Applying science in practice: the optimization of biological therapy in rheumatoid arthritis

Most authorities recommend starting biological agents upon failure of at least one disease-modifying agent in patients with rheumatoid arthritis. However, owing to the absence of head-to-head studies, there is little guidance about which biological to select. Still, the practicing clinician has to decide. This review explores the application of published evidence to practice, discussing the goals of treatment, the (in) ability to predict individual responses to therapy, and the potential value of indirect comparisons. We suggest that cycling of biological agents, until remission is achieved or until the most effective agent for that individual patient is determined, deserves consideration in the current stage of knowledge.


Introduction
Th e development of specifi c biological therapies has resulted in a remarkable improvement in the treatment of rheumatoid arthritis (RA) and also in the understanding of its complex pathogenesis. We better recognize the multitude of cells and biological pathways involved in the disease process. We have also become more aware of the individual variability in disease features and in patterns of response to therapy. A large array of new treatment opportunities is currently under development and soon will be available as new biological agents. While enjoying these fruits of research, rheumatologists face the challenge of defi ning the best therapeutic plan for patients who have failed classical disease-modifying antirheumatic drugs (DMARDs).

Remission is now a realistic therapeutic goal in every patient
It is certainly desirable that our patients feel better and have improved function and acute-phase reactants as measured by response criteria, but the remaining infl ammatory activity (status) seems decisive: 'It is good to feel better but it is better to feel good' [1]. Aletaha and colleagues [2] have demonstrated, in a pooled analysis based on data from several clinical trials in RA involving anti-tumor necrosis factor (anti-TNF), that within the ACR50 (American College of Rheumatology 50% improve ment criteria) and ACR70 responder groups, the most important determinant of progression is the fi nal disease state and not the relative degree of improvement. In fact, functional ability was best and radiographic progression was lowest in patients who had attained disease remission at 1 year compared with those who had attained only low or moderate disease activity. Furthermore, among patients attaining the same disease activity category, physical function and radiographic progression did not diff er signifi cantly by the level of response. Even with low disease activity, damage progresses and only sustained remission is capable of abrogating progression of joint destruction [3]. Moreover, optimal disease control is associated with less work disability [4], lower mortality rates [5][6][7], and better quality of life [8,9]. Even if low disease activity is achieved, work productivity, quality of life, and health states are still signifi cantly worse when compared with remission [9]. Remission used to be a 'guiding utopia' but now, thanks to biological therapy, is a very realistic therapeutic objective. Now that we have in our hands a variety of safe and effi cacious medications to achieve it, remission should be our goal in every patient with RA and we should try to achieve that goal as soon as possible [10].

Defi ning remission
Remission, our elected goal, should be understood as a near-complete suppression of disease activity or an absence of discernable disease activity [11]. Which of the current defi nitions of remission should we adopt for practice and for evaluation of the effi cacy of diff erent treatment regimes? Remission defi nitions (Table 1) and their diff erences have been assessed and reviewed in detail [12][13][14][15][16][17]. As expected, the proportion of patients achieving remission is dependent on how it is defi ned [17].
Molenaar and colleagues [18] found that some patients in clinical remission, defi ned according to the modifi ed American Rheumatism Association (ARA) criteria [19][20][21] or the disease activity score (DAS) criteria [20], still showed radiographic progression during a 2-year followup, although to a lesser extent than patients having an exacer bation. Th ese fi ndings suggest that DAS/disease activity score with 28-joint assessment (DAS28) and ARA remission criteria may actually describe a low disease activity state rather than a true remission state [22]. In recent trials, DAS28 remission rates exceeded ACR70 response rates [23][24][25][26], meaning that more patients achieved a state of DAS28 remission than the proportion of patients reach ing a decrease of 70% or higher in tender and swollen joints. However, other reports have suggested that DAS remission is a more strict criterion for remission and that an ACR70 response should not be used as a surrogate for remission [27]. Th is should not be surprising given that ACR20/ACR50/ACR70 is designed to measure improve ment in disease activity, not to defi ne remission. Impor tantly, some recent reports have suggested that the clinical disease activity index (CDAI) and simplifi ed disease activity index (SDAI) [28] may represent remis sion criteria that are more stringent than those by the DAS/DAS28 and the modifi ed ARA response criteria because the latter allow for signifi cant residual disease activity [27,29]. Consequently, smaller proportions of patients may be classifi ed as in remission by SDAI and CDAI criteria than by DAS/DAS28 and modifi ed ARA criteria, and this also has an impact on the percentage of patients showing radiological progression and on patients' mean health assessment questionnaire (HAQ) scores [16,30,31]. One limitation of all of these remission defi nitions (ARA, DAS/DAS28, and CDAI/ SDAI) is that they omit the eff ects of RA on functional disability and structural joint damage [28].
Owing to these limitations in the defi nitions of remission and recognizing its importance as a crucial goal in current management of RA, ACR and the European League Against Rheumatism (EULAR) set up a task force to redefi ne the concept. Th e main conclusions from the fi rst meeting are that the new remission defi nition should be strict, based on no or very low disease activity, and should be validated against long-term outcomes, specifi cally physical function and radiographic progres sion [14]. Th ose in remission should have a stable level of joint damage over time and should have less deterioration or more improvement in functional status over time (remission defi nition should have predictive validity) [14].
It is our opinion that, while we await this new defi nition, the practicing rheumatologist should do the following: 1. Choose persistent remission as the primary goal for every patient with RA. 2. Always measure disease activity, using any of the available tools (DAS/DAS28, CDAI/SDAI) to guide therapy toward remission (benchmarking). 3. If remission has been achieved, be critical about it.
Check whether the patient fulfi lls available remission criteria and if the patient does, ask yourself whether that is 'true remission' (for example, a very low erythro cyte sedimentation rate [ESR] or very low/ absent number of tender joints may be leading you to categorize the patient in a 'false state of remission').
Also, be critical about other disease states (for example, a patient with chronically elevated acute-phase reactants not related to RA or a patient with RA and fi bromyalgia may never fulfi ll current defi nitions of remission but may nonetheless be in 'true remission'). 4. If the patient is not in 'true' and persistent remission, continue the search for that goal. Th e remainder of this paper will discuss data on DAS/ DAS28 remission and ACR70 response rates.  [20,21] No joint pain by history ARA; 2 months required No joint tenderness or pain on motion No soft tissue swelling in joints or tendon sheaths Morning stiff ness for not more than 15 minutes ESR of less than 30 mm/hour in women and less than 20 mm/hour in men DAS [20] [0. 54

Do biologics vary in their ability to induce remission?
It can be expected that 10% to 30% of unselected patients with RA will achieve remission (natural history data) [32]. According to current guidelines in most countries, biologics should be considered if patients do not respond to traditional DMARDs, including methotrexate (MTX), during the fi rst few months of treatment [33]. Most clinicians who now face a traditional DMARD failure encounter the problem of selecting among currently approved biological therapies (adalimumab, etanercept, infl iximab, abatacept, rituximab, tocilizumab, certolizumab pegol, golimumab, and anakinra). According to the current methodological paradigm, no formal judgment on the relative merits of drugs can be made in the absence of head-to-head trials. Th e only published head-to-head trial with biologics had three arms: abatacept, infl iximab, and placebo. Th e relative effi cacies of the biologics were similar at 6 months, with signifi cant diff erences favoring abatacept at 1 year in DAS28, good EULAR response, low disease activity, and health-related quality of life [34]. One additional head-tohead trial recently started recruiting patients to be randomly assigned to tocilizumab or adalimumab and we are looking forward to the results [35]. Other trials of this kind are lacking and would be welcome. While waiting for more information from clinical trials, the clinician still needs to make a decision. How?
Looking independently at individual trials is not very informative. ACR20, ACR50, and ACR70 response rates in randomized trials are not very diff erent between diff erent drugs, and a crude analysis is not elucidative. Several aspects, including diff erences in placebo response rates, trial designs, inclusion and exclusion criteria, and safety profi le, should be taken into account. In the absence of any evidence of relevant diff erences between biologics, the clinician may leave the choice to the patient, allowing convenience of administration, access to medication, and safety concerns to serve as guides.
Indirect comparisons produced through formal statistical methodologies have been proposed in an attempt to help clinicians make a choice. We could fi nd three such analyses. Singh and colleagues [36] systematically reviewed the existing updated Cochrane systematic reviews of six biologic DMARDs for RA (abatacept, adalimumab, etanercept, infl iximab, rituximab, and anakinra). Th is review included biologic DMARDs alone used in standard approved doses or in combination with other biologic/traditional DMARDs compared with placebo alone or with placebo plus biologic/traditional DMARDs. Th e authors anticipated that the observed 'control event rate' (that is, the placebo eff ect) and the trial duration would be important eff ect modifi ers and adjusted for these factors in the analysis (Table 2). Following this method ology, fi ve biologics (abatacept, adalimumab, etaner cept, infl iximab, and rituximab) showed signifi cant superiority to placebo but did not diff er among themselves. A summary of statistics is presented for the main effi cacy outcome of the study, ACR50, as a global 'number needed to treat' (NNT) to benefi t from each drug: abatacept NNT = 5 (95% confi dence interval [CI] 3 to 10), adalimumab NNT = 4 (95% CI 3 to 6), etanercept NNT = 3 (95% CI 3 to 5), infl iximab NNT = 5 (95% CI 3 to 18), and rituximab NNT = 4 (95% CI 3 to 8).
Bergman and colleagues [37] conducted a mixedtreatment comparison of biologic DMARD effi cacy at 24 weeks, among traditional DMARD inadequate respon ders, in order to make treatment-to-treatment compari sons. Th is analysis included data from tocilizumab trials, and besides establishing comparisons with placebo, the authors further estimated the effi cacy of tocilizumab in comparison with other biologics (anti-TNF being considered a block). A signifi cant diff erence was found between all biologic DMARDs and placebo in the three outcomes (ACR20, ACR50, and ACR70) ( Table 3). Relative risks (RRs) compared with placebo were similar for ACR20, but for ACR50 and especially ACR70, tocilizumab had a higher probability of response than other biologic DMARDs. Tocilizumab had an effi cacy similar to that of other biologic DMARDs for ACR20 and ACR50 responses, but a signifi cantly higher ACR70 response compared with TNF-α inhibitors (RR 1.8, 97.5% CI 1.2 to 2.6) and abatacept (RR 2.0, 97.5% CI 1.3 to 3.1) ( Table 3).
Gartlehner and colleagues [38] also performed a metaanalysis and indirect comparisons between biologics in populations that had residual disease activity despite MTX treatment (that is, excluded MTX-naïve patients), but only the three anti-TNF treatments were included, and no signifi cant diff erences were found.
Th ese comparisons lack suffi cient data on remission because not all of the randomized controlled trials (RCTs) provide that information. Consequently, the most ambitious outcome one can derive from these studies is ACR70. Overall, the abovementioned studies suggest that the effi cacies of diff erent biologics regarding this endpoint are very similar. Th e only exception seems to be a higher response rate, for ACR70, for tocilizumab (anti-TNFs being considered together). Notably, however, whether or not a patient will experience an ACR70 response is dependent not only on the effi cacy of the intervention but also on the baseline level of disease activity, and this duality makes ACR70 responses tricky for indirect comparisons.
With respect to radiographic progression, a comparison between biologics is more diffi cult because of the heterogeneity of the methodology applied, but in general, all biologic drugs have shown the potential of successfully inhibiting structural damage progression.
Can the physician, having reviewed these data, distinguish biologics on the basis of published evidence for effi cacy? Th ese indirect analyses look promising and are very appealing to the reader but actually have a number of methodological drawbacks, and therefore fi ndings need to be interpreted with caution. What indirect compari sons actually do is adjust for the placebo response across trials under the assumption that the placebo response is a generic refl ection of the characteristics of the study population without active intervention. A second assumption is that there is a clear and linear relationship between the placebo response and the response in the active treatment group. In the 2010 EULAR recommendations for the management of RA [39], the expert team deliberately refrained from including these indirect comparisons. Having reviewed the literature systematically, the authors concluded that it was impossible to prioritize the several biologics.

Can biologics be distinguished on the basis of safety?
Singh and colleagues [36] evaluated withdrawals due to adverse events and concluded that, compared with patients receiving placebo, those receiving adalimumab and infl iximab were at signifi cantly higher risk of withdrawals due to adverse events (odds ratio [OR] ranging from 1.54 to 2.21). Patients receiving abatacept, etanercept, and rituximab did not diff er signifi cantly from those receiving placebo in this aspect. Indirect comparisons revealed that adalimumab was more likely to lead to withdrawals compared with etanercept (OR 1.89, 95% CI 1.18 to 3.04) and etanercept was less likely than infl iximab (OR 0.37, 95% CI 0.19 to 0.70). Th e results were also translated into an absolute value, in this case number needed to harm (NNH): adalimumab NNH = 39 (95% CI 19 to 162) and infl iximab NNH = 18 (95% CI 8 to 72). Th is comparison, unfortunately, does not include tocilizu mab, a drug that was shown to have a rate of withdrawals due to adverse events of 5.8 per 100 patientyears, driven mainly by elevated liver enzyme levels, infections, and benign and malignant neoplasms [40]. Additionally, there seem to be diff erences in the risk of tuberculosis (TB) among diff erent biologics, and this might contribute to the selection of the biological agent,

Table 2. Biologics combined 3-, 6-, and 12-month outcome data (ACR20/ACR50/ACR70), adjusted for control event rate
Combined 3-, 6-, and 12-month outcome data: relative risk (95% confi dence interval) versus placebo ACR20 Abatacept  especially in countries with a high prevalence of TB. Monoclonal antibodies (that is, infl iximab and adalimumab) are known for a higher risk of TB compared with soluble TNF receptor therapy (that is, etanercept) [41,42]. Minimal data on TB risk in patients treated with non-TNF biological therapies exist, but to date, this risk seems to be limited [42]. Overall, the safety profi le of biologics in long-term registries has been very satisfactory, and it is accepted that diff erences in safety profi le would not warrant a major impact in the selection of medication, given the potential benefi ts at stake [43].
In summary, the clinician is still faced with considerable diffi culty in performing an evidence-based selection of the best possible biologic to add to MTX. At the moment, little evidence of any signifi cant diff erence between the biologics exists. Th e clinician will weigh benefi ts against risks specifi c to each biologic in addition to considering the common adverse events of all of the biologics [44]. Examples of specifi c adverse events might be neutropenia, increases in total cholesterol, lipoproteins, and triglycerides associated with tocilizumab [26,40]; or progressive multifocal leukoencephalopathy associated with rituximab [45]. Of course, such diffi culties would be overcome if the individual response to each biologic could be reliably predicted. Th is would allow us to hope that we can do better than average.

Can response to individual therapeutic agents be predicted?
Th e fact that biologics are quite similar in terms of response rate at the group level does not preclude them from being completely diff erent at the individual level. Evaluation of RCTs is based on averages, and surely the majority of patients will not behave like the average; it may even happen that not a single one will. In a hypothetical scenario in which 20% of the population responds to one out of fi ve diff erent medications but to none of the others, all fi ve medications would have the same response rate at the group level (20%) and be considered indistinguishable. However, this conclusion would be wrong to each and every patient! In fact, diff erent agents may be eff ective in diff erent people, and if this concept is applied to the RA therapeutic arsenal, there might be such a thing as a disease primarily responsive to MTX, an anti-TNF biologic, an anti-interleukin-6 (IL-6) biologic, and so on. Diff erent 'sensitivity' to diff erent targeted agents is actually to be expected in a complex multifactorial disease like RA, as a function of the relative dominance of one biological pathway over the others in a particular individual, depending on environmental and genetic factors. It is conceivable that the dominant pathogenic mechanisms (and therefore drug responses) may even vary within a patient in the course of the disease. Th is concept cannot be proven at the moment, because the data available have been driven essentially by historical opportunity: anti-TNF were the fi rst to become available and for that reason they became fi rst line, and the alternatives used upon their failure. Moreover, current methodological wisdom is based on average responses and subgroup analysis is precluded.
Th e optimal selection among these medications for an individual patient would require that we be able to identify the subset of patients who would respond better to each drug. In this case, initial treatment could be tailored to the individual and we could aim at shortening the time to onset of eff ective treatment, improving the cost-benefi t and risk-benefi t ratios of these agents, and eventually achieving 100% response rate with minimal toxicity.
Several demographic and clinical characteristics as well as serological biomarkers have been studied as predictors of treatment response. Large-scale genetic and proteome studies are now available and have led to the study of genetic polymorphisms (pharmacogenetics) and screening of large amounts of gene transcripts (transcriptomic analysis of pharmacogenomics) and proteins (proteomic analysis of pharmacogenomics) as candidate biomarkers. Several polymorphisms in genes of MTX transporters and the folic acid and adenosine pathways have been studied for MTX response, whereas for anti-TNF-α response, major histocompatibility complex and Fcrecep tor polymorphisms have been the main candidates.
So far, however, no clear-cut relationships between demo graphic, clinical, biochemical, or genetic factors and RA response to biological therapy have been established [46,47]. Th is may refl ect diff erences in study design, diffi culty in controlling for confounders (such as ethnicity, age, disease duration, concomitant therapy, and smoking), or simply the inadequacy of proposed markers [47][48][49]. It is likely that sensitivity to therapies depends on a conjunction of factors whose study will require complex models combining genetic and non-genetic factors, as proposed by Wessels and colleagues [50] for predicting the effi cacy of MTX monotherapy.
In regard to response to anti-TNF agents, most of the studies performed to date have been small, underpowered, and restricted to the analysis of single candidate genes. Th e only replicated and validated genetic predictor of anti-TNF response is the 308GA single-nucleotide polymorphism in the TNF promoter region, but the amount of variation in response accounted for by this marker is probably modest and was questioned in a recent meta-analysis [51]. It is still unknown whether variation in treatment response is determined by several genes that each have a small eff ect size or by small numbers of genes with large eff ect sizes. Authors agree on the need for a large-scale, non-hypothesis-driven approach to identify further genetic markers of anti-TNF response [47]. In summary, at present, there is no robust biomarker to allow the prediction of responsiveness of individual patients to each biological agent.

Conclusions
Now what? How should the practicing rheumatologist select biologics upon failure of MTX or classical DMARD combination? Let us summarize the problem. Th e need for early introduction and rapid escalation of therapy in a response-driven strategy leaves little room for doubt. Th ere is also little doubt that the physician should treat to target and that the elected target should be remission, unless this is proven impossible. Th e number of biologics available and their effi cacy make this target potentially attainable for every patient. At the moment, there is no way to predict individual response to each of these agents.

How should the clinician proceed?
In our view, the best strategy in the absence of the ability to predict response to specifi c biologics consists of trial and error and is based on three main principles: (a) start with an eff ective agent, (b) move to another eff ective agent unless persistent remission is achieved with acceptable toxicity, and (c) consider going back to the most eff ective agent if none of the biologic DMARDs results in remission. We could name this strategy 'cycling for remission' . Th is process could develop at a relatively fast pace, thus avoiding the risk of leaving a patient for too long with ineff ective medication. Aletaha and colleagues [52] showed that a patient's response to treatment during the fi rst 3 months of biological therapy determined the level of disease activity at 1 year. So, quite soon after therapy has been started, the clinician can assess the effi cacy of each biological agent in controlling disease activity at 1 year.

Where should the clinician start?
With the currently available evidence, the order in which available biological agents should be used cannot be established on evidence-based grounds. Th is view is also adopted by the 2010 EULAR recommendations for the management of RA; the authors refrain from taking a position with regard to the preferred biologic drug [39]. We fully agree with this position in general. It will obviously require adaptation according to characteristics and preferences of individual patients, safety concerns, access to medications, and local policies. Etanercept and abatacept may be considered for a higher order of priority if the safety profi le is given a higher importance, and the opposite may happen with adalimumab and infl iximab. Th e priority of rituximab may be infl uenced by rheumatoid factor status since patients with positive rheumatoid factor seem to be the best candidates for rituximab [53]. Approval status of a drug (that is, as fi rst or second line) also infl uences treatment selection. Th e individual risk profi le of a patient should also be taken into account and balanced with an individual's relative and absolute risk of an adverse event with each biological agent. In many cases, the lack of these data may lead the clinician to make the best clinical judgment. TB prevalence and risk shall also be considered, and when the risk of TB is high, non-TNF inhibitors or etanercept may be preferred over monoclonal antibodies.
Th e EULAR 2010 recommendations [39] also state that it is 'current practice' to prescribe a TNF blocker fi rst, implying that the newer biologicals (rituximab, abatacept, and tocilizumab) come thereafter. Th e accumulated experi ence with anti-TNFs may be invoked to support their use as fi rst line. However, it could be argued that more recent biologics have far more controlled data than the original anti-TNFs. Furthermore, the argument of greater experience will retain face value forever unless the current paradigm is questioned. So, we hope that in the near future the clinician will be provided with data regarding the use of the newer biologics as fi rst line and we can progress to a more evidence-based selection among these agents.
Last but not least, the costs of each must be taken into account while choosing the optimal biological treatment. Costs vary among countries and cannot be easily compared. However, the clinician should keep them in mind, and in the absence of other signifi cant diff erences in benefi t and risks between various treatment options, cost considerations may infl uence therapeutic choices.
Th ere are still several unmet needs in RA. Th e search for valid and reliable biological and clinical markers to predict responsiveness to a particular targeted therapy and optimize treatment success for a particular individual must continue. Th e ideal sequence of use of biologics cannot be clearly established on the basis of available data, but the strategy cannot be based solely on issues of historical opportunity: the main reason why anti-TNFs are considered fi rst line is that they were the fi rst to appear. Dynamic treatment strategies avoiding treatment delay should be compared, with newer biologics started early in the disease course. Switching and rotating among biologics until the best possible option is established for each individual will allow the accumulation of data on their respective effi cacies and facilitate crucial studies on predictors of response.