographics + baseline score
		Coefficient (95% CI)	p-value	Coefficient (95% CI)	p-value
Months 1–6	DAS28	−0.68 (−0.99 to −0.37)	< 0.001	−0.63 (−0.93 to −0.34)	< 0.001
	Tender joint count	−2.42 (−4.22 to −0.63)	0.008	−1.79 (−3.31 to −0.26)	0.022
	Swollen joint count	−1.35 (−2.76 to 0.07)	0.062	−1.16 (−2.20 to −0.12)	0.029
	ESR (mm/hour)	−6.46 (−10.23 to −2.68)	0.001	−7.18 (−10.60 to −3.76)	< 0.001
	VAS	−6.97 (−13.10 to −0.84)	0.026	−6.41 (−11.66 to −1.15)	0.017
Months 7–12	DAS28	−0.31 (−0.69 to 0.07)	0.111	−0.19 (−0.55 to 0.18)	0.317
	Tender joint count	−1.10 (−3.22 to 1.01)	0.307	−0.13 (−1.79 to 1.53)	0.879
	Swollen joint count	−0.69 (−2.27 to 0.88)	0.388	−0.31 (−1.36 to 0.75)	0.570
	ESR (mm/hour)	−1.63 (−5.88 to 2.62)	0.452	−2.15 (−5.73 to 1.44)	0.241
	VAS	0.60 (−6.47 to 7.67)	0.867	2.04 (−4.08 to 8.17)	0.513
Months 1–12	DAS28	−0.48 (−0.79 to −0.17)	0.002	−0.40 (−0.69 to −0.10)	0.009
	Tender joint count	−1.69 (−3.50 to 0.11)	0.066	−0.93 (−2.36 to 0.51)	0.205
	Swollen joint count	−0.86 (−2.27 to 0.55)	0.233	−0.63 (−1.57 to 0.31)	0.186
	ESR (mm/hour)	−4.04 (−7.67 to −0.40)	0.029	−4.62 (−7.77 to −1.47)	0.004
	VAS	−2.83 (−8.85 to 3.20)	0.358	−1.96 (−7.04 to 3.11)	0.448

Changes in Disease Activity Score for 28 Joints components

Intention-to-treat population

Baseline tender joint counts, swollen joint counts, ESR and patient global assessments were similar in both groups and they all improved when patients received either cDMARDs or TNFis. The patterns of change are shown in Table 13 and Figures 10 and 11.

FIGURE 10.

Mean tender and swollen joint counts (95% CIs) by treatment group in the ITT population. (a) Tender joints; and (b) swollen joints.

FIGURE 11.

Mean ESR (mm/hour) and patient global VAS (95% CIs) by treatment group in the ITT population. (a) ESR; (b) patient global VAS.

Longitudinal analysis (see Table 14) showed that in the overall adjusted model changes in ESR were significantly different between patients randomised to cDMARDs and those randomised to TNFis; the decrease in ESR was significantly larger in the TNFis group (coefficient −4.62, 95% CI −7.77 to −1.47; p = 0.004). There were no significant differences between treatment groups in the other components over the whole 6 months.

In the first 6 months of treatment the adjusted mean treatment effects for all of the components were significantly greater in patients randomised to TNFis than in those randomised to cDMARDs. In the second 6 months there were no statistically significant differences between the groups.

The speed of onset of changes was particularly marked for the ESR in patients randomised to TNFis. With cDMARDs the ESR fell from an initial mean of 33.1 (95% CI 28.1 to 38.2) mm/hour to 32.4 (95% CI 27.4 to 37.5) mm/hour by 1 month. With TNFis the ESR fell from an initial mean of 30.1 (95% CI 25.7 to 34.6) mm/hour to 19.6 (95% CI 15.8 to 23.3) mm/hour by 1 month.

The effects of patients in the combination disease-modifying antirheumatic drug group switching to tumour necrosis factor inhibitors

Patients were selected to switch from cDMARDs to TNFis after 6 months if the change in DAS28 was < 1.2. As a consequence, the mean DAS28 for the switchers would be expected to be more than that in those who remained on cDMARDs. This difference is shown in Table 15, together with changes in the individual components of the DAS28, and is also illustrated in Figure 12. The difference is confirmed to be significant in the longitudinal analysis shown in Table 16. The adjusted model showed a significant reduction in DAS28 in the switchers. The same effect was seen for the components of the DAS28 and was most marked for tender joint count and patient global VAS score.

TABLE 15 - Mean DAS28 and component scores (95% CIs) in the ITT population cDMARD group (n = 104) by treatment status

Month of assessment	Stayed on cDMARDs (n = 58)					Changed to TNFis (n = 46)
	DAS28	Tender joint count	Swollen joint count	ESR (mm/hour)	VAS	DAS28	Tender joint count	Swollen joint count	ESR (mm/hour)	VAS
0	6.10 (5.84 to 6.35)	15.05 (13.11 to16.99)	9.95 (8.45 to 11.44)	34.07 (27.29 to 40.85)	69.50 (64.19 to 74.81)	6.36 (6.12 to 6.60)	18.07 (16.24 to 19.89)	11.22 (9.34 to 13.10)	31.98 (24.26 to 39.70)	66.41 (60.88 to 71.95)
1	5.06 (4.69 to 5.42)	10.14 (7.78 to 12.49)	5.72 (4.39 to 7.06)	32.76 (26.01 to 39.50)	52.23 (45.54 to 58.92)	5.66 (5.26 to 6.05)	13.92 (11.39 to 16.46)	8.07 (5.93 to 10.21)	32.05 (24.35 to 39.74)	58.21 (50.74 to 65.68)
2	4.74 (4.40 to 5.07)	8.17 (6.33 to 10.00)	4.70 (3.49 to 5.92)	30.11 (24.05 to 36.18)	49.69 (43.34 to 56.05)	5.37 (4.99 to 5.75)	11.97 (9.64 to 14.30)	7.01 (5.38 to 8.64)	30.38 (22.77 to 37.98)	54.81 (47.50 to 62.13)
3	4.64 (4.27 to 5.01)	8.54 (6.58 to 10.50)	4.59 (3.08 to 6.09)	28.94 (21.86 to 36.03)	48.80 (41.44 to 56.16)	5.27 (4.81 to 5.72)	11.77 (9.12 to 14.42)	6.35 (4.69 to 8.00)	31.55 (23.45 to 39.66)	53.65 (46.32 to 60.98)
4	4.51 (4.15 to 4.87)	7.04 (5.44 to 8.64)	3.80 (2.59 to 5.00)	31.11 (23.79 to 38.43)	44.47 (36.92 to 52.02)	5.01 (4.54 to 5.49)	10.13 (7.78 to 12.47)	6.32 (4.52 to 8.12)	31.32 (24.17 to 38.47)	46.52 (37.63 to 55.41)
5	4.27 (3.90 to 4.65)	6.43 (4.64 to 8.23)	3.64 (2.46 to 4.81)	29.31 (22.87 to 35.76)	39.13 (31.94 to 46.32)	5.15 (4.68 to 5.62)	11.13 (8.73 to 13.53)	7.72 (5.85 to 9.59)	30.57 (22.88 to 38.27)	49.28 (40.01 to 58.55)
6	4.01 (3.61 to 4.40)	6.27 (4.35 to 8.19)	3.20 (1.78 to 4.62)	27.66 (20.92 to 34.40)	37.26 (29.75 to 44.78	5.76 (5.32 to 6.19)	15.07 (12.47 to 17.66)	9.78 (7.65 to 11.91)	31.24 (23.35 to 39.14)	62.11 (54.51 to 69.71)
7	4.09 (3.71 to 4.46)	5.85 (3.79 to 7.90)	3.91 (2.40 to 5.41)	28.87 (21.57 to 36.18)	38.04 (30.16 to 45.92)	5.18 (4.74 to 5.61)	11.49 (8.97 to 14.01)	8.11 (6.25 to 9.97)	27.37 (20.00 to 34.74)	48.15 (40.36 to 55.93)
8	4.13 (3.68 to 4.58)	6.26 (4.51 to 8.00)	3.55 (2.23 to 4.86)	24.88 (18.17 to 31.58)	42.35 (34.19 to 50.50)	4.41 (3.91 to 4.92)	8.87 (6.54 to 11.21)	5.79 (3.97 to 7.60)	23.20 (16.01 to 30.40)	39.97 (31.60 to 48.34)
9	4.15 (3.69 to 4.62)	5.75 (3.87 to 7.63)	3.46 (2.12 to 4.81)	26.58 (20.08 to 33.07)	43.31 (34.89 to 51.73)	4.29 (3.78 to 4.80)	8.43 (6.02 to 10.84)	4.64 (3.03 to 6.26)	24.07 (16.40 to 31.74)	39.36 (31.28 to 47.44)
10	3.91 (3.48 to 4.34)	5.23 (3.51 to 6.96)	3.35 (2.11 to 4.59)	25.67 (18.55 to 32.80)	37.16 (29.52 to 44.81)	4.23 (3.75 to 4.72)	7.35 (5.03 to 9.67)	4.12 (2.77 to 5.48)	25.10 (18.29 to 31.92)	39.81 (32.33 to 47.29)
11	3.87 (3.48 to 4.25)	5.44 (3.54 to 7.33)	2.63 (1.40 to 3.85)	22.65 (16.69 to 28.61)	40.56 (31.84 to 49.27)	4.23 (3.77 to 4.69)	8.22 (5.66 to 10.78)	4.09 (2.53 to 5.65)	24.08 (16.79 to 31.38)	40.74 (32.36 to 49.13)
12	3.91 (3.52 to 4.31)	5.37 (3.66 to 7.08)	2.87 (1.83 to 3.91)	26.39 (20.45 to 32.33)	38.33 (29.89 to 46.78	4.19 (3.73 to 4.66)	7.52 (5.08 to 9.97)	4.04 (2.93 to 5.15)	23.33 (15.89 to 30.76)	40.32 (31.89 to 48.75)

FIGURE 12.

Disease Activity Score for 28 Joints mean scores (95% CIs) in patients in the cDMARD group (n = 104) by treatment status in the ITT population.

TABLE 16 - Longitudinal analysis of treatment effects on disease activity in the cDMARD group (n = 104) by treatment status: analysis of changes in DAS28 and its components using GEEs in the ITT population

Demographics variables are age, sex, ethnicity, disease duration and region. The reference group is no switch.

Variable	Model 1 (unadjusted): treatment + region + time		Model 2 (adjusted): treatment + demographics + baseline score
	Coefficient (95% CI)	p-value	Coefficient (95% CI)	p-value
DAS28	0.35 (−0.03 to 0.74)	0.071	0.51 (0.16 to 0.86)	0.005
Tender joint count	0.69 (−1.47 to 2.85)	0.532	2.42 (0.75 to 4.10)	0.004
Swollen joint count	0.97 (−1.04 to 2.97)	0.344	1.94 (0.65 to 3.22)	0.003
ESR (mm/hour)	2.54 (−2.65 to 7.73)	0.338	2.44 (−1.97 to 6.84)	0.278
VAS	10.26 (2.70 to 17.83)	0.008	8.29 (2.14 to 14.44)	0.008

Achieving a clinical response

Time to achieve a response

An exploratory analysis examined the time taken to achieve a clinically meaningful response – a decrease in DAS28 of ≥ 1.2. The time to achieve a clinically meaningful response was compared between the groups using Kaplan–Meier plots (Figure 13). In total, 98 of 104 patients (94%) randomised to receive cDMARDs and 94 of 101 patients (93%) randomised to receive TNFis achieved such responses. The responses occurred sooner in the patients randomised to TNFis and this difference was significant in a log-rank test (p = 0.035). Patients randomised to receive cDMARDs who had a clinically meaningful DAS28 response achieved it within a mean of 3 months. Patients randomised to receive TNFis who had a clinically meaningful DAS28 response achieved it within a mean of 2 months.

FIGURE 13.

Time to achieve a response by treatment group. Kaplan–Meier plot of patients showing a reduction in DAS28 of ≥ 1.2 using all observed data.

Persistence of response

There was a complex pattern of achieving responses. In some patients the response was persistent and in others it was unsustained. Examples of these variations are shown in Figure 14 for four patients randomised to the TNFi group. As a consequence of these variations we evaluated the frequency of response each month in the two treatment groups (Figure 15). There was a different pattern of response in the two groups. Patients randomised to cDMARDs showed a gradual increase in the rate of response from ≤ 45% at 3 months or earlier to > 70% by 10 months. By contrast, patients randomised to TNFis achieved a response rate of > 70% by 2 months and the response rate remained at > 70% thereafter, with the highest response rate achieved in this group being 84% (achieved at month 11).

FIGURE 14.

Examples of persistent and unsustained responses in four patients randomised to the TNFi group.

FIGURE 15.

Frequency of response by treatment group: patients with a reduction in DAS28 of ≥ 1.2 using all observed data.

Impact of switching from combination disease-modifying antirheumatic drugs to tumour necrosis factor inhibitors

There was a difference in response rate between the patients randomised to cDMARDs who remained on cDMARDs and those who were randomised to cDMARDs but who switched to TNFis (Figure 16). Patients remaining on cDMARDs had a response rate of > 50% from 2 months onwards and after 6 months the response rate increased to > 70%. Those patients who switched to TNFis had an initial response rate of < 50% and the response rate did not increase to 70% until 10 months.

FIGURE 16.

Frequency of response in cDMARD patients by treatment status: patients with a reduction in DAS28 of ≥ 1.2 using all observed data.

Achieving a Disease Activity Score for 28 Joints of ≤ 2.6

Time to achieve a Disease Activity Score for 28 Joints of ≤ 2.6

The time taken to achieve remission (DAS28 of ≤ 2.6) was compared using Kaplan–Meier plots (Figure 17). In total, 36 out of 104 patients (35%) randomised to receive cDMARDs and 44 out of 101 patients (44%) randomised to receive TNFis achieved remission at any time. There was no evidence that the speed of onset of remission was significantly different between groups (p = 0.085). Those patients randomised to receive both cDMARDs and TNFis who achieved DAS28 remission achieved it within a mean of 4 months.

FIGURE 17.

Time to achieve DAS28 remission by treatment group. Kaplan–Meier plot of time to achieve a DAS28 of ≤ 2.6 using all observed data.

Persistence of Disease Activity Score for 28 Joints of ≤ 2.6

There was a complex pattern of achieving remission. In some patients remission was persistent and in others it was unsustained. Examples of these variations are shown in Figure 18 for four patients randomised to the TNFi group. As a consequence of these variations we have also evaluated the frequency of response each month for each group (Figure 19). There was a different pattern of response between the groups. Patients randomised to cDMARDs showed a gradual increase in the rate of response from ≤ 5% at 3 months or earlier to a maximum of 20% by 12 months. By contrast, those patients randomised to TNFis had achieved a remission rate of 16% by 3 months, which gradually increased to a maximum of 32% by 11 months.

FIGURE 18.

Examples of persistent and unsustained responses in four patients randomised to the TNFi group.

FIGURE 19.

Frequency of response by treatment group: patients with a DAS28 of ≤ 2.6 using all observed data.

Impact of switching from combination disease-modifying antirheumatic drugs to tumour necrosis factor inhibitors

There was a difference in response rate between the patients randomised to cDMARDs who remained on cDMARDs and those randomised to cDMARDs who switched to TNFis (Figure 20). In both groups < 10% of patients achieved a DAS28 of ≤ 2.6 at ≤ 5 months. From 6 to 12 months between 13% and 24% of patients remaining on cDMARDs and between 5% and 21% of patients who switched to TNFis achieved a DAS28 of ≤ 2.6.

FIGURE 20.

Frequency of remission in cDMARD patients by treatment status: patients with a DAS28 of ≤ 2.6 using all observed data.

Serious adverse events

In total, 10 patients in the cDMARDs group had a serious adverse event, eight in the first 6 months and two in the second 6 months (Table 17). Seven of these serious adverse events involved or prolonged inpatient treatment. In the TNFis group, 19 patients had a serious adverse event, six in the first 6 months and 13 in the second 6 months; 13 of these serious adverse events involved or prolonged inpatient treatment. One patient in the TNFi group died from pneumonia and multiple organ failure during the second 6 months of treatment. Cardiovascular, respiratory, digestive and genitourinary systems were most commonly involved. Although there were more serious adverse events in the TNFi group, there was no evidence of major clinically important differences between the treatment groups and the frequency of adverse events was not significantly different (Fisher’s exact test p = 0.110).

Stopping treatment because of adverse events

In total, 10 out of 104 patients (10%) in the cDMARD arm and six out of 101 patients (6%) in the TNFi arm stopped treatment because of toxicity (see Figure 1). Although more patients withdrew from treatment because of toxicity in the cDMARDs arm, there was no evidence of major clinically important differences between the treatment groups and the frequency of withdrawals because of adverse events as this was not significantly different (Fisher’s exact test p = 0.441).

Individual adverse events

There were 635 different adverse events reported by patients in the cDMARD group. The most frequent events are listed in Table 18 and they are grouped by system involved in Table 19. All reported events are listed in Appendix 4.

There were 465 different adverse events reported by patients in the TNFis group. The most frequent events are listed in Table 18 and they are grouped by system involved in Table 19. All reported events are listed in Appendix 4.

Chest infections (46 events), headaches (45 events), diarrhoea (42 events), nausea (34 events), sore throats (28 events), colds (28 events), elevated liver enzymes (alanine aminotransferase) (23 events) and fatigue (16 events) were the most common adverse events across both groups. Some types of adverse events spanned systems, in particular infections, which accounted for 112 adverse events in the cDMARDs group and 117 in the TNFis group.

There was no evidence of any major clinically important differences between the two treatment groups. However, the cDMARDs group had 37% more adverse events overall (635 vs. 465). This difference was mainly due to there being 88 more adverse events related to the digestive system (148 vs. 60) and 20 more adverse events related to the nervous system (61 vs. 41) in the cDMARDs group.

Response rates

The response rates for the CSRI and outcome questionnaires and the availability of trial medication data are summarised in Tables 20–22 respectively. These were > 90% and were similar for all of the questionnaires at baseline and 6 and 12 months and across both trial arms.

Table 23 summarises the joint availability of both cost and outcome data (a requirement for the construction of CEACs) by outcome measure. In total, 191 of the 205 study participants (93%) had both cost and outcome data at 6 months’ follow-up and 186–188 of the 205 study participants (91–92%) had both cost and outcome data at 12 months’ follow-up. There were thus very few cases excluded from the available case analyses.

Tables 24–26 suggest that there were no notable differences in characteristics between the subsamples included in the available case analyses and the full sample.

Resource use

Resource use differences between the groups were not compared statistically, first, because the economic evaluation was focused on costs and cost-effectiveness/utility and, second, to avoid problems associated with multiple testing. Therefore, resource use patterns are described in Tables 27–29 without statistical comparisons. Use of services appeared similar in both groups at all three time points. General practitioner (GP) surgery visits, practice nurse surgery visits, repeat prescription requests and hospital outpatient appointments were common in both groups at all time points, with other service use being relatively rare. The number of participants using non-trial medications was also similar in both groups at all time points.

Data on the use of NHS/social services-funded transport, equipment and home adaptations (costs of which are excluded from cost calculations) are presented in Table 30.

Costs

Cost components at baseline, 6 months and 12 months are summarised in Table 31. Costs for both groups were equivalent at baseline. Costs of social security benefits and employment losses are small compared to the cost of health and social care. At 6 and 12 months’ follow-up all cost components remained equivalent between groups except for the cost of trial medications, which was significantly lower in the cDMARDs group (6-month adjusted mean difference −£3637, 95% CI −£3838 to −£3420; 12-month adjusted mean difference −£1894, 95% CI −£2320 to −£1427). The additional trial medication cost in the TNFis group overshadowed all other cost components in that group.

The increase in trial medication costs in the cDMARDs group between 6 and 12 months was due to a significant proportion of this group switching to the more expensive TNFis at 6 months because of non-response to cDMARDs by 6 months. Switching in the reverse direction was uncommon (a total of four participants) and so trial medication costs in the TNFis group did not fall a great deal between 6 and 12 months.

Table 32 shows total costs at 6 and 12 months from a health and social care perspective and the two societal perspectives that we adopted (with and without social security benefit costs). All figures (including those for trial medication) represent a 3-month period. The cDMARDs group has significantly lower total costs from all perspectives at both follow-up points. The difference is greater at 6 months than at 12 months because of the greater trial medication cost differential before switching taking place. Costs from each of the societal perspectives are similar to those from a health and social care perspective because of the dominance of trial medication costs.

For the purpose of combining cost and outcome data for the cost-effectiveness/cost–utility analyses, all costs were equivalised to 6-month values. Trial medication costs were available for the 0- to 6-month and 7- to 12-month periods so all other costs were multiplied by 2 to represent 6-month rather than 3-month periods. The extrapolated figures are shown in Table 33. Imputing missing cost data (based on the extrapolated costs) for those lost to follow-up confirmed the findings from the unimputed available case analysis (Table 34).

Outcomes

The cDMARDs arm had an advantage of four points based on the SF-36-based utility scores at baseline but this did not carry through as an advantage in (baseline-adjusted) utility scores at either of the two follow-up points or in the resulting QALY estimates (Table 35). The cDMARDs group did, however, show advantages in terms of the HAQ and EQ-5D-based utility scores at 12 months, although the latter did not translate into an advantage in terms of the QALYs estimated from the EQ-5D. As with cost data, imputing missing outcome data for those lost to follow-up did not alter the conclusions from the available case analyses (Table 36).

Cost-effectiveness and cost–utility analyses

Table 37 presents the ICERs for the cost-effectiveness and cost–utility analyses based on costs from each perspective (based on extrapolations representing 6-month periods) and outcomes at 6 and 12 months. For the ICERs, the mean difference in the HAQ score was reversed (negatives turned to positives and vice versa) as a reduction in HAQ score indicates a better outcome. Of the 18 cost–outcome combinations, three showed statistically significant between-group differences for both costs and outcomes: at 12 months, the cDMARDs group dominated with the group having better outcomes (mean difference −0.16, 95% CI −0.32 to −0.01) and lower costs from a health-care perspective (mean difference −£1930, 95% CI −£2599 to −£1301), societal perspective excluding benefits (mean difference −£1974, 95% CI −£2648 to −£1334) and societal perspective including benefits (mean difference −£1977, 95% CI −£2644 to −£1338). These translated into ICERs of −£12,063, −£12,338 and −£12,356 per QALY respectively. All other cost–outcome combinations suggest that the cDMARDs group is superior, with equivalent outcomes achieved at a significantly lower cost. The conclusions remained the same when costs and outcomes for those lost to follow-up were imputed. It was not necessary to compute any ICERs as none of the combinations suggested a significantly better outcome at a significantly lower cost.

Figures 21 and 22 show the probability that the cDMARDs group is cost-effective compared with the TNFis group for each outcome from a health and social care perspective at 6 and 12 months respectively. Both EQ-5D- and SF-36-based QALYs at each time point suggest that the probability that the cDMARDs group is cost-effective is ≥ 99% at all willingness-to-pay thresholds that were examined.

FIGURE 21.

Cost-effectiveness acceptability curves at 6 months from a health and social care perspective for all outcomes. (a) HAQ; (b) SF-36; and (c) EQ-5D.

FIGURE 22.

Cost-effectiveness acceptability curves at 12 months from a health and social care perspective for all outcomes. (a) HAQ; (b) SF-36; and (c) EQ-5D.

The probability that the cDMARDs group is cost-effective at 6 months based on the HAQ is 100% for willingness-to-pay thresholds of up to £10,000 per point improvement on the HAQ but decreases at higher willingness-to-pay thresholds. At 12 months the probability of cost-effectiveness is 100% for willingness-to-pay thresholds of up to £10,000 per point improvement on the HAQ and remains at 99% up to a threshold of £50,000.

Early rheumatoid arthritis

Trials

The preliminary search identified 463 papers, of which 36 were potentially relevant trials and were selected for full-text review (Figure 23). Of these, four trials were excluded: one included patients with a disease duration of > 3 years, two used treatment strategies in which the same approaches were included in both arms and one used steroids with methotrexate in the control monotherapy arm. The remaining 32 trials68,111,147,171,192–219 formed the basis of this systematic review.

FIGURE 23.

Selection of trials in early RA.

The baseline characteristics of the 32 RCTs are summarised in Table 38. The trials randomised between 20 and 1049 patients and enrolled over 8400 patients. In total, 19 trials compared cDMARDs with methotrexate,68,111,171,192–207 10 trials compared TNFis/methotrexate with methotrexate monotherapy208–217 and three trials compared cDMARDs with TNFis/methotrexate directly (head-to-head trials). 147,218,219 The Optimal Protocol for Methotrexate and Adalimumab Combination Therapy in Early Rheumatoid Arthritis (OPTIMA)217 and High Induction Therapy with Anti-Rheumatic Drugs (HIT-HARD)216 studies withdrew anti-TNF treatment from 24 and 26 weeks, respectively; therefore, only outcomes at 24 and 26 weeks, respectively, were considered. The BeSt220 and Swedish Farmacotherapy (Swefot)221 trials initially published 12-month results and subsequently 24-month results.

TABLE 38 - Characteristics of included RCTs in early RA

CsA, ciclosporin; HCQ, hydroxychloroquine; IA, intra-articular; MTX, methotrexate; SSZ, sulfasalazine.

Study	Year	Design	Cases	Max RA duration	Quality assessments				Follow-up	Combinations
					Allocation	Blinding	Analysed in original groups	Analysed for bias
cDMARDs
Boers et al.68	1997	Step-down	156	2 years	Unstated	Double	Yes	Yes	1 year	SSZ/MTX/prednisolone vs. SSZ
Haagsma et al.192	1997	Parallel	105	1 year	Unstated	Double	Yes	No	1 year	MTX/SSZ vs. SSZ vs. MTX
van den Borne et al.193	1998	Step-up	88	3 years	Central block randomisation	Double	Yes	No	0.5 years	Chloroquine/CsA vs. chloroquine
Dougados et al.194	1999	Parallel	209	1 year	Unstated	Double	Yes	No	1 year	MTX/SSZ vs. SSZ vs. MTX
Mottonen et al.195	1999	Step-down	199	2 years	Cards	Open	Yes	Yes	2 years	SSZ/MTX/HCQ/prednisolone vs. SSZ
Proudman et al.196	2000	Parallel	82	1 year	Centralised randomisation list	Open	Yes	Yes	1 year	CsA/MTX/IA methylprednisolone vs. SSZ
Ferraccioli et al.197	2002	Step-up	126	–	Unstated	Open	Yes	No	1.5 years	MTX/CsA vs. SSZ
Gerards et al.198	2003	Parallel	120	3 years	Computer generated	Double	Yes	No	1 year	CsA/MTX vs. CsA
Marchesoni et al.199	2003	Parallel	61	2 years	Randomisation list	Open	Yes	No	1 year	CsA/MTX vs. MTX
Capell et al.200	2004	Parallel	167	3 years	Unstated	Double	Yes	No	2 years	SSZ/prednisolone vs. SSZ
Grigor et al.111	2004	Step-up	111	5 years	Computer generated	Single	Yes	No	1.5 years	Intensive combinations vs. routine care
Miranda et al.201	2004	Parallel	149	2 years	Computer generated	Double	Yes	No	1 year	CsA/chloroquine vs. CsA
Ichikawa et al.202	2005	Parallel	71	2 years	Randomised by test drug number	Double	Yes	No	1.8 years	MTX/bucillamine vs. MTX vs. bucillamine
Sarzi-Puttini et al.203	2005	Parallel	105	3 years	Unstated	Open	Yes	Yes	1 year	CsA/HCQ vs. CsA/MTX vs. CsA
Svensson et al.204	2005	Parallel	259	1 year	Computer generated	Open	Yes	No	2 years	Prednisolone/DMARD vs. DMARD
Wassenberg et al.205	2005	Parallel	192	2 years	Computer generated	Double	Yes	No	2 years	Prednisolone/DMARD vs. DMARD
Hetland et al.206	2006	Parallel	163	6 years	Computer generated	Double	Yes	Yes	1 years	MTX/CsA/IA beclamethasone vs. MTX/IA beclamethasone
O’Dell et al.207	2006	Parallel	66	1 year	Cards	Double	Yes	No	2 years	Doxycycline/MTX vs. MTX
Choy et al.171	2008	Step-up	467	2 years	Computer generated	Double	Yes	Yes	2 years	MTX vs. MTX + CsA vs. MTX + prednisolone vs. triple
TNFi/MTX
Breedveld et al.208	2004	Parallel	82	3 years	Unstated	Double	Yes	No	2 years	Infliximab/MTX
St Clair et al.209	2004	Parallel	1049	3 years	Interactive voice response	Double	Yes	Yes	1 year	Infliximab/MTX
Taylor et al.210	2004	Parallel	24	3 years	Pharmacist randomisation	Double	Yes	No	1 year	Infliximab/MTX
Quinn et al.211	2005	Parallel	20	1 year	Adaptive stratified randomisation technique	Double	Yes	No	1 year	Infliximab/MTX
Breedveld et al.212	2006	Parallel	799	3 years	Unstated	Double	Yes	No	2 years	Adalimumab/MTX
Durez et al.213	2007	Parallel	44	1 year	Unstated	Open	Yes	No	1 year	Infliximab/MTX
Emery et al.214	2008	Parallel	542	2 years	Computer generated	Double	Yes	Yes	1 year	Etanercept/MTX
Soubrier et al.215	2009	Parallel	65	0.5 years	Unstated	Open	Yes	Yes	1 year	Adalimumab/MTX
Detert et al.216	2013	Parallel	172	1 year	Unstated	Double	Yes	Yes	0.5 years	Adalimumab/MTX
Kavanaugh et al.217	2013	Parallel	1032	1 year	Interactive voice response	Double	Yes	Yes	0.5 years	Adalimumab/MTX
Direct comparisons
Goekoop-Ruiterman et al.147	2005	All three	508	2 years	Variable block randomisation	Open	Yes	Yes	1 year	Step-up vs. step-down vs. infliximab
van Vollenhoven et al.218	2009	Step-up	487	1 year	Computer generated	Open	Yes	No	1 year	MTX/SSZ/HCQ vs. infliximab/MTX
Moreland et al.219	2012	Parallel	755	3 years	Computer generated	Double	Yes	Yes	2 years	MTX/SSZ/HCQ vs. etanercept/MTX

American College of Rheumatology responses and withdrawals for inefficacy

Indirect comparisons showed that in trials of DMARD combinations (Table 40 and Figure 24) more patients achieved ACR20–70 responses with combination therapy (OR 1.76–2.81) and less patients withdrew because of inefficacy with combination therapy (OR 0.47, 95% CI 0.34 to 0.64). In trials of TNFi/methotrexate combinations more patients achieved ACR20–70 responses with combination therapy (OR 1.88–2.22) and fewer patients withdrew because of inefficacy with combination therapy (OR 0.44, 95% CI 0.22 to 0.85). Sensitivity analysis of trials using only methotrexate monotherapy showed similar results.

TABLE 40 - Indirect comparisons in early RA: summary of meta-analysis of key outcomes – ACR responses, patient withdrawals, HAQ score and radiological progression

Outcome	Treatment regimen	Studies	Random-effects analyses
			OR (95% CI)
Categorical outcomes
ACR20	DMARD combinations	14	1.76 (1.26 to 2.46)
	DMARD combinations (methotrexate only)	5	2.01 (1.08 to 3.72)
	TNFi/methotrexate	8	1.88 (1.61 to 2.19)
ACR50	DMARD combinations	13	2.34 (1.40 to 3.91)
	DMARD combinations (methotrexate only)	5	1.64 (1.15 to 2.34)
	TNFi/methotrexate	7	2.09 (1.80 to 2.43)
ACR70	DMARD combinations	8	2.81 (1.48 to 5.33)
	DMARD combinations (methotrexate only)	4	2.00 (1.32 to 3.02)
	TNFi/methotrexate	7	2.22 (1.78 to 2.76)
Inefficacy withdrawals	DMARD combinations	15	0.47 (0.34 to 0.64)
	DMARD combinations (methotrexate only)	7	0.52 (0.33 to 0.82)
	TNF/methotrexate	9	0.44 (0.22 to 0.85)
Toxicity withdrawals	DMARD combinations	15	1.86 (1.42 to 2.44)
	DMARD combinations (methotrexate only)	7	2.69 (1.49 to 4.83)
	TNFi/methotrexate	9	1.42 (0.87 to 2.34)
			WMD (95% CI)
Continuous outcomes
Disability (HAQ)	DMARD combinations	7	−0.03 (−0.12 to 0.07)
	DMARD combinations (methotrexate only)	2	−0.17 (−0.33 to −0.01)
	TNFi/methotrexate	2	−0.16 (−0.24 to −0.08)
Radiological progression	DMARD combinations	7	−0.99% (−1.11% to −0.87%)
	DMARD combinations (methotrexate only)	4	−1.21% (−1.37% to −1.04%)
	TNFi/methotrexate	4	−0.61% (−0.79% to −0.43%)

FIGURE 24.

Forest plots of ACR20 responses in early RA: results for (a) cDMARD vs. methotrexate trials; and (b) TNFi/methotrexate vs. methotrexate trials.

Direct comparisons showed that there were no differences between DMARD combinations (Table 41) and TNFi/methotrexate with regard to ACR20 outcomes or patient withdrawals because of inefficacy. However, fewer patients achieved ACR50 and ACR70 responses using cDMARDs than using TNFi/methotrexate (ORs 0.54 and 0.53 respectively). A more detailed analysis of data from each of these trials is shown in Figure 25. Overall, there were small differences in favour of TNFi/methotrexate compared with cDMARDs at most time points but these were not always significant. There were also marked differences in response rates in the different trials.

TABLE 41 - Direct comparisons (cDMARDs vs. TNFi/methotrexate) in early RA: summary of meta-analysis of all outcomes – ACR responses, patient withdrawals and radiological progression

Note

TNFi/methotrexate was the control group and cDMARDs was the comparison group (OR < 1 indicates reduced response.

Outcome	Studies	Random-effects analyses
		OR (95% CI)
Categorical outcomes
ACR20	2	0.74 (0.42 to 1.29)
ACR50	1	0.54 (0.33 to 0.90)
ACR70	2	0.53 (0.36 to 0.79)
Inefficacy withdrawals	3	3.28 (0.51 to 21.28)
Toxicity withdrawals	3	1.63 (0.78 to 3.43)
		WMD (95% CI)
Continuous outcome
Radiological progression	2	0.22 (−0.02 to 0.45)

FIGURE 25.

American College of Rheumatology responses in head-to-head trials in early RA. (a) ACR20; and (b) ACR70 responses from the BeSt,147 Treatment of Early Aggressive Rheumatoid (TEAR)219 and Swefot trials. 218 The trials variously reported outcomes at 6, 12 and 24 months.

Established rheumatoid arthritis

Trials

The preliminary search identified 3642 papers, of which 28 were potentially relevant and were selected for full-text review (Figure 26). Of these, nine studies were excluded: in four patients were treatment naive, in two patients had not received DMARDs for > 3 months, two did not specify previous DMARD treatment and one was a duplicate of an included study. The remaining 19 studies66,67,170,222–237 were included in the systematic review and are summarised in Table 42. In total, 10 trials compared cDMARDs with DMARD monotherapy,66,67,222–229 of which six used methotrexate monotherapy as the control arm,66,67,222,223,227,229 and eight compared TNFi/methotrexate with methotrexate monotherapy,170,230–236 with one involving infliximab,170 two etanercept,230,232 one adalimumab,231 two golimumab233,236 and two certolizumab pegol. 234,235 For the Trial of Etanercept and Methotrexate with Radiographic Patient Outcomes (TEMPO)232 and Rheumatoid Arthritis Prevention of Structural Damage 1 (RAPID1)234 trials, 2-year follow-up data were subsequently published. 238,239 Finally, one trial made a direct comparison between methotrexate/sulfasalazine/hydroxychloroquine and etanercept/methotrexate. 237

FIGURE 26.

Selection of trials in established RA.

TABLE 42 - Characteristics of RCTs in established RA

AZA, azathioprine; CsA, ciclosporin; HCQ, hydroxychloroquine; IM, intramuscular; MTX, methotrexate; PGT, gold aurothiomalate; SSZ, sulfasalazine.

Study	Year	Cases	Refractory to treatment	Quality assessments				Follow-up	Treatment arms
				Allocation	Blinding	Analysed in original groups	Analysed for bias
cDMARDs
Ferraz et al.222	1994	82	Failed more than one DMARD	Unstated	Double	Yes	No	0.5 years	MTX vs. MTX/chloroquine
Tugwell et al.66	1995	148	MTX > 3 months	Unstated	Double	Yes	No	0.5 years	MTX vs. CsA/MTX
Willkens et al.223	1995	209	Resistant to penicillamine/gold	Permutated blocks	Double	Yes	Yes	1 year	MTX vs. MTX/AZA
Bendix et al.224	1996	40	Insufficient response to gold	Computerised	Double	Yes	No	0.5 years	PGT vs. PGT/CsA
O’Dell et al.67	1996	102	More than one DMARD	Cards	Double	Yes	Yes	2 years	MTX vs. MTX/SSZ/HCQ
Kremer et al.225	2002	263	MTX > 6 months	Computerised	Double	Yes	Yes	0.5 years	Leflunamide vs. leflunamide/MTX
Dougados et al.226	2005	106	Inadequate response to leflunomide	Unstated	Double	Yes	No	0.5 years	SSZ vs. leflunamide/SSZ
Lehman et al.227	2005	65	MTX > 12 weeks	Random number table	Double	Yes	Yes	1 year	MTX vs. MTX/IM gold
Karanikolas et al.228	2006	106	Refractory to more than one DMARD	Unstated	Open	Yes	No	1 year	Leflunamide vs. CsA/leflunamide
Capell et al.229	2007	191	SSZ > 6 months	Computerised	Double	Yes	No	1 year	MTX vs. MTX/SSZ
TNFis/MTX
Weinblatt et al.230	1999	89	MTX > 6 months	Unstated	Double	Yes	No	0.5 years	MTX vs. etanercept/MTX
Lipsky et al.170	2000	428	MTX > 6 months	Unstated	Double	Yes	No	1 year	MTX vs. 3 mg infliximab/MTX
Weinblatt et al.231	2003	271	MTX > 6 months	Unstated	Double	Yes	No	0.5 years	MTX vs. 40 mg adalimumab/MTX
Klareskog et al.232	2004	686	More than one DMARD other than MTX	Centralised telephone	Double	Yes	Yes	0.5 years	MTX vs. etanercept/MTX
Kay et al.233	2008	172	MTX > 3 months	Unstated	Double	Yes	No	0.3 years	MTX vs. 50 mg golimumab/MTX
Keystone et al.234	2008	982	MTX > 6 months	Unstated	Double	Yes	No	1 year	MTX vs. 200 mg certolizumab/MTX
Smolen et al.235	2009	619	MTX > 6 months	Unstated	Double	Yes	Yes	0.5 years	MTX vs. 200 mg certolizumab/MTX
Kremer et al.236	2010	643	MTX > 3 months	Interactive voice response	Double	Yes	Yes	0.3 years	MTX vs. 2 mg/kg golimumab/MTX
Direct comparison
O’Dell et al.237	2013	353	MTX > 3 months	Unstated	Double	Yes	No	1 year	MTX/SSZ/HCQ vs. etanercept/MTX

American College of Rheumatology responses and withdrawals for inefficacy

In trials of DMARD combinations more patients achieved ACR20–70 responses with combination therapy (OR 2.75–5.07), as shown in Table 44 and Figure 27. More patients withdrew with combination therapy (OR 1.51, 95% CI 1.02 to 2.25). Sensitivity analysis of RCTs that included a methotrexate monotherapy arm showed that more patients achieved ACR20–70 responses with combination therapy (OR 3.55–4.74) but few patients withdrew because of inefficacy (OR 0.34, 95% CI 0.20 to 0.59).

TABLE 44 - Summary of outcomes in established RA trials

Outcome	Treatment	Studies	Random-effects analyses
			OR (95% CI)
Categorical outcomes
ACR20	DMARDs (all)	6	2.75 (1.79 to 4.22)
	DMARDs (methotrexate only)	4	3.55 (2.43 to 5.17)
	TNFi/methotrexate	8	5.32 (3.03 to 9.34)
ACR50	DMARDs (all)	6	5.07 (3.10 to 8.29)
	DMARDs (methotrexate only)	4	4.70 (2.40 to 9.20)
	TNFi/methotrexate	8	8.13 (4.26 to 15.52)
ACR70	DMARDs (all)	5	4.85 (2.34 to 10.05)
	DMARDs (methotrexate only)	3	4.74 (1.65 to 13.61)
	TNFi/methotrexate	8	5.36 (2.92 to 9.83)
Inefficacy withdrawals	DMARDs (all)	10	0.38 (0.24 to 0.62)
	DMARDs (methotrexate only)	7	0.34 (0.20 to 0.59)
	TNFi/methotrexate	8	0.12 (0.06 to 0.25)
Toxicity withdrawals	DMARDs (all)	10	1.51 (1.02 to 2.25)
	DMARDs (methotrexate only)	7	1.58 (0.97 to 2.59)
	TNFi/methotrexate	8	0.94 (0.62 to 1.41)
			WMD (95% CI)
Continuous outcome
Disability (HAQ)	DMARDs (all)	3	−0.19 (−0.27 to −0.10)
	DMARDs (methotrexate only)	1	−0.30 (−0.42 to −0.18)
	TNFi/methotrexate	1	−0.35 (−0.56 to −0.14)

FIGURE 27.

Forest plots of ACR20 responses in established RA: results for (a) cDMARD vs. DMARD monotherapy trials; and (b) TNFi/methotrexate vs. methotrexate trials.

In trials of TNFi/methotrexate combinations more patients achieved ACR20–70 responses with combination therapy (OR 5.32–8.13), as shown in Table 44. Fewer patients withdrew because of inefficacy with combination therapy (OR 0.12, 95% CI 0.06 to 0.25).

The trial comparing triple DMARD therapy with etanercept/MTX237 showed no statistical difference between groups in ACR20 (57% vs. 66%), ACR50 (35% vs. 43%) and ACR70 (18% vs. 26%). This study did not report patient withdrawals for inefficacy.

Key findings

Patients with active established RA who meet current NICE criteria to receive TNFis achieve equivalent reductions in disability and improvements in quality of life over 12 months by treating initially with cDMARDs and reserving TNFis for cDMARD non-responders. The cDMARD strategy costs substantially less. However, neither treatment strategy was ideal. The majority of patients in both groups failed to achieve a DAS28 of ≤ 2.6, which is often considered to indicate remission.

The DAS28 improved more rapidly in patients receiving TNFis. Overall, monthly DAS28 were lower in patients receiving TNFis and more patients receiving TNFis achieved a DAS28 response (decrease in score of ≥ 1.2) within the first 6 months. This benefit of TNFis with regard to the DAS28 response particularly reflected rapid and sustained reductions in ESR in this group. However, the benefits of TNFis with regard to the DAS28 response were small and did not result in improvements in disability or quality of life. There was also no evidence that patients who received cDMARDs had more erosive progression. Larsen scores showed that both groups had comparable, minimal radiological progression.

Serious adverse events and withdrawals because of toxicity were equally common with cDMARDs and TNFis. However, the total number of adverse events, spanning both serious and more minor events, was higher with cDMARDs. This was most marked for adverse reactions involving the digestive system.

As TNFis are more expensive than cDMARDs, the economic evaluation showed that the cDMARD group was substantially more cost-effective, whatever approach was taken to assessing costs and relevant outcomes. This included incorporating societal costs such as lost time from work and social security benefit claims into the calculations.

In total, 44% of the patients in the cDMARD group were recommended to switch to a TNFi because their disease activity had not improved after 6 months of treatment. However, there was no evidence that patients who switched in this way had a worse quality of life, more disability or more erosive progression. There was therefore no evidence that these ‘switchers’ had any long-term disadvantages from taking cDMARDs for 6 months.

Limitations and sources of bias

Not all patients invited to participate agreed to do so; overall, 192 out of 432 patients (44%) declined to take part in the trial. We cannot be certain that the patients who did not consent to the trial would have responded in the same way as those who took part. 240 However, this is only one of a number of causes of bias in trials of long-term diseases241 and does not seem a crucial factor compared with the range of issues influencing such trials. In addition, patient choice is of crucial importance and accepting that not everyone will agree to participate is an inevitable consequence of informed choice around clinical trials. In addition, as considered below, TACIT patients receiving TNFis were similar to those in the UK national register (see pp. 90).

Those patients who did not respond to cDMARDs in the TACIT trial were treated with TNFis. It could therefore be argued that over time the two arms of the trial become very similar if not identical. However, only a minority of patients were involved in switching, with 44% of patients who started cDMARDs switching to TNFis in the second 6 months. As a consequence, the two trial arms remained sufficiently different to make this a genuine comparison. Furthermore, the 6-month comparisons did not include any patients randomised to cDMARDs who had received TNFis (because no-one switched until after 6 months). Therefore, at 6 months there was a genuine head-to-head comparison within the TACIT trial. Switching between treatment strategies is normal clinical practice and over time many RA patients starting DMARDs or biologics will switch to other treatments.

The cDMARD treatment was not standardised and it could be argued that the therapy given was too heterogeneous, making it an intervention that could be difficult to reproduce. This is an intellectual challenge as the only way to standardise cDMARD treatment is to study early RA patients who are DMARD naive or study methotrexate non-responders. These patients do not meet existing NICE criteria for receiving TNFis. If anything, the cDMARD treatments used were too conservative. We had hoped that patients would receive more intensive treatment and more short-term steroids in the cDMARD arm. However, supervising clinicians and patients placed more emphasis on slowly changing treatment to limit toxicity rather that giving maximal-dose therapy as soon as possible. Over time we anticipate that the use of cDMARDs will increase and concerns about toxicity may consequently lessen. We also accept that some combinations may be more effective, although this could not be resolved in the TACIT trial. More trials of different cDMARDs would be needed to answer this question.

Steroid use in the cDMARD group, including intramuscular injections, was less than anticipated when designing the TACIT trial based on our previous experience with steroids in established RA. 242 UK rheumatologists may have concerns about treating many patients with steroids because of the risk of adverse events. However, the relatively limited use of steroids would serve to reduce rather than magnify the impact of cDMARDs. More intensive steroid use could make cDMARD treatment even more effective.

It could be argued that the same results could have been obtained from starting another DMARD monotherapy and that the use of intensive DMARD combinations in the TACIT trial was not needed. This is a theoretical rather than a practical issue as there is no reason to stop one DMARD and start another in active RA in the absence of adverse events. DMARDs often have long half-lives, particularly agents such as leflunomide. Consequently, washing out current DMARDs and then starting a new DMARD monotherapy has limitations for patients as well as being of limited interest as the toxicity of modern DMARDs used in combination is not excessive.

The use of DMARDs other than methotrexate in combination with TNFis could have reduced the efficacy of these treatments in some patients. However, to do otherwise would be to move away from current UK practice, in which a range of DMARDs are given with TNFis. There is some evidence supporting the use of these different DMARDs in combination, as shown in our systematic reviews.

The use of the HAQ as the primary outcome measure might be viewed by some experts as being inappropriate, as the opportunity to reverse HAQ scores decreases with increasing disease duration. 243,244 Although this is theoretically correct, both of our groups showed clinically relevant reductions in HAQ scores over 12 months. In addition, the disease duration of patients in the TACIT trial (median duration of < 6 years in both groups) was below that in the Phase III trials that have led to the approval of the different TNFis. Furthermore, the degree of reduction in HAQ scores in the TACIT trial was similar to that reported in previous trials of biologics. In our view, if TNFis do not substantially reduce HAQ scores compared with other treatments then their potential clinical value is limited.

The TACIT trial was not a blinded trial. It could be argued that being unblinded influenced patients and clinicians to favour cDMARDs inappropriately. However, it was impractical to have a fully blinded treatment strategy involving multiple drugs, all of which need careful monitoring. Given the enthusiasm of most clinicians and most patients for receiving high-cost biological treatment, we consider that unblinding would, if anything, benefit TNFis. The issue of blinding is related to a number of ethical matters; these are considered in detail in the subsequent section on ethical issues.

The TACIT trial and similar trials of efficacy are not able to completely assess the relative impacts of different types of adverse event on clinical outcomes. Many adverse events reported in patients taking cDMARDs in the TACIT trial, particularly gastroenterological events, may have been relatively minor. Many patients might have felt that the treatment was ‘worth it’, irrespective of these adverse events. Although we collected detailed information about such adverse events, we did not assess their specific impacts on patient outcomes and whether they affected patients’ quality of life. We are therefore not able to fully determine the clinical consequences of such adverse events. Large long-term observational studies are needed to fully assess the impact of adverse events on treatment outcomes. Despite this limitation, the data that we collected in the TACIT trial on serious adverse events and adverse events linked to stopping treatment provided no evidence that there were more such events in the cDMARDs group than in the TNFis group.

The trial involved dividing patients into two groups after 6 months based on change in DAS28: responders and non-responders. This approach resulted in 46 out of 104 (44%) patients switching to TNFis after starting with intensive DMARDs. However, more than half (56%) of patients did not switch. This simple concept hides a more complex problem. Patients can fall on one side and then the other of such an arbitrary response without there being a major change in their condition, and the duration that patients remain within a defined state is not captured using such as approach, a problem discussed by Farewell and Su. 245 This general issue applies whenever arbitrary cut-offs are used at single time points in longitudinally collected data. It shows the difficulty of comparing the extent to which patients benefited from treatment.

Analytical issues

The TACIT trial was analysed on an ITT basis using multiple imputations and the primary outcome was compared using logistic regression methods. The HAQ is a complex assessment and it does not invariably behave as a conventional numerical scale. 246 There are identical issues with regard to the linearity of the scales of other key outcome measures, including the EQ-5D and Larsen scores. As both trial arms gave very similar outcomes using the HAQ, EQ-5D and Larsen scores, there is little merit in such an argument. In addition, the overwhelming balance of advice that we received favoured the analytical method that we preselected.

Not all of the outcomes confirmed equivalence. Changes in DAS28 and ESR favoured TNFis, particularly within the first 3–6 months. In part, this reflects the rapid onset of response with TNFis and the slow onset of response with DMARDs, which historically, and probably more accurately, used to be known as slow-acting drugs. Most DMARDs show maximal effects only by 6 months.

Measurement issues

There are several relevant measurement issues when measuring HAQ scores, radiological progression and DAS28. The HAQ was the primary outcome measure and its validity as an assessment instrument is therefore of most concern. The validity, reliability and responsiveness of the HAQ were reviewed by Linde et al. 158 Overall, the HAQ has appropriate measurement properties for a patient-generated outcome measure. It is not necessarily a simple scale and Rasch analysis by Wolfe247 and Tennant et al. 246 have highlighted some relative weaknesses. Our own previous research contributions suggest that the HAQ is the best available measure to use in trials such as the TACIT trial. 145,248,249

There is debate about the minimum clinically important difference in HAQ score in routine practice as opposed to clinical trials. Pope et al. 250 suggest that this is smaller than the difference that is considered important in trials. From this perspective the change in HAQ score in patients randomised to receive cDMARDs, which was overall 0.15 greater than that seen in patients randomised to receive TNFis, might be clinically relevant. However, this perspective appears questionable. We predetermined the minimum clinically important difference in HAQ score in trial settings and believe that it is inappropriate to change it retrospectively. It is also a theoretical rather than a practical issue as our aim was to show that the TNFi strategy was not better than the cDMARD strategy; showing that the cDMARD strategy has benefits is an identical conclusion in terms of its influence in clinical practice.

The use of Larsen scores to assess erosive damage merits consideration. The radiographs were all scored using the modified Larsen method by one investigator (DS), who has contributed to a number of published trials using this method and has achieved appropriate reproducibility of scoring. 171,242,251–260 There is debate about the relative merits of different approaches to scoring radiographs. The TACIT trial followed the approach taken in the British Rheumatoid Outcome Study Group (BROSG) trial in established RA261 and we have no reason to doubt its validity or appropriateness.

The TACIT trial used the DAS28, which was scored by many clinicians in different clinics. Each centre received detailed information during initiation on the methods of scoring the DAS28. 262 However, we did not give either explicit standardisation training nor did we retrain observers periodically to assess whether or not they maintained consistent standards. There are challenges in assessing patients using the DAS28. 263 Although training increases clinicians’ short-term agreement when measuring joint counts,264 the overall benefit of such training is uncertain. 265 Training is useful within the national context to deliver high-quality care, but its value in an individual clinical trial is limited. As the TACIT trial is a strategy trial within routine care settings we consider that it needs to replicate standard methods and should not adopt more stringent approaches because these would limit its generalisability.

Strengths of the Tumour necrosis factor inhibitors Against Combination Intensive Therapy trial

The TACIT trial was undertaken in outpatient rheumatology clinics in England in conditions that, as far as is possible within a clinical trial, mirrored routine practice. The patients enrolled were typical of those treated within England and included patients from a range of ethnicities and levels of deprivation. They are similar to those reported in the British Society for Rheumatology (BSR) Biologics Register. 266 A comparison between patients enrolled in the BSR Biologics Register and those enrolled in the TACIT trial is shown in Table 46. As there is evidence that the patients enrolled in the BSR Biologics Register have changed over time, data from all available years are shown. For the HAQ and DAS28, patients in the TACIT trial had similar initial scores and similar changes in scores (in the TNFi group) to those of patients most recently enrolled in the BSR Biologics Register.

The TACIT trial focused on patient-centred outcomes. We consider this vital because such patient-centred outcomes have a central place in clinical trials in RA. Changes in measures such as the ESR, although of interest to clinicians, are of limited value to patients. There are also concerns about the interobserver reproducibility of assessing joint counts.

The TACIT trial was of sufficient size to provide robust assessments of the changes in measures. In addition, it showed benefits favouring cDMARD treatment. In other words, cDMARDs give somewhat better outcomes than just achieving equivalence. Although we do not think that the trial shows that cDMARDs are preferable, the chance of the conclusions being incorrect and of TNFis being better appears remote.

The TACIT trial showed that only a minority of patients randomised to TNFis achieved DAS28 of ≤ 2.6 and the use of cDMARDs also resulted in relatively few DAS28 of ≤ 2.6. These low scores are often considered to reflect remission although, as discussed earlier, defining remission is an ongoing challenge. The frequency of such ‘remission scores’ in the TACIT trial is similar to that reported by both the BSR Biologics Register266 and other international registers of patients receiving TNFis in routine clinical practice267–271 (Table 47). In addition to achieving few single low DAS28 of ≤ 2.6, few TACIT trial patients achieve sustained remission. There is a need for more research on the nature and predictors of sustained low DAS28 and other indicators of remission, but this problem lies outside the remit of the TACIT trial.