Introduction Clinical remission is today the treatment goal for rheumatoid arthritis (RA), which requires fast and assertive therapeutic decisions for a tight control of disease activity. Few objective parameters are available to guide clinical decisions, particularly in switcher patients. We designed a preliminary algorithm introducing immunogenicity assessment in the current approach to patients with RA receiving tumour necrosis factor inhibitors (TNFi).
Objective To evaluate the concordance between the new algorithm and current clinical practice, comparing the effectiveness of ‘immunogenicity-based’ versus ‘empirical-based’ switches in a cohort of patients with established RA receiving biologics.
Methods EULAR therapeutic response was evaluated in 105 patients with RA (naive or switchers) over one year, through generalised estimation equation (GEE) analyses. Serum drug trough levels were assessed by ELISA and antidrug antibodies (ADAb) by Bridging ELISA.
Results During follow-up, 48.6% of patients had therapeutic decisions concordant with the proposed algorithm (Group A), and 51.4% had discordant decisions (Group B). One year after the therapeutic decision, patients from Group A had a higher probability of achieving response (OR=7.91, p<0.001, 95% CI 3.27 to 19.13) and low disease activity (OR=9.77, p<0.001, 95% CI 4.69 to 20.37) than patients in Group B.
Conclusions Immunogenicity assessment might help to optimise therapeutic decisions, leading to a better control of disease activity with significantly better clinical outcomes in patients with RA receiving TNFi.
- DMARDs (biologic)
- Rheumatoid Arthritis
Statistics from Altmetric.com
The ‘treat-to-target’ strategy, now part of the European League Against Rheumatism (EULAR) recommendations, has revealed the importance of an early tight control of disease activity among patients with rheumatoid arthritis (RA).1–4 To reach clinical remission or at least low disease activity, fast and assertive therapeutic decisions are required. No formal recommendations exist to guide the order of the sequence of biologics, particularly after the failure of previous tumour necrosis factor inhibitors (TNFi), which represents a common situation in daily practice. The effectiveness of cycling between different TNFi is controversial and currently, the decision to switch mechanism of action (MOA) is largely empirical.
Drug immunogenicity has been proposed as one of the main mechanisms behind biologic therapeutic failure.5–8 We recently conducted a systematic review and meta-analysis concluding that antidrug antibodies (ADAb) are clinically relevant and lead to significant decrease of therapeutic response rates.9
The presence of ADAb results in functional neutralisation of the drug and formation of immune complexes that promotes a faster clearance of the drug from circulation.10 ,11 ADAb-positive patients exhibit very low or undetectable serum drug trough levels, in contrast with ADAb-negative patients, who often have normal or even high serum drug trough levels.8 ,12–14
Non-responder patients, who exhibit adequate serum drug levels and no detectable ADAb, have lower probability of response to another agent with the same MOA, and may benefit in switching to a drug with a different MOA.15 Non-responders, who have no detectable serum trough levels and detectable ADAb, may benefit in switching to a less immunogenic drug.16 These patients may have a higher probability of developing ADAb against the new biopharmaceutical.15 Neutralising ADAb against etanercept or abatacept have not been detected.9 ,10 ,17
The added value of assessing immunogenicity in current clinical practice has been questioned. Based on available evidence, we designed a preliminary algorithm that introduces immunogenicity assessment in the current clinical approach to patients with RA receiving biologic therapies—figure 1. We propose to evaluate the concordance between the new algorithm and current clinical practice, comparing the effectiveness of ‘immunogenicity-based’ versus ‘empirical-based’ switches. The combination of clinical and immunogenicity data may provide a tool to optimise the use of biologic therapies.
This study aims to evaluate how concordant rheumatologists’ current clinical practice was with our proposed treatment algorithm, and to compare therapeutic response rates between patients who followed the proposed algorithm and those who followed other therapeutic strategies.
Therapeutic responses over one year in non-responders, who switched according to the following two main branches of our algorithm were compared with other strategies on both occasions: (1) switching to non-TNFi if serum drug levels were detectable; (2) switching to a less immunogenic drug if serum drug levels were undetectable and ADAb testing positive.
Secondarily, we evaluated the role of ADAb as a mediator of therapeutic response.
During a period of 2 years (January 2010–December 2011), we followed all adult patients (≥18 years) with established RA, receiving TNFi (infliximab 3 mg/kg intravenous at 0, 2, 6, 14 weeks and every 8 weeks thereafter, adalimumab 40 mg subcutaneous every other week, or etanercept 25 mg subcutaneous twice a week or 50 mg once a week) in monotherapy or with concomitant immunosuppressors, at the Department of Rheumatology, Hospital Garcia de Orta, Portugal. All patients fulfilled the American College of Rheumatology 1987 revised criteria for RA, and followed the Portuguese recommendations for the management of patients with RA receiving biologic therapies.18 ,19 The study was approved by the hospital ethics committee. All patients gave written informed consent.
For the concordance between rheumatologists’ current clinical practice and our proposed treatment algorithm, disease activity was evaluated in all patients every 3 months, using the Disease Activity Score in 28 joints (DAS28), according to the rheumatologists’ standard of care. Therapeutic response was defined as EULAR good and moderate responses (improvement DAS28>1.2 and DAS28≤3.2; improvement DAS28>1.2 and DAS28>3.2 or improvement 0.6>DAS28≤1.2 and DAS28≤5.1) and low disease activity as a DAS28≤3.2, according to national and international guidelines.18 ,20
Drug immunogenicity was assessed every 3 months through serum samples collected just before the next administration of a biologic. Serum drug trough levels were measured by ELISA, using a setup as described before.12 ,21 Limit of detection was determined by testing 100 sera of patients with RA before treatment. The mean ±6 times SD was chosen as a cut-off. Limit of detection was about 2 ng/ml for all TNFi. ADAb were tested by Bridging ELISA as described before.22 The sensitivity of these assays depends on the affinity of the ADAb. We used a series of patient-derived monoclonal antibodies to establish the sensitivity of the assay. A monoclonal antibody to adalimumab with the median affinity showed reached half-maximal extinction at 10 ng/ml and a detection limit <1 ng/ml. However, non-specific binding of serum components, such as rheumatoid factor or C1q sets the limit of detection for sera at about 20 ng/ml.
Clinicians were blind to immunogenicity test results, and therapeutic decisions were undertaken according to the Portuguese recommendations for the management of patients with RA.18
We classified patients into Group A if, during the entire study follow-up, they followed empirically any of the branches proposed in our algorithm, and Group B, those who followed different therapeutic strategies. Therapeutic responses were evaluated over one year after therapeutic decision, which may have included to switch or maintain therapy.
For comparison of therapeutic responses over one year in non-responder patients who switched according to the two main branches of our algorithm, therapeutic response was assessed before the switch and at 3, 6, 9 and 12 months thereafter. The proportion of patients with therapeutic response and the proportion of patients with low disease activity were compared over one year after the switch, between: (a) non-responders with detectable serum trough levels who switched to another TNFi, and non-responders with adequate serum trough levels who switched to a non-TNFi; (b) non-responders with undetectable serum drug trough levels, ADAb-positives, who switched to therapeutic monoclonal antibodies, and non-responders with undetectable serum drug trough levels, ADAb-positives who switched to etanercept or abatacept.
To evaluate the role of ADAb as a mediator of therapeutic response, the proportion of patients with therapeutic response was assessed at study beginning, between ADAb-positive and ADAb-negative patients.
Confounders or effect modifiers
To assess the impact of the proposed algorithm on therapeutic response and the influence of ‘immunogenicity-based’ switches versus ‘empirical-based’ switches on therapeutic response and low disease activity rates, we defined as potential confounders, or effect modifiers: age, disease duration, disease duration before biologic DMARD introduction, exposure time to biologic, concomitant immunosuppressors, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), DAS28 and response status (responder vs non-responder) at the study beginning or time of therapeutic decision.
Differences in patient characteristics between groups at study beginning were analysed by χ2 test (binary variables), t test or Mann–Whitney U test (continuous, variables), as appropriate. The threshold for significance was set at a p value (p) of less than 0.05, two-sided.
To estimate the course of therapeutic response and low disease activity status over time in patients who followed specific therapeutic strategies, we used generalised estimation equation (GEE) with binary response. All the variables previously defined as potential confounders or effect size modifiers were tested in our model. Statistical software SPSS V.17.0 (Chicago, Illinois, USA) and R 2.14.2 platform package ‘GEE’ were used.23
Characteristics of the cohort
The study included 105 patients with RA, 45% switchers, with a median (IQR) disease duration of 10.1 (4.6–12.6) years. At the study beginning, 22.9% patients were receiving infliximab, 31.4% adalimumab and 45.7% etanercept. During the follow-up period, 48.6% of patients (Group A) had therapeutic concordant decisions with the proposed algorithm, whereas 51.4% (Group B) had discordant. Therapeutic decisions concordant with the proposed algorithm were undertaken with a median delay of 249 days (IQR 116–388). Patient's baseline characteristics of both cohorts are listed in table 1.
The impact of the proposed algorithm on clinical response over one year after therapeutic decision
GEE demonstrated that patients from Group A had significantly higher probability of achieving therapeutic response (OR=7.91, p<0.001, 95% CI 3.27 to 19.13) and low disease activity over one year after the therapeutic decision (OR=9.77, p<0.001, 95% CI 4.69 to 20.37), in comparison with patients from Group B—table 2.
By excluding ADAb-positive patients, the probability of achieving therapeutic response (OR=5.97, p<0.001, 95% CI 2.38 to 14.99) and low disease activity (OR=7.88, p<0.001, 95% CI 3.57 to 17.39) remained significantly higher in patients in Group A when compared with those in Group B.
Response status (responder vs non-responder) at the time of therapeutic decision had a significant effect modification on both outcomes: therapeutic response (OR=9.59, p<0.001, 95% CI 4.21 to 21.84) and low disease activity (OR=5.23, p<0.001, 95% CI 2.30 to 11.89). In Group A, higher DAS28 scores at the time of therapeutic decision were associated with a lower probability of achieving low disease activity over the following year (OR=0.40, p<0.001, 95% CI 0.27 to 0.61), whereas in group B the DAS 28 scores did not affect the probability of achieving low disease activity (OR=0.84, p=0.33, 95% CI 0.58 to 1.20). In both situations, either response status or DAS28 scores at the time of therapeutic decision did not abrogate the association between therapeutic decision and the studied outcomes, which remained highly significant and nearly unchanged (less than 10%). The same result was verified after controlling for all other considered confounders or effect modifiers assessed at the time of therapeutic decision.
Therapeutic responses over one year in non-responder patients who followed the two main branches of our algorithm
Non-responders to a TNFi in the presence of detectable serum drug trough levels and no detectable ADAb had higher probability of achieving response by switching to a drug with different MOA, rather than another TNFi, even after adjusting for potential confounders, such as DAS28 at the time of switch (OR=6.76, p=0.004, 95% CI 1.82 to 25.04). Despite a trend toward higher probability of achieving low disease activity, statistical significance was not reached (regression coefficient=0.24, p=0.73).
By analysing the 18 non-responder patients with undetectable serum drug trough levels, ADAb-positives, we verified that: (1) only two patients switched to a less immunogenic drug, both achieving response and low disease activity after the switch; (2) 15 patients maintained immunogenic drugs (14 remained within the same MOA and one switched to a different MOA), none was able to achieve therapeutic response; (3) one patient, despite ADAb-positivity, was considered as a responder and maintained the same therapy, achieving low disease activity and even clinical remission over the following year.
ADAb as a mediator of therapeutic response
At the study beginning, ADAb was detected in 37.5% of infliximab-treated patients, and in 27.3% of adalimumab-treated patients. No detectable ADAb were verified among patients receiving etanercept. All ADAb-positive patients had undetectable serum drug trough levels, which were not verified in patients without detectable ADAb. Patient's baseline characteristics among ADAb-positive and ADAb-negative are listed in online supplementary table 1.
Lower proportion of responders was observed among ADAb-positive patients than ADAb-negatives (22.2% vs 62.1%, p=0.003). One patient with detectable ADAb had low disease activity, in comparison with 34.5% of patients without detectable ADAb (5.6% vs 34.5%, p=0.02) (table 3). ADAb-positive patients had significantly higher mean (SD) CRP values, compared with ADAb-negatives: 5 (3–10) vs 2.15 (1–4) mg/l, p=0.001 (figure 2).
Our results demonstrate that therapeutic decisions according to the proposed algorithm lead to better disease control with significantly better clinical outcomes in patients with RA receiving biologic therapies. These results were independent of ADAb status and DAS28 scores at the time of therapeutic decision. Less than half the patients empirically followed therapeutic strategies concordant with the proposed algorithm. Had our algorithm been followed, about 8 months would have been gained, with important clinical and economic impact. Serum drug trough levels predict therapeutic responses in switchers. Non-responders to etanercept, adalimumab or infliximab, in the presence of detectable serum drug trough levels, were most likely to respond when switched to a drug with different MOA, in comparison with those who switched to another TNFi. In this subset, TNF might not play a central role in disease pathogenesis. Our results are in agreement with two previous studies, where switchers who failed previous TNFi (infliximab or adalimumab) in the absence of ADAb had poor responses to a second TNFi.15 ,16
Non-responders to TNFi in the presence of undetectable serum drug trough levels, ADAb-positives, were poorly represented in our study, and no robust conclusions should be drawn exclusively from our data. Our data suggest that those patients might benefit in switching to a less immunogenic drug, TNFi or non-TNFi. Previous studies have revealed that patients who discontinued a TNFi due to immunogenicity were able to achieve response to another TNFi if free of significant immunogenicity.16 By contrast, lower therapeutic responses were verified when the second TNFi was a monoclonal antibody.15 ,16
Previous studies have revealed that TNFi therapy is more efficient in patients who discontinued the first TNFi due to secondary failure or adverse events, rather than by primary failure.24 ,25 Therapeutic algorithms based on the clinical distinction between primary and secondary failures have been proposed by some authors who have underestimated the added value of immunogenicity assessment in clinical practice.26 Such a clinical distinction is not straightforward, as therapeutic response assessment is based on subjective criteria that can be strongly affected by certain biases. The placebo effect, ‘regression to the mean’, or optimisation of concomitant therapies may be the cause of initial clinical response.
A decision tree algorithm based on immunogenicity monitoring was recently proposed by others.27 In this work, contrarily to ours, the first branching implicitly concerns clinical response as the study addresses specifically patients with primary and secondary failures to TNFi. We propose a first branching according to drug level that offers the possibility to readily identify patients who are overtreated or uselessly treated. The previous algorithm also considers dose escalation in non-responders presenting low drug levels and undetectable ADAb. In some of these cases, ADAb might be hidden by the presence of drug, and dose escalation may boost ADAb production with serious adverse events.8 ,21 ,22 ,28 In the previous algorithm, a switch to another TNFi is recommended for non-responder patients who present optimal serum drug levels and ADAb-positive. Such ADAb titres which have no significant impact in serum drug concentrations, are unlikely to fully neutralise a drug's bioactivity.10 ,21 The absence of clinical response, despite TNF neutralisation, may warrant switching to an agent with a different MOA. We did not include this category of patients in our own algorithm, as in presence of optimal drug levels, ADAbs are unlikely to be revealed by the most common assays, such as Bridging ELISA or RIA assays.8 ,14 ,22 ,29
The inclusion of responders in our algorithm should help identify patients for whom drug dose reduction or increased interval between drug administrations might increase treatment cost-effectiveness. Correlation between DAS28 improvement and serum drug trough levels has been verified up to a threshold of drug level, above which no significant DAS28 changes occur.30
EULAR recommendations, according to expert opinion, suggest that tapering a biological DMARD should only be considered in patients in remission for at least 12 months.4 Biologic withdrawal might be considered earlier among patients in remission despite undetectable drug levels and high titres of ADAb. Subclinical synovitis might lead to bone damage in patients who clinically seem to have controlled RA.31 ,32 Image techniques, including ultrasonography or MRI, have revealed higher sensitivity and reproducibility than clinical evaluation in assessing active synovitis, and may be used to confirm remission.33
Our experience suggests that high ADAb titres in the trough are associated with undetectable drug levels during most of the interval between two drug administrations. It seems reasonable to postulate that in these cases, remission is not maintained by the therapy. Further double-blind, randomised, controlled trials may better clarify this point.
At the study beginning, 37.5% of infliximab-treated patients and 27.3% of adalimumab-treated patients had detectable ADAb. These proportions might be underestimated, since many patients with obvious non-responses, potentially ADAb-positives, had already switched from initial therapy. Interestingly, in some cases, we were still able to detect ADAb against the previous biologic, several years after drug discontinuation. We did not find antietanercept antibodies which is in agreement with previous studies that used more specific methods to detect ADAb.9 ,16 ,34
Patients with detectable ADAb had higher mean CRP values, reflecting the poor control of inflammation. Previous studies have revealed higher baseline CRP values in ADAb-positive patients, but its association with the development of ADAb is not clear.6 Concomitant immunosuppressive therapies, particularly methotrexate, have been associated with decreased ADAb frequencies.9 ,35 In our study, almost all patients were receiving concomitant methotrexate and low-dose corticosteroids, which limited us to confirm this.
Our study has important limitations, despite reflecting real-world evidence. This is a small-scale study, conducted at one single centre, and not powered to assess all the branches described in our algorithm, as this was not our main purpose. We did not categorise detectable serum drug trough levels from low to high, since no robust studies have clearly defined those cut-offs. ADAb are unlikely to be detected in the presence of a circulating drug by the most common assays, namely Bridging ELISA. We assessed immunogenicity of three TNFi approved for RA treatment. Extrapolations to other agents in the same class, such as golimumab and certolizumab, should be done cautiously. We treated non-TNFi agents as a homogeneous group and did not evaluate the potential differences among them due to limited data. Etanercept and abatacept did not reveal clinically significant immunogenicity and were both considered ‘less immunogenic’. Further studies, using the same methodology, should be conducted to better compare the immunogenic profile of biopharmaceuticals.
Many questions regarding immunogenicity remain to be elucidated. We are proposing strategies that are already approved for patients with RA. This algorithm represents a preliminary tool to aid decision making among clinicians, and how immunogenicity assessment can be integrated in the care for these patients leading to personalised and more cost-effective strategies to RA treatment.
We are grateful to nurse Lurdes Barbosa for blood sample collection; to Margreet Hart for the technical support with the assays; to Ana Cordeiro, MD, and Francisca Moraes-Fontes, MD PhD, for the critical review of the article.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online table
Handling editor Tore K Kvien
Contributors All authors fulfil the authorship criteria.
Funding The Gulbenkian Programme for Advanced Medical Education is sponsored by Fundação Calouste Gulbenkian, Fundação Champalimaud, Ministério da Saúde and Fundação para a Ciência e Tecnologia, Portugal. Marília Antunes’ research is partially funded by project Pest-OE/MAT/UI0006/2011.
Competing interests None.
Patient consent Obtained.
Ethics approval Regional Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.