Predictors of Response to Anti-Tumor Necrosis Factor Alpha Therapies in Rheumatoid Arthritis

In the 10 years since anti—tumor necrosis factor alpha (anti–TNF-a) antagonists have been introduced, well over 1 million patients have received these agents (infliximab, etanercept, and adalimumab). Anti–TNF-a agents have demonstrated efficacy and are approved for several rheumatologic diseases, but arguably have had the greatest impact on the treatment of rheumatoid arthritis (RA). It is now clear that anti–TNF-a therapy combined with diseasemodifying antirheumatic drugs (DMARDs) provides symptomatic relief, decreases inflammation, curbs radiographic progression of disease, and improves both quality of life and function.^1-8 This impact has been so great that the treatment paradigm for RA has shifted from obtaining a “good enough” response to achieving robust disease remission soon after anti—TNF-a introduction.^9,10

Despite revolutionizing RA therapy, anti—TNF-a medications are not universally successful in achieving significant clinical improvement or disease remission by validated measures of disease activity including the Disease Activity Score in 28 joints (DAS28) and American College of Rheumatology (ACR) 20%, 50%, and 70% responses (ACR20, ACR50, and ACR70). One-half to one-third of patients with RA fail to respond to anti–TNF-a therapy.^9-11 Another 50% have a good clinical response, but do not achieve remission.⁹ Interestingly, some patients who have a suboptimal response to the first anti—TNF-a medication will achieve better control once switched to another anti–TNF-a medication as measured by the DAS28, Health Assessment Questionnaire, ACR20, and European League Against Rheumatism (EULAR) responses.^12-15 However, some patients fail to respond to all anti—TNF-a therapies. Complicating the treatment with anti–TNF-a therapy are patients who initially benefit from the medication but subsequently suffer a relapse.

Being able to predict which patients will be anti—TNF-a nonresponders would allow the clinician to bypass anti–TNF-a therapy for other biologic medications. It also would thwart the significant costs of unsuccessful anti–TNF-a treatment, improve time to disease suppression or remission, and avoid progression of irreversible joint destruction while reducing the risks inherent to this class of medication, including infections.¹⁶

A previous meta-analysis of clinical trials using traditional DMARDs in the treatment of RA identified previous DMARD failure, longer disease duration, greater disability, and female sex with lower response rates.¹⁷ To date, a paucity of published literature describes reliable clinical parameters that can be used to accurately predict anti—TNF-a treatment outcome in RA.¹⁸ Our aim is to identify readily available clinical data that may be used to predict anti—TNFa treatment failures in patients with RA in two independent tertiary care rheumatology outpatient populations.

SUBJECTS AND METHODS

Subjects with RA were identified in both the Wake Forest University (WFU) Baptist Medical Center and the Medical University of South Carolina (MUSC) rheumatology clinics. We used International Classification of Diseases, Ninth Revision codes 714 (RA and other inflammatory arthropathies), 714.1 (Felty’s syndrome), 714.3 (juvenile chronic polyarthritis), 714.31 (acute polyarticular juvenile RA), 714.32 (pauciarticular juvenile RA), and 714.9 (unspecified inflammatory polyarthropathy). More than 500 electronic and paper medical records met these criteria in each center. A random sample of these medical records was assessed for patients who had received at least one anti—TNF-a medication (infliximab, etanercept, adalimumab) for RA for any duration of time in combination with any DMARD or prednisone. Records were reviewed by a single physician reviewer at each center, Dr Daniel (MUSC) and Dr Bourne (WFU), between April 2007 and May 2008. Subjects with a previous diagnosis of juvenile idiopathic arthritis and comorbid connective tissue diseases (systemic lupus erythematosus, scleroderma, inflammatory myopathy, and vasculitis) except secondary Sjögren syndrome were excluded. Subjects younger than age 18 years, with active malignancy, with complement deficiency, or with documented noncompliance to anti–TNF-a therapy also were excluded.

Demographics, features of disease, medication history, family history of connective tissue disease, comorbidities, smoking history, physical exam features, laboratory results, and presence of erosive disease documented before the initiation of the first anti—TNF-a agents were recorded. Subjects who met inclusion criteria were assigned to one of three outcome groups. The Responder group was defined as subjects who remained on the initial anti–TNF-a medication for more than 90 days and who had not received other anti–TNF-a therapies at the time of chart review. The Failure group included subjects who were no longer on an anti–TNF-a medication regardless of reason for discontinuation and duration of therapy. This definition of failure was chosen as opposed to calculated measures of disease activity because of its more practical application. The Switch group contained subjects who were exposed to more than one anti–TNF-a medication. Subjects were assigned to the Switch group provided that the most recently prescribed anti–TNF-a agent was used for at least 90 days; otherwise, they were assigned to the Failure group. If a patient with RA was started on a second or third anti–TNF-a agent with fewer than 90 days of documented follow-up at the time of study completion, the subject was assigned to the Switch group if efficacy was documented in the medical record. Increasing doses or frequency of anti–TNF-a medications did not alter group status.

For analysis purposes, we classified the covariates into five groups: (1) patient characteristics, (2) symptoms, (3) laboratory values, (4) comorbid conditions, and (5) medications. Patient characteristics included age (continuous and categorical [>50 years]), sex, ethnicity, history of ever-smoked cigarettes, and obesity. The symptoms category included disease duration (months), morning stiffness (minutes), presence of erosive disease, presence of elbow flexion contractures, upper and lower extremity synovitis, and number of rheumatoid nodules (ordinal: 0, 1-2, 3-4). The continuous laboratory values included were erythrocyte sedimentation rate (ESR; dichotomous [>50 mm/hour], continuous, and continuous log-transformed), white blood cell count, platelets, hemoglobin, and hematocrit. The categorical laboratory values included rheumatoid factor, anti—cyclic citrullinated peptide (CCP), and antinuclear antibody. The comorbid conditions (all categorical) included thyroid dysfunction, osteoporosis, osteopenia, and diabetes. All dosages of RA medications were recorded as continuous variables and included prednisone, methotrexate (MTX), sulfasalazine, hydroxychloroquine, azathioprine, leflunomide, and minocycline. All medications except for prednisone and MTX were combined into a non-MTX DMARD variable. Indicator variables were included in the model to control for the variable availability of the three drugs in the marketplace over time. Logistic regression was performed using SAS version 9.1 (SAS Institute Inc, Cary, NC).

RESULTS

A total of 713 and 600 medical records were reviewed at MUSC and WFU, respectively. Of the 179 electronic and paper medical records that met inclusion criteria, 84 and 95 medical records were reviewed at MUSC and WFU, respectively. Subjects were classified as 110 Responders, 48 Switchers, and 21 Failures. Subject demographics stratified by treatment center and group assignment are summarized in

Table 1

and Table 2, respectively. Anti—TNF-a switching order is summarized in

Table 3

. The indicator variables included to control for the variability of the 3 drugs in the marketplace were noted to be insignificant and did not change the overall results.

Of the 64 patients who received adalimumab as their initial anti—TNF-a therapy, 53 (83%) did not switch or discontinue the medication and were classified as Responders. Two of these patients required weekly adalimumab. For the patients who initially received infliximab, 52% (27/52) did not switch or discontinue the drug. Of the patients who initially received etanercept, 51% (32/63) did not switch or discontinue the drug.

In the initial analysis, the groups were independently assessed for the ability of each covariate to predict failure of anti—TNF-a therapy. Next, we combined variables across groups with predictive ability (P <.20) in order to test their predictive value. From these models we removed—one at a time&mdash;variables that added little to the predictive ability of the model. Once we isolated a valid combination of covariates, we tested their interactions sequentially. The best final explanatory model contained three variables and one interaction term and is summarized in

Table 4

. In this model, patients taking MTX who had an ESR >50 mm/hour at the time of starting anti—TNF-a therapy had an increased likelihood of anti–TNF-a therapy failure (odds ratio [OR] = 23.6; P = .0144). The relationship between anti–TNF-a therapy and an ESR >50 mm/hour suggests that patients with more active inflammation and presumably less responsiveness to DMARDs have an increased likelihood of anti–TNF-a therapy failure. Thyroid dysfunction or osteoporosis controlled for bias and was thus included in the final model, but was not significant in the bivariate analysis (OR = 1.53, 95% confidence interval [CI] = 0.55, 4.25; P = .42). Patients with thyroid dysfunction or osteoporosis, regardless of MTX use and ESR value at baseline, had a 3.9-fold higher risk of failing anti–TNF-a therapy (P = .0363) than patients without either of these comorbid conditions. The mean prednisone dose was not significantly different between subjects with thyroid disease or osteoporosis and subjects without thyroid disease or osteoporosis (4.15 ± 5.2 mg vs 3.78 ± 5.3 mg daily; P = .43).

We examined the effect of the type of first anti—TNF-a medication prescribed on predicting failure and found that they were not statistically significant predictors of treatment failure. However, there were differences in the association between the type of anti–TNF-a drug and treatment failure, with patients receiving adalimumab having the least likelihood of failure. To control for selection bias (ie, anti–TNF-a allocation was not randomized), we constructed a categorical variable, the propensity score, which predicts the probability of receiving adalimumab as a first anti–TNF-a therapy. The propensity score for receiving adalimumab as the first anti–TNF-a therapy was then used to examine and control for selection bias in the final model.^19-21 The model controlling for the propensity of receiving adalimumab first did not change the ability of ESR >50 mm/hour and MTX use to predict failure, nor did it modify the increased risk of failure for patients with thyroid disease or osteoporosis.

In a second analysis we examined the same covariates that predicted switching from one anti—TNF-a therapy to a second anti–TNF-a therapy. In this analysis MTX was dichotomized to <15 mg and >15 mg weekly. In order to examine the risk of switching, we excluded all patients who failed anti–TNF-a therapy from this analysis. The same model-fitting techniques and statistical thresholds were used in this analysis. Patients who started on adalimumab first had the lowest likelihood of switching (OR = 0.293, 95% CI = 0.117, 0.734), whereas those starting etanercept first had the greatest likelihood of switching (OR = 2.51, 95% CI = 1.129, 5.580). The model with infliximab was not statistically significant. The differences in switching rates could be due to selection bias in this retrospective sample, if those least likely to switch anti–TNF-a medications were initially prescribed adalimumab. We thus used the method described by Rubin et al to construct a categorical variable with five strata of probabilities of receiving adalimumab (propensity score categories) to control for bias in the initial anti–TNF-a selection.^20,21 The final model indicates that receiving adalimumab first reduced the likelihood of switching by 71% (P = .0088) regardless of MTX dose and the propensity of receiving adalimumab as a first anti—TNF-a therapy. Also, patients on low-dose MTX (15 mg/week at baseline, regardless of initial anti—TNFa therapy. Final model results are presented in

Table 5

.

DISCUSSION

Although anti—TNF-a therapies have begun to find a niche for the treatment of RA, our ability to predict treatment outcomes with these medications is still limited. Given the side effect profile and high cost of anti–TNF-a therapies, and the relatively high prevalence of RA, it is imperative that a cost-effective algorithm for the implementation of biologic agents be developed. The success of such an algorithm will depend on the physicians’ ability to predict anti–TNF-a treatment outcomes. Our study was the first non-European, multicenter, university-based investigation to examine clinical and serologic measures for their abilities to predict which patients will fail anti–TNF-a therapy and those who will switch to another anti–TNF-a agent. Although our data are limited, the resulting models may provide an important foundation for further work in anti–TNF-a outcomes in RA.

Multiple randomized controlled trials, mouse models, and other studies have demonstrated the increased efficacy of an anti—TNF-a therapy plus MTX versus MTX alone.^2,3,22-27 The synergistic relationship between anti—TNF-a therapy and MTX implies that patients are more apt to continue on an anti–TNF-a agent if MTX background therapy has been instituted. These findings are consistent with a large retrospective analysis performed on the British Society for Rheumatology Biologics Register where the concurrent use of MTX (dose not reported) was a strong predictor of response to anti–TNF-a therapy.¹⁸

Our study supports this analysis as we found that patients on MTX <15 mg/ week were almost three times more likely to switch to another anti—TNF-a therapy than patients on MTX >15 mg/week regardless of the initial biologic agent prescribed. These may represent a subset of RA patients who are biologically responsive to the synergistic interactions of MTX with anti–TNF-a drugs. Patients with RA may have a common inflammatory pathway that is susceptible to MTX >15 mg/week; as a result, the more aggressively treated MTX group is more apt to respond to anti–TNF-a therapy.

Poor outcomes with infliximab despite concomitant up-titration of MTX were reported in a small observational cohort of 22 patients with RA.²⁸ In this study 95% of patients had active disease by EULAR criteria (DAS28 >3.2) and disease remission was not observed (DAS28 <2.6) at 16 weeks despite a mean ± SD weekly MTX dose that increased from 9.9 ± 3.9 mg to 15 ± 4.3 mg. However the DAS28 did improve in one-third of the patients, supporting the role for concomitant MTX with anti—TNF-a therapy. Although results of many trials have been published documenting the efficacy of MTX with anti–TNF-a therapy, a paucity of data are dedicated to investigation of the effects of MTX dose on the outcome of combination therapy. The major trials that document the efficacy of MTX plus anti–TNF-a therapy used protocols that attempted to maximize MTX dose and the doses were generally >15 mg/week.^{1,3,5-7,29-31} Our findings and those of Ornetti et al bring into question the efficacy of MTX <15 mg/week in combination with anti—TNF-a agents.²⁸ We suggest the existence of a critical MTX dose threshold that can predispose a patient to poorer anti—TNF-a therapy responsiveness if the MTX dose is below this threshold. These studies support the theory that MTX should be increased to at least 15 mg/week before initiation of anti–TNF-a therapy as an inexpensive and appropriate strategy to derive maximal benefit from the anti–TNF-a agent. MTX has been shown to improve mortality, which further supports its key role in the treatment of RA with or without anti–TNF-a therapy.^32,33 Further research is needed in larger populations taking combination therapy to investigate this relationship.

Eighty-three percent (53/64) of patients who received adalimumab as the initial anti—TNF-a therapy experienced an improvement in disease activity that warranted continuation of the drug and did not switch to another class of drug or fail anti–TNF-a therapy. These findings are congruent with the ARMADA 4-year extended study (n = 262) where 87%, 79%, 71%, and 64% of the patients remained on adalimumab with a sustained clinical response at 1, 2, 3, and 4 years of treatment, respectively.³⁴ Response rates for etanercept and infliximab as the initial anti—TNF-a therapy were significantly lower. Patients who were initially prescribed infliximab and etanercept continued on these medications 52% and 51% of the time, respectively, implying that approximately half of all patients who initially received infliximab or etanercept eventually switched or failed these anti–TNF-a therapies at the time of chart review. Buch et al reported a first-year discontinuation rate for infliximab primary responders of 55%,35 which is consistent with our findings.

Our final model suggests that patients who receive adalimumab as the initial anti—TNF-a drug are 71% less likely to switch to a subsequent therapy (P = .0088). Interestingly, patients on etanercept were more apt to switch to another anti–TNF-a agent (P = .0240). The explanation for this trend is not clear, but may stem from the unique pharmacokinetic and pharmacodynamic mechanisms of each drug.^36-39 One difference between these two therapies is their molecular targets: etanercept inhibits soluble TNF-a and lymphotoxin a (TNF-ß), whereas adalimumab inhibits soluble and membrane-bound TNF-a but not lymphotoxin. Structurally, etanercept is a dimeric fusion protein of the extracellular TNF-a receptor linked to the Fc portion of immunoglobulin G; adalimumab is a recombinant fully humanized antibody that targets TNF-a.¹¹ It is important to realize that we were unable to distinguish between patients who switched therapies because of inefficacy, adverse events, preference, or other reasons. However, the ORs for switching from adalimumab or etanercept as initial therapies were statistically significant and warrant further exploration.

Long-term retrospective and prospective studies have confirmed the utility and tolerability of MTX, the DMARD widely considered the cornerstone of RA therapy.^33,40-42 The 1-year continuation rate of MTX is estimated at 65% to 86% but drops significantly to 46% to 79% after 5 years.^33,40-42 Our finding that 83% of patients with RA who receive adalimumab as the initial anti—TNF-a are likely to remain on this class of anti–TNF-a therapy is comparable, and potentially superior, to previously published continuation rates for MTX.

Adalimumab had the lowest discontinuation rate of the commercially available anti—TNF-a therapies, and its tolerability was at least comparable to that of MTX, with 71% of patients who received adalimumab as the initial anti–TNF-a less likely to switch to another anti–TNF-a therapy. Therefore, an argument can be made that it should be designated as the “first-line” biologic agent for RA. However, such an indication would require more rigorous randomized, blinded, head-to-head trials of all three approved anti–TNF-a therapies. With new biologic therapies and anti–TNF-a drugs in development, the practicality and financial feasibility of such a study remain unlikely.

In our logistic regression model, anti-CCP antibody status and titers were not predictive of switching anti—TNF-a therapies. Moreover, these antibodies did not predict anti–TNF-a failure. This finding conflicts with previous work by Braun-Moscovici et al, who concluded that anti-CCP may be predictive of outcome.⁴³ However, the patient populations and type of anti—TNF-a therapies between the two studies differed significantly. The Braun-Moscovici study was limited to 30 consecutive, seropositive RA patients who had failed at least three DMARDs and anti–TNF-a efficacy was assessed using infliximab only. The present study included both seropositive and seronegative patients who did not have to fail prior DMARD therapy.

We found a 3.9-fold increased risk of anti—TNF-a failure in patients with either thyroid dysfunction or osteoporosis. These two comorbidities were combined in our model, and an explanation for the increased propensity to fail anti–TNF-a therapy was not obvious. Prednisone use is known to cause osteoporosis and may be linked to refractory RA; therefore, osteoporosis may serve as a proxy for anti–TNF-a–resistant disease. However, prednisone doses did not differ statistically between the subjects with thyroid disease or osteoporosis and subjects lacking these comorbidities, which makes this theory less plausible. We were not able to determine the duration of prednisone therapy or cumulative corticosteroid exposure, and these may be contributory variables. Rheumatoid arthritis, a recognized risk factor for osteoporosis, confounds this relationship.^44,45 Anti—TNF-a metabolism may be altered in patients with thyroid dysfunction; however, this has not been previously demonstrated. Additionally, our database is unable to distinguish between hyperthyroid and hypothyroid disease, preventing subanalysis of these two groups. Our findings suggest that larger studies to determine the efficacy of anti–TNF-a therapies in patients with thyroid disease and osteoporosis may have clinical merit.

This study has significant limitations intrinsic to retrospective design, including missing data, reliance on the medical record, and investigator interpretation of documents. Inherent in this approach is the reliance on the treating rheumatologist’s assessment for medical decision making, including the perceived need to change therapy. Ideally, this study would be optimally performed in a prospective manner with clear end points for discontinuation or switching of a drug, including primary or secondary lack of response, as measured by validated measures such as the Disease Activity Score, Routine Assessment of Patient Index Data Score, or the Health Assessment Questionnaire. Such a study would be complex, costly, and time consuming, and would require the coordination of several healthcare facilities to acquire a significant sample size, raising the question of feasibility.

We recognize that our definition of “Responder” is based on the premise that the continuation of an anti-TNF medication implies that the treating rheumatologist noted improvement in disease activity after starting the anti-TNF drug. Similarly our definitions of “Failure” and “Switcher” imply a lack of response or an adverse event that should dictate mandatory changes or discontinuation of anti-TNF therapy by the treating rheumatologist. Previous large retrospective studies examining anti—TNF-a therapy outcomes also were limited by the definition of therapy failure, inability to quantify reasons for switching agents, and inability to distinguish primary nonresponse from secondary nonresponse.^18,46-48 Our study shared these limitations. Our sample size was relatively small, limiting our ability to detect weak predictors; thus, larger populations need to be studied. However, because this was a multicenter study, the results may be more generalizable, and we implemented a propensity scoring method to control for anti—TNF-a selection bias. The tertiary care referral patient population may have led to a selection for more DMARD-refractory disease.

To date definitive clinical and serologic predictors of anti—TNF-a therapies have yet to be established despite large reviews of European biologic registries.^18,46-48 Establishment of a valid predictor to biologic therapy, specifically anti—TNF-a therapy in RA, may require gene polymorphism studies rather than reliance on clinical findings and serum biomarkers.^39,49-51 Although this information is convenient to measure and obtain in the clinic, clinical findings and commercially available serum biomarkers have not been demonstrated to be reliable predictors of anti—TNF-a therapy. Further research is needed to identify readily available biomarkers with clinical predictive utility, especially when prescribing costly biologic therapies.

In conclusion, patients with RA were less likely to continue initial anti—TNF-a therapy if the MTX dose was <15 mg/week, which may represent a therapeutic threshold. Attempts should be made to decrease the ESR prior to anti–TNF-a therapy, as levels >50 mm/hour were associated with a higher rate of anti–TNF-a therapy discontinuation. Patients who received adalimumab as the initial anti–TNF-a therapy were 71% more likely to remain on this drug than patients receiving other classes of anti–TNFa medications. Our study will need validation in other RA populations before widespread generalization. Further investigation is warranted to better delineate reliable and available predictors for anti–TNF-a therapy in RA.