Influence of disease-modifying therapy on radiographic outcome in inflammatory polyarthritis at five years: Results from a large observational inception study

Authors


Abstract

Objective

To determine the effect of early treatment with disease-modifying antirheumatic drugs (DMARDs) in reducing radiographic progression over a 5-year period in patients with new-onset inflammatory polyarthritis.

Methods

Three hundred thirty-five consecutive patients with paired radiographs obtained 1 year and 5 years after enrollment in a population-based arthritis register were studied. Logistic regression was used to model differences in baseline factors associated with the start of DMARDs. The time from symptom onset to first use of DMARDs was stratified to represent 4 groups: no DMARD use, <6 months, 6–12 months, and >12 months. Radiographs of the hands and feet were scored using the Larsen method. Progression in the Larsen score was evaluated as a 5-year score adjusted for the first film score. Negative binomial regression was used to compare Larsen score progression for each of the 3 treatment groups with that for patients not receiving DMARDs. Results were then adjusted for severity, based on propensity modeling.

Results

Patients who received treatment had more radiographic progression than did patients who were untreated. Coefficients (95% confidence intervals), expressed as a multiple of the Larsen score in DMARD-treated patients compared with untreated patients, were as follows: 1.6 (1.1–2.3) for <6 months, 2.4 (1.5–3.6) for 6–12 months, and 2.0 (1.4–2.8) for >12 months. As expected, patients receiving treatment had more severe disease at baseline. Using the propensity score as a method of adjusting for disease severity, the influence of treatment on outcome became attenuated as follows: 1.1 (0.8–1.7) for <6 months, 1.6 (1.0–2.6) for 6–12 months, and 1.5 (1.0–2.2) for >12 months. This effect was also seen in the crude Larsen score at year 5.

Conclusion

In this observational study, DMARD treatment was a marker not only of worse disease at presentation but also of the radiographic state and radiographic progression at 5 years. After adjustments were made for baseline disease severity, earlier therapy was shown to have a beneficial effect on outcome.

Radiologic erosions are considered to be one of the key features associated with a poor outcome in rheumatoid arthritis (RA) (1). Although the cross-sectional relationship between current disease activity and radiographic severity is modest, the relationship between radiographic severity and markers of cumulative damage, such as joint deformity and physical disability, is stronger (2–7). It is thus not surprising that in clinical trials of disease-modifying antirheumatic drugs (DMARDs) in RA, the rate of radiographic progression is one of the primary outcome measures (8–19).

In RA, clinical trials are the optimal approach to investigating treatment efficacy. Such trials are difficult to undertake when long-term followup is required, however, because of the problems involved in maintaining blinding and participation. Large-scale trials are also expensive to conduct. More importantly, it is hard to extrapolate results from a clinical trial to routine clinical practice. Patients selected for clinical trials are not a random sample of the population of patients with RA. Furthermore, the care received by patients during clinical trials is more intensive than that used in routine practice, which possibly enhances the beneficial effect of treatment.

There is thus considerable interest in the use of well-conducted longitudinal observational studies (LOS) to complement the results of randomized controlled trials (20, 21). The influence of treatment on outcome in RA is, however, seldom considered in LOS. The major problem in examining the role of treatment as a predictor of (a good) outcome is that the intensity of therapy is related to the severity of disease. Therefore, when evaluating the role of treatment, it is necessary to allow for this bias in treatment allocation.

We recently demonstrated that it is possible to use a propensity model to adjust for the influence of disease severity on treatment decisions in observational studies (22). With this approach, an attempt is made to model the treatment decision using all available clinical and laboratory data. The resultant model gives a probability of being treated (propensity score). Among patients with the same probability, the treatment decision is then considered to be independent of disease severity and hence quasi-random. The efficacy of treatments can then be compared, analogous to a true randomized clinical trial (23, 24).

In our earlier study, we evaluated the role of early treatment with DMARDs in reducing longer-term disability in an inception cohort of patients with inflammatory polyarthritis. We showed that patients who received early treatment (within 6 months of symptom onset) achieved a severity-adjusted disability outcome at 5 years comparable with that of patients who were judged clinically as not requiring treatment (22). Starting DMARD treatment later (≥6 months) did not overcome this disadvantage.

In the present study, we extended this methodology to consider the impact of early DMARD treatment on radiographic outcome 5 years after symptom onset. Specifically, we tested the following hypotheses: (a) patients for whom DMARD therapy was prescribed have a more severe radiographic outcome because of increased disease severity, (b) adjusting for these severity differences will demonstrate a protective effect of treatment, and (c) this effect is maximal in patients given DMARD treatment early in the course of disease. We tested these hypotheses in a large inception cohort of patients with inflammatory polyarthritis.

PATIENTS AND METHODS

Patients.

Patients were recruited from the Norfolk Arthritis Register (NOAR) study, details of which have been described elsewhere (25). Briefly, patients visiting their general practitioners are referred to NOAR if they have a history of swelling of ≥2 joints lasting for ≥4 weeks. After giving consent, patients are visited at home within 2 weeks by a trained research nurse who completes a structured interview and performs a clinical examination. Patients who have joint inflammation that is not subsequently explained by another specific diagnosis given by a rheumatologist (excluding psoriatic arthritis or postviral arthritis) are eligible for inclusion in NOAR.

Baseline evaluation.

At the initial interview, the date of symptom onset (defined as the first time that joint swelling was reported), and the nature of current joint symptoms and morning stiffness are documented. Each patient completes the Health Assessment Questionnaire (HAQ) (26). Fifty peripheral joints are examined for the presence of swelling, tenderness, or both, and the presence of rheumatoid nodules is also recorded. A blood sample is obtained, and C-reactive protein (CRP) and rheumatoid factor (RF) levels are measured by rate nephelometry and by latex agglutination, respectively.

Clinical followup.

Patients were followed up annually, and the American College of Rheumatology (ACR; formerly, the American Rheumatism Association) 1987 classification criteria for RA (27) were applied, using both a point and cumulative evaluation as described elsewhere (28). At each visit, information was collected regarding any DMARD therapy, including oral steroids, that had been started or stopped in the interval between visits. When necessary, the patient's clinical records were used to verify treatment start and stop dates.

Radiographic followup.

The protocol for requesting radiographs is described elsewhere (29). Patients were asked to undergo radiographic examination of the hands and feet at the first annual followup visit if they met at least 3 ACR criteria (excluding the radiographic criterion) at registration, or if they met at least 2 ACR criteria at the first followup visit. At the second annual followup visit, patients were asked to undergo radiographic examination if they met at least 2 ACR criteria and did not have radiographic erosions at the first followup or had not attended the first followup visit. All patients were asked to undergo radiographic examination at the fifth followup visit. Radiographs were scored using the Larsen method (30). Radiographs were read by 2 of 3 investigators (DPMS, BH, MB), and major disagreement on whether a film showed erosions was settled by arbitration by a third investigator (DGIS or BH).

Analysis.

All patients who underwent radiographic examination at the first, second, and fifth followup visits were included in the analysis. Results for the first 335 eligible subjects in NOAR are presented. The interval since self-reported symptom onset and the first prescribing of DMARDs or steroids was calculated for each patient treated. All of the clinical and laboratory variables gathered at baseline (described above) were utilized in a logistic regression model to predict the probability of ever use of DMARDs (before the 5-year radiograph) versus never use of DMARD treatment. This model was then used to calculate the predicted probability of treatment for each patient. These predicted probabilities, or propensities, were divided by quintiles. Thus, patients were divided into 5 groups on the basis of the probability of being treated.

For the purpose of this analysis, individuals receiving DMARD treatment were divided into 3 groups as follows: those who received treatment within 6 months of symptom onset, those who were first treated between 6 and 12 months after symptom onset, and those who were treated at least 12 months after symptom onset. The median Larsen score at 5 years, both before and after adjustment for Larsen score at first film, in the 3 treatment groups was compared with that in the group that did not receive treatment. A negative binomial regression model was then used to estimate the influence of treatment group on Larsen score. This method is similar to Poisson regression, which is the conventional approach to modeling count data. However, negative binomial regression modeling has the advantage of not underestimating standard errors and significance levels, which is the problem with Poisson regression if there is more variation between individuals than can be explained by the measured covariates.

Univariate analysis estimating the association between DMARD use and Larsen score was undertaken initially, followed by subsequent analysis after adjusting for propensity quintile. Finally, it was assumed that patients in whom erosive changes were already present at the time of the first film might be less likely to benefit from treatment; therefore, a subgroup analysis was performed on the influence of timing of treatment in patients who were free of erosions at time of the first film.

RESULTS

The baseline clinical and demographic characteristics of the cohort are shown in Table 1. Although only 47% of the patients satisfied the ACR criteria for RA at the first visit, this proportion increased to 93% by 5 years, using a cumulative application of the criteria (28). At 5 years, 218 patients (approximately two-thirds) had received either a disease-modifying drug or steroids. The first drugs received by these patients are shown in Table 2. In the majority of these patients (57%), sulfasalazine was the first drug administered, reflecting local practice during the period of study, and 27% of the patients received oral prednisolone first. As expected, there was a relationship between the time to the start of DMARD therapy and disease severity. Patients who received no DMARD during the study had milder disease (Table 3).

Table 1. Baseline characteristics of the cohort (n = 335)*
  • *

    IQR = interquartile range; HAQ = Health Assessment Questionnaire; CRP = C-reactive protein.

Male, no. (%)100 (30)
Age at symptom onset, years, mean ± SD55.2 ± 14.1
Delay to presentation, months, median (IQR)5.0 (2.1–10.1)
No. of swollen joints, median (IQR)8 (4–14)
No. of tender joints, median (IQR)9 (4–17)
No. of swollen and tender joints, median (IQR)4 (2–10)
Nodules, no. (%) of patients29 (9)
HAQ score, median (IQR)0.88 (0.38–1.50)
CRP, gm/dl, median (IQR)7 (2–17)
Rheumatoid factor titer, no. (%) of patients 
 <1:40230 (69)
 1:40–1:16047 (14)
 >1:16058 (17)
Table 2. First DMARD/steroid prescribed to patients who received treatment (n = 218)*
TreatmentNo. (%) of patients
  • *

    DMARD = disease-modifying antirheumatic drug.

Sulfasalazine125 (57)
Prednisolone58 (27)
Hydroxychloroquine12 (5.5)
Intramuscular gold11 (5)
Methotrexate10 (4.6)
Other3 (1.4)
Table 3. Differences in baseline characteristics between patients who did and did not start DMARDs/steroids*
VariableDid not receive DMARDs (n = 117)Started DMARDs <6 months after symptom onset (n = 90)Started DMARDs 6–12 months after symptom onset (n = 46)Started DMARDs >12 months after symptom onset (n = 82)P
  • *

    DMARD = disease-modifying antirheumatic drug; IQR = interquartile range; HAQ = Health Assessment Questionnaire; CRP = C-reactive protein; RF = rheumatoid factor; ACR = American College of Rheumatology.

  • Cumulative for 5 years.

Male, no. (%)34 (29)32 (36)10 (22)24 (29)0.41
Age at symptom onset, years, mean ± SD53 ± 14.359.3 ± 13.854.1 ± 14.853.3 ± 14.00.01
No. of swollen and tender joints, median (IQR)4 (1–6)7 (3–15)5 (2–10)3.5 (1–9)0.0006
HAQ score, median (IQR)0.5 (0.13–1.0)1.25 (0.75–2.00)1.25 (0.50–1.75)0.75 (0.38–1.38)0.0001
CRP level, gm/dl, median (IQR)3 (1–7)11.5 (7–38)12 (5–26)5 (3–15)0.0001
Delay to presentation, months, median (IQR)3.7 (1.9–9.5)3.3 (1.7–5.4)5.9 (3.7–8.0)11 (5.1–18.3)0.0001
RF positive, no. (%)18 (15)29 (32)26 (57)32 (39)0.0001
Propensity score, mean ± SD0.46 ± 0.220.78 ± 0.200.81 ± 0.190.69 ± 0.230.0001
No. (%) fulfilling ACR criteria99 (85)89 (99)46 (100)80 (98)

There was also a difference in the choice of first drug between these delayed-treatment groups. In both the group starting treatment within 6 months of symptom onset and the group starting treatment between 6 and 12 months after symptom onset, the proportion of patients that started treatment with sulfasalazine was identical (63%). Only 48% of patients who experienced the longest delay before treatment started treatment with sulfasalazine. In contrast, methotrexate was the first drug used in 10% of the group with the longest delay before treatment but was used first in only 1 patient in each of the other delayed-treatment groups.

The 335 patients were divided by quintile into 5 groups based on the results of the logistic regression analysis used to predict treatment propensity. The propensity model used the following 10 variables: age at onset, delay to presentation, sex, maximal early morning stiffness, RF titer, HAQ, CRP, number of swollen joints, number of tender joints, and number of swollen and tender joints. The sensitivity and specificity of the model were 81% and 64%, respectively. As expected, there was a strong relationship between these treatment propensity groups and disease severity (Table 4). Patients with the highest probability of being treated had substantially more inflamed joints, were more likely to have rheumatoid nodules, had higher HAQ scores, and were more likely to have an elevated CRP level and to be RF positive.

Table 4. Baseline characteristics within propensity quintiles*
 Quintile
12345
  • *

    Each quintile represents 67 patients. Quintile 1 is lowest, and quintile 5 is highest. IQR = interquartile range; HAQ = Health Assessment Questionnaire; CRP = C-reactive protein; RF = rheumatoid factor.

Male, no. (%)17 (25)23 (34)20 (30)21 (31)21 (30)
Age at symptom onset, years, mean ± SD53.0 ± 12.954.5 ± 15.752.5 ± 14.659.3 ± 13.756.7 ± 12.5
Delay to presentation, months, median (IQR)4.3 (2.0–9.5)5.0 (2.4–8.5)5.9 (3.2–11.5)5.0 (1.9–11.8)4.8 (2.1–9.0)
No. of swollen joints, median (IQR)4 (2–8)6 (3–11)7 (4–11)14 (7–21)13 (7–22)
No. of tender joints, median (IQR)6 (2–13)7 (3–13)9 (4–14)12 (5–23)10 (6–20)
No. of swollen and tender joints, median (IQR)2 (0–6)3 (1–6)5 (2–9)8 (3–17)7 (3–14)
Nodules, no. (%) of patients3 (4)4 (6)4 (6)5 (7)13 (19)
HAQ score, median (IQR)0.1 (0.0–0.5)0.6 (0.3–1.1)1.0 (0.5–1.4)1.3 (0.8–1.8)1.6 (1.1–2.1)
No. (%) with CRP level >13 gm/dl1 (1)7 (10)16 (24)23 (34)41 (61)
No. (%) RF negative63 (94)60 (90)49 (73)34 (51)24 (36)
No. (%) of patients with positive RF titer     
 1:40–1:1604 (6)7 (10)13 (19)14 (21)9 (13)
 >1:1600 (0)0 (0)5 (7)19 (28)34 (51)

These modeled propensities were then compared with the actual treatment decision made. The data show that 28% of patients in the lowest quintile were treated with DMARDs, compared with 90% or more of those in the 2 highest quintiles (Table 5). Thus, modeling for severity still did not predict treatment decision completely, providing the quasi-randomization element needed for then evaluating the impact of therapy.

Table 5. Relationship between propensity score and actual treatment received*
Propensity score quintileNo. of patients actually treatedObserved probability of treatment
  • *

    Each quintile represents 67 patients. Quintile 1 is lowest, quintile 5 is highest.

1190.28
2340.51
3410.69
4600.90
5640.96

The descriptive radiographic severity data are shown in Table 6. In general, patients had low Larsen scores, reflecting the fact that these individuals were recruited from a primary care–based cohort. In this group, there was evidence of radiographic progression (either newly developed erosions or increased severity of existing erosions) in an important number of patients between the initial and the 5-year films. The radiographic scores according to treatment group are shown in Table 7. The results demonstrate that both the number of patients with erosions and the median Larsen scores were greatest in the groups receiving treatment, and that the radiographic severity was greater in those patients who began their DMARD treatment more than 6 months after symptom onset. This is true both for the absolute Larsen score at 5 years and for the change in Larsen score between the first and final films.

Table 6. Comparison of radiographic severity at initial and followup films*
GroupFirst filmSecond filmChange
  • *

    IQR = interquartile range.

Whole cohort (n = 335)   
 No. (%) with erosions105 (31)161 (48)+56 (17)
 Larsen score, median (IQR)2 (0–10)8 (1–25)+3 (0–15)
Patients with erosions at baseline (n = 105)   
 Larsen score, median (IQR)8 (2–18)26 (13–43)+15 (6–27)
Patients erosion-free at baseline (n = 230)   
 No. (%) with erosions69 (30)+69 (30)
 Larsen score, median (IQR)0 (0–3)3 (0–11)+1 (0–7)
Table 7. Progression of Larson score*
Larsen scoreDid not receive DMARDs (n = 117)Started DMARDs <6 months after symptom onset (n = 90)Started DMARDs 6–12 months after symptom onset (n = 46)Started DMARDs >12 months after symptom onset (n = 82)
  • *

    Values are the median (interquartile range). DMARDs = disease-modifying antirheumatic drugs.

Initial score0 (0–4)3 (0–12)6 (0–17)5 (0–13)
5-year score3 (0–9)10.5 (2–22)20.5 (4–40)13 (2–42)
Change+1 (0–5)+4 (0–13)+11.5 (1–28)+6 (0–21)

The results of the negative binomial regression for Larsen score at 5 years and for the change in Larsen score are shown in Tables 8 and 9, respectively. Univariate analysis confirmed that Larsen scores were indeed higher in patients receiving treatment. However, after adjusting for propensity quintile there was, as anticipated, a reduction in this effect. More importantly, the influence of treatment after adjusting for treatment propensity was less in patients whose treatment was started more than 6 months after symptom onset. This was also true in the subgroup of patients who had no erosions at baseline. Similar results were observed when the outcome of interest was change in Larsen score. Thus, patients treated within 6 months of symptom onset had a change in Larsen score that might have been predicted on the basis of baseline severity. In contrast, a treatment delay of more than 6 months had less of an effect; that is, allowing for disease severity by adjusting for treatment propensity did not explain as much of the greater radiographic severity than that seen in the early-treatment group.

Table 8. Influence of treatment on Larsen score at 5 years*
Larsen scoreDid not receive DMARDs (n = 117)Started DMARDs <6 months after symptom onset (n = 90)Started DMARDs 6–12 months after symptom onset (n = 46)Started DMARDs >12 months after symptom onset (n = 82)
  • *

    Values are the multiplier (95% confidence interval) of the Larsen score in the group that did not receive disease-modifying antirheumatic drugs (DMARDs) (referent).

  • Analysis was restricted to patients who were erosion-free at baseline.

Crude1.02.5 (1.7–3.7)3.9 (2.4–6.3)3.3 (2.2–4.9)
Adjusted for propensity1.01.5 (0.9–2.3)2.3 (1.4–3.9)2.2 (1.4–3.5)
Adjusted for propensity, in patients erosion-free at baseline1.01.3 (0.7–2.3)1.8 (0.8–3.8)1.7 (0.9–3.0)
Table 9. Influence of treatment on progression of Larsen score between initial and 5-year film*
Larsen scoreDid not receive DMARDs (n = 117)Started DMARDs <6 months after symptom onset (n = 90)Started DMARDs 6–12 months after symptom onset (n = 46)Started DMARDs >12 months after symptom onset (n = 82)
  • *

    Values are the multiplier (95% confidence interval) of the Larsen score in the group that did not receive disease-modifying antirheumatic drugs (DMARDs) (referent).

  • Analysis was restricted to patients who were erosion-free at baseline.

Crude1.01.6 (1.1–2.3)2.4 (1.5–3.6)2.0 (1.4–2.8)
Adjusted for propensity1.01.1 (0.8–1.7)1.6 (1.0–2.6)1.5 (1.0–2.2)
Adjusted for propensity, in patients erosion-free at baseline1.01.3 (0.7–2.2)1.8 (0.9–3.6)1.4 (0.8–2.5)

DISCUSSION

This study examines the influence of DMARDs on the natural history of radiographic disease in a large, unselected inception cohort of patients with inflammatory polyarthritis. The major aim of the analysis was to assess the influence of timing of first DMARD therapy after adjusting for the severity bias, which is that patients with more aggressive disease are more likely to be given treatment. Several observations can be made based on these data. First, patients in this cohort who received treatment had worse disease in terms of radiographic outcome at 5 years, although the effect was greatest in patients whose treatment was delayed. Second, after adjusting for disease severity, the beneficial impact of treatment on disease outcome was greatest in patients treated within 6 months of symptom onset, despite the fact that patients for whom treatment was delayed the longest had milder disease at baseline. Third, early (versus later) treatment also appears to be beneficial in terms of influencing radiographic outcome at 5 years, even in patients who were erosion-free at the time of the first film.

A number of methodologic issues must be considered. First, there are concerns regarding the external validity of these findings. This was a primary care–derived cohort, comprising an unselected group of individuals with inflammatory polyarthritis derived from a whole population. Decisions pertaining to the timing of DMARD therapy and its impact may not be applicable in other settings. However, this is true for all studies, both LOS and RCTs, given the variability in clinical practice not only between but also within countries. Second, given the inherent instability in applying the label “rheumatoid arthritis” to patients with early disease (22), we chose to study all patients who had inflammatory polyarthritis. As shown, however, the overwhelming majority of individuals did satisfy ACR criteria for RA at the fifth followup examination. Third, as discussed elsewhere (29), the findings reported are not derived from a complete cohort of subjects. Patients were eligible for a followup radiograph only if they would potentially satisfy the ACR criteria for RA. Thus, patients at the milder end of this community-derived spectrum of cases were excluded.

Because our major hypothesis related to the timing of the first DMARD rather than to specific agents, we combined all DMARDs for the analysis. Therefore, we did not attempt subgroup analyses according to which drug was actually given first. In addition, the small numbers in most subgroups, except those for sulfasalazine, precluded such an analysis. It is interesting to note that methotrexate use was greatest in the group that experienced the longest delay before treatment. This group, as mentioned above, also had milder disease at baseline. Thus, it seems unlikely that the worse outcome in the group with the longest delay was attributable to greater use of less effective treatments. It should also be emphasized that, as might be expected, patients' DMARD treatment changed during the period of followup. Given substantially larger numbers of patients, the propensity approach could be used to assess the role of differing strategies for DMARD therapy.

In this analysis, prednisolone was considered to be a DMARD. This designation is perhaps not unreasonable given the beneficial effects of prednisolone suggested by some investigators (10, 31), although others have reported that prednisolone does not influence radiographic progression (32). Interestingly, the use of steroids as the first “DMARD” was identical (29%) in patients who received treatment earliest and in those who received treatment latest. We did, however, repeat the analyses, excluding consideration of steroids, and this exclusion had no important impact on the results. Additionally, we examined only the timing of the start of treatment and did not attempt to examine the effect of duration of treatment. The effect of treatment duration is also extremely complex to investigate in an observational study, because the decision to continue treatment is substantially confounded by disease severity.

The probability of receiving treatment was derived from data gathered at a single time point (baseline) and did not necessarily represent each patient's clinical condition at the time the treatment decision was made. Thus, if we had been able to capture the actual disease status at that time, the fit of the propensity model might have been better. Furthermore, the clinician may have used other unmeasured (or indeed immeasurable) items of clinical data in reaching a decision to start treatment, and such data were not collected. Clinicians might reasonably base their treatment decisions on factors such as the patient's general sense of well-being and satisfaction, data that are difficult to obtain. Theoretically, refinement of our propensity model may have improved the fit. It is therefore likely that our propensity model does not remove all of the confounding by indication, and thus the benefit of treatment may be underestimated. However, as shown, the quintiles generated did demonstrate the expected trend in treatment likelihood in the patients studied, as well as the expected distribution of the key clinical laboratory variables used.

The current data support results from other types of investigations suggesting a beneficial effect of early DMARD therapy. The studies by Egmose et al and Rich et al (33, 34) showed that early DMARD therapy confers some protection against the development of erosions, although it does not completely suppress the occurrence of erosions. Patients with longer disease duration do not respond well to DMARD therapy (35), which possibly indicates that early DMARD therapy might suppress inflammation better. Stenger et al (36) reported that early aggressive DMARD therapy (with sulfasalazine and methotrexate) resulted in reduced radiographic progression in 76 patients with early RA.

In summary, patients treated with DMARDs had a worse radiographic outcome at 5 years compared with patients who did not receive such treatment. Much of this difference is explained by greater disease severity at baseline in patients for whom DMARDs were prescribed. The important conclusion is that this severity disadvantage is harder to overcome when treatment is delayed more than 6 months after symptom onset.

Acknowledgements

We acknowledge the contribution of Mrs. Bett Barrett, the NOAR manager, and the NOAR metrologists in the collection of the clinical data, Dr. Beverley Harrison for reading of some of the radiographs, and David Carthy for the CRP assays. We thank the rheumatologists and general practitioners for their support.

Ancillary