Abstract
Objective To compare the Clinical Disease Activity Index (CDAI) with the Routine Assessment of Patient Index Data 3 (RAPID3) from 2 large United States registries.
Methods Using a cross section of clinic visits within 2 registries, we determined whether the outcome of each metric would place the patient in remission (REM), low (LDA), moderate (MDA), or high disease activity (HDA) using the CDAI, with the assumption that a patient in MDA or HDA would be a candidate for acceleration of treatment.
Results We identified significant disparities between the 2 indices in final disease categorization using each index system. For patients identified in LDA by CDAI, RAPID3 identified 20.4% and 28.3% as LDA in Corrona and the Brigham and Women’s Rheumatoid Arthritis Sequential Study (BRASS), respectively. For patients identified as MDA by CDAI, RAPID3 identified 36.2% and 31.1% as MDA in Corrona and BRASS, respectively, with the greatest disparities within each system identified for LDA and MDA activity by the CDAI (20.4% and 36.2% agreement of RAPID3 with CDAI, respectively, in Corrona and 28.3% and 31.1% agreement in BRASS). Overall comparison between CDAI and RAPID3 in the 4 disease categories resulted in estimated κ = 0.285 in both. The RAPID3 scores indicated the potential for treat-to-target acceleration in 34.4% of patients in REM or LDA based on CDAI in Corrona and 27.7% in BRASS, respectively.
Conclusion The RAPID3, based on patient-reported outcomes, shows differences with CDAI categories of disease activity. The components of CDAI are not highly correlated with RAPID3, except for patient global assessment. These differences could significantly affect the decision to advance treatment when using a treat-to-target regimen.
Metrics are essential for evaluating disease activity in patients with rheumatoid arthritis (RA). Given that rheumatologists have accepted the widely held approach of “treating to target,”1,2 it is apparent that measures accurately reflecting disease activity in RA are critical for management decisions regarding maintaining, changing, or adding treatment regimens in order to achieve the desired target of low disease activity (LDA) or remission.1,2
The Disease Activity Score in 28 joints based on erythrocyte sedimentation rate (DAS28-ESR) and DAS28 based on C-reactive protein (DAS28-CRP) are accepted as a gold standard, along with the American College of Rheumatology (ACR) 20, 50, and 70 responses.3 However, it is widely recognized that these measures required for regulatory approval are not widely utilized in routine clinical practice in the United States given their complexity and the diligence needed to perform them in the course of routine care. In addition, the acute-phase reactant is not available at the time of the clinic visit.
The Clinical Disease Activity Index (CDAI) is another validated metric that can actually be calculated at the same time as a clinical encounter as it is the simple sum of the number of tender and swollen joints along with the numerical value of both a patient and physician global activity on a visual analog scale.4,5 Another metric that has become widely used is the Routine Assessment of Patient Index Data 3 (RAPID3),6,7which employs values obtained directly from the patient without physician input.
However, several publications have questioned the contribution of the patient-derived pain and global activity,8–18 with the latter measure found in both metrics. The RAPID3 will of course derive its final score from patient evaluations as this metric is not associated with either a physician examination or an acute-phase reactant, but only a patient pain and global arthritis activity score along with a Health Assessment Questionnaire. The CDAI is derived from a simple summary of a 28-joint tender (TJC) and swollen joint count (SJC) along with a physician (PGA) and patient global assessment (PtGA).
It is relevant that a significant disparity has been found between patient assessment of pain and global arthritis activity and other measures of disease activity.8–17 Several publications have found that factors other than disease activity affect these core components of the RAPID3 metric. These elements include depression, life satisfaction, and anxiety that may be ongoing and independent of control of inflammation.18,19,20,21,22 It has been suggested that patient pain and global disease activity scores measure different domains of patient welfare.23–30
We thus sought to compare the CDAI and RAPID3 scores on patients along with analyses of the individual components of each metric to better understand their contribution to the final score within each system. We used 2 different registries in the US, including one from primarily geographically diverse community rheumatologists and one from an academic practice collecting these prospective metrics at the time of a clinical encounter. We focused on how the different measures would inform treatment decisions if they were used in everyday clinical practice while also determining which components of each final metric are potentially problematic for making decisions to treat ongoing inflammation.
METHODS
All participating investigators were required to obtain full board approval for conducting research involving human subjects. Sponsor approval and continuing review was obtained through a central institutional review board (IRB; New England Independent Review Board, NEIRB No. 120160610). For academic investigative sites that did not receive a waiver to use the central IRB, full board approval was obtained from the respective governing IRBs and documentation of approval was submitted to the sponsor prior to initiating any study procedures. All registry subjects were required to provide written informed consent prior to participating.
Data sources. The registry, known as the Brigham and Women’s Rheumatoid Arthritis Sequential Study (BRASS), began recruitment in 2003. Subjects older than 18 years of age were recruited from the arthritis center practices of attending rheumatologists and fellows. All diagnoses of RA were either verified according to the 1987 ACR criteria by a rheumatologist or met the rheumatologists’ criteria for a diagnosis of RA. Subjects were evaluated by their rheumatologists yearly, where information about demographics, medication use, disease activity, functional status, and comorbidities were obtained. Currently, 1343 subjects enrolled in the study have baseline data. Details related to the participation and protocol in the BRASS Registry have been reported elsewhere.31
The Corrona RA registry, founded in 2000, includes a network of > 700 participating academic and community rheumatologists at over 180 sites in > 40 states within the US. All patients with a diagnosis of RA treated by participating rheumatologists are eligible to be included in the Corrona registry. Data are gathered at clinical visits from patient and provider forms, as has previously been described.32
Data from the most recent visits of 48,255 patients were extracted from 184 Corrona registry sites (83% private and 17% academic) in 42 states to perform a cross-sectional analysis. The visit dates ranged from October 2, 2001, to August 30, 2019.
The Corrona RAPID3 (cRAPID3) is a modified RAPID3 scale used for this study.It does not include the following 2 items that are not collected by the Corrona registry: (1) Are you able to “walk 2 miles or 3 km, if you wish”; and (2) Are you able to “participate in recreational activities and sports as you would like, if you wish?” The scoring of the remaining 8 items (“Over the last week, were you able to…”) of the RAPID3 were weighted to ensure the cRAPID3 score was on the same scale as the RAPID3 (0–10). The patient pain and patient global health components of the RAPID3 were not reweighted, and the cRAPID3 score ranged from 0 to 30. The cRAPID3 was also computed in the BRASS registry for comparative purposes.
The rate of moderate disease activity (MDA)/high disease activity (HDA) by cRAPID3 for those in CDAI remission (REM)/LDA were compared between biologic-experienced and -naïve using a logistic regression model and adjusted for duration of disease (early ≤ 2 yrs) vs late (> 2 yrs) at time of the measurement.
Disease activity categories. RA disease activity was measured using CDAI and cRAPID3 in Corrona. A CDAI of ≤ 2.8 was classified as REM, > 2.8 but ≤ 10 was classified as LDA, > 10 but ≤ 22 was classified as MDA, and > 22 was classified as HDA. Similarly, a RAPID3 and cRAPID3 of ≤ 3 was classified as REM, > 3 but ≤ 6 was LDA, > 6 but ≤ 12 was MDA, and > 12 was HDA.
Statistical methods. Spearman correlation coefficients (no assumptions of normality) were calculated to assess the correlation between CDAI, cRAPID3, and RAPID3, as well as individual measurements of clinical characteristics and disease activity levels. κ statistics were calculated to determine the relationship between CDAI and cRAPID3 classifications (REM, LDA, MDA, and HDA). Additionally, κ statistics were also calculated to determine the relationship between the potential treat-to-target groups (REM/LDA vs MDA/HDA) measured by CDAI and cRAPID3.
RESULTS
There are 48,255 Corrona RA patients with CDAI and cRAPID3 measures, and 1343 BRASS RA patients with CDAI and RAPID3 measures. Patient demographics and clinical disease measures are shown in Table 1. BRASS patients with RA are younger (55.9 yrs vs 61.7 yrs) and have higher mean CDAI (19.5 vs 11.2), but slightly lower cRAPID3 (7.3 vs 8.2). PGAs are higher in BRASS (30.6 vs 19.3), but PtGAs are lower (31.7 vs 32.6).
Patient characteristics at index date in each registry.
We found that the RAPID3 and cRAPID3 have a high correlation (rs = 0.998) and overall agreement across disease activity categories is 94% (Supplementary Table 1, available with the online version of this article), as measured by a κ statistic that accounts for chance agreement (κ = 0.92).
Comparison of CDAI and cRAPID3 disease categories in both registries is illustrated in Figure 1 and Table 2 with κ = 0.24 for both Corrona and BRASS. Both registries show a high proportion of cRAPID3 REM within CDAI remission group (78% Corrona, 87% BRASS) but more disagreement at other disease activity levels. For example, only 32% (Corrona) and 41% (BRASS) of CDAI MDA are classified as MDA by cRAPID3. As seen in Figure 1, Table 2, and Table 3, results are similar for comparison of CDAI and RAPID3 in BRASS.
Distribution of cRAPID3 categories within each CDAI category for the Corrona registry. Darker section shows area of agreement. For example, for CDAI remission, 78.4% cRAPID3 also indicated remission. cRAPID3 is RAPID3 computed within Corrona registry without 2 of the HAQ questions, rescaled to 0–30. CDAI: Clinical Disease Activity Index; cRAPID3: Corrona RAPID3; HAQ: Health Assessment Questionnaire; HDA: high disease activity; LDA: low disease activity; MDA: moderate disease activity; RAPID3: Routine Assessment of Patient Index Data 3.
Agreement of disease activity levels of cRAPID3 and CDAI from Corrona and BRASS registries.
Comparison of potential treat-to-target indication in Corrona and BRASS. CDAI and cRAPID3 disease categories dichotomized into REM/LDA and MDA/HDA.
Collapsing categories combining MDA and HDA for indication of potential treatment acceleration and REM and LDA for indication of no potential treatment acceleration agreement of CDAI and cRAPID3 are shown in Table 3 (κ = 0.49 Corrona and 0.39 BRASS). Both registries show close to one-third (34% Corrona and 28% BRASS) cRAPID3 indicating potential acceleration, whereas CDAI indicates REM/LDA.
The correlation of CDAI, cRAPID3, and RAPID3 along with components of the measures are shown in Table 4 for Corrona and BRASS. CDAI and cRAPID3 have estimated Spearman correlation of 0.72 (Corrona) and 0.58 (BRASS). As would be expected given the lesser contribution of this metric to the calculation of the CDAI, PtGA has a higher correlation with RAPID3 and cRAPID3 (0.94 Corrona; 0.91 BRASS for both cRAPID3 and RAPID3) than CDAI with PGA (0.78 Corrona, 0.58 BRASS cRAPID3, 0.57 BRASS RAPID3). PGA, not found in the RAPID3, correlated more highly with CDAI (0.78 Corrona, 0.81 BRASS) than cRAPID3 (0.47 Corrona, 0.56 BRASS) and RAPID3 (0.57 BRASS). TJC and SJC, not found in the RAPID3, correlated more highly with CDAI (TJC 0.80, SJC 0.74 Corrona; TJC 0.90, SJC 0.85 BRASS) than cRAPID3 or RAPID3 (TJC 0.48, SJC 0.31 Corrona cRAPID3; TJC 0.44, SJC 0.29 BRASS cRAPID3; TJC 0.44, SJC 0.30 BRASS RAPID3). However, patient pain, not found in the CDAI but present in the RAPID3 metrics, correlated more highly with cRAPID3 and RAPID3 (0.94 Corrona cRAPID3; 0.92 BRASS for both cRAPID3 and RAPID3) than CDAI (0.67 Corrona, 0.50 BRASS).
Spearman rank correlations of CDAI, cRAPID3a, and components within the Corrona and BRASS registries. P values testing correlations equal to zero are all < 0.001.
The comparison of CDAI to cRAPID3 was examined by disease duration and by biologic-naïve vs -experienced (data not shown). Rates of cRAPID3 in MDA/HDA among those with CDAI remission/LDA were slightly higher in the biologic-experienced vs -naïve group (38% vs 30% Corrona, 30% vs 28% BRASS, respectively). Adjusted for duration of disease, the difference by biologic experience was significant (P < 0.001 Corrona, P = 0.046 BRASS).
DISCUSSION
We found poor correlations in both registries between RAPID3 (or cRAPID3) and CDAI scores for patients in CDAI LDA and MDA, whereas the congruence for both between the metric final scores of REM and HDA were better. When considered together, these data indicate that patient-derived measures contribute differently to the metrics dominating the RAPID3, while SJC, TJC, and PGA dominate the CDAI (rs = 0.74, 0.80, and 0.78, respectively, in Corrona, and 0.85, 0.90, and 0.81 in BRASS, respectively; Table 4).
While it might at first appear to be somewhat circular reasoning, or an obvious predetermined outcome, to derive correlations of components found selectively within only one of the 2 metrics with final scores of both metrics, we believe that it serves a larger purpose. Both the CDAI and the RAPID3 are presently being used to inform treatment decisions. If patients are to be managed based upon the results of these scores, then an understanding of what the components represent, and their relative contribution to the final score, is relevant.
Our approach in comparing the CDAI with the RAPID3 (or cRAPID3) was to compare the different final metrics to each other. We believed that it was appropriate to identify the CDAI as the acceptable core clinical outcome metric, as has previously been established.4,5 Because we had simultaneous outcome metrics from both registries obtained at the same visit from all patients, there was an unusual juxtaposition of circumstances to facilitate comparison of the real-world performance of the RAPID3 to the CDAI.
A rich recent literature has described the psychological and life satisfaction factors that contribute so strongly to patient-derived measures of the RAPID3, including the patient global and pain scores.9–23 It is apparent that the RAPID3 can provide a clinician with potentially complementary information on the welfare of their patient. But, as has been demonstrated in these recent publications, because the RAPID3 is dominated by variables with strong psychological and psychosocial derivations, the scores should not be used in isolation to either evaluate ongoing RA inflammation or to adjust treatment. Ferreira and colleagues have suggested a “dual-target approach” in which psychological outcomes are derived from patients, whereas inflammation-related variables are derived from physician joint counts as found in the CDAI.33
We believe that it is thus important to recognize that even though certain key variables such as tender and swollen joints on examination as well as a physician global evaluation are not found in the RAPID3, their absence should not release this metric from the burden of achieving the goal of reflecting inflammatory disease activity as captured in a validated metric such as the CDAI. As several authors have described, these patient-reported outcomes reflect different domains of patient welfare.23–30 Additional compelling evidence from a Dutch society published by Boone and colleagues has described similar disparities between the RAPID3 and both the CDAI34 and the DAS28.35
The high correlation and agreement of RAPID3 and cRAPID3 provides evidence that the cRAPID3 used in Corrona is a good proxy for RAPID3.We thus employed the cRAPID3 in Corrona to derive correlations with the CDAI after confirming the very high correlation with the full RAPID3 in BRASS (Supplementary Table 1, available with the online version of this article). The metrics from the Corrona registry were collected from 706 rheumatologists at 184 sites in 42 different states with a distribution of private vs academic sites (83% vs 17%). The results for each metric were compared both within and across registries. The BRASS registry represents data that are entirely derived from a single academic center. We found that there were differences in the evaluation of disease activity between registries. The PtGAs are quite similar whereas the PGAs differ (19.26 [SD 20.40] Corrona, and 30.28 [SD 21.79] BRASS), as do the CDAI scores (11.24 [SD 11.50] Corrona, and 19.47 [SD 16.45] BRASS). The reasons for the differences are speculative but may reflect either the likelihood that a tertiary academic center might attract more challenging patients or the fewer numbers of evaluators in BRASS. We did not have an a priori expectation that the disease activity would be overwhelmingly similar in both Corrona and BRASS given the differences in the makeup of the registry sites, and we believed that this was a virtue when comparing the CDAI with the RAPID3. Nevertheless, the reason(s) for the differences in the PGA remain speculative. The differences in these ratings may indeed be site-specific and reflect a shared approach for disease assessment of providers at a single site. Differences in PGA might be a topic for future research.
While the RAPID3 is convenient and saves physicians’ time compared with a 28-joint count and PGA, it is apparent that the outcome scores are frequently divergent from the CDAI. In addition, it has previously been published that the same RAPID3 can be used to reflect osteoarthritis (OA) disease activity.36 Given that patients with RA often have concomitant OA, it would not be possible to determine the contribution of OA to the patient’s rating of pain and global arthritis activity when the same metric is used for RA.
This study has several strengths. We studied a very large number of patients and compared the CDAI and cRAPID3 measures obtained at the time of the same clinic visit. We compared data across 2 different registries, including one from an academic health science center (BRASS) and one from predominantly private practitioners (Corrona). In addition, we were able to examine the correlation of the patient global and pain measures with each metric and confirm the discordance with an array of clinical disease measures obtained by a physician. For what we believe is the first time, we extend earlier findings on patient pain and global evaluations as they apply to a commonly used patient-derived metric, the RAPID3, in comparison with a CDAI.
There are also some potential weaknesses. Physician TJCs and SJCs were typically performed by the same clinician but not always. The Corrona registry consists of multiple sites with inevitable variability across sites, whereas BRASS is a single academic site with fewer evaluators.It is possible that differences in results between the registries reflect these differences in site composition. These analyses are derived from a cross-sectional, retrospective review of prospectively collected observational data. Nevertheless, the inclusion of the possibility of different investigators evaluating the same patient might actually add to the representativeness of the observations across different physicians, thus hypothetically buttressing the external validity of the data. We believe that the heterogeneity of the data sources and very large number of clinical evaluations from 2 different registries serve to support the clinical conclusion. It is also possible that these US-based findings may not be representative of other societies, although we believe that this is not likely as much of the data we cite on the discordance of the PtGA with actual disease activity are derived from European authors.9–12,14–17,19,20–23
In conclusion, the RAPID3 should not be used as an exclusive measure to evaluate clinical status and inform treatment decisions, as the individual components of this metric are highly associated with noninflammatory conditions such as depression, anxiety, and quality of life19,20,21,22,23 and are discordant with CDAI evaluations. Nevertheless, the RAPID3 can add valuable information on patient psychometrics, which complement the CDAI outcomes that are more reflective of inflammatory outcomes. We believe that it is important for treating clinicians to recognize and acknowledge the core clinical themes that are actually being measured within each metric. A patient who is doing well on a CDAI, but not on a RAPID3, should be further evaluated for the contribution of both psychosocial factors and OA to this score. Further, the RAPID3 should not be relied upon to inform decisions when treating to target.
Footnotes
This study was supported by the Corrona Research Foundation.
The authors declare no conflicts of interest.
- Accepted for publication December 9, 2020.
- Copyright © 2021 by the Journal of Rheumatology
