Disease remission and sustained halting of radiographic progression with combination etanercept and methotrexate in patients with rheumatoid arthritis


  • ClinicalTrials.gov identifier: NCT00393471.



The Trial of Etanercept and Methotrexate with Radiographic Patient Outcomes (TEMPO) is a 3-year, double-blind, multicenter study evaluating the efficacy and safety of etanercept, methotrexate, and the combination of etanercept plus methotrexate in patients with active rheumatoid arthritis (RA). The results after 1 and 2 years of the study have been previously reported. Here we provide the 3-year clinical and radiographic outcomes and safety of etanercept, methotrexate, and the combination in patients with RA.


In this randomized, double-blind, multicenter TEMPO study, 682 patients received etanercept 25 mg twice weekly, methotrexate ≤20 mg weekly, or the combination. Key efficacy assessments included the Disease Activity Score (DAS) and the DAS in 28 joints.


Combination therapy resulted in significantly greater improvement in the DAS and in more patients with disease in remission than either monotherapy. This finding was confirmed by longitudinal analysis; patients receiving combination therapy were more than twice as likely to have disease in remission than those receiving either monotherapy. Independent predictors of remission included male sex, lower disease activity, lower level of joint destruction, and/or better physical function. Combination and etanercept therapy both resulted in significantly less radiographic progression than did methotrexate (P < 0.05). Etanercept and combination treatment were well tolerated, with no new safety findings.


Etanercept plus methotrexate showed sustained efficacy through 3 years and remained more effective than either monotherapy, even after adjustment for patient withdrawal. Combination therapy for 3 years led to disease remission and inhibition of radiographic progression, 2 key goals for treatment of patients with RA.

Rheumatoid arthritis (RA) results in chronic pain, loss of function, and disability. Although the newer antirheumatic medications are highly effective in randomized, well-controlled studies (1–9), there are few reports of the long-term therapeutic effects of these medications.

The Trial of Etanercept and Methotrexate with Radiographic Patient Outcomes (TEMPO) (2, 3) was designed to further evaluate the risk–benefit relationship of etanercept. This randomized, parallel, controlled study of patients with RA provides 3-year efficacy and safety results of the 3 treatment regimens. In contrast to previous longer-term studies with etanercept (6) or other anti–tumor necrosis factor agents (4, 5), in this study patients were not necessarily receiving methotrexate at baseline.

The results from years 1 and 2 of the TEMPO study have been previously reported (2, 3). Briefly, combination therapy resulted in a significantly greater reduction in disease activity, improvement in physical function, and a higher rate of remission than did either monotherapy. Radiographic results suggested that joint damage was halted in patients receiving combination therapy with etanercept plus methotrexate (2, 3).

The interim clinical results of the TEMPO study included conventional time point analyses using imputation methods such as the last observation carried forward (LOCF) or completers approach. Despite randomization, long-term trials face a number of methodologic weaknesses, such as missing data, patient attrition, and the phenomenon of “bias by completers”(10) that may impact data reliability and increase over time. Additionally, repeated univariate measurements of any variable over time do not address the high correlation among the separate measurements and may lead to overinterpretation of contrasts among treatment groups (10).

To address our concerns regarding the potential biases associated with the conventional analyses of long-term study results, we analyzed 2 key outcomes, disease activity and remission, using multivariate longitudinal analyses. The generalized estimating equation (GEE) approach (11, 12) using pattern-mixture modeling (13) is a statistical technique used to analyze longitudinal data that may not be normally distributed. The GEE approach adjusts for correlated observations and the effects of unequal time intervals, whereas pattern-mixture modeling provides an adjustment for biases invoked by unbalanced withdrawal (bias by completers). Additionally, we also used this model to identify predictors of long-term disease remission. For comparison with other trials and earlier reports, we also present the results using conventional time point analysis.


Study design.

The TEMPO study was designed as a 3-year, double-blind, randomized, parallel-group study in Europe, Australia, and Israel to evaluate the efficacy and tolerability of etanercept and methotrexate, combined and alone, for patients with RA for whom therapy with at least 1 disease-modifying antirheumatic drug (DMARD) other than methotrexate had previously failed. A detailed description of the study design has been reported previously (2). A list of TEMPO study investigators, in addition to the authors of this article, is provided in Appendix A.


Patients enrolled in the study were age ≥18 years and had active adult-onset RA (American College of Rheumatology [ACR] functional class I–III) (14), defined as ≥10 swollen and ≥12 painful joints, and either an erythrocyte sedimentation rate (Westergren) of ≥28 mm/hour, C-reactive protein (CRP) level ≥20 mg/liter, or morning stiffness for ≥45 minutes. In addition, patients had to have a disease duration of >6 months and <20 years. Details of the inclusion/exclusion criteria for the TEMPO study have been described previously (2, 3).

Study protocol.

Throughout the 3-year duration of the study, both investigators and patients remained blinded to the study treatment. Patients continued their randomly assigned treatment throughout the study (etanercept [25 mg twice weekly subcutaneously], methotrexate [up to 20 mg once weekly orally], or etanercept plus methotrexate).

The study protocol and informed consent documents were approved by an independent ethics committee, and the trial was conducted in accordance with the International Conference on Harmonization Guidelines for Good Clinical Practice. Written informed consent was obtained from all patients at the time of enrollment.

Interim analyses were conducted at the end of year 1 and year 2. Investigators were notified of the interim results after each analysis, and patients reconsented to continuation of year 2 and year 3 of the blinded study. Patients and investigators could withdraw from or discontinue the study at any time.

Clinical outcome assessments.

The primary end points were reported previously (2). These included the numeric index of the ACR response area under the curve over the first 6 months and the change from baseline to week 52 in total joint damage assessed with the modified Sharp/van der Heijde score (TSS) (6).

For this 3-year report, the key analyses focused on disease activity and remission as assessed by the Disease Activity Score (DAS) and the DAS in 28 joints (DAS28) (15, 16). Disease remission was defined as a DAS <1.6 or a DAS28 <2.6. Response to therapy was assessed using the ACR 20%, 50%, and 70% improvement criteria (ACR20, ACR50, and ACR70, respectively) (17–19) at regular intervals.

Radiographic outcomes.

Progression of joint damage was evaluated using radiography. Details of the methodology have been discussed previously (2, 3). At the end of year 3, 4 sets of images (from baseline and from 1, 2, and 3 years) were scored for each patient; the baseline, year 1, and year 2 images were read for a third time. For patients who did not continue into year 3 or did not have radiographs from year 3, raters reread the baseline, month 6, year 1, year 2, or last available radiographic images. Two physicians who were blinded to patient identity and treatment read the images in a randomized time sequence. Changes from baseline to 3 years in total TSS (20), erosions, and joint space narrowing (JSN) were calculated. Nonprogression was assessed as a TSS change ≤0.5.

Cumulative probability plots of all radiographic data were created to fully visualize each patient's changes from baseline by treatment regimen. Frequency distributions of the observed cumulative proportion (i.e., change scores ranked from the lowest to the highest values per treatment group and presented as a cumulative proportion of all scores) were plotted against the actual values of the variables (3, 21).

Safety assessments.

Safety assessments, including treatment-emergent adverse events (AEs), serious AEs, and serious and medically important infections, were based on reports of AEs and results of routine physical examinations and laboratory determinations (2, 3, 22).

Statistical analysis.

The GEE approach is a statistical method that adjusts for within-patient correlation and allows for the evaluation of intervariable relationships in longitudinal studies (11, 12). Pattern-mixture modeling (13) allows patients to be grouped by their missing data patterns and adjusts for patient attrition. The GEE approach using pattern-mixture modeling was used to estimate status scores (DAS remission, defined as a DAS <1.6) as well as continuous scores (DAS). To address the possible bias on DAS and remission estimates due to patient dropouts in this long-term trial, a variable indicating whether year 3 was reached was added to the base model; further partitioning did not significantly improve the model. Therefore, a base pattern-mixture model was fit that included treatment, time, and withdrawal pattern. Treatment by time interactions were evaluated and found to be nonsignificant. Therefore, the model applying constant differences over time between treatments was appropriate. Additionally, the effect of early withdrawal was evaluated; while there was an effect, it was similar across treatment groups.

Subsequently, a stepwise procedure was used to build the covariate model adding the most significant predictor; at each step, the next most significant predictor was added until no additional predictor was significant. Predictors included baseline variables, such as number of previous DMARDs, disease duration, physical function, sex, and the presence of rheumatoid factor (RF). Using the base and covariate models, treatment differences in mean response and the odds of remission were estimated.

For the analysis, the last raw DAS and remission status in each 3-month interval from 6 through 36 months (up to 12 observations per patient) were included. GEE models were fit using SAS version 9 software PROC GENMOD (SAS Institute, Cary, NC) with a logit link function for binary data and a normal link for continuous data. To estimate the within-patient correlation needed for the longitudinal model, we selected an autoregressive pattern, which assumes that visits closer together are more correlated than visits further apart.

For estimation of the DAS, ACR response rates, and Health Assessment Questionnaire (HAQ) (23) mean changes, conventional analyses (LOCF and completers analyses) were also performed on the modified intent-to-treat (ITT) population, which included the patients who had received at least 1 dose of study drug. Linear extrapolation was used for patients who were missing a radiographic image at the 1-, 2-, or 3-year time point. Only patients with an acceptable baseline image and at least 1 postbaseline image were included in the analysis of radiographic end points.

The efficacy variables (ACR20, ACR50, and ACR70 response rates; DAS and DAS28; and HAQ scores) were analyzed using an analysis of covariance (ANCOVA) model, including factors for study center, treatment, and previous methotrexate use. ACR response rates, the DAS, and the DAS28 were analyzed using a Mantel-Haenszel approach. Radiographic end points (TSS, total erosions, and total JSN) on the ranks of modified Sharp scores were analyzed with an ANCOVA model that included factors for baseline radiographic score rank, study center, treatment, and previous methotrexate use. In addition, 95% confidence intervals (95% CIs) for the estimates were provided. To supplement the analyses on progression rate, radiographic nonprogression was analyzed using a Mantel-Haenszel approach. The relationship between DAS remission and radiographic remission (TSS change ≤0) was evaluated for patients with year 3 radiographs and DAS. The incidences of treatment-emergent AEs among treatment groups were compared using Fisher's exact test.


Patient characteristics and disposition.

Of the 686 patients randomly assigned, 682 (99%) received study drug and were considered the modified ITT population. Baseline demographics and disease characteristics were similar among the 3 groups (2). Not all patients who completed year 1 continued into year 2, and, similarly, not all patients completing year 2 continued into year 3 (Figure 1A). Forty-one patients, classified as having withdrawn for “other nonmedical” events during years 2 and 3, were withdrawn because of the administrative closure of 13 sites.

Figure 1.

A, Patient disposition. B, Kaplan-Meier curve of time to patient discontinuation for any reason.

During the 3 years, 350 patients were withdrawn from the study (Figure 1A); significantly more patients withdrew from the methotrexate group than from the combination (P < 0.001) and etanercept (P < 0.05) groups. Throughout the 3 years of the study, fewer patients in the combination group withdrew from the study compared with either monotherapy group (Figure 1B). Significantly fewer patients in the combination group (5%) were withdrawn from the study because of lack of efficacy than in either monotherapy group (17% and 16% for methotrexate and etanercept, respectively) (P < 0.001).

Clinical outcomes: DAS and remission.

Conventional time point analysis.

The efficacy end points of interest were first analyzed using both the LOCF approach and a completers analysis. By LOCF analysis, patients in the combination therapy group improved significantly compared with patients in the methotrexate or etanercept group at all time points during the 3 years. Mean DAS results over the 3-year study duration analyzed using the LOCF approach are presented in Figure 2A. By completers analysis, which included patients who completed year 3, the mean DAS at year 3 was similar for the groups receiving etanercept (2.2) and methotrexate (2.1); the combination group had a significantly lower mean DAS (1.8) than the methotrexate group (P < 0.01).

Figure 2.

Disease Activity Score (DAS) and DAS remission over time. A, Mean DAS according to univariate (last observation carried forward [LOCF]) analysis. B, Mean DAS according to multivariate (generalized estimating equation [GEE] approach using pattern-mixture modeling) analysis. C, Patients in DAS remission according to univariate (LOCF) analysis. D, Patients in DAS remission according to multivariate (GEE approach using pattern-mixture modeling) analysis. † = P < 0.05 versus methotrexate; ‡ = P < 0.05 versus etanercept; ∗ = P < 0.05 versus methotrexate.

After 3 years of treatment, a significantly higher proportion of patients treated with the combination therapy had low disease activity (64.5% with a DAS <2.4 and 56.3% with a DAS28 <3.2) compared with patients in either monotherapy group (DAS <2.4, 44.4% and 38.6% for etanercept and methotrexate, respectively; DAS28 <3.2, 33.2% and 28.5% for etanercept and methotrexate, respectively) (P < 0.01). The numbers of patients achieving remission as assessed by DAS and DAS28 criteria (DAS <1.6 and DAS28 <2.6) are presented in Table 1.

Table 1. Remission at years 1, 2, and 3 according to the DAS and DAS28*
 LOCF analysisCompleters analysis
MTXEtan.Etan. + MTXMTXEtan.Etan. + MTX
  • *

    Values are the percent of patients with disease in remission. Values were derived by last observation carried forward (LOCF) and completers analysis. For the LOCF analysis, there were 228 patients in the methotrexate (MTX) group, 223 patients in the etanercept (Etan.) group, and 190 patients in the combination (Etan. + MTX) group. For the completers analysis, at year 1, there were 156 patients in the MTX group, 164 patients in the etanercept group, and 190 patients in the combination group; at year 2, there were 117 patients in the MTX group, 135 patients in the etanercept group, and 162 patients in the combination group; at year 3, there were 82 patients in the MTX group, 98 patients in the etanercept group, and 124 patients in the combination group. DAS = Disease Activity Score; DAS28 = DAS in 28 joints.

  • P < 0.01 versus MTX.

  • P < 0.01 versus etanercept.

  • §

    P < 0.05 versus MTX.

DAS <1.6      
 Year 114.017.537.219.221.342.1
 Year 215.823.3§40.724.831.151.2
 Year 317.521.540.728.030.851.5
DAS28 <2.6      
 Year 117.117.538.121.822.043.2
 Year 218.922.442.425.629.653.7
 Year 318.920.640.336.030.851.5

Longitudinal analysis.

DAS and DAS remission results were also analyzed using the GEE approach with pattern-mixture modeling. The results of the longitudinal analysis confirmed those of the conventional time point analysis. The DAS was statistically significantly lower during the entire study period in the combination therapy group compared with both the etanercept group and the methotrexate group, irrespective of time, within-patient correlation, and unbalanced withdrawal (Table 2 and Figure 2B). The longitudinal analysis results for DAS remission were similar (Table 2 and Figure 2D).

Table 2. Longitudinal analysis for DAS and DAS remission*
 Estimated treatment difference or OR95% CIP
  • *

    95% CI = 95% confidence interval (see Table 1 for other definitions).

  • Estimated treatment differences are given for the DAS analysis, and odds ratios (ORs) are given for the DAS remission analysis.

 Base model   
  Combination vs. MTX−0.49−0.70, −0.27<0.0001
  Combination vs. etanercept−0.50−0.70, −0.29<0.0001
  Etanercept vs. MTX−0.13−0.19, 0.220.9053
 Covariate model   
  Combination vs. MTX−0.48−0.65, −0.30<0.0001
  Combination vs. etanercept−0.45−0.62, −0.28<0.0001
  Etanercept vs. MTX−0.03−0.20, 0.150.7767
DAS remission   
 Base model   
  Combination vs. MTX2.391.64, 3.48<0.0001
  Combination vs. etanercept2.171.52, 3.12<0.0001
  Etanercept vs. MTX1.100.74, 1.640.6435
 Covariate model   
  Combination vs. MTX2.731.87, 4.01<0.0001
  Combination vs. etanercept2.411.63, 3.57<0.0001
  Etanercept vs. MTX1.130.74, 1.740.5625

In both the base and covariate models for mean DAS, estimated treatment differences between combination therapy and either monotherapy were significant (P < 0.001), but the differences between the monotherapies were not (Table 2 and Figure 2B). In the models for DAS remission, patients in the combination group had a 2.3–2.6-fold greater chance to be in remission at any time point during the 3 years than patients in either the etanercept group or the methotrexate group (P < 0.0001) (Table 2 and Figure 2D).

The covariate model describes the same relationships as the base model with adjustment for a set of potentially important covariates including sex, DAS, number of prior DMARDs, TSS, HAQ score, and disease duration at baseline. Although these covariates were all significant predictors of long-term remission, there were no significant interactions between any covariate and treatment effect, which further supports the validity of the conventional time point results (Table 2).

The effect of baseline variables on mean response and the odds of remission were also assessed using the GEE approach with pattern-mixture modeling. Independent predictors of DAS remission (stepwise P < 0.05) included male sex (odds ratio [OR] 1.92; 95% CI 1.32, 2.77), lower DAS (OR 1.68; 95% CI 1.43, 1.96), lower TSS per 10 units (OR 1.08; 95% CI 1.03, 1.11), and lower HAQ score (OR 1.59; 95% CI 1.19, 2.04) at baseline. Baseline CRP level (P = 0.08), RF (P = 0.86), previous methotrexate use (P = 0.18), longer disease duration (P = 0.26), and age (P = 0.39) did not contribute significantly to achieving DAS remission.

Other efficacy measures.

We also reported responses assessed by the ACR20, ACR50, and ACR70 criteria and analyzed them using conventional methods. At year 3, an LOCF analysis showed that ACR20 responses were achieved by 85.3% of patients receiving the combination compared with 70.9% and 70.2% of patients receiving etanercept and methotrexate, respectively (P < 0.01 for the combination versus either monotherapy). ACR50 responses were achieved by 67.1% of patients receiving the combination and by 45.7% and 43.9% of patients receiving etanercept and methotrexate, respectively (P < 0.01 for the combination versus either monotherapy), and ACR70 responses were achieved by 47.2% of patients receiving the combination compared with 26.0% and 21.1% of patients receiving etanercept and methotrexate, respectively (P < 0.01 for the combination versus either monotherapy). Similar trends were observed in the analysis that included only those patients who completed year 3.

Physical disability as assessed by improvement in the HAQ score was significantly improved in both the combination and etanercept monotherapy groups after 2 weeks of therapy compared with the methotrexate monotherapy group (P < 0.01), and the improvement was sustained throughout the 3 years of the study. At 3 years, the improvement (55%) seen with combination therapy was significantly greater than that seen with either monotherapy (37.0% and 33.3% for etanercept and methotrexate, respectively) (P < 0.01). Furthermore, a significantly greater percentage of patients in the combination group (48.1%) had no disability (HAQ score 0) compared with patients in either monotherapy group (35.4% and 32.9% for etanercept and methotrexate, respectively) (P < 0.01).

Radiographic progression.

A total of 638 patients (217 in the combination group, 211 in the etanercept group, and 210 in the methotrexate group) who had baseline radiographs and at least 1 set of postbaseline radiographs were included in the radiographic analyses. At baseline, the mean TSS for patients in the methotrexate, etanercept, and combination groups were similar (30.7, 28.6, and 31.9, respectively).

Mean changes from baseline in the TSS at all 3 time points (year 1, year 2, and year 3) were significantly lower for patients receiving combination or etanercept therapy than for those receiving methotrexate therapy (P < 0.05); the TSS change score was lower for the combination group than for the etanercept group (P < 0.05) (Figure 3A). At year 3, mean TSS change scores were –0.14 (95% CI –1.07, 0.78) for combination therapy versus 1.61 (95% CI 0.41, 2.81) and 5.95 (95% CI 2.96, 8.94) for etanercept and methotrexate, respectively (P < 0.01). Similarly, the mean changes from baseline in erosion scores for the combination and etanercept groups were significantly lower compared with the methotrexate group (P < 0.05) (Figure 3C). At 3 years, the combination-treated patients had significantly lower erosion change scores (–0.67; 95% CI –1.05, –0.28) than patients treated with etanercept alone (0.39; 95% CI –0.44, 1.22) or methotrexate alone (3.25; 95% CI 1.50, 5.01) (P < 0.05). The mean change from baseline in JSN at year 3 was significantly lower for patients in the combination group (–0.67; 95% CI –1.05, –0.28) compared with those in the methotrexate (2.70; 95% CI 1.26, 4.13) (P < 0.01) and etanercept (1.22; 95% CI 0.59, 1.84) (P < 0.01) groups (Figure 3E).

Figure 3.

Mean change in radiographic progression and corresponding distribution of individual change scores by treatment. A, Mean change from baseline in total modified Sharp/van der Heijde scores (TSS). B, Probability plots of TSS change scores. C, Mean change from baseline in erosion scores. D, Probability plots of erosion scores. E, Mean change from baseline in joint space narrowing (JSN). F, Probability plots of JSN. ∗ = P < 0.05 versus methotrexate; † = P < 0.05 versus methotrexate; ‡ = P < 0.05 versus etanercept.

The distribution of individual changes from baseline in the TSS, erosion scores, and JSN scores at 3 years was charted by treatment on probability plots (Figures 3B, D, and F). The majority of patients in the 3 treatment groups had values close to zero; a zero or negative change from baseline was seen in more patients receiving combination therapy than in those receiving either monotherapy.

After 3 years, significantly more patients in the combination (76%) and etanercept (61%) groups had radiographic remission of their disease (defined as a TSS change score ≤0.5 units) than did patients in the methotrexate group (51%) (P < 0.05). The comparison between the combination and etanercept groups was also significant (P < 0.05).

Sensitivity analyses confirmed the primary ITT analysis using the linear extrapolation approach. The completers analysis included only those who had a radiograph at 36 months (136 in the combination group, 109 in the etanercept group, and 91 in the methotrexate group). Overall, the progression rate was lower in the completers group compared with the ITT population in all 3 groups, but the differences between treatment groups remained. For the sensitivity analysis that included all patients, the mean changes from baseline to 36 months in the TSS, erosion scores, and JSN were used for the 44 patients who did not have acceptable data in the clinical ITT population (total n = 682). The mean baseline data were assigned to any patients not included in the primary analysis. The results for the TSS, erosion scores, and JSN generated using the primary- and mean-imputation methods were similar (data not shown).

DAS and radiographic remission.

In patients who achieved DAS remission, radiographic remission (TSS ≤0.5) was also attained in 85.3%, 80.0%, and 55.6% of patients in the combination, etanercept, and methotrexate groups, respectively. Conversely, more patients in the group not achieving DAS remission showed radiographic progression (TSS >0.5) (33.9%, 45.2%, and 58.0% of patients in the combination, etanercept, and methotrexate groups, respectively).


Patient exposure to the combination therapy (554.2 patient-years) was considerably higher than that for either methotrexate (437.2 patient-years) or etanercept (479.1 patient-years) monotherapy. This was because of the higher retention rate seen with combination therapy during the 3 years of the study.

Adverse events reported during the first 2 years of the TEMPO study have been reported previously (2, 3). No new or unexpected safety findings were reported during year 3, and the proportions of patients reporting ≥1 treatment-emergent AE or infection were similar across treatment groups (22). No significant differences were seen among the groups in the incidence of serious AEs; noninfectious serious AEs were reported for 23.4%, 22.9%, and 18.9% of patients, and serious infections were reported for 7.4%, 6.7%, and 8.3% of patients in the combination, etanercept, and methotrexate groups, respectively. All serious infections that occurred more than once during the 3-year duration of the study are listed in Table 3.

Table 3. Serious infectious events (>1 event) during the 3-year duration of the study*
 MTX (n = 228)Etan. (n = 223)Etan. + MTX (n = 231)
  • *

    Values are the number (%) of patients. There were no significant differences (P ≥ 0.05) between groups in the number of serious infectious events. See Table 1 for definitions.

  • Includes infections of toes, herpes zoster, and erysipelas.

  • Includes cholecystitis, diverticulitis, cholangitis, hepatitis B infection, and abdominal wall incisional purulence.

  • §

    Includes salmonellosis, gastroenteritis, and diverticulitis.

Any infection19 (8.3)15 (6.7)17 (7.4)
Pneumonia4 (1.8)4 (1.8)6 (2.6)
Septic arthritis1 (0.4)3 (1.3)3 (1.3)
Skin infection3 (1.3)2 (0.9)0 (0.0)
Cellulitis/abscess1 (0.4)2 (0.9)1 (0.4)
Postoperative wound infection2 (0.9)1 (0.4)1 (0.4)
Abdominal2 (0.9)1 (0.4)1 (0.4)
Gastrointestinal tract§1 (0.4)2 (0.9)0 (0.0)
Sepsis syndrome2 (0.9)0 (0.0)0 (0.0)

A total of 13 patients (3 in the methotrexate group, 5 in the etanercept group, and 5 in the combination group) with a history of tuberculosis were enrolled in this study; reactivation of tuberculosis developed in none of these patients. During year 3, tuberculosis was diagnosed in 1 patient receiving combination therapy. A 2-cm abnormal lesion was detected in the lung on a chest radiograph, and histopathology findings revealed tuberculosis. The patient was withdrawn from the study and subsequently underwent surgery and received antituberculosis treatment (isoniazid, rifampicin, ethambutol, and pyrazinamide). Results of specific stains, cultures, and a nuclear probe examination were negative for tuberculosis 2 months after treatment.

Five patients died during the 3-year TEMPO study; 3 deaths occurring before the year 1 interim analysis were reported previously (2). No patients died during year 2 of the study. During year 3, 1 patient receiving etanercept died from acute pulmonary edema, and 1 patient receiving combination therapy died from cardiac arrest.


The TEMPO study is the first long-term, double-blind, controlled study to examine the relative efficacy of etanercept and methotrexate, given alone and in combination, for the treatment of patients with RA who were not receiving background methotrexate at the beginning of the study. In this study, treatment with the combination of etanercept and methotrexate resulted in marked improvement in the key treatment outcomes, such as disease activity, radiographic progression, and physical function, compared with treatment with either monotherapy early during the first year of the study (2), and the therapeutic benefit was maintained during ∼3 years. The inhibitory effect of etanercept on radiographic progression was superior to that of methotrexate; however, the comparative clinical efficacies of the monotherapies were not significantly different over the 3-year duration of the study. Regarding safety outcomes, more patients receiving methotrexate monotherapy withdrew from the study due to lack of efficacy and adverse events than those receiving etanercept monotherapy. One of the important objectives of long-term studies is to reliably determine whether efficacy is maintained. Conventional time point analysis may be acceptable for shorter-term studies but may not appropriately provide a true estimate of efficacy, because the impact of missing data, patient withdrawals, and responder bias may confound the analyses, especially in long-term trials.

Although one common practice is to evaluate only patients who have completed the full duration of a trial, this approach is not without complexities; statistical analyses of completers may be difficult to interpret, because patients who continue remain in the study because they are responding to the treatment. Rigorous trial blinding as in the TEMPO study does not preclude this “bias by trial completion” (although it eliminates the concern that a particular drug is preferentially withdrawn based on knowledge of that drug). Another important question that cannot be answered by conventional time point analyses is whether a given disease state induced by the study drug (i.e., remission) is maintained, fluctuates, or is lost over time.

The mixed-model repeated-measures analysis has an advantage over observed case analyses, because all data points are used in modeling the response over the entire observation period. Unlike LOCF or nonresponse imputation, the mixed-model repeated-measures analysis does not fill in missing time points with values that may not be valid. Mixed-model repeated-measures analysis alone, like the completers analysis, may be biased at later time points, because the subjects with complete data have more influence at later time points than do early dropouts. The pattern-mixture approach modifies the mixed-model repeated-measures model by evaluating the model by the time of dropout (pattern) and weighing the overall model according to the proportions of patients with each pattern (mixture). Pattern-mixture modeling was used to adjust for bias that may potentially occur by unbalanced patient withdrawal regardless of reason. Additionally, this model of remission was used to determine the predictors of disease remission for the 3 treatment regimens.

This longitudinal analysis confirmed the results obtained using the traditional LOCF time point analysis. These data appear to validate the reliability of the traditional LOCF analysis, confirming the robustness of the traditional approaches. Both approaches showed that combination therapy reduced disease activity to a significantly greater extent than either monotherapy, as measured by the DAS; the GEE approach using pattern-mixture modeling showed that at any time point in the 3-year followup, the probability of being in DAS remission was ∼2.5 times higher in patients receiving combination therapy, and that this superiority over time was not an artifact of unbalanced withdrawal (selection of the responders) or an “overweighted” initial positive clinical response. The concordance between the conventional and longitudinal approaches further reinforces the reliability of the remission findings in this trial.

The disease remission modeling predicted that patients receiving combination therapy with etanercept plus methotrexate and patients continuing to receive therapy for the duration of the study had a significantly greater probability of achieving disease remission. Men and patients with lower disease activity, less joint destruction, and/or better physical function at baseline were more likely to reach remission. These observations suggest that early and long-term treatment with combination therapy leads to disease remission, the ultimate goal of treatment for patients with RA.

The 3-year reading of the radiographic results included all of the radiographic data through the end of year 3. The year 3 radiographic results are remarkably consistent with the radiographic results in the year 1 and year 2 reports, as were the results from the rereading of the year 1 and year 2 radiographic images, adding to the robustness of these radiographic findings and confirming that they are driven by true differential treatment effects rather than by other effects, such as inter- and intrareader variability, that can potentially influence scoring and analysis of such end points.

Therapy with the combination of etanercept and methotrexate continued to show a negative mean TSS change from baseline. Patients in this study had evidence of significant damage at baseline, with a mean TSS of ∼30.4. This damage had accumulated over a mean disease duration of 6.4 years. It was estimated that if joint damage had continued at the same rate as it had before study entry, these patients would have had an increase of 14.4 TSS units in 3 years. Treatment with a combination of etanercept and methotrexate halted this progression, and there was even a small decrease in joint damage at 12, 24, and 36 months compared with baseline, although the decrease at 36 months was not significantly different. Further analysis of these data illustrates that this improvement is primarily driven by an effect on erosions: the entire 95% CI for erosions remained below zero at all 3 time points. Reductions in the numbers of erosions in the combination group were statistically significant throughout the 3 years of this study, suggesting that improvement in joint damage may be due to a reduction in the size and/or number of erosions, perhaps because of filling in of erosions. The significance of this observation is not known.

In the probability plots, the change from baseline was zero for the majority of patients in this 3-year study, indicating that joint damage progression was undetectable for the majority of patients, in all 3 treatment arms. In the minority of patients showing a change, the probability plots showed that progression was more frequent with methotrexate and least frequent with combination therapy.

As observed in the 1-year report on remission (24), significantly more patients receiving etanercept in combination with methotrexate or as monotherapy achieved DAS remission than did patients receiving methotrexate alone. After 3 years, at least 80% of patients receiving etanercept or combination therapy whose disease was in remission by DAS criteria also did not have radiographic progression. As previously demonstrated in a comprehensive analysis using year 2 data from this study, the association between disease activity and radiographic progression was treatment dependent (25). An association was detected in the methotrexate group; however, there was a clear disconnect between these 2 measures in patients receiving combination therapy with etanercept plus methotrexate.

During the 3 years, treatment with etanercept, both alone and in combination with methotrexate, was well tolerated and did not result in any unexpected safety findings. As seen in earlier studies, combining the 2 therapies resulted in an acceptable safety profile. During the 3 years, there was 1 report each of tuberculosis and bronchopulmonary aspergillosis (3), both in patients receiving combination therapy. The total number of deaths (n = 5) observed in this study is far lower than the 14 deaths expected, based on total patient exposure using the general US population adjusted for age and sex.

In conclusion, conventional time point analyses as well as longitudinal analyses have clearly shown that the response to combination treatment was durable and maintained for at least 3 years of continuous therapy, with no unexpected safety findings. The probability of achieving and maintaining remission was considerably and statistically higher with the combination compared with either monotherapy during the entire 3 years of the study; patients who receive continued therapy with etanercept plus methotrexate are more likely to achieve disease remission. The continuous and significant decrease in erosion scores suggests that improvement of joint damage may be possible in patients treated with the combination of etanercept and methotrexate.


Dr. van der Heijde had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Van der Heijde, Klareskog, Herrero-Beaumont, Pedersen, Wajdula, Fatenejad.

Acquisition of data. Bruyn, Cantagrel, Herrero-Beaumont, Molad, Codreanu, Valentini, Zahora, Wajdula, Fatenejad.

Analysis and interpretation of data. Van der Heijde, Klareskog, Landewé, Bruyn, Durez, Codreanu, Pedersen, MacPeek, Wajdula, Fatenejad.

Manuscript preparation. Van der Heijde, Klareskog, Landewé, Bruyn, Durez, Herrero-Beaumont, Codreanu, Valentini, Pedersen, MacPeek, Wajdula, Fatenejad.

Statistical analysis. Landewé, Pedersen, MacPeek, Wajdula, Fatenejad.

Patient recruitment/care. Bruyn.


Wyeth Pharmaceuticals was responsible for the collection and analysis of data. The authors and Wyeth Pharmaceuticals were involved in the study design, data interpretation, manuscript writing, and the decision to publish.


We would like to acknowledge the following individuals from Wyeth Research: R. Lavielle (Paris, France), L. Grassnickel (Munster, Germany), D. Simon (Zug, Switzerland), G. de Crescenzo (Aprilia, Italy), T. Nash (Baulkham, Australia), J.-M. Kroodsma (Hoofddorp, The Netherlands), A. Zareba (Warsaw, Poland), R. Murillo (Madrid, Spain), and G. Skoglund (Solna, Sweden). R. Pereira in the Wyeth Publications and External Communications Group is acknowledged for her writing support.



The TEMPO study investigators, in addition to the authors of this article, are as follows: H. D. Bolosiu (Cluj-Napoca, Romania), P. Bourgeois (Paris, France), J. Cunha Branco (Lisbon, Portugal), J. Braun (Herne, Germany), J. Broell (Wien, Austria), J. Brzezicki (Elblag, Poland), G. Burmester (Berlin, Germany), B. Canesi (Milan, Italy), X. Chevalier (Creteil, France), H. Chwalinska-Sadowska (Warsaw, Poland), L. Cleland (Adelaide, Australia), F. Radulescu (Bucharest, Romania), L. Coster (Linkoping, Sweden), M. Cutolo (Genoa, Italy), R. Dahl (Uppsala, Sweden), P. Dawes (Burslem Stoke-on-Trent, UK), J. Dehais (Bordeaux, France), D. J. de Rooij (Nijmegen, The Netherlands), D. Doyle (London, UK), L. Euller-Ziegler (Nice, France), F. Fantini (Milan, Italy), G. Ferraccioli (Udine, Italy), A. Filipowicz-Sosnowska (Warsaw, Poland), S. Freiseleben-Sorensen (Copenhagen, Denmark), P. Geusens (Diepenbeek, Belgium), J. Melo-Gomes (Lisbon, Portugal), J. J. Gómez Reino (Santiago de Compostela, Spain), A. Gough (North Yorkshire, UK), J. P. de Jager (Southport, Australia), M. Janssen (Arnhem, The Netherlands), H. Julkunen (Vantaa, Finland), R. Juvin (Grenoble, France), H. Haentzschel (Leipzig, Germany), W. Hissink-Muller (Tilburg, The Netherlands), J. Kalden (Erlangen, Germany), C. Kaufmann (Drammen, Norway), J. Kekow (Vogelsang, Germany), A. Harju (Stockholm, Sweden), T. Kvien (Oslo, Norway), A. Laffón (Madrid, Spain), H. Lang (Plauen, Germany), P. Lanting (Doetinchem, The Netherlands), X. LeLoet (Rouen, France), B. Lindell (Kalmar, Sweden), F. Liote (Paris, France), R. Luukkainen (Satalinna, Finland), M. Malaise (Liege, Belgium), X. Mariette (Le Kremlin Bicetre, France), E. Martín Mola (Madrid, Spain), B. Masek (Venlo, The Netherlands), Z. Mencel (Kalisz, Poland), O. Meyer (Paris, France), C. Montecucco (Pavia, Italy), R. Myllykangas-Luosujarvi (Kuopio, Finland), H. Nielsen (Herlev, Denmark), G. Papadimitriou (Athens, Greece), K. Pavelka (Prague, Czech Republic), A. Perniok (Cologne, Germany), V. Rodriguez Valverde (Santander, Spain), A. Rubinow (Jerusalem, Israel), P. Sambrook (St. Leonards, Australia), R. Sanmartí (Barcelona, Spain), J. Sany (Montpellier, France), A. Saraux (Brest, France), M. Schattenkirchner (Munich, Germany), U. Serni (Florence, Italy), L. Settas (Thessaloniki, Greece), J. Sibilia (Strasbourg, France), H. Stehlikova (Ceska Lipa, Czech Republic), J. Szechinski (Wroclaw, Poland), J. Szerla (Krakow, Poland), C. Tanasescu (Bucharest, Romania), H. Tony (Wurzburg, Germany), J. Tornero-Molina (Guadalajara, Spain), S. Transo (Jonkoping, Sweden), F. Trotta (Ferrara, Italy), T. Tuomiranta (Tampere, Finland), E. Veys (Ghent, Belgium), H. Van den Brink (Alkmaar, The Netherlands), M. Van de Laar (Enschede, The Netherlands), P. Vitek (Zlin, Czech Republic), R. Westhovens (Leuven, Belgium).

The evaluators of the radiographic images were A. van Everdingen, Deventer, The Netherlands, and T. H. M. Schoonbrood, Rheumatology Department, University Hospital, Maastricht, The Netherlands. D. van der Heijde, Maastricht, The Netherlands, assisted as a consultant in the radiographic methodology applied in this study.