Article Text

Extended Report
Evaluating processes underlying the predictive value of baseline erosions for future radiological damage in early rheumatoid arthritis
  1. Jessica A B van Nies1,
  2. Hanna W van Steenbergen1,
  3. Annemarie Krabben1,
  4. Wouter Stomp2,
  5. Tom W J Huizinga1,
  6. Monique Reijnierse2,
  7. Annette H M van der Helm-van Mil1
  1. 1Department of Rheumatology, Leiden University Medical Center, Leiden, The Netherlands
  2. 2Department of Radiology, Leiden University Medical Center, Leiden, The Netherlands
  1. Correspondence to Jessica A B van Nies, Department of Rheumatology, Leiden University Medical Center, P.O. Box 9600, Leiden 2300 RC, The Netherlands; j.a.b.van_nies{at}lumc.nl

Abstract

Objectives Baseline erosions are characteristic for rheumatoid arthritis (RA) and predictive for a severe disease course. The mechanisms leading to baseline erosions being a strong predictor for radiological progression are unknown. We aimed to increase this understanding by mediation analyses in an observational cohort and a cross-sectional MRI study.

Methods 3256 hands and feet radiographs of 653 early RA patients assessed during 7 years of disease were scored using the Sharp–van der Heijde method. Mediation models and multivariate regression analyses were used to explore the association between baseline erosions, other predictors and radiological damage over time. 603 joints (MCP2-5 and MTP1-5) of 67 RA patients underwent 1.5 T MRI at baseline. Data on MRI inflammation were compared with clinical inflammation and baseline radiological erosions.

Results Patients with baseline erosions had, at any point in time during 7 years, 3.45 times more joint damage than patients without erosions (p<0.001, 95% CI 3.00 to 3.98). Baseline erosions were an independent predictor and not a mediator between symptom duration, systemic or local clinical inflammation (erythrocyte sedimentation rate (ESR), swollen joint count (SJC)) or autoantibodies (anti-citrullinated-peptide antibodies, rheumatoid factor) and radiological damage. Subclinical MRI inflammation was studied in relation to erosions, revealing that 83% of the non-swollen joints with baseline erosions had subclinical MRI inflammation compared with 25% of the non-swollen joints without baseline erosions (OR 15.2 95% CI 3.1 to 102.1). The association between MRI inflammation and baseline erosions was independent of symptom duration, ESR, SJC and autoantibodies.

Conclusions Baseline erosions are a predictor for future joint damage, independent of known predictors as time, autoantibodies or clinical measurable inflammation. Subclinical inflammation is suggested as an underlying mechanism.

  • Outcomes Research
  • Rheumatoid Arthritis
  • Early Rheumatoid Arthritis
  • Magnetic Resonance Imaging

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Erosions are characteristic for rheumatoid arthritis (RA) and occur in a majority of RA patients during the disease. A proportion of patients have erosions on hands or feet radiographs already at first presentation. For decades, information on the presence of bone erosions is used to classify RA. In the 1958 ARA criteria, radiographic changes typical of RA were part of the criteria; this concerned decalcification of joints that was not confined to hands or feet.1 Presence of erosions on hand or wrist radiographs was one of the seven 1987 ACR classification criteria.2 Here it was not defined what type of erosive lesions or what number of lesions were required for satisfying this criterion. According to the 2010 American College of Rheumatology/European League Against Rheumatism (ACR/EULAR) criteria, RA can be classified in two ways: by scoring six points on the basis of these criteria or by having RA-specific erosiveness.3 ,4 According to the definition developed by an EULAR task force, RA-specific erosiveness concerns three or more metacarpal phalangeal (MCP), wrist or metatarsophalangeal (MTP) joints with a cortical break.4 Altogether, baseline erosions are considered typical for RA and are relevant to classifying RA.

Presence of erosions at first presentation is also a predictor for the further progression of structural damage.5 ,6 It has thus far not been thoroughly explored whether the association between baseline erosions and disease outcome is mediated by associations between other predictors and radiological progression. In a study on 112 RA patients, multivariable analyses were performed and it was concluded that the baseline erosion score was the most important prognostic factor for the severity of joint damage after 10 years.7 Nonetheless, it remains undetermined whether baseline erosions have a direct relationship with structural damage or that it acts in the path of other risk markers for a severe destructive disease course.

In other words, although it is common practice to screen early arthritis patients for the presence of erosions with hand and foot radiographs, it has to the best of our knowledge not yet been thoroughly explored by what mechanism baseline erosions are associated with the development of further joint damage. Several possible mechanisms can be hypothesised. It is possible that RA patients with erosions at first presentation have a more advanced disease than patients without baseline erosions. This would imply that patients with baseline erosions do not have a more severe disease but they just visit rheumatologists at a later point in time. An alternative explanation is that the presence of baseline erosions is a hallmark of severe disease. Then baseline erosions are expected to be more frequently present in RA patients with high levels of inflammation or in RA patients carrying autoantibodies. Interesting in this respect are the recent observations suggesting that anticitrullinated peptide antibodies (ACPA) themselves stimulate osteoclast differentiation and bone loss.8 Other hypotheses for the method by which baseline erosions reflect a marker for more severe progression may also be proposed.

We aimed to increase the understanding of mechanisms leading to baseline erosions being a strong predictor for structural damage. We addressed this question by evaluating the relationships between different predictors for structural damage using mediation models and multivariate regression analyses. Data of two sets of RA patients were explored to this end: a set of 653 RA patients with radiological follow-up over 7 years of disease and a set of 67 RA patients with baseline radiographs and MRI data.

Methods

Patients

All RA patients that were studied fulfilled the 2010 ACR/EULAR criteria by scoring six points and were included in the Leiden Early Arthritis Clinic (EAC) Cohort, a large inception cohort that started in 1993.9 Inclusion took place when synovitis was confirmed by physical examination and symptom duration was <2 years. At baseline, patients were asked about their joint symptoms and subjected to physical examination, including a 66 swollen joint count (SJC). Blood samples were taken, among others, for determination of the erythrocyte sedimentation rate (ESR), rheumatoid factor (RF) and ACPA. Follow-up visits occurred yearly. Radiographs of hands and feet were taken at baseline and thereafter annually. Since August 2010, extremity MRI was added to the baseline visit.

Two sets of RA patients were extracted from the EAC database. First, in order to evaluate structural damage over 7 years of disease, we studied RA patients consecutively included between 1993 and 2006 (dataset A). Different treatment strategies were used over time. Patients included between 1993 and 1995 were initially treated with non-steroidal anti-inflammatory drugs and disease-modifying antirheumatic drugs were started later after a median of 5 months (and generally within 2 years’ time).10 Patients included between 1996 and 1998 were initially treated with chloroquine or sulfasalazine, and patients included after 1999 were promptly treated with methotrexate or sulfasalazine.10 In the analyses, the inclusion period was used as a proxy for the applied treatment strategy as described previously.11

The second dataset consisted of consecutive RA patients who underwent extremity MRI during the first visit in addition to the regular baseline visit. These patients were included between 2010 and 2012 (dataset B). In this cross-sectional study, radiographic erosions at baseline were related to MRI findings.

All patients gave informed consent. The study was approved by the local Medical Ethics Committee.

Radiological measurements

Baseline erosions were defined similar to the 2010 criterion for RA-specific erosiveness.4 ,12 Patients with ≥3 erosive small joints were considered ‘positive’, and patients with 0–2 erosive joints were considered ‘baseline erosions negative’.

All radiographs obtained during 7 years in dataset A were scored according to the Sharp–van der Heijde (SHS) method by two readers with known time order and blinded to clinical data. Within-reader intraclass-observer correlation coefficients (ICC) were 0.91 and 0.87 and the between-reader ICC was 0.89. The baseline radiographs of dataset B were SHS scored by one reader (ICC 0.86). In the analyses that were done on joint level, a joint was considered erosive in case of a SHS erosion score ≥1.

MRI measurements

MRI examinations were performed on a MSK Extreme 1.5 T dedicated extremity scanner (GE, Wisconsin, USA). The recommended RAMRIS imaging set was acquired for the wrist, MCP and MTP joints. See the online supplementary methods for a detailed scanning protocol. Joints were scanned at the most painful, or if indifferent, dominant side. MRIs were scored according to RAMRIS by two readers independently, blinded to clinical data.13 The within-reader ICCs for the total RAMRIS score were 0.98 and 0.83; the between-reader ICC was 0.82. For all analyses, the synovitis and bone marrow oedema scores were assessed; the sum was called MRI inflammation. For analyses on joint level, MRI inflammation data on MCP(2–5) and MTP (1-5)-joints were extracted; MRI inflammation data on wrists were not used, as part of the carpal bones assessed with RAMRIS cannot be accurately evaluated on radiographs, prohibiting direct comparisons. For analyses on joint level, MRI inflammation data were categorised. MRI inflammation was considered to be present when both readers scored ≥1 for synovitis and/or bone marrow oedema at that joint or bone. An independent third reader was used in case of discordant scores, if 2/3 readers scored ≥1 MRI-inflammation was considered present. For analyses on patient level, the total MRI inflammation score, the sum of all scanned joints, was evaluated.

Analyses

For comparisons between groups, Student t test, Mann–Whitney test or χ2 test were used when appropriate. Multivariate normal regression analysis for repeated measurements with radiological damage as response variable was used on log-transformed radiological data as described elsewhere.11 ,14 This model takes advantage of within-patient correlations of serial radiographs and allows inclusion of patients with missing radiographs at certain time points. All analyses were adjusted for age, gender and treatment strategy. First, the association between baseline erosions and joint damage over time was assessed. Then mediation analyses were performed to investigate whether baseline erosions acted as a mediator between presumed predictors and structural damage over time. The presumed predictors for structural damage were symptom duration (in weeks) as representative of the disease duration, presence of the autoantibodies ACPA and RF, the level of systemic inflammation as represented by the ESR and the extent of clinical local inflammation reflected by the SJC. A path diagram was used as described by Baron and Kenny to depict a causal chain (see figure 1).15 This illustrates the two causal paths that can lead to the outcome; a direct path from the independent to the outcome (c) and an indirect path from the mediator to the outcome (b). Finally, a path exists from the independent to the mediator (a). To test for mediation, three regression analyses need to be performed.15 (1) Regress the mediator on the independent variable (a); the independent should significantly affect the mediator. (2) Regress the dependent (outcome) on the independent variable (c); the independent variable should significantly affect the outcome. (3) Regress the dependent on both the independent and the mediator. In case of mediation, the mediator is significantly associated with the outcome and the effect of the independent variable on the outcome is less than in step 2 (partial mediation) or absent (full mediation). Baseline radiographic and MRI data were analysed by logistic regression. SPSS V.20.0 was used. p Values <0.05 were considered significant.

Figure 1

Schematic overview of the causal paths that were studied using mediation models as described by Baron and Kenny. The diagram illustrates the two causal paths that can lead to the outcome; a direct path from the independent to the outcome (c) and an indirect path from the mediator to the outcome (b). Finally, a path exists from the independent to the mediator (a). According to the description of Baron and Kenny, to test for mediation the following three regression analyses need to be performed.15 (1) Regress the mediator on the independent variable (a); the independent should significantly affect the mediator. (2) Regress the dependent (outcome) on the independent variable (c); also here the independent variable should significantly affect the outcome. (3) Regress the dependent on both the independent and the mediator; in case of mediation the mediator is significantly associated with the outcome and the effect of the independent variable on the outcome is less than in step 2 (partial mediation) or there is no effect at all (full mediation). In this study, we tested the following hypothesis; Do baseline erosions act as a mediator in the path of symptom duration, or in the path of autoantibodies (anticitrullinated-peptide antibodies (ACPA) and rheumatoid factor (RF)), or in the path of systemic inflammation (measured using the erythrocyte sedimentation rate), or in the path of local clinical inflammation (measured using the swollen joint count)? Suppose baseline erosions act in the path of symptom duration, symptom duration is associated with baseline erosions (a) and with structural damage over 7 years (c) but when baseline erosions and symptom duration are both included in the model, symptom duration is no longer associated with structural damage (full mediation) or the effect size of symptom duration in relation to structural damage over 7 years has diminished (partial mediation). The β for baseline erosions on structural damage over 7 years was 3.45 (95% CI 3.00 to 3.98, p<0.001). Thus patients with baseline erosions had 3.45 times more radiological damage at any point during follow-up than patients without baseline erosions. SHS, Sharp–van der Heijde score.

Results

Baseline characteristics

Of the 653 RA patients studied in dataset A, 53% had baseline erosions (≥3 erosive joints). Table 1 presents the baseline characteristics. Overall patients with baseline erosions were older, more often male (36.3% vs 26.1%), more often RF-positive (65.4% vs 55.9%) and had higher ESR (median 39.5 vs 27.0 mm/1 sthr). The median symptom duration was similar in both groups.

Table 1

Baseline characteristics of RA patients with and without baseline erosions (dataset A) and of RA patients who underwent an extremity MRI (dataset B)

Baseline erosions are a predictor for future radiological damage

First, baseline erosions were studied in relation to radiological damage over 7 years. This demonstrated that patients with baseline erosions had 3.45 times (95% CI 3.00 to 3.98, p<0.001) more radiological damage at any point in time than patients without baseline erosions.

Baseline erosions are not a mediator of symptom duration

We then hypothesised that patients who have baseline erosions are more advanced in their disease and that baseline erosions are a mediator between symptom duration at baseline and radiological damage over time. Step 1 of the mediation analysis showed no significant association between symptom duration and baseline erosions (table 2). Patients with baseline erosions had symptoms for median 19.8 weeks and patients without erosion for median 19.0 weeks. Thus, baseline erosions did not act as a mediator in the path of symptom duration.

Table 2

Results of the individual steps of the performed mediation analyses

Baseline erosions are not a mediator of the presence of RA-related autoantibodies

Since ACPA and RF are associated with severe radiological joint damage, we then speculated that these autoantibodies are associated with baseline erosions and more subsequent joint damage. We therefore sought to determine whether baseline erosions acted as a mediator in the path of ACPA or RF. In step 1, ACPA had no significant effect on baseline erosions, whereas RF was significantly associated with baseline erosions (OR 1.49, 95% CI 1.08 to 2.05, p=0.02, table 2). Step 2 showed that RF had a significant effect on radiological damage over 7 years (β 1.08, 95% CI 1.05 to 1.10, p<0.001). Finally, RF and baseline erosions both had a significant effect on radiological damage over time in step 3 (β 1.10, 95% CI 1.07 to 2.12, p<0.001, and β 3.39, 95% CI 2.94 to 3.91, p<0.001, respectively). However, the conditions on mediation did not hold; namely, the effect size of RF was not diminished in the analysis of step 3 (β=1.08 and 1.10, respectively) and statistically significant. ACPA was not significantly associated with baseline erosions in step 1, and step 3 revealed that baseline erosions were independently associated with radiological damage when adjusted for ACPA (table 2). Therefore, baseline erosions were not a mediator between RA-related autoantibodies and radiological damage over 7 years.

Baseline erosions are not a mediator of systemic or local clinical inflammation

The next hypothesis was that baseline erosions occurred in patients with high levels of inflammation and that this actually drives the association between baseline erosions and radiological joint damage over time. The ESR and SJCs were considered to reflect systemic and local inflammation. In step 1, only a significant effect was found for ESR (OR 1.01, 95% CI 1.01 to 1.02, p<0.001, table 2) However, in step 2, no significant effect was observed between ESR and long-term structural damage; therefore, the conditions on mediation were not met. Altogether baseline erosions had an effect on radiological joint damage independent of systemic or clinically measured local inflammation.

Baseline erosions independent in multivariate model

Subsequently, all potential predictors for severe joint damage (symptom duration, ACPA, RF, SJC and ESR) were included in a multivariate regression model with radiological damage over 7 years as outcome (table 3). Patients with baseline erosions had 3.23 times more joint damage at any point during follow-up versus patients without baseline erosions (95% CI 2.79 to 3.74, p<0.001, figure 1, table 3). Thus presenting with baseline erosions was associated with a worse radiological outcome over time, and this could not be explained by the studied known predictors.

Table 3

The results of a multivariate analysis including nine variables and joint damage over 7 years as outcome

Joints with baseline erosions often show subclinical inflammation on MRI

To further explore mechanisms underlying baseline erosions as predictor for future joint damage, we evaluated whether they were associated with subclinical inflammation, which can be visualised by MRI in clinically non-swollen joints. Of the 603 MCP and MTP joints studied, 491 were clinically not swollen. Of these, MRI and radiographic data were available on 485 joints (for technical reasons 6/491 joints could not be evaluated on MRI). Joints with subclinical inflammation (MRI inflammation in clinically non-swollen joints) had radiographic erosions substantially more often compared with joints without subclinical inflammation (83% vs 25%, OR 15.2, 95% CI 3.1 to 102.1, table 4).

Table 4

Frequency of subclinical inflammation in 485 clinically non-swollen MCP and MTP joints in relation to the presence of baseline erosions on hand and foot radiographs in 67 RA patients

MRI inflammation is independently associated with baseline erosions

Finally, we studied, on patient level, whether MRI inflammation at disease presentation was associated with baseline radiographic erosions independent of other possible predictors for erosions. In a logistic regression analysis with the presence of baseline erosions as outcome, the MRI inflammation score was associated with an increased odds of erosions (OR 1.11, 95% CI 1.02 to 1.20) independent of SJC, ESR, ACPA, RF or symptom duration (table 5).

Table 5

Associations with radiographic baseline erosions (≥3 erosive joints) as outcome in 67 RA patients that had both radiographic and MRI data at baseline

Discussion

Baseline erosions are used in classification criteria for RA1–3 and are one of the most potent predictors for future joint damage.16 Surprisingly, thus far no studies thoroughly explored potential mechanisms underlying this association. During the present study, we observed that baseline erosions are not located in the causal path of disease duration (symptom duration), autoantibodies (ACPA, RF) or commonly used markers of inflammation (SJC, ESR). Furthermore, we observed that the large majority of joints presenting with baseline erosions also showed subclinical inflammation. Combined together, these data suggest that subclinical inflammation is relevant for the development of baseline erosions.

Association analyses as performed in the present study provide only an indication of causality. More conclusive evidence on causal relationships can be obtained from animal models where specific proteins can be knocked out or in relation to the current manuscript where inflammation can be induced and subsequent MRIs and radiographs of extremities can reveal the induced changes in joints and bone.17 In humans, however, such approaches cannot be adopted and mediation models were used as substitute. An advantage of mediation analysis is that known predictors were assessed separately. This yielded a higher certainty to conclude on causal paths than could be obtained from multivariable analysis.

No uniform definition of baseline erosions exists. Previous studies used different definitions, such as any radiological evidence of erosions, a cortical break of ≥2 mm, presence of two or three erosions, or a certain Larsen or SHS score.7 ,18–22 In this study, erosive disease was defined as having ≥3 erosive joints, which is the same as the recently proposed definition of RA-specific erosiveness in light of the 2010 ACR/EULAR criteria.4 Notably, when applying another cut-off for erosive disease (≥2 erosive joints), similar results were obtained (data not shown). A limitation of any definition of erosions visualised with radiographs is that these are depicted only two-dimensionally.

The presence of baseline erosions was not a reflection of a longer symptom duration, suggesting that the duration of processes underlying the disease was similar in patients with and without erosions. Furthermore, differences in inflammation measured using ESR and SJC were insufficient to explain the association between baseline erosions and subsequent damage. No difference was found in the mediation analyses when CRP was used instead of ESR to reflect systemic inflammation (data not shown). In other words, these known risk factors for long-term joint damage (symptom duration, ESR and SJC) did not act via mechanisms that also promote baseline erosions. The biological processes underlying the association between symptom duration and radiological outcome,23 which is also called the ‘window-of-opportunity’, are unknown. Apparently, these processes do not promote baseline erosions. Of note, there might be risk factors for baseline or long-term structural damage that were not assessed in this study.

Intriguingly we observed no significant difference in the percentage of ACPA positivity in patients with and without baseline erosions (58% vs 51%). Because a recent study showed that ACPA itself is capable of activating osteoclast activity, even in the preclinical phase,8 we assumed that ACPA-positive patients had more frequent erosions at baseline. The absence of this finding does not imply that ACPA does not activate osteoclasts. Still, radiological visible bone erosions at disease presentation cannot be explained by the effect of ACPA.

We observed that inflammation measured with MRI is associated with the presence of baseline erosions independently of ESR, SJC, ACPA and symptom duration. More importantly, 83% of the non-swollen joints with baseline erosions had subclinical swelling. These data suggest that previous or still present subclinical inflammation is important to develop baseline erosions and further joint damage. This finding at disease onset is in line with observations of others, showing the relevance of MRI inflammation of RA patients in remission.24 ,25 There are however several limitations. The number of MCP and MTP joints that had baseline erosions in dataset B was small. Furthermore, no long-term radiological outcome data were available of these joints. Thus, although a strong association was observed between subclinical inflammation and baseline erosions (step 1 of the mediation analyses), step 3, revealing that baseline erosions are a mediator in the causal path between subclinical inflammation and future joint damage, could not be evaluated. Nonetheless, despite these limitations, the finding that all but two clinically non-swollen joints with baseline erosions also had subclinical inflammation at MRI is suggestive. Since the MRIs were made at the same time as the radiographs, it cannot be excluded that the baseline erosions in the two non-swollen joints without subclinical inflammation were related to previous local inflammation that resolved at the time of disease presentation.

In conclusion, this study is the first to explore mechanisms mediating the association between baseline erosions and future structural damage in early RA by taking advantage of both a large observational cohort and high-quality extremity MRI data. Present data suggest that local subclinical inflammation is relevant. Further studies on the long-term outcome of subclinical inflammation in the early clinical phase and presumably also in the preclinical phase of RA are warranted.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

Footnotes

  • Handling editor Tore K Kvien

  • Contributors JABvN analysed the data. JABvN and AHMvdH-vM contributed to the conception, design and interpretation of the data and drafting of the article. HWvS, AK and WS scored all MR images. MR, AHMvdH-vM and TWJH revised the article for important intellectual content. All authors gave final approval of the version to be published.

  • Funding The research leading to these results was funded by a Vidi grant of the Netherlands Organisation for Scientific Research, a grant of the Dutch Arthritis Foundation (Reumafonds), The European Community Seventh Framework Program FP7 Health-F2-2008-223404 (Masterswitch) and by the IMI JU funded project BeTheCure, contract no. 115142-2.

  • Competing interests None.

  • Ethics approval Ethics committee LUMC.

  • Provenance and peer review Not commissioned; externally peer reviewed.