Rheumatoid arthritis follows a variable course and the lack of reliable prognostic factors that are useful at disease onset is well recognised by practising rheumatologists. Outcome in this disease is complex and relates to several entities such as disease activity, functional status, joint damage, and patients’ perception of general health. Synovitis may explain most of the early symptoms and is also considered to contribute to the development of joint damage and disability. The correlation between inflammation and joint damage has been studied extensively, especially the relevance of inflammatory variables such as C reactive protein and erythrocyte sedimentation rate (ESR).^1–4 Although there is a link between inflammation and the development of joint damage it is well established that damage may progress in spite of decreased inflammatory activity, and erosions may develop in patients who have few clinical signs of inflammation. Thus it has been suggested that pathological processes other than inflammation are involved in the destructive process.^5–7

In this study we focused on the destructive process and used joint damage in the hands and feet, evaluated on radiographs, as the outcome variable. We selected several laboratory variables with suggested prognostic potential and tested their value in a well defined cohort of patients with early rheumatoid arthritis.⁸ The variables were ESR and C reactive protein, reflecting inflammation; cartilage oligomeric matrix protein (COMP), a marker of cartilage turnover with putative roles in disease chronicity⁹; and a set of autoantibodies: rheumatoid factor (RF) of IgG, IgA, and IgM type, anti-cyclic citrullinated peptide antibodies (anti-CCP), interleukin 1α antibodies (anti-IL1α), and shared epitope.

Molecular markers from joint tissue matrix could be used to reflect the destructive process in rheumatoid arthritis.¹⁰ Increased serum levels of COMP have been found to correlate with large joint destruction in patients with this disease.¹¹ It has also been shown that COMP is a measure of tissue processes that are distinct from the reactions in inflammation.^12–14

RF and the more recently described anti-CCP,^15–17 which is reported to be more specific for rheumatoid arthritis than RF, have both been shown to be associated with more severe joint damage.^18–26 The different subtypes of RF have also been examined in relation to joint damage, and some studies have found IgA RF to be more related to damage,^27–29 but the findings are conflicting.^30,31

IL1α is an important proinflammatory cytokine in rheumatoid arthritis, and naturally occurring anti-IL1α antibodies were first reported in 1989.³² These antibodies bind with high affinity to IL1α, thereby preventing binding of IL1α to its receptors and thus neutralising the biological activity of the cytokine.^32,33 It has been shown that patients with arthritis who carry the anti-IL1α develop less destructive disease.^34–37 The reported frequency of anti-IL1α in the population varies from 10% to 60%, depending on age, sex, and assay sensitivity.³⁸

Genetic predictors for rheumatoid arthritis, mainly the HLA-DRB1 alleles, have been investigated but the results are inconsistent in relation to prognosis for joint damage, disability, and disease persistence.^24,39–41

In this report, we sought to determine whether a combination of ESR, C reactive protein, COMP, shared epitope, RF (IgG, IgA, and IgM types), anti-CCP, and anti-IL1α provides prognostic information on the development of joint damage in the hands and feet after five and 10 years in a prospective study of early rheumatoid arthritis.

METHODS

All new patients with definite rheumatoid arthritis according to the 1958 American Rheumatism Association (ARA) criteria were included in a prospective study from 1985 to 1989 at the department of rheumatology at Lund University Hospital in southern Sweden. All consecutive patients aged 18 years or older with duration of symptoms for less than two years were included. Most patients were referred from primary health care units as a result of a special campaign to recruit cases of recent onset. The study comprises 183 patients (116 women and 67 men). Their mean (SD) age was 51.2 (12.4) years and the mean duration of symptoms at inclusion in the study was 11.1 (6.1) months. All patients, irrespective of disease activity, were included and followed prospectively at least annually at a team care unit. Patients with active disease were offered treatment with disease modifying antirheumatic drugs (DMARDs) throughout the study according to general clinical practice, which changed during the study period. Early in the study D-penicillamine and antimalarial agents were the most commonly used DMARDs, while at the 10 year follow up 23% of the patients were being treated with methotrexate, which was then the most commonly used drug. During the first five years, 13 patients were treated with methotrexate, while from year 5 to 10, 46 patients were treated with methotrexate. The whole cohort has been followed for at least 10 years, and the clinical outcome is presented extensively elsewhere.⁴²

Radiographic evaluation

In the present study radiographic findings in hands and feet at years 5 and 10 were used as outcome variables. Radiographic outcome over the first 10 years and the scoring methodology were described in detail previously.⁴³ In brief, radiographs of hands and feet (standard film in postero-anterior projection) were obtained annually from inclusion to year 5 and at year 10. Joint damage caused by rheumatoid arthritis was evaluated according to Larsen et al.⁴⁴ Thirty two joints in the hands and feet were assessed. Each joint was compared with a standard reference film and changes were graded from 0–5, where a score of ⩾2 represents erosive disease. A joint damage score was calculated by adding all scores, the wrist multiplied by 5, resulting in a range of 0 to 200.⁴³ The scoring was made by one of two assessors; years 0–5 were scored in chronological order and the 10 year evaluations were scored separately. Scoring reliability (inter- and intraobserver, chronological order versus separately) was evaluated using the intraclass correlation coefficient and varied from 0.92 to 0.99.⁴³

Biochemical analyses

Blood samples were collected from all patients at inclusion in the study. ESR was determined according to Westergren and C reactive protein by an electroimmunoassay.⁴⁵ HLA-DRB alleles were typed by restriction fragment length polymorphism analysis with sequence specific primers, as previously described.³⁹ EDTA treated plasma and sera were frozen in aliquots and stored at −80°C. Serum COMP was measured with a commercial sandwich enzyme linked immunosorbent assay (ELISA) using two monoclonal antibodies directed against separate antigenic determinants on the human COMP molecule (AnaMar Medical, Lund, Sweden). The detection limit was <0.1 U/l and the intra- and interassay coefficient of variation was <5%. RF (IgG, IgA, and IgM subtypes) was analysed by ELISA using commercial kits (Inova Diagnostics, San Diego, California, USA) according to the manufacturer’s instructions. Rabbit IgG was used on the solid phase. The limit for a positive result was set to >6 IU. Anti-CCP was analysed by ELISA using a commercial kit (Inova Diagnostics) in which a synthetic cyclic citrullinated peptide is used as antigen. The lowest value for a positive outcome was set to ⩾20 U. Anti-IL1α were analysed by radioimmunoassay as described previously.³⁷ The limit for positive results for anti-IL1α was defined as a binding of >10%.

The study was approved by the ethics committee of Lund University.

Statistical analysis

Comparisons between groups were analysed by the Mann–Whitney U test, Wilcoxon’s test, or the χ² test as appropriate. Pearson correlation coefficients expressed the relations between the assessed variables. Multiple linear regression with backward elimination was carried out to decide the degree of explanation of the different variables to the Larsen score at years 5 and 10. The following independent variables were entered as continuous variables: ESR, C reactive protein, COMP, and age at inclusion. Sex, RF (IgG, IgA, and IgM subtypes), anti-CCP, anti-IL1α, and presence of shared epitope were entered as dichotomised independent variables. The variance inflation factor (VIF) was used to check for colinearity between the independent variables. All tests were two tailed and the limit value for significance and elimination in the regression analysis was set at p<0.05.

RESULTS

The baseline characteristics of the patients are given in table 1. After 10 years 168 patients were still being followed and 157 underwent the radiographic evaluation. Of the 24 patients missing the 10 year radiographic evaluation, 15 had died, five had moved from the area, three were excluded because of old age or other diseases, and one refused to undergo the examination. The median (interquartile range) Larsen score at year 5 was 42 (17 to 60), with a range from 0 to 152, and at year 10 it was 54 (28 to 80), with a range from 0 to 162.

At inclusion, 117 of 176 patients (66%) were positive for IgG RF, 138 of 176 (78%) for IgA RF, 139 of 176 (79%) for IgM RF, 140 of 176 (80%) for anti-CCP, and 40 of 182 (22%) for anti-IL1α. The concordance between the various types of RF and anti-CCP antibodies was high. Over half the patients (101 of 176 (57%)) had all three types of RF and anti-CCP. In all, 170 of the patients were genotyped and of these 145 (85%) carried the shared epitope. Fifty four (32%) carried the epitope on both alleles.

Correlations

Table 2 shows the correlation coefficients for the variables assessed. ESR and C reactive protein were strongly correlated. The various subtypes of RF showed relatively strong correlation. Anti-CCP had the highest correlation with IgA RF and in descending order with IgM and IgG RF. ESR, C reactive protein, and IgA RF were correlated significantly but rather weakly with the Larsen score, while COMP showed only a weak correlation with ESR, C reactive protein, and Larsen score at year 5.

Regression analyses

ESR, COMP, IgA RF, anti-CCP, and anti-IL-1α were found to explain 44% of the variance in Larsen score at year 5. The last step of the regression analysis is given in table 3. Increased ESR and COMP levels and the presence of IgA RF and anti-CCP antibodies were significantly associated with more severe joint damage; the presence of anti-IL1α was associated with less severe joint damage. Presence of shared epitope, age at inclusion in the study, and sex did not influence the results after either five or 10 years. The variance inflation factor (VIF) being close to 1 shows that the measurements represent different entities, except for IgA RF and anti-CCP, where a VIF of 1.6 indicates a certain degree of multicolinearity.

Thirty two per cent of the variance in Larsen score year 10 could be explained by C reactive protein and anti-CCP. Anti-IL1α, IgA RF, and COMP were not significantly associated with Larsen scores at year 10 (p values 0.06, 0.11, and 0.14, respectively). The last step of the regression analyses is given in table 4.

To be able to use these findings to predict the development of joint damage in individual patients, an interpretation of the results from the analysis with the Larsen score for year 5 as the dependent variable is given in table 5A and for year 10 in table 5B. Importantly, if more than one risk factor was present in the same patient, the effect was additive.

Owing to missing data in one or more variables, only 107 of 177 patients at year 5 and 119 of 168 at year 10 could be included in the regression analyses. At year 5, patients not included in the analyses had higher C reactive protein values (p<0.05) and tended to be IgA RF positive less often (p = 0.09). At year 10, patients not included in the analyses were older (p<0.01) and more often men (p<0.05). An increased mortality among older men contributed to this.⁴⁶ No other significant differences between assessed variables were found at any time point.

DISCUSSION

In this study we have shown that laboratory variables explained about half the variance in the Larsen score after five years. Serum COMP, anti-CCP, and anti-IL1α added some prognostic information to ESR and IgA RF, which were the strongest predictors of joint damage in the hands and feet after five years. After 10 years, anti-CCP was the only significant predictor, together with C reactive protein. We found that inflammatory variables were related to joint destruction, in agreement with many earlier reports.^3,4,43 The various RF subtypes were intercorrelated. However, as in many previous studies, IgA RF turned out to have the best prognostic power after five years. The lack of significant association for IgA RF at year 10 could be caused by the existing colinearity between anti-CCP and IgA RF. We have earlier reported that IgG RF showed the strongest relation to joint damage after two years of follow up in this cohort.⁴⁷ The discrepant findings might be explained by the shorter follow up period, different assessment methods, and fewer patients.

Furthermore, the presence of anti-CCP predicted a more destructive form of rheumatoid arthritis after both five and 10 years. The 5 year findings are in concordance with other studies.^19–21 The current study extends these results by showing that this also holds true in a longer perspective. Anti-CCP appears early in the disease course and has a high specificity for rheumatoid arthritis.⁴⁸ Anti-CCP might therefore be a valuable additional predictive tool in the management of this disease. The strong concordance between RFs and anti-CCP means that very few patients lacking RF will be positive for anti-CCP. This limits the clinical value of using both antibodies.

We also found that COMP, a marker reflecting another aspect of the disease process, had additive predictive value. COMP shows very weak associations with both inflammatory indices and autoantibodies. This protein was originally isolated and characterised as a cartilage matrix component but has subsequently also been found in other tissues—for example, synovium, tendon, and meniscus (for references, see Larsson et al¹⁴). However, numerous studies in human and experimental arthritis clearly indicate that changes in serum levels of COMP are related to processes in cartilage.^10,12,14 Thus serum COMP is a potential marker of changes in cartilage turnover, and increased serum levels may occur early in the course of rheumatoid arthritis as a sign of cartilage involvement. The present study lends further support to this hypothesis by showing that the serum concentration of COMP at inclusion was predictive of future small joint damage. The original COMP assay and the one used in the present study measure both intact and fragmented COMP, which limits the possibility of discriminating between matrix synthesis and degradation.⁹ Refinement of the technology enabling specific measurement of select fragments will most probably increase the prognostic value of COMP. The observations regarding COMP in this study are at variance with findings in a previous study of ours, where COMP was not found to be predictive of small joint damage in a subset of patients from the early rheumatoid arthritis cohort.⁴⁹ A possible explanation for this discrepancy is that a different COMP assay with a polyclonal antibody was used.⁹

The naturally occurring anti-IL1α antibodies were found to predict less severe radiographic outcome in this cohort. This finding is in agreement with some previous studies,^34–37 but it may be apparent only with prolonged disease, because a two year follow up in an early rheumatoid cohort failed to show this relation.⁵⁰ This latter study did not correct for the presence of RF and anti-CCP, which may have contributed to the discrepant results.

The presence of shared epitope in single or double dose did not provide any predictive information on the development of joint damage in this study. We have earlier found that the presence of shared epitope was not related to functional outcome.⁴² These findings are in agreement with several other reports indicating that genetic screening should not be done routinely in early arthritis clinics.⁴⁰

The variance of radiographic damage that could be explained in our multivariate models was 44% after five years and 32% after 10 years by baseline laboratory assessments. Even if the degree of explanation is fairly moderate our findings can be used as guidance when trying to predict the risk of joint damage in individual patients, as shown in table 5. So far no other better predictors have been recognised. The prognostic value of our laboratory assessments was somewhat less apparent for the 10 year outcome. One explanation for this might be that about 75% of the joint damage had already occurred after five years.⁴³ Furthermore, the therapeutic strategy used in this cohort changed over time. During the first five years only 7% of the patients (n = 13) were treated with methotrexate, compared with 27% (n = 46) during the following five years. Although this figure is low compared with the current strategy, it could still influence the results, as methotrexate has been shown to modify progression of joint damage.⁵¹ Modification of the disease course by effective treatment strategies will increasingly hamper the evaluation of new potential prognostic markers in the future.⁵² Thus our cohort of conservatively treated patients (in comparison with current standards) will be valuable for such studies.

Conclusions

In this prospective early rheumatoid arthritis cohort, ESR/C reactive protein, COMP, and the presence of IgA RF, anti-CCP, and anti-IL1α antibodies assessed at presentation provided prognostic information on future joint destruction. The laboratory measures used were selected to reflect different aspects of the disease process. The combination of markers was found to yield additive prognostic information.