Periarticular osteoporosis and joint erosion are both known as radiographic hallmarks of rheumatoid arthritis (RA).1 Evidence from animal2 3 and human studies4 support that erosions and osteoporosis are caused by an increased activation of osteoclasts. Hand bone loss has been shown to take place in early RA,5 even in the undifferentiated stage of the RA disease process.6 7 On hand radiographs periarticular osteoporosis has been shown to precede the development of erosions.8

Measures of quantitative hand bone loss eg, by dual energy x ray (DXA) and digital x ray radiogrammetry (DXR) in early RA have been proposed as an outcome measure for bone involvement.9^–11 However, there is a lack of data on the relationship between hand bone loss and radiographic joint damage. Cross-sectional studies have shown a moderate correlation between low hand bone mineral density (BMD) and radiographic damage10 12^–17 and two small longitudinal studies have indicated that early hand bone loss may predict subsequent radiographic joint damage.18 19

Thus, the objective of this study was to examine if cortical hand bone loss in early RA, as assessed by DXR in the first year of follow-up, could predict radiographic joint damage at 5-year and 10-year follow-up.

MATERIALS AND METHODS

Study design and study population

The current analysis is a part of the European Research on Incapacitating Disease and Social Support (EURIDISS) longitudinal observational study. The patients and methods have previously been described in detail elsewhere.20 21 In short, all patients had disease duration of maximum 4 years at inclusion, were aged 20–70 years and fulfilled the American College of Rheumatology (ACR) criteria for RA.1 Clinical, laboratory and radiographic data were collected at baseline and at 1, 2, 5 and 10 years. During the observation period, patients were treated according to clinical judgement by their rheumatologist. Sera from the baseline visit were stored at −70 degrees for later analysis of micro-C-reactive protein (CRP) (Dade Behring, Newark, New Jersey, USA), anti-cyclic citrullinated peptide (CCP) (ELISA, Inova Diagnostics, San Diego, California, USA) and rheumatoid factor IgM (IgM RF) (in-house ELISA).22

Patients with hand radiographs at baseline, 1-year follow-up and either 5-year or 10-year follow-up were included in the present analyses. The number of patients with baseline radiographs available for scoring were 163, and 15 of these were excluded due to missing radiographs at 5 and 10 years. Of the remaining 148 patients, baseline radiographs from 5 patients could not be analysed for DXR-BMD because the radiographs were underexposed and 7 patients were excluded at the 1-year follow-up (5 radiographs were missing, 1 was underexposed and 1 patient had surgical material in the metacarpal bone). A total of 136 patients were included in the final analyses.

Missing radiographic scores at 5 years (n = 11) were replaced by a projected score based on the radiographic progression from baseline to the 10-year assessment (horizontal imputation). Any 10-year missing data (n = 18) were replaced by last observation carried forward23 to avoid overestimation of the radiographic damage. Missing DXR-BMD values were not imputed. To test the robustness of the results, all analyses were repeated without imputation.

Radiographic analyses

The radiographs were scored according to the van der Heijde modification of the Sharp score (vdH Sharp score),24 and were read in known time order by one experienced reader.21 In all, 16 joint areas for erosions (0–5) and 15 for joint space narrowing (JSN) (0–4) were evaluated in each hand, and the maximum score was 280. Conventional radiographs were available at baseline, 1, 2 and 5 years, while radiographs at 10 years were digitised. Scoring of digitised and conventional radiographs have been shown to yield similar results.25

Analyses were performed for continuous and dichotomised data. Cut-off for an important increase in vdH Sharp score was defined as an increase of 1 unit per year (ie, 5 units at the 5-year and 10 units at the 10-year follow-up assessment).22 The smallest detectable change (SDC) in radiographs read in known time order has been found to be 2.9 units.26 The conventional radiographs were acquired by a Siemens Multix Polymat equipment (film: AGFA Curix (AGFA, Mortsel, Belgium); film focus distance (FFD): 100 cm; x ray tube voltage: 55 kV; exposure dose: 6 mAs). The digital radiographs were acquired by an AGFA ADC Compact (computed radiography) and a Siemens (Erlangen, Germany) tube (AGFA ADCC HR image plate, x ray tube voltage 50 kV, FFD 100 cm, exposure dose 5 mAs).

Cortical hand bone density

The conventional hand radiographs used for radiographic scoring of joint damage was also used for hand BMD measures at 1, 2 and 5 years. Cortical hand BMD was measured by DXR Pronosco X-posure system, V. 2.0 (Sectra, Linköping, Sweden),27 which is a development of the traditional technique of radiogrammetry.28 On hand radiographs the computer automatically recognises regions of interest around the narrowest part of the second, third and fourth metacarpal bone. In each region, cortical thickness, bone width and porosity is measured 118 times per cm. The BMD estimate is defined as: BMD-DXR = c×VPA_comb×(1–p), were c is a constant (empirically determined so DXR-BMD on average was equal to the mid-distal forearm region of the Hologic QDR-2000 densitometer; Hologic, Bedford, Massachusetts, USA); VPA is volume per area and p is porosity. This method has been described in detail elsewhere.27 29

Precision was calculated from 28 healthy individuals who underwent duplicate hand radiographs with repositioning of the hand between each measure. The coefficient of variation (CV%) was found to be 0.28%, and least significant change (LSC) was 0.78%30

The LSC was used as cut-off to define a DXR-BMD loss exceeding the measurement error on the individual level. We applied the mean values of both hands to avoid bias regarding dominant and non-dominant hand, and this approach has been shown to improve precision.19

Statistics

The statistical analyses were performed using SPSS V.14 (SPSS, Chicago, Illinois, USA).

Because of skewed data, non-parametric tests were used. Hand BMD loss was tested as a continuous and a dichotomised variable. The following methods were used: Spearman correlation, group comparisons using the Mann–Whitney U test and multivariate analyses (linear and logistic regression analyses).

A linear regression model was developed to investigate if absolute hand BMD loss during the first year could predict subsequent radiographic outcome (dependent variable). The model was adjusted for age and gender as well as for other potential predictors of radiographic damage: anti-CCP (cut-off: 25 U/ml), IgM RF (cut-off: 25 U/ml), CRP, Health Assessment Questionnaire (HAQ) and baseline vdH Sharp score. Separate models for 5 and 10 years were developed.

In a logistic regression model patients were dichotomised as progressors and non-progressors of radiographic damage with an annual increase of 1 unit as cut-off value (dependent variable) and stratified into patients with and without DXR-BMD loss with LSC as the cut-off value (independent variable). This model was adjusted for the variables that were statistically significant in the linear regression model together with age and gender. For CRP, we used a cut-off of 10 mg/litre. From the logistic regression model a probability score was calculated to assess the risk of radiographic progression dependent on the combination of hand BMD loss, early radiographic damage and anti-CCP. All tests were two-sided and conducted at the p = 0.05 significance level.

Ethics and legal aspects

The study was approved by the regional committee for ethics and medical research and data collection was approved by the Data Inspectorate.

RESULTS

Patient characteristics at baseline are shown in table 1. No significant differences were observed between the 136 examined patients in the present analyses and the 102 patients who were excluded. At 5 years none of the patients were using anti-tumour necrosis factor (TNF)α therapy and 55% were using disease-modifying antirheumatic drugs (DMARDs); at 10-year follow-up the corresponding figures were 12% and 50%, respectively.

BMD change and radiographic damage

The median loss in hand DXR-BMD expressed in absolute values (g/cm²) were 0.009 after 1 year, 0.016 after 2 years and 0.031 after 5 years. The percentage changes of hand DXR-BMD are displayed as a cumulative probability plot in fig 1. The hand bone change was similar in males and females (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Cumulative probability plot of digital x ray radiogrammetry bone mineral density (DXR-BMD) percentage change at 1-year, 2-year and 5-year follow-up in patients with rheumatoid arthritis (RA).

The mean (median) change in the vdH Sharp score were 3.6 (1.0), 7.1 (3.0), 15.5 (10.0) and 26.4 (16.0) units after 1, 2, 5 and 10 years, respectively. The correlation coefficient (r) between change in DXR-BMD and vdH Sharp score increased from −0.35 (p<0.01) at the first year, to −0.47 (p<0.01) at the second year and −0.56 (p<0.01) at the fifth year.

On the individual level, 67% of the patients had a hand BMD loss exceeding LSC in the first year of follow-up. The proportions of patients with radiographic progression in the hands (defined as 1 unit per year) were 46% after 1 year and 60% after 5 and 10 years. The radiographic progression (median values) was significantly higher in patients who lost DXR-BMD vs patients who did not lose DXR-BMD at 5 (12 vs 2, p = 0.001) and 10 years (22 vs 4 p = 0.002). These results are shown as cumulative probability plots in fig 2.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Change in van der Heijde Sharp score at 5 years (A) and 10 years (B) in patients with rheumatoid arthritis stratified for hand bone mineral density (BMD) loss at 1 year (cut-off >least significant change (LSC)). Presented as a cumulative probability plot were each symbol represent a patient (hand BMD loss n = 91, no hand BMD loss n = 45). (A) 5-year change in vdH Sharp score dependent on DXR-BMD loss or not at 1 year. p = 0.001 (group comparison Mann–Whitney U). (B) 10-year change in vdH Sharp score dependent on DXR-BMD loss or not at 1 year. p = 0.002 (group comparison Mann–Whitney U).

Multivariate linear regression models

Change at 1 year in DXR-BMD was a significant and independent predictor of vdH Sharp score at 5 and 10 years, adjusting for other relevant factors as baseline vdH Sharp score, anti-CCP, IgM RF, baseline micro-CRP, HAQ-score, age and gender (table 2). In this model a loss of 0.1 g/cm² would on average give an increase in vdH Sharp score of 20 units after 5 years and 28 units after 10 years (B values, not shown). Other significant predictors of radiographic damage were anti-CCP, radiographic damage at baseline and inflammatory activity assessed by CRP (the latter only at 5 years) (table 2).

Multivariate logistic regression models

Logistic regression models were created to see whether DXR-BMD loss could be used as a predictor in a clinical situation. The models were adjusted for the baseline variables that turned out to be statistically significant in the linear regression model, ie, baseline vdH Sharp score (cut-off 1 unit), anti-CCP (cut-off 25 U/ml) and CRP (cut-off 10 mg/litre) together with age and gender.

Loss of hand DXR-BMD exceeding the LSC during the first year of follow-up was an independent risk factor for radiographic progression at 5 and 10 years with an odds ratio (95% CI) of 3.5 (1.4 to 8.8) and 3.5 (1.4 to 8.4), respectively (table 3).

Radiographic damage at baseline and positive anti-CCP were also significant risk factors for subsequent radiographic damage (table 3). For these three predictors the probability for radiographic damage was calculated based on the odds ratios (OR). The probability of developing radiographic progression (for 5 years) was calculated as follows: log (p/1–p) = −1.61 (constant)+1.26 DXR-BMD loss+1.60 Radiographic damage at baseline+1.44 anti-CCP positive, where p is the probability and each variable is entered as a categorical variable (0 or 1) weighted by the importance of each variable. From this algorithm the probability of radiographic damage at 5 years could be calculated as 94% for patients with bone loss, radiographic damage and positive anti-CCP and as 17% for a patient with none of these risk factors. These results, together with the 10-year data are depicted in fig 3. For the 5-year and 10-year data, hand bone loss adds additional information to the subsequent risk for radiographic damage.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Probability of radiographic progression in patients with rheumatoid arthritis at 5 (A) and 10 years (B). The probability (%) is based on the odds ratio from the logistic regression model according to the combined presence of hand bone loss at 1 year, radiographic damage and anti-cyclic citrullinated peptide (CCP) at baseline.

Robustness

The multivariate models were also tested without imputation of missing values. The number of available patients at 5-year and 10-year follow-up was 126 and 118, respectively. In the linear regression model a loss of 0.1 g/cm² DXR-BMD would give an increase in vdH Sharp score of 19.8 units (p = 0.01) at 5 years and 23.2 units (p = 0.07) at 10 years. In the logistic regression model the OR (95% CI) for radiographic damage among those with hand bone loss were 3.0 (1.2 to 7.7) at 5 years and 2.9 (1.1 to 7.4) at 10 years. All group comparisons and correlation analyses remained statistically significant.

DISCUSSION

This study shows for the first time that hand bone loss is an independent predictor of progression of radiographic joint damage and that the predictive power is comparable to biomarkers that are well known predictors of radiographic joint damage as anti-CCP and CRP.

Algorithms have been proposed to identify patients with poor prognosis by the presence of various predictors.22 31 32 Up to now early hand bone loss as a risk factor has not been included in any of these algorithms. The two characteristics of the DXR-BMD method, high sensitivity to change29 and ability to predict joint damage, suggest that hand DXR-BMD may be used in prediction models of poor outcome in patients with early RA. In the multivariate analysis and in our algorithm based on presence or absence of the three risk factors, anti-CCP, erosive disease at baseline and 1-year change in bone loss, we showed that hand DXR-BMD gives additionally information to the other risk factors for predicting radiographic progression.

As illustrated in fig 3, a patient with radiographic damage at baseline, but negative anti-CCP and no hand bone loss during the first year will have a 34% probability of an important radiographic progression after 10 years, whereas a patient with erosions and hand BMD loss has a probability of 64%. With additional presence of anti-CCP this probability increases to 87% after 10 years. DXR-BMD has most additional predictive value in patients with only one risk factor (either baseline radiographic damage or anti-CCP). These data illustrate the potential of DXR-BMD as a predictive tool, but we find it important to recommend that our results also should be confirmed in patients with very early RA (ie, in patients without radiographic erosions).

The fact that bone loss in early RA occurs more rapidly than the development of erosions is not only supported by the multivariate analysis but also from correlation analysis. The correlation coefficient between hand bone change and radiographic damage increased from −0.35 at the first year to −0.56 at the fifth year. However it is important to emphasise that erosions are pathognomonic for bone involvement in RA, while bone loss may also occur in normal individuals. This limitation may, however, be a minor problem in studies with short observation time.19 DXR-BMD values for the 10-year follow-up are not presented for methodological reasons as these radiographs were digitised and DXR-BMD assessed from digital printouts has been shown to have a poorer precision than conventional radiographs.33 Further, the precision regarding direct analysing of digitised DXR-BMD has not been fully investigated.

One limitation of this study is that data on generalised bone loss was not available. From the literature we know that generalised bone loss has also been found to be associated with disease activity34 and radiographic damage.35 36 Few studies have compared periarticular and generalised osteoporosis among patients with RA.7 13 17 37 Hand bone loss in early RA has been shown to occur more rapidly than bone loss in hip and spine.7 37 Radiographic joint damage has been shown to be more strongly correlated with low hand DXR-BMD than DXA-BMD at hip and spine.13 17 In a randomised placebo controlled trial among patients with early RA, use of prednisolone reduced hand bone loss.38 These studies suggest that the effect of inflammation on hand bone in RA may be greater than the effect on other bones (eg, spine and hip). In addition the precision of BMD hand, and particularly DXR-BMD, is better than the precision of DXA measurements of the hip and spine. This difference also indicates that a longer observation period may be required to detect a true loss of bone in hip and spine.19 29

This study examined the relation between early hand bone loss and radiographic progression at 5 and 10 years. Horizontal imputations were used for the missing 5-year data and last observation carried forward used for the missing 10-year data. This approach was chosen because the literature is not consistent as to whether radiographic damage increases in a linear fashion during the first year of the disease and declines afterwards,39 or increases linearly over a period of 10 years.40 41 By using this conservative imputation the radiographic damage was not overestimated. No imputations regarding DXR-BMD were performed. All analyses were also repeated without imputations and hand bone loss turned out to be a significant predictor of radiographic outcome in all analyses except for the linear regression model at 10 years, which showed a borderline significance.

Although DXR-BMD shows convincing result as a predictor for radiographic damage, there are a few patients with no DXR-BMD bone loss who have a high increase in radiographic damage (fig 2). The reason for this is not clear. One explanation may be that RA is a heterogenic disease. However, this emphasise the importance of further research on predictors for radiographic damage.

In conclusion, we have shown that hand bone loss measured by DXR-BMD is an independent predictor of subsequent radiographic damage in patients with RA. Our findings support that quantitative hand bone measures may be a complimentary approach in the study of changes in bone involvement in RA and may be an important tool in the daily clinical work together with anti-CCP, markers of inflammation and radiographs to identify patients at high risk of developing progression in radiographic joint damage.