SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Objective

To study the distribution of clinical responses to treatment with the tumor necrosis factor α (TNFα) antagonists etanercept and infliximab, and in particular, to determine whether there is a biologically meaningful distinction between responders and nonresponders.

Methods

Among patients in the Stockholm TNFα Followup Registry, we analyzed the clinical responses to etanercept and infliximab, using the American College of Rheumatology (ACR) core set of outcome measures. For each parameter, the absolute change (value at baseline − current value) and the percentage change ([absolute change]/[value at baseline] × 100) from baseline were calculated. The results were plotted as histograms and inspected visually, and the distributions were statistically compared with computer-generated normal distributions.

Results

Absolute and relative changes in outcomes on the ACR core set of measures in 406 patients receiving etanercept or infliximab were studied. All but a few of these analyses yielded normal or somewhat skewed distributions. The statistical analyses did not detect any non-normal distributions, and visually, the distributions did not appear to be bimodal.

Conclusion

The clinical response to TNFα blockade displays a normal or skewed, but not bimodal, distribution. The frequently encountered perception that a clear distinction can be made between responders and nonresponders is not borne out. These relatively straightforward findings imply that the biologic mechanisms determining responsiveness to TNFα blockade are multifactorial and may also have important implications for regulatory guidelines pertaining to treatment with these biologic agents.

Clinical trial results are frequently presented as dichotomies, most notably, the 20% improvement in disease activity according to the American College of Rheumatology criteria (ACR20 response) (1) and the European League Against Rheumatism response (2) based on either the Disease Activity Score (DAS) (3) or the DAS in 28 swollen and 28 tender joints (DAS28) (4), separating patients into responders and nonresponders according to those criteria. This manner of presentation may suggest that the responses to treatments are, in fact, distributed bimodally and that a sharp and biologically meaningful distinction between “responders” and “nonresponders” can be made. Therefore, we wished to determine whether clinical responses to tumor necrosis factor α (TNFα)–blocking therapies are bimodally distributed, using the Stockholm TNFα Followup Registry (STURE).

PATIENTS AND METHODS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

The STURE database collects efficacy and safety data for all patients starting biologic treatments at the major hospitals in Stockholm, as part of the nationwide registry of biologic agents, called Anti-Rheumatic Therapies in Sweden (ARTIS). The assessments are done at 0, 3, 6, and 12 months, and annually thereafter, and include the ACR core set of outcome measures (5) (modified to include the 28 swollen and 28 tender joint count [4], visual analog scale [VAS] assessments for global health and for pain, the disability index of the Health Assessment Questionnaire [HAQ] [6], the erythrocyte sedimentation rate and C-reactive protein level [considered 1 ACR core measure, the acute-phase reactant value], and, in a slight deviation from the ACR core set of measures, a 5-point Likert scale for physician's assessment of global disease activity), the DAS28, a record of concurrent medications, employment status, and treatment side effects. While the STURE database is part of the ARTIS national biologicals safety registry, this study was performed using only data collected and analyzed at the Karolinska Hospital.

We used data on 406 patients in the STURE database, 124 of whom were treated with etanercept and 282 with infliximab. We analyzed the distribution of absolute and relative (percentage) changes in outcomes on the ACR core set of measures by creating histograms of these parameters. The histograms were inspected visually, and normal-distribution curves matching the data were computer generated. In uncertain instances, the following test for normality was used: a computer-generated, normally distributed set of values was created with the same mean and SD as the outcome of interest. This outcome was then compared with the matching set of normally distributed values by a Kolmogorov-Smirnov test run on StatView 5.0.1 software (SAS Institute, Cary, NC).

RESULTS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

In this patient population, 64% of patients achieved an ACR20 response, 33% achieved an ACR50 response (7), and 8% achieved an ACR70 response. The absolute changes from baseline for 5 of the 7 ACR core outcomes as well as the DAS28 exhibited normal-appearing distributions; that is, visual inspection revealed that the distributions followed the computer-generated normal distribution (examples are shown in Figure 1). Changes in the VAS scores for pain and global health were the only outcomes with a slight appearance of bimodality (Figure 2). However, a formal test for normality did not show the distribution to be non-normal (P > 0.50). Additional analysis assuming the existence of 2 peaks in this outcome did not yield better fits to the normal distribution for each of the “peaks” compared with the whole. Relative (percentage) changes in outcomes on the ACR core set of measures exhibited distributions that appeared normal (e.g., for the HAQ score) (Figure 3A) or skewed but unimodal (e.g., for the swollen joint count) (Figure 3B). Again, the statistical test for normality did not demonstrate that any distributions were significantly different from the normal distribution. Additional analyses were performed for all absolute and relative changes in outcomes after 6 or 12 months and separately for all outcomes according to TNFα antagonist used, diagnosis, and sex. No different results for any of these were seen (data not shown).

thumbnail image

Figure 1. Normal distribution of absolute changes in outcomes on the American College of Rheumatology (ACR) core set of measures. The histograms show the distribution of absolute changes in A, ACR28 swollen joint count, B, ACR28 tender joint count, C, erythrocyte sedimentation rate (ESR), and D, Health Assessment Questionnaire (HAQ) score, each calculated as the value at baseline minus the value at 3 months. The normal curves are computer generated. Normally distributed histograms are seen, without any suggestion of bimodality.

Download figure to PowerPoint

thumbnail image

Figure 2. Absolute changes in pain and global health by visual analog scale (VAS). The histograms show the absolute changes in A, pain and B, global health by VAS, both calculated as the value at baseline minus the value at 3 months. The normal curves are computer generated. Broad histograms are seen, with a possible slight suggestion of bimodality. In a statistical analysis for deviation from a matching set of normal values by Kolmogorov-Smirnov test, P values were greater than 0.50 for each outcome. A similar analysis after dividing the data into 2 groups, with a VAS score of 25 as the cutoff, revealed similar statistical nonsignificance, that is, treating the data as if they were derived from 2 sets of patients (“responders” and “nonresponders”) did not result in better “fits” to the matching, normally distributed sets.

Download figure to PowerPoint

thumbnail image

Figure 3. Normal or skewed, but not bimodal, relative (percentage) changes in outcomes on the ACR core set of measures. The histograms show the distribution of the numeric values for relative changes in A, the HAQ score and B, the ACR28 swollen joint count, each calculated as follows: (value at baseline – value at 3 months)/(value at baseline) × 100. The normal curves are computer generated. A, Normal-appearing histogram, without any suggestion of bimodality. B, Skewed histogram, without any suggestion of bimodality. See Figure 1 for definitions.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

We wished to determine whether there was validity to the frequently encountered perception that in the case of TNFα blockers, a sharp distinction between responders and nonresponders could be made. To do this, we examined the distribution of absolute and relative changes in the ACR core set of clinical responses and the DAS28 during TNFα-blocking treatment. In our analyses, no such distribution was shown to deviate significantly from the normal distribution. However, just as commonly used statistical tests for comparisons of 2 groups can provide statistical significance to the effect that the groups are different, but cannot with certainty demonstrate that they are similar, so could it be argued that the analyses used here could not truly exclude the possibility that a distribution was, in fact, not normal (a Type II error). While this is in and of itself correct, we nonetheless believe that our data show that within the range of biologically meaningful distinctions, the distributions of these responses can be regarded as unimodal and very nearly normal.

We conclude that clinical outcomes with TNFα-blocking therapies do not show bimodal distributions. Thus, the frequently encountered perception that a clear distinction can be made between responders and nonresponders is not borne out in this analysis. The theoretical basis for the normal (Gaussian) distribution is an infinite regression of dichotomies. However, in biologic systems, variables determined by ≥6 dichotomies can often be considered “normally” distributed. Therefore, our results are consistent with the view that responsiveness to TNFα-blocking therapy is determined by multiple factors rather than by a single factor. This puts significant minimum requirements on the number of patients that are needed in studies to identify such factors, for instance, genetic markers for responsiveness. These findings also have consequences for the development of clinical guidance documents or regulatory requirements that would place specific “cutoff” values on clinical responses in order to determine eligibility for continued treatment with such agents.

REFERENCES

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. 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: 72735.
  • 2
    Van Gestel AM, Prevoo MLL, van 't Hof MA, van Rijswijk MH, van de Putte LBA, van Riel PLCM. Development and validation of the European League Against Rheumatism response criteria for rheumatoid arthritis: comparison with the preliminary American College of Rheumatology and the World Health Organization/International League Against Rheumatism criteria. Arthritis Rheum 1996; 39: 3440.
  • 3
    Van der Heijde DMFM, van 't Hof MA, van Riel PLCM, van de Putte LBA. Development of a disease activity score based on judgment in clinical practice by rheumatologists. J Rheumatol 1993; 20: 57981.
  • 4
    Prevoo MLL, van 't Hof MA, Kuper HH, van Leeuwen MA, van de Putte LBA, van Riel PLCM. Modified disease activity scores that include twenty-eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995; 38: 448.
  • 5
    Felson DT, Anderson JJ, Boers M, Bombardier C, Chernoff M, Fried B, et al. The American College of Rheumatology preliminary core set of disease activity measures for rheumatoid arthritis clinical trials. Arthritis Rheum 1993; 36: 72940.
  • 6
    Fries JF, Spitz P, Kraines RG, Holman HR. Measurement of patient outcome in arthritis. Arthritis Rheum 1980; 23: 13745.
  • 7
    Felson DT, Anderson JJ, Lange MLM, Wells G, LaValley MP. Should improvement in rheumatoid arthritis clinical trials be defined as fifty percent or seventy percent improvement in core set measures rather than twenty percent? Arthritis Rheum 1998; 41: 156470.