DAVID T. FELSON, JENNIFER J. ANDERSON, MARY L. M. LANGE, GEORGE WELLS. and MTCHAEL P. LAVALLEY

Transcription

1 1564 ARTHRITIS 8z RHEUMATISM Vnl. 41, No. 9, September 1998, pp , American Cnllcgc of Rhcumatology SHOULD IMPROVEMENT IN RHEUMATOID ARTHRITIS CLINICAL TRIALS BE DEFINED AS FIFTY PERCENT OR SEVENTY PERCENT IMPROVEMENT IN CORE SET MEASURES, RATHER THAN TWENTY PERCENT? DAVID T. FELSON, JENNIFER J. ANDERSON, MARY L. M. LANGE, GEORGE WELLS. and MTCHAEL P. LAVALLEY Objective. To determine whether improvement of more than 20% in core set parameters should be required before patients are characterized as improved in rheumatoid arthritis (RA) clinical trials. Methods. Data from 6 RA trials were reanalyzed to evaluate the discriminant validity (ability to differentiate active treatment from control) of 4 proposed definitions of improvement: the current American College of Rheumatology (ACR) definition (a 20% threshold for core set parameters [ACR 20]), a 50% threshold (ACR SO), a 70% threshold (ACR 70), and an ordinal definition in which a patient could be classified in any of 3 categories (unimproved, ACR 20, or ACR 50). To evaluate the discriminant validity of these 4 definitions of improvement, we characterized each patient in each trial as improved or not, based on each definition, and computed a chi-square value differentiating the active treatment group from the control group, with the corresponding P value. Results. With an increase in the threshold for improvement, the percentage of placebo-treated patients who were classified as experiencing response dropped dramatically in all trials, as did the percentage of patients receiving active therapy (second-line drug, combination therapy, tumor necrosis factor p75-fc fusion protein) who were classified as experiencing response. Generally, the drop in active treatment re- ~. ~ global Supported by NIH grant AR David T, Pelson, MD, MpH, Jennifer J. AI1derson, PhD, Michael P. LaVallev, PhD: Boston University Arthritis Centcr and School of Public Health, Boston, Massachuseits; Mary I,. M. Lange, MS: Immunex Corporation, Seattle, Washington; George Wells, PhD: Loeb Research Institute. Ottawa. Ontario. Canada. Address reprint requests to David T. Felson, MD, MPH, Arthritis Center, Room A-203, Boston University School of Medicine, 715 Albany Street, Boston, MA sponse rates was greater than the drop in placebo response rates, leaving the difference between the 2 groups less at the higher thresholds. Therefore, chisquare values fell as the threshold for response was raised. The ordinal definition of improvement yielded chi-square values similar to those obtained using ACK 20 alone. Conclusion. Adopting a definition of efficacy in RA trials that requires 50% or 70% improvement in core set parameters would likely compromise statistical power and make it more difficult to distinguish between 2 treatments with different efficacy. ACR 20 should continue to be the primary measure of eflicacy in RA trials, with higher thresholds for improvement being determined and reported as secondary efficacy measures. The American College of Rheumatology (ACR) preliminary definition of improvement in rheumatoid arthritis (RA), published in 1995 (1,2), has been adopted widely as a primary outcome measure in RA clinical trials. Validation studies (3) have corroborated its ability to efficiently differentiate results of active treatment versus placebo. To reach improvement according to the ACR definition, a patient with RA must improve by at least 20% in the tender and swollen joint count and by at least 20% in 3 of 5 other core set measures (patient assessment, physician global assessment, physical disability score, acute-phase reactant, and patient pain assessment). Whie the ACR 20% definition accomplishes what was intended, i.e., discriminating between 2 treatments, the minimum threshold of 20% improvement is not a high one. Thus, many recent trials in RA have Submitted for publication December 31, 1997; accepted in measured both *'% and a higher Of revised form March 19, improvement (43) or have focuscd on an ACR (or

2 THRESHOLD FOR IMPROVEMENT IN RA CLINICAL TRIALS 1565 Table 1. Trials included in the present study Treatments compared (ref.) MTX vs. GST vs. DP vs. AUR vs. MTX vs. MTWCSA vs. TNFR:Fc vs. placeho (1l)t$ placeho (13)t$ placebo (12)t$ placebo (13)s AUR (14)$ MTWplaceho (15) placebo (4) No. of RA patients n n analyzed Duration of trial. weeks 18-30# Average duration of RA in 10# >5 patients studied, years * MTX = methotrexate; GST = gold sodium thiomalate; DP = D-penicillamine; AUR = auranofin; CSA = cyclosporin A, TNFR:Fc = tumor necrosis factor p75-fc fusion protein; RA = rheumatoid arthritis. t Trial conducted by the Cooperative Systematic Studies of Rheumatic Diseases group (CSSRD). $ Trial data also used to validate the American College of Rheumatology (ACR) preliminary definition of improvement (1). 0 These 2 arms were from the CSSRD trial of GST, AUR, and placeho. The GST vs. placebo comparison, which was used to develop thc ACR definition of improvement, is listed separately here and in Table 2. l Two hundred seventy-four patients were analyzed in these 3 trials combined. #Data from 3 trials combined. Paulus criteria [6])-based definition of improvement, utilizing only a higher threshold of improvement (7,8). With potent new therapies on the horizon for RA, it may be reasonable to raise the bar and define RA improvement by more substantial changes in RA core set measures, such as requiring at least 50% or even 70% improvement. The validity of higher thresholds to define improvement has not been evaluated formally. In increasing the threshold for defining improvement, we anticipated that one desirable corollary would be a drop in the percentage of placebo-treated patients classified as having experienced improvement. A drop in response (improvement) rates in patients receiving active treatment should also occur. The net effect of these 2 changes on the treatment/placebo difference is unknown. One of the most important components of validity in a definition of improvement is its discriminant validity, which has been defined as the ability of a measure to distinguish clinically significant differences between treatments (9). The higher the discriminant validity, the greater the statistical power in distinguishing 2 treatment groups. Use of definitions of improvement with higher discriminant validity would enable one to detect the difference between active treatment and placebo with fewer patients. If discriminant validity is low, trials require the recruitment of more patients. Using a diverse set of RA clinical trials including second-line drug-versus-placebo trials, a comparative trial of second-line drugs, a recent combination therapy trial, and a trial of a promising biologic agent, we evaluated the comparative discriminant validity of different ACR definition-based definitions of improvement in RA. METHODS We reanalyzed data from 6 controlled clinical trials of RA treatments. All trials have been published, and all studied adult patients who met the ACR (formerly, the American Rheumatism Association) criteria for RA (10). Thrcc of the trials were performed by the Cooperative Systematic Studies of Rheumatic Diseases group (CSSRD). These include a study of low-dose methotrexate (MTX) versus placebo (ll), a study of D-penicillamine (DP) versus placebo (12), and a study of injectable gold versus auranofin (AUR) versus placebo (13). Also, we analyzed data from a large multicenter trial of MTX vcrsus AUR (14). Data from these trials were used to develop and validate the ACR core set and definition of improvement. Dlfferent percentage thresholds were not compared as part of this work. We also analyzed data from other trials not previously analyzed. First, we analyzed the AUR versus placebo comparison from the CSSRD trial of injectable gold versus AUR versus placebo (13). Next, we analyzed data from a combination therapy trial of combined cyclosporine (CSA) and MTX versus MTX alone in severe RA (15), and a recently published trial of tumor necrosis factor p75-fc fusion protein (TNFR: Fc) versus placebo (4). In the latter trial, a phase I1 trial in which 3 different doses of TNFR:Fc were compared with placebo, we focused only on the high-dose group, which experienced the best response. We did this because it provided the maximal discrimination between active treatment and placebo, and high-dose TNFR:Fc constitutes one of the best examples so far of newly developed treatments in RA. Details of the trials are shown in Table 1. For the purpose of comparing highly effective treatments versus placebo, we removed treatment arms from the CSSRD trials that we believed, based on meta-analysis data (1 6,17), constituted weaker therapies. Thus, we removed the AUR arm from the gold trial (which we then analyzed separately) and the low-dose DP arm (125 mg/day) from the DP trial. This afforded the maximal discrimination between active treatment and placebo. We treated the MTX versus AUR trial as a trial of a potent second-line drug versus a relatively weak one (17).

3 1566 FELSON ET AL Using these trial data, we tested several different definitions of improvement based on those being used in trials presented at recent ACR meetings and in published reports (4, 8). First, as the standard definition, we used the ACR preliminary definition of improvement (1). The ACR definition requires that the patient has improved by at least 20% in both tender and swollen joint counts and also has improved by at least 20% in 3 of 5 other core set parameters (patient global assessment, physician global assessment, self-reported physical disability, acute-phase reactant, and patient pain assessment). This definition will be referred to as ACR 20. We developed an analogous ACR 50 definition. To achieve ACR 50, a patient must have improved by at least 50% in both tender and swollen joint count and by at least 50% in 3 of the 5 other core set parameters. We also developed a similar definition of 70% improvement, called ACR 70. Realizing that improvement according to the ACR 20 definition likely corresponds to a minimally important clinical improvement and that the higher threshold levels may signify major improvement, we also created an ordinal variable in which patients in a trial could be classified in 1 of 3 categories based on their response: They could fail to meet ACR 20. They could meet criteria for ACR 20 but not meet criteria for ACR 50 (i.e., achievement of at least 20% improvement in both tender and swollen joint counts and at least 20% improvement in 3 of 5 other parameters, but without achievement of the criteria for ACR 50 described above). Finally, they could have achieved ACR 50. Thus, within each treatment group in a trial, there would be 3 response groups (<ACR 20, ACR 20, ACR SO), and an ordinal outcome could be evaluated. ACR 70 was not included as one of the ordinal outcomes because a preliminary examination of data from the 3 CSSRD trials suggested that since so few patients reached this threshold, little additional information would be provided by adding an ACR 70 level or substituting ACR 70 for ACR 50. In analyzing the data from the trials, we used only the end-of-trial data and an intent-to-treat design. Our statistical measure of discriminant validity was the chi-square, which tested treatment response rate of one treatment group versus the other. In the main statistical analysis, we used a Mantel- Haenszel chi-square because it is computed the same way for both dichotomous and ordinal outcomes. This allowed us to compare the discriminant validities of ACR 20, ACR 50, and ACR 70 and the ordinal outcome proposed above. Despite small numbers of patients in some cells, we did not use continuity corrections, since this would bias against definitions of improvement in which few patients improved. Since these definitions were likely to be those with higher thresholds for improvement, we did not want to bias against these newly proposed definitions. Thus, the results reported here are forgiving for definitions with a higher threshold for improvement. These would actually have less statistical power (or discriminant validity) than we are suggesting. Generally speaking, continuity corrections should be used if the expected number in any cell in a 2 X 2 table is 5 or less, a situation that would hold in a placebo-controlled trial if the total number of patients who reached a high threshold of improvement were low. To show the effect of the small numbers, we did compute P values derived from a Fisher s exact test for analyses of ACR 20, 50, and 70. To project when a higher threshold for improvement Table 2. Discriminant validity of different definitions of KA improvement, by trial* 3 CSSRD trials Placebo (n = 119) High-dose DP, GST, MTX (n = 155) J J p MTX vs. AUR trial AUR (n = 118) MTX (n = 119) AUR vs. placebo trial Placebo (n = 38) AUR (n = 56) J x p MTWCSA vs. MTX/placebo trial MTWplacebo (n = 73) MTX/CSA (n = 75) x J p TNFR:Fc (high-dose only) vs. placebo trial Placebo (n = 44) TNFR:Fc (n = 44) Definition of improvement ACR ACR ACR ACR (ordinal) ( < < <O.(H)l < : c * Values for improvement are the percent of patients in each treatment group reaching this threshold. See Table 1 for definitions. (e.g., ACR 50 or ACR 70) might offer enhanced discriminant validity, we computed the sample size in each arm of a 2-arm trial that would be needed in order to achieve 80% power with a 2-sided 0.05 significance level test of the difference between active treatment and placebo response rates. These sample sizes are based on the assumption that there is equal allocation of subjects into the 2 arms of the trial. The arcsin transformation (18) was used with version 4.1 of STPLAN (available from Dr. Barry Brown, M. D. Anderson Cancer Center, Houston, TX) to compute sample sizes. A probability of response of was used to approximate the sample sizes required, with a placebo response probability of 0. RESULTS The percentage of patients in each treatment group in each trial who were responders or who experienced improvement according to these different definitions is shown in Table 2. Findings with regard to response rates with placebo or weak treatment will be addressed first. In the CSSRD trials, 8.4% of placebo-

4 THRESHOLD FOR IMPROVEMENT IN RA CLINICAL TRIALS 1567 treated paticnts experienced improvement at the ACR 20 level. No placebo-treated patients experienced improvement at ACR SO or ACR 70. In the AUR versus MTX trial, 28.8% of AUR-treated patients experienced improvement at the ACR 20 level, whereas the percentage of patients who experienced higher thresholds of improvement dropped dramatically (to 17.8% for ACR 50 and 5.9% for ACR 70). Raising the threshold reduced placebo response in the AUK versus placebo trial to 0. A similar dramatic drop in response rates was seen in the MTXiCSA combination trial among those treated with MTX alone, in whom the percentage responding dropped from 12.2% using the ACR 20 definition to 2.7% using ACR 50 and to 0% using ACR 70. Similarly, in the TNF receptor trial, the placebo group response rate dropped from 13.6% using ACR 20 to 6.8% using ACR 50 to 2.3% (1 patient) using ACR 70. Therefore, if one intends to diminish the likelihood of placebo response, raising the threshold for response is an effective method. Data on rates of response with effective therapy are also shown in Table 2. The percentage of responders with active and effective therapy also dropped precipitously when the threshold for improvement was raised. For example, in the 3 CSSRD trials, the percentage of patients receiving active second-line drugs who improved at the ACR 20 lcvel was 40.3%, and this dropped to 9.0% using the ACR 50 definition and 0.6% (1 patient) using the ACR 70 definition. In the AUR versus MTX trial, the percentage of MTX-treated patients experiencing improvement dropped sharply with a rise in the threshold for improvement, from 64.7% (ACR 20) to 35.3% (ACR 50) to 9.2% (ACR 70). Similar drops in the percentage of patients reaching the improvement definition were seen in the AUR versus placebo comparison, in the combination therapy trial among those treated with the active combination MTXKSA, and in the TNFR.Fc trial in the high-dose active treatment group. Thus, while the rate of response to placebo or weak treatment dropped dramatically with raising of the threshold, the rate of response to active treatment, if anything, dropped even more. When the placebo response rate was subtracted from the active treatment response rate in placebo-controlled trials, the difference in response rates diminished with an increase in threshold. For example, in the CSSRD trials, the difference in response rates between active treatment and placebo went from 31.9% to 9% to 0.6% as the threshold was raised from ACR 20 to ACR 50 to ACR 70. In the TNFR:Fc trial, the difference in response rates using ACR 20 was 61.4%, but it dropped to 18.2% using ACR 70. Focusing on the chosen measure of discriminant validity, in evcry trial the chi-square dropped as the threshold for improvcment was raised. For example, in the CSSRD trials the chi-square dropped from 36.0 using ACR 20 to 0.8 using ACR 70. Using ACR 70, even if one combined all 3 trials of second-line drugs, investigators would have been unable to detect an effect of second-line drugs compared with placebo. Similarly, in the MTX versus AUR trial, the chi-square dropped as the threshold for improvement increased. The samc was true in the MTXiCSA combination trial and in the TNFR:Fc trial. Because of the high ACR 50 response rate among the high-dose group in the TNFR:Fc trial, the fall in chi-square was not as great as in the other trials. When we testcd ACR 20,50, and 70 using Fisher s exact test (the test that would be used to analyze trials with small numbers of improved patients), the results were the same for each trial in that ACR 20 had thc highest discriminant validity and ACR 70 the lowest. ACR 50 results were either generally equivalent to ACR 20 results for some trials (both P < 0.001) or were less significant. We then tested an ordinal definition of improvement. Even though this provided more information about how patients did through the trial, it did not discriminate between active treatment and either weak or placebo treatment better than did the simpler ACR 20 definition (see Table 2). For example, in the CSSRD trials, the ACR 20 dcfinition yielded a chi-square of 36.0, versus a chi-square of 36.8 when an ordinal definition was used. In the AUR versus MTX trial, the chi-square of 30.6 fell to 23.1 when an ordinal definition was used. In the AUR versus placebo trial, the chi-square rose slightly when the ordinal definition was used. In the other trials, the difference in the chi-square was not substantial. We projected sample size requirements taking into account both thc placebo response rate and the difference between treatments in response rates (a rcsponse rate equals the percentage experiencing improvement) (see Figure 1). The lowest placebo response rate seen was higher than 0%, and rates of response up to 30% have been seen. Given the same absolute difference in treatment response, a lower placebo response rate would theoretically provide more power (i.e., discriminant validity). This suggests that raising the bar for defining improvement by decreasing the placebo response rate would increase statistical power. However, when the threshold for improvement is raised, the

5 1568 FELSON ET AL S E 100 aa.z 80 v) aa 60 I 40 G t Difference in Response Rate Between Groups Placebo Rate x * Figure 1. Sample sizes required, in each arm of a 2-arm trial with equal allocation of subjects, to reach 80% power to reject the null hypothesis of no difference in response rate. difference in treatment response falls more than does the placebo response rate, thus compromising power. An example might be CSSRD trials in which ACR 20 yields a placebo response rate between 5% and 10% and a difference in response rates of 31.9%. Roughly 30 patients would be needed in each treatment group to demonstrate this difference. For ACR 50 with 0% placebo response but a responder difference of 8%, patients would be needed in each treatment group. DISCUSSION The results of this study confirm the discriminant validity of ACR 20 and suggest that use of ACR 50 or ACR 70 as the primary definition of response in an RA trial would compromise power and necessitate a larger sample size. It should be noted that ACR 20 was found to have greater discriminant validity than ACR 50 or ACR 70 in every trial we reviewed, including a recent trial of a new biologic agent. While our results were generated in part by reanalysis of data used to develop the ACR definitions of improvement, neither of these higher thresholds was ever tested during that process. Paulus et a1 (6) did evaluate higher thresholds for improvement and chose the 20% threshold (for most outcomes). Also, in the present study of different thresholds for improvement, we analyzed other new data sets and found the same trend. In each case, when the threshold was raised, the response rate dropped more in the active treatment group than in the placebo group. Thus, raising the bar for defining improvement would likely compromise power. Why does ACR 20 work better than ACR 50 or 70? Placebo response rates do indeed drop when one raises the threshold for improvement, and this is certainly appealing. Unfortunately, response rates for active treatment drop even more, leaving the difference between active treatment response rates and placebo (or weak treatment) response rates diminished. Thus, discriminant validity falls. While development of a definition of major improvement in RA is intuitively appealing, it should be remembered that the current preliminary definition was developed not just with discriminant validity in mind. Using paper patients, the ACR committee tried to determine what clinicians who treat RA patients would characterize as clinical improvement. Based on a large number of scenarios and surveyed respondents, the committee drew this improvement line at roughly a 20% consistent improvement across outcome measures. The ACR definition as currently constituted requires at least

6 THRESHOLD FOR IMPROVEMENT IN RA CLINICAL TRIALS 1569 a 20% improvement in a series of parameters. Consistency of improvement across parameters in RA trials is relatively unusual, especially among placebo-treated patients. Most patients who have at least 20% improvement in all of these parameters actually experience much greater improvement in most of them. An alternative to using ACR 50 or ACR 70 as primary definitions of efficacy in RA trials would be to measure them as secondary measures of efficacy and include these results in the report of the trial. This would provide information to readers and to clinicians involved in the study regarding the number of patients who experienced higher levels of response. One of the main advantage5 of ACR 20 is to promote standardization of measurement across trials and we recommend, based on its cfficiency as documented here, that ACR 20 continue to be the primary outcome measure used in RA trials and that ACR 50 recults might also be consistently reported acroys trials so that rates of improvement can be compared. Another alternative might be to use the ordinal measure of improvement tested here, which appears to have discriminant validity comparable with that of ACR 20. While not as simple as ACR 20, the ordinal outcome, because it provides more information on the distribution of improvement, is more informative. To permit comparison of results across trials, if an ordinal measure were used, we suggest that authors report on the percentage of patients in each treatment group achieving each response threshold. We discourage readers from using the data in Table 2 to make comparisons of the efficacy of treatments. Rates of improvement in the trials reported here cannot be readily compared. For example, in the CSSRD trial of MTX, the response rates were similar to the response rates reported in Table 2 for all second-line drugs, whereas in the comparative trial of MTX versus AUR, response rates in the MTX-treated group were much higher. Thus, although treatments studied here were similar, response rates vary considerably across trials, suggesting that other factors, such as patient characteristics, may have substantial effecl5 on the response rates. Such comparisons should adjust for differences in patient characteristics and other factors that may affect response. Some have argued for a definition of response that cannot detect modest treatment efficacy. Such definitions (e.g., ACR 50) havc less discriminant validity than ACR 20, and their use would make it difficult to compare the efficacy of 2 active treatments and would require trials with larger sample sizes to document the efficacy of treatments. We suggest that an instrument with high discriminant validity provides more accurate information about the differences between treatments in almost all circumstances and should be the optimal choice. In summary, the definition of RA improvement proposed by the American College of Rheumatology, using a threshold of at least 20% improvement in a variety of core set parameters, has better discriminant validity than definitions of improvement that use higher thresholds, such as 50% or 70%1. We recommend that ACR 20 continue to be used as the primary measure of efficacy in RA trials. REFERENCES 1. Felson DT, Anderson JJ, Boers M, Bombardier C, Furst D, Goldsmith C, et al. American College of Rheumatology preliminary definition of improvement in rheumatoid arthritis. Arthritis Rheum 1995;38: Felson DT, Anderson JJ, Boers M, Bombardier C, Chernoff M, Fried B, et al. The American College of Rheumatology preliminary core set of discasc activity measures for rheumatoid arthritis clinical trials. Arthritis Rheum 1993;36: Pillemer SK, Fowler SE, lilley BC, Alarcon GS, Heyse SP, Trentham DE, et al. Meaningful improvement criteria sets in a rheumatoid arthritis clinical trial. Arthritis Rhcum 1997;40: Moreland LW, Baumgartner SW, Schiff MH, Tindall EA, Fleischmann RM, Weaver AL, et al. Treatment of rheumatoid arthritis with a recombinant human tumor necrosis factor receptor (p75)-fc fusion protein. N Engl J Med 1997;337: Van den Borne BEEM, Landewe RBM, Goei The HS, Breedveld FC, Dijkmans BAC. Low dose cyclosporine in early rheumatoid arthritis: effective and safe after two years of therapy when compared with chloroquine. Scand J Rheumatol 1996;25: Paulus HE, Egger MJ, Ward JR, Williams HJ, and the Cooperative Systematic Studies of Rheumatic Diseases Group. Analysis of improvement in individual rheumatoid arthritis patients treated with disease-modifying antirheumatic drugs, based on the findings in patients treated with placebo. Arthritis Rheum 1990;33: O'Dell JR, Haire CE, Palmer W, Drymalski W, Wees S, Blakely K, et al. Treatment of early rheumatoid arthritis with minocycline or placebo: results of a randomized, double-blind, placebo-controlled trial. Arthritis Rheum 1997;40: Ferracciolo GF, Di Poi E, Damato R, Casatta L, De Vita S, Bartoli E. Analysis of remission and full responses in rheumatoid patients treated with combination therapy for three years [abstract]. Arthritis Rheum 1997;40 Suppl 9:S Bombardier C, Tugwell P. A methodological framework to develop and select indices for clinical tasks: statistical and judgmental approaches. J Rheumatol 1982;9: Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31: LVilliams HJ, Willkens RF, Samuelson CO Jr, March GS, Guttadauria M, Yarboro C, et al. Comparison of low-dose oral pulse methotrexate and placebo in the treatment of rheumatoid arthria controlled clinical trial. Arthritis Rheum 1985;28: lliams HJ, Ward JR, Reading JC, Egger MJ, Grandone JT,

7 1570 FELSON ET AL Samuelson CO, et al. Low-dose D-penicillamine therapy in rheumatoid arthritis: a controlled double-blind clinical trial. Arthritis Rheum 1983;26: Ward JR, Williams HJ; Egger MJ, Reading JC, Boyce E, Altz- Smith M, et al. Comparison of auranofin, gold sodium thiomalate, and placebo in the treatment of rheumatoid arthritis: a controlled clinical trial. Arthritis Rheum 1483;26: , 14. Wcinblatt ME, Kaplan H, Germain BF, Merriman RC, Solomon SD, Wall B, et al. Low-dose methotrexate compared with auranofin in adult rheumatoid arthritis: a thirty-six-week, double-blind trial. Arthritis Rheum 1990;33: Tugwell P, Pincus T, Yocum D, Stein M, Gluck 0, Kraag G, et al. Combination therapy with cyclosporine and methotrexate in severe rheumatoid arthritis. N Engl J Med 1945;333: Felson DT, Anderson JJ, Meenan KF. Thc comparative efficacy and toxicity of second-line drugs in rheumatoid arthritis: results of two metaanalyses..&thritis Rheum 1990;33: Felson DT, Anderson JJ, Meenan RF. Use of short-term efficacy/ toxicity tradeoffs to select second-line drugs in rheumatoid arthritis: a metaanalysis of published clinical trials. Arthritis Rheum 1992;35: Cohen J. Thc arcsine transformation and the effect size index. In: Statistical power analysis for the behavioral science. 2nd ed. Hillsdale (NJ): Lawrence Erlbaum Associates; p