Gender stratified adjustment of the DAS28-CRP improves inter-score agreement with the DAS28-ESR in rheumatoid arthritis.

OBJECTIVES: To evaluate determinants of discordance between DAS28-ESR and DAS28-CRP and resulting impact on disease activity stratification in RA.
METHODS: Paired DAS28-ESR and DAS28-CRP readings (n = 31 074) were obtained from the British Society for Rheumatology Biologics Register for RA. Factors influencing discordance between DAS28-ESR and DAS28-CRP were evaluated alongside the resulting effect on disease activity stratification. The impact of gender adjustment to the DAS28-CRP was evaluated.
RESULTS: DAS28-CRP scores were ∼0.3 lower than DAS28-ESR overall, with greatest differences for women (-0.35) and patients over 50 years old (-0.34). Mean male DAS28-CRP scores were 0.15 less than corresponding DAS28-ESR scores. Discordance between DAS28-ESR and DAS28-CRP significantly impacted disease activity stratification at low disease activity and remission thresholds (32.0% and 66.6% concordance, respectively). Adjusting DAS28-CRP scores by gender significantly (P < 0.001) improved agreement with the DAS28-ESR.
CONCLUSION: Discordance between DAS28-ESR and DAS28-CRP is greatest for women and patients over 50 years of age, and influences disease activity stratification. The proposed gender-adjusted DAS28-CRP improves inter-score agreement with DAS28-ESR, supporting more reliable disease activity stratification in treat-to-target approaches for RA.

Rheumatology key messages

On average, DAS28-ESR generates scores 0.3 greater than the DAS28-CRP.

The difference between the DAS28-CRP and DAS28-ESR is greater for women and patients over 50.

Adjusting the DAS28-CRP according to gender significantly reduces inter-score differences with the DAS28-ESR.

Introduction

Disease activity in RA is commonly measured using the 28-joint count DAS (DAS28), a composite index incorporating a tender joint count, swollen joint count, patient global assessment and markers of inflammatory response. The original DAS28 was developed and validated using the ESR [1, 2]. Original development of the DAS28-CRP followed assessment of paired samples obtained from a relatively small cohort of 334 patients with subsequent wide adoption in clinical practice and trial settings [3]. ACR and EULAR do not differentiate between the two versions of the score when using disease activity thresholds. This results in the two versions of the score being used interchangeably in clinical practice (and observational studies that rely on data captured in clinical practice), with identical disease activity stratification thresholds adopted in assessment of disease activity, treatment response and treat-to-target approaches. Disparity in DAS28-CRP and DAS28-ESR values could also influence patient management where high-cost drug reimbursement is only permitted if specific disease activity thresholds are reached. In the UK access to biologics is dependent on patients having a DAS28 score of >5.1 on two successive occasions at least 1 month apart [4]. Therefore, discrepancies between scoring methods may delay the availability of biologic DMARDs to patients, with potential negative consequences on quality of life, function and radiographic progression. Many studies have highlighted consistently lower DAS28-CRP compared with DAS28-ESR scores, at both lower levels of disease activity that form the focus of treat-to-target management [5-9] and high disease activity [10]. It is acknowledged that ESR values are generally higher in females and increase with age [11], while the CRP is not affected by these factors. ESR and CRP levels may also be differentially affected by BMI. Understanding the relative impact of gender, age and BMI on inter-score differences may help improve agreement between the DAS28-CRP and DAS28-ESR such that both instruments could be used interchangeably.

We report an analysis from the British Society of Rheumatology Biologics Register for RA (BSRBR-RA) using real-world data, exploring discordance between the DAS28-CRP and DAS28-ESR, and resulting impact on disease activity stratification in RA. We propose that the DAS28-CRP should be adjusted depending on gender.

Methods

Study population

The BSRBR-RA is a national, prospective, longitudinal, observational study examining long-term safety of biologic agents in patients with RA in the UK. Ethical approval was obtained from the Multicentre Research Ethics Committee for the North-West of England. All patients enrolled provided written informed consent.

Subject selection and data collection

The methods of the BSRBR-RA have been described previously [12]. Patients treated with biologic therapy with concurrent measures of ESR and CRP were identified from the BSRBR-RA, enabling paired calculation of DAS28-ESR and DAS28-CRP using existing formulae [13]. Because of NICE requirements, baseline DAS28 scores for participants of the BSRBR-RA are mostly >5.1. Therefore, data obtained at baseline and following treatment with biologic agents were used in the initial cohort analysis. Data from patients taking tocilizumab were excluded due to specific effects of IL-6 on serum CRP levels [14].

Statistical analysis

The impact of age, baseline BMI and gender on concordance between DAS28-ESR and DAS28-CRP was assessed by dichotomizing the group for age (⩾ or <50 years) and gender, and stratifying BMI according to World Health Organization thresholds [15]. This analysis focused on the agreement between two methods of measuring an outcome (i.e. the DAS28-CRP and DAS28-ESR), rather than changes in an outcome with repeated within-subject measurements. Accordingly, repeated measurements were not adjusted for. A random effects model was used to allow for the possibility that ESR and CRP were not measured from the same blood sample (e.g. at the same time). Age at enrolment to the BSRBR-RA was used where it was not possible to calculate the age at time when the DAS28 score was measured.

Agreement between the scores was compared using Bland–Altman statistics. Descriptive analysis was applied to compare disease stratification within accepted DAS28 disease activity thresholds. The cohort was subsequently subdivided according to gender. DAS28-CRP scores were differentially adjusted according to the inter-score differences identified in the initial analysis, and subsequent inter-score differences were compared. Kappa values and root mean squared error quantified agreement and mean error of DAS28-CRP and subsequent adjusted DAS28-CRP scores [16]. The differences in mean errors between the DAS28-ESR and DAS28-CRP or adjusted DAS28-CRP were compared using the Wilcoxon signed rank test.

Results

Subject characteristics

Paired ESR and CRP values were available for 8509 subjects, with 31 074 paired assessments (Supplementary Table S1, available at Rheumatology online). The majority of subjects were female (76%), with mean (s.d.) age of 57.3 (12.2) years and a mean (s.d.) baseline disease duration of 12.7 (9.6) years.

Discordance between DAS28-ESR and DAS28-CRP

The DAS28-CRP was on average 0.3 points lower than the corresponding DAS28-ESR for the whole cohort. When stratifying by age and gender, differences between the two scores were more pronounced for women and patients aged over 50 (0.35 points for both). The mean DAS28-CRP score in males was only 0.15 points less than the corresponding DAS28-ESR score. Mean inter-score differences did not alter when categorized by baseline BMI (Table 1).

1

Comparative mean difference between DAS28-CRP and DAS28-ESR (by Bland–Altman statistics), and effect of gender and age

	Whole cohort			Female subgroup			Male subgroup
Characteristic	DAS28- ESR	DAS28- CRP	Mean difference (95% CI; n)	DAS28- ESR	Adjusted DAS28-CRP	Mean difference (95% CI; n)	DAS28- ESR	Adjusted DAS28-CRP	Mean difference (95% CI; n)
Overall	4.44	4.13	0.30 (0.30, 0.31; 31 074)
Male	4.17	4.02	0.15 (0.13, 0.16; 7380)				4.16	4.17	−0.01 (−0.02, 0.01; 7380)
Female	4.52	4.17	0.35 (0.35, 0.36; 23 694)	4.52	4.52	<0.01 (<−0.01, 0.01; 23 694)
Age
<50 years	4.27	4.09	0.17 (0.16, 0.19; 7786)	4.30	4.41	−0.12 (−0.13, −0.10; 7118)	3.83	3.98	−0.16 (−0.19, −0.14; 1859)
>50 years	4.50	4.15	0.35 (0.34, 0.35; 23 288)	4.62	4.57	0.05 (0.05, 0.06; 16 576)	4.28	4.23	0.05 (0.03, 0.06; 5521)
Underweight (BMI <18.5)	4.51	4.19	0.32 (0.29, 0.35; 1054)	4.59	4.61	−0.02 (-0.05, 0.01; 843)	4.18	4.04	0.14 (0.06, 0.22; 211)
Normal (BMI 18.5–<25)	4.34	4.04	0.30 (0.29, 0.31; 12 348)	4.42	4.42	−0.01 (−0.02, <−0.01; 9903)	4.04	4.04	0.01 (−0.02, 0.03; 2445)
Overweight (BMI 25–<30)	4.43	4.15	0.29 (0.28, 0.30; 9988)	4.33	4.33	0.02 (<0.01, 0.03; 6981)	4.14	4.19	−0.05 (−0.07, −0.03; 3007)
Obese (BMI ≥30)	4.59	4.27	0.33 (0.32, 0.34; 7684)	4.64	4.64	0.02 (0.01, 0.03; 5967)	4.37	4.34	0.04 (0.01, 0.06; 1717)

Mean values are presented. DAS28: 28-joint count DAS ; n: number of paired scores.

Impact of disparity between DAS28-ESR and DAS28-CRP on disease activity stratification

Disparity between the DAS28-ESR and DAS28-CRP had a significant impact on disease stratification, particularly within the low disease activity category where the two scores only agreed in 32.0% of cases (Table 2). The DAS28-ESR classified fewer patients in remission compared with the DAS28-CRP, and more in high disease activity (Supplementary Table S2, available at Rheumatology online).

2

Inter-score agreement between DAS28-ESR and unmodified DAS28-CRP, and gender-adjusted DAS28-CRP scores

	Percentage agreement with DAS28-ESR (n)
	Whole cohort		Female subgroup		Male subgroup
Disease activity	DAS28-CRP	Adjusted DAS28-CRP	DAS28-CRP	Adjusted DAS28-CRP	DAS28-CRP	Adjusted DAS28-CRP
Remission (<2.6)	66.6 (4040)	78.2 (3366)	62.2 (2680)	76.4 (2131)	77.5 (1360)	81.6 (1235)
LDA (2.6–≤3.2)	32.0 (1302)	40.3 (1494)	31.0 (965)	41.5 (1144)	35.1 (337)	36.6 (350)
MDA (3.2–≤5.1)	73.7 (8880)	77.7 (9827)	73.5 (6940)	78.4 (7787)	74.2 (1940)	75.3 (2040)
HDA (>5.1)	96.3 (8560)	93.0 (9684)	96.8 (6614)	92.9 (7626)	94.1 (1946)	93.4 (2058)

DAS28: 28-joint count DAS; HDA: high disease activity; LDA: low disease activity; MDA: moderate disease activity.

Adjusting DAS28-CRP according to gender

Subdividing the cohort by gender and adjusting DAS28-CRP scores by +0.35 for females and +0.15 for men significantly reduced inter-score differences overall (P < 0.001 for females and males), and for age and BMI strata (Table 1). Inter-score disease activity classification improved for remission, low and moderate disease activities, with only a very minor reduction in agreement at high disease activity (3.9% and 0.7% for females and males respectively; Table 2). Root mean squared error reduced from 0.62 to 0.51 DAS28 points for females and from 0.58 to 0.56 for men and kappa was increased for females (from 0.61 to 0.69) and males (from 0.66 to 0.69) when baseline DAS28-CRP scores were adjusted according to gender.

Discussion

This study agrees with existing evidence suggesting DAS28-ESR and DAS28-CRP should not be viewed as interchangeable outcome measures, highlighting the limitations of doing so, with important consequences for clinical and research practice [5-10]. We have proposed a novel and practical method for improving agreement between the two indices that could be applied in future observational studies and clinical practice where both methods have been applied at different time points.

Initial development of the DAS28-CRP by Fransen et al. [3], which included data from 334 patients, used linear regression and high Pearson correlation coefficient to suggest equivalence with the DAS28-ESR. However, at that time, agreement analysis was not undertaken. Subsequent analyses by Wells et al. [17] demonstrated the DAS28-CRP to be a valid outcome measure, although inter-score discrepancies were identified. Consequently, while correlation between the scores was high, equivalence and interchangeability were not demonstrated. Disparity between DAS28-ESR and DAS28-CRP has led some to propose applying lower disease activity thresholds for the DAS28-CRP [6, 7, 10]. However, this approach does not take into account the differential impact of age and gender between the two DAS28 scores. Furthermore, changing disease activity thresholds of the DAS28-CRP would necessitate updating all guidance where reference is made only to the ‘DAS28’ without specification as to the version used, and require clinicians to familiarize themselves with two different stratification systems.

We have demonstrated that variation in equivalence between the two scores is most pronounced for older patients and women (demographics representing the majority of the RA population) and at lower disease activity levels (the target for most treatment strategies). This discordance precludes easy comparison of outcomes of studies that have adopted different versions of the DAS28. The impact of BMI was lower than expected and correction for BMI to improve agreement between the two scores is not necessary. It is possible that other factors (including comorbidities) may affect the relationship between ESR and CRP measurements. However, limiting the additional data required (gender) to improve inter-score agreement means these findings can be translated to use in clinical and research settings easily.

We propose that adjusting DAS28-CRP scores by +0.35 points for females and +0.15 points for males would significantly improve inter-score agreement, and allow existing disease activity thresholds to be used without modification. This adjustment takes into account observed biological differences in ESR levels between females and males and is straightforward and practical.

Improvement in agreement of disease activity stratification at lower disease activity thresholds is achieved at the expense of a minor reduction in agreement at high disease activity; the effect of which would be to encourage more active treatment for patients with higher disease activity, in line with current treatment paradigms. Adjustment of the DAS28-CRP based on age was considered, but the time-varying nature of age makes this a less practical adjustment to the score.

A potential limitation is that we used a single cohort for our study. However, patients included in the BSRBR-RA are enrolled from across the UK, representing a broad population and spectrum of RA management. The cohort is mainly of Caucasian ethnicity, which may influence ESR and CRP relationships differently compared with other ethnicities [8, 9]. The main cohort included patients on biologic agents recruited to a registry, which may introduce selection bias, although it is unlikely this would impact on ESR/CRP comparisons. It is possible unknown confounders may influence whether an individual has both an ESR and CRP test undertaken rather than only one. There were some missing data, although there were no significant demographic differences between missing and complete groups (Supplementary Table S3, available at Rheumatology online).

Our findings suggest the DAS28-ESR and DAS28-CRP should not be used interchangeably when stratifying disease activity. Gender influences inter-score agreement, and adjustment of the DAS28-CRP according to gender significantly improves inter-score reliability.

Click here for additional data file.