Despite better disease suppression with combination disease-modifying antirheumatic drugs (DMARDs), some patients with rheumatoid arthritis (RA) have progressive erosive disease. The objective of this study was to determine whether hand bone mineral density (BMD) loss in the first 6 months of treatment indicates increased risk of erosions at 12 months.
Patients with DMARD-naive early RA receiving treat-to-target therapy were studied (n = 106). Hand BMD was measured at baseline and 6 months by dual x-ray absorptiometry. Hand and feet radiographs were performed at baseline and 12 months and scored using the van der Heijde modification of the Sharp method. A K-means clustering algorithm was used to divide patients into 2 groups: the BMD loss group or the no loss group, according to their absolute change in BMD from baseline to 6 months. Multiple regression analysis (hurdle model) was performed to determine the risk factors for both erosive disease and erosion scores.
Hand BMD loss at 6 months was associated with erosion scores at 12 months (P = 0.021). In a multiple regression analysis, hand BMD loss (P = 0.046) and older age at onset (≥50 years; P = 0.014) were associated with erosive disease, whereas baseline erosion scores (P = 0.001) and anti–cyclic citrullinated peptide (P = 0.024) were correlated with erosion severity/progression.
In RA patients receiving treat-to-target therapy, early hand BMD loss could identify patients who are at risk of developing erosive disease at 12 months, potentially allowing intensification of treatment to prevent erosive damage.
Early intensive treatment of rheumatoid arthritis (RA) within the “window of opportunity” leads to better patient outcomes with fewer patients developing erosive disease and subsequent disability (). Treatment of RA in its very early phase (first 12 weeks) is the best predictor of no erosion progression at 12 months (), but despite better disease suppression with disease-modifying antirheumatic drugs (DMARDs), some patients have progressive erosive disease ([3-5]). Early identification of patients at risk of progressive joint damage would allow them to be selected for more intensive therapy, including earlier addition of biologic agents. Markers of disease activity in early disease do not predict erosions at 12 months ([6, 7]), therefore other measures are needed to identify especially at-risk patients early in their disease course before they develop joint damage.
The earliest radiographic change in RA, prior to the appearance of erosions, is periarticular osteopenia ([8, 9]). Hand bone mineral density (BMD) loss (), measured by dual x-ray absorptiometry (DXA), can differentiate between RA and other inflammatory joint diseases as well as noninflammatory arthritis in patients presenting with early inflammatory joint symptoms (). Hand BMD loss in the first year after RA diagnosis, measured using digital x-ray radiogrammetry (DXR), has been shown to be associated with subsequent joint destruction at 4 years (), 5 and 10 years (), and over a 20-year followup period (). However, DXR requires specialized software to calculate an estimate of BMD in the metacarpals based on digitized radiographs and, although a useful measure in the research setting, it is not readily available in clinical practice. Furthermore, an inverse correlation between baseline hand BMD measured by DXA and radiographic progression at 12 months has also been reported (). Measurement of hand BMD using DXA has been shown to be a highly reproducible measure of periarticular osteopenia in early RA () and can be made available readily utilizing instruments already available in most clinical centers treating RA.
This study looks at the possibility that hand BMD loss measured by DXA in the first 6 months after diagnosis could predict early erosive change.
Box 1. Significance & Innovations
Previous studies have demonstrated that bone mineral density (BMD) loss in early rheumatoid arthritis (RA) is associated with subsequent joint destruction.
These earlier studies measured BMD with digital x-ray radiogrammetry, which requires specialized software to calculate an estimate of BMD in the metacarpals based on digitized radiographs. Although it is a useful measure in the research setting, it is not readily available in clinical practice.
Measurement of hand BMD using dual x-ray absorptiometry (DXA) has been shown to be a highly reproducible measure of periarticular osteopenia in early RA and can be made available easily in most clinical centers treating RA.
This study demonstrates that in a cohort of disease-modifying antirheumatic drug (DMARD)–naive early RA patients treated with response-driven DMARD therapy, early BMD loss measured using DXA may help identify patients who are at special risk of developing erosive disease at 12 months.
PATIENTS AND METHODS
A total of 106 consecutive patients with recent-onset RA treated at the Royal Adelaide Hospital Early Arthritis Clinic from September 2001 to October 2008 were included in the study. Ethics approval was obtained from The Royal Adelaide Hospital Research Ethics Committee. All patients gave written informed consent for their clinical data to be collected and used for research purposes. This cohort of patients and their treatment protocol has been described in detail previously (). Briefly, patients were ages >18 years and their RA diagnosis was based on the revised 1987 American College of Rheumatology (ACR) criteria (). Inclusion criteria were symptoms of polyarthritis of more than 6 weeks and less than 24 months, a swollen joint count ≥3, a tender joint count ≥6, an erythrocyte sedimentation rate (ESR) >28 mm/hour, and/or a C-reactive protein (CRP) level >10 mg/dl. Patients were DMARD naive at enrollment and treated with a combination of methotrexate, sulphasalazine, and hydroxychloroquine with dose adjustments according to disease activity and toxicity using a treat-to-target strategy.
Baseline laboratory measures included ESR, CRP level, rheumatoid factor (RF), anti–cyclic citrullinated peptide (anti-CCP), and HLA–DR–binding groove shared epitope. A 28-joint count Disease Activity Score (DAS28) based on the ESR was calculated at baseline, 6 months, and 12 months. A single assessor determined the tender and swollen joint counts. Smoking status, sex, age at onset of RA, and exposure to bone-modifying therapy such as vitamin D, calcium, and bisphosphonates was recorded.
Bone density measurement
BMD of the hand was determined using DXA (LUNAR, model DPX-IQ). Measurements were taken at baseline and 6 months. The change in BMD (ΔBMD) between baseline and at 6 months was calculated.
Bone density precision and accuracy
From 2001 to March 2006 the coefficient of variation percentage (CV%) for the spine phantom was 0.56%. The spine phantom was changed in March 2006 and the new CV% until December 2008 was 0.48%, indicating that the DPX-IQ machine was accurate and quality control standards were met throughout the duration of the study. In regard to precision, the least significant change (LSC) for the total hand (using a 95% confidence interval [95% CI]) in healthy controls (n = 4 with 3 scans each) was 0.0058.
Radiographs of the hands and feet were obtained at baseline and 12 months. They were scored in pairs (baseline and 12 months) using the van der Heijde modification of the Sharp method (), with hands and feet scored separately. Following an initial calibration session, each set of radiographs was scored by 2 of 3 assessors (RJB, CLH, or SMP). All assessors were blinded to the patient's identity and the time sequence of the radiographs. In the case of discrepancy, a final score was achieved by consensus. Films were assessed for both erosions and joint space narrowing.
ΔBMD values were clustered into 2 groups using the K-means clustering procedure in Statistica, version 6.1 (Statsoft). Cluster 1 was assigned as “no loss” and cluster 2 as “BMD loss” between baseline and 6 months.
Multiple logistic regression analysis was used to determine whether BMD loss was associated with conventional risk factors for bone loss. The nonparametric 2-sample Wilcoxon test was used to compare erosion scores between the 2 groups. The chi-square test was used to compare the proportion of patients with erosive disease and erosion progression between each group.
A hurdle model regression analysis was carried out to determine predictors of erosion scores at 12 months. This enabled analysis of erosion scores at 12 months as a mixture of 2 distributions: 1) the “zero” component of the model, which reflects the absence or presence of erosive disease (scores 0 versus ≥1) and 2) the “count” component, which reflects the erosion score in patients with erosive disease (score ≥1). Hurdle models were estimated using R, version 2.12.1 (), and the pscl package ([20, 21]), with the binomial distribution (logit link) for the zero component and the negative binomial distribution (overdispersed Poisson distribution, with log link) for the zero-truncated count component.
Baseline characteristics of the RA patients are shown in Table 1. There were 56 of 106 (53%) patients in the no loss group and 50 of 106 (47%) patients in the BMD loss group. Baseline hand BMD measures were not significantly different between the 2 groups (P = 0.33). The mean ± SD change in BMD (between 0 and 6 months) for the no loss group was 0.006 ± 0.014 gm/cm2 and for the BMD loss group was −0.012 ± 0.009 gm/cm2. All patients in the BMD loss group had a ΔBMD ≤ −0.004 gm/cm2. BMD loss was associated with the conventional risk factors for bone loss of age (odds ratio [OR] 1.04 [95% CI 1.01–1.07], P = 0.013), female sex (OR 3.2 [95% CI 1.2–8.4], P = 0.020), and smoking (OR 2.4 [95% CI 1.0–5.6], P = 0.041). These associations validate the statistical partitioning of the 2 groups.
Table 1. Baseline characteristics of patients included in the study (n = 106)*
Values are the number/total number (percentage) unless indicated otherwise. IQR = interquartile range; anti-CCP = anti–cyclic citrullinated peptide; BMD = bone mineral density; DAS28 = Disease Activity Score in 28 joints.
Age at symptom onset, mean ± SD years
57 ± 15
Duration of polyarthritis, median (IQR) weeks
Rheumatoid factor positivity
Shared epitope positivity
No loss group
BMD loss group
DAS28 score, mean ± SD
5.9 ± 1.1
3.7 ± 1.5
3.4 ± 1.7
Hand BMD at baseline, median (IQR)
Erosion scores at baseline did not correlate significantly with subsequent categorical BMD loss (P = 0.15) (Figure 1A). However, when compared with patients in the no loss group at 6 months, patients with hand BMD loss had higher erosion scores (P = 0.021) (Figure 1B) and greater increases in erosion scores (P = 0.033) (Figure 1C) at 12 months. Comparable results were observed for the proportion of patients with erosive disease, or erosion progression, over 12 months. There was no difference between the no loss and BMD loss groups in the proportion of patients with ≥1 erosions at baseline (P = 0.15) (Figure 2A), yet a higher proportion of patients with hand BMD loss at 6 months had ≥1 erosions at 12 months (24 of 50 [48%] patients versus 14 of 56 [25%]; P = 0.013) (Figure 2B) and also a higher risk of erosion progression (17 of 50 [34%] patients versus 9 of 56 [16%]; P = 0.032) (Figure 2C).
A hurdle model regression approach was employed to investigate the context of hand BMD loss in relation to other potential predictor variables for erosion scores at 12 months. This approach enabled the simultaneous yet separate assessment of risk factors for initiation of erosive disease (zero component) and risk factors for erosion severity, as measured by erosion scores in patients with erosive disease (count component). An initial univariate screening process was employed for potential predictor variables; however, baseline erosion score was always included in the “count” component of the model. The results of the univariate analysis for predictors of erosion scores at 12 months (with a nominal P value of < 0.10) are reported in Table 2. Interestingly, predictor variables for erosion scores at 12 months were different for the 2 components of the model. Candidate predictors of erosive disease at 12 months were age at onset ≥50 years (P = 0.005), lower baseline BMD (P = 0.034), and categorical BMD loss at 6 months (P = 0.015). Baseline erosion score was the most significant predictor of erosion scores (in erosive patients) at 12 months (P = 0.001). After adjustment for baseline erosion scores, anti-CCP positivity also contributed independently to the erosion score in erosive patients (P = 0.024). The analysis also suggests that disease control in the first 6 months (as measured by the change in DAS scores and CRP levels, or CRP levels at 6 months) may influence erosion scores at 12 months, although these variables did not reach significance at the nominal level of 0.05 (Table 2). Other variables tested, which were not associated with erosive disease at 12 months (P > 0.10), were sex, current smoking status, duration of polyarthritis, RF and shared epitope status, baseline DAS or CRP level or BMD, and DAS scores or responder status at 6 months.
Table 2. Univariate predictors (with a nominal P < 0.10) for erosion scores at 12 months by hurdle model regression analysis*
The hurdle model analyzes erosion scores as a mixture of 2 distributions: the zero component is dichotomous and partitions around the presence/absence of erosive disease (susceptibility), and the count component is the distribution of erosion scores (severity) in patients with erosive disease. While variables were screened individually, baseline erosion scores were included as a predictor in all models. NA = not assessed; BMD = bone mineral density; anti-CCP = anti–cyclic citrullinated peptide; DAS28 = Disease Activity Score in 28 joints; CRP = C-reactive protein.
aLog transformed prior to analysis because of skewness. Identified predictor variables were significant in one or the other components of the model, but not both.
The univariate predictor variables identified in Table 2 were subsequently combined into a final multiple regression model, and the best model was selected according to the Akaike Information Criteria. Baseline BMD, although not a significant univariate predictor (P > 0.5 in both components of the univariate model), was also included in the multiple regression analysis because of its potential relevance to BMD loss. In this model (Table 3), categorical BMD loss at 6 months (P = 0.046), older age at onset (≥50 years; P = 0.014), and lower baseline BMD (P = 0.072) were retained as independent predictors of erosive disease at 12 months.
Table 3. Multiple regression analysis for predictors of erosion scores at 12 months by hurdle model regression*
Predictor variables identified in a univariate analysis were retained in the multiple regression model on the basis of the Akaike Information Criteria. NA = not assessed; BMD = bone mineral density; anti-CCP = anti–cyclic citrullinated peptide; CRP = C-reactive protein.
aLog transformed prior to analysis because of skewness.
Predictions from the fitted model (estimated at the mean baseline BMD) suggest that approximately 70% of patients age ≥50 years with evidence of hand BMD loss may develop erosive disease within 12 months of diagnosis. In addition to the baseline erosion score (P = 0.001), anti-CCP positivity (P = 0.024) and CRP level at 6 months (P = 0.055) were retained as independent predictors for erosion scores at 12 months in patients with erosive disease.
The analysis demonstrates that erosive disease is clearly a multifactorial process in early RA patients. A further implication is that appropriate algorithms for prediction of patients at risk of erosion “progression” may differ according to whether the patient is, or is not, erosive at baseline. For example, the analysis indicates for patients with erosive disease at baseline that baseline erosion scores, anti-CCP positivity, and inadequate disease control will each contribute to progression of erosion scores over 12 months, and BMD loss has no apparent utility in this regard. However, BMD loss may have utility in identifying “progression” in patients who are initially erosion free, but who are at risk of developing erosive disease. This latter group of patients can be expected to carry multiple risk factors for erosive disease as well as for erosion severity. While numbers were small, and only 9 patients developed erosive disease between baseline and 12 months, this indeed appeared to be the case as all 9 patients had at least 2 risk factors in terms of age ≥50 years, BMD loss, and anti-CCP positivity. Individually, 8 of 9 patients were age ≥50 years (with the remaining patient age 46 years), 7 of 9 were classified in the BMD loss group, and 9 of 9 were positive for anti-CCP.
Despite early and intensive treatment, some patients with RA still develop progressive joint damage. Safe and cost-effective measures are needed to guide decision making with regard to dose escalation of DMARDs and possible introduction of biologic therapy in these patients. Ideally, such measures would help clinicians reach a balance between adequate treatment of those at risk of progressive joint damage, while avoiding overtreatment and unnecessary expenditure for those who would naturally run a more benign disease course. Approximately 17–21% of patients with RA remain erosion free over a 10-year time course ().
Radiographic damage at baseline appears to be one of the most important predictors of subsequent joint damage ([2, 7]), but with the strong focus on early diagnosis fewer patients have erosions at baseline. This is evident in our cohort where only 27% of patients had radiographic erosions at baseline. Magnetic resonance imaging and ultrasound studies suggest that conventional radiographs underestimate the prevalence of erosions. Even so, the majority of patients who develop erosive disease may demonstrate radiographic evidence of erosions within the first 12 months, as with one study that followed 55 patients with RA over a 3-year period and reported that 74% of those who developed erosions (64% of the whole cohort) had radiographically evident erosive disease within the first 12 months (). For many patients, radiographic erosions will only appear after the so-called “window of opportunity” when therapy is likely to be most effective. The ideal biomarker would identify those patients at risk of joint damage prior to the development of radiographically evident erosive disease.
In this study we evaluated BMD loss as a predictor of erosions at 12 months, utilizing a prospective study design in RA patients with very recent disease onset (median duration of polyarthritis 16 weeks). These patients were all DMARD naive at diagnosis, and subsequent treatment followed a standardized treat-to-target algorithm that was informed by systematic assessments of disease activity. We observed that BMD loss measured at 6 months was associated with erosive disease at 12 months postdiagnosis. Upon further analysis, it became evident that erosive disease is a multifactorial process with different factors influencing susceptibility and severity. BMD loss together with age was associated with increased susceptibility to erosive disease, and therefore BMD loss may have the most utility in the identification of erosion-free patients at risk of developing erosive disease. In contrast, anti-CCP status and ongoing inflammation (as measured by CRP levels at 6 months) contribute to erosion scores and their progression in patients with erosions at baseline.
Several previous studies have identified BMD loss as a predictor of long-term joint damage over periods of up to 20 years ([12-14]). More recently, Bejarano and coworkers () also demonstrated an association between DXA-measured hand BMD loss in the first year and long-term radiographic progression (6.4-year followup) in a smaller (n = 64) early RA cohort. However, they concluded that hand BMD loss did not provide any additional information over and above the presence of erosions on baseline radiographs of the hands and feet. Given that neither variable was significantly associated with erosion progression in a multiple logistic regression model, the suggestion is that baseline erosions and hand BMD loss were confounded (i.e., associated) in this study and/or there was a lack of statistical power to properly determine the independent contributions of both variables to erosion progression. We did not observe a relationship between baseline erosions and BMD loss in our study and it is probable that this may reflect the relatively short median disease duration at the baseline of 16 weeks.
Haugeberg et al () studied 74 patients with undifferentiated hand arthritis of less than 12 months' duration, treated according to a dose escalation protocol. More than 50% were found to have noninflammatory joint disease, making this study population quite different from our population of early RA patients fulfilling ACR criteria. The Haugeberg et al study reported significant 12-month hand BMD loss in RA patients, which was not observed in patients with non-RA inflammatory arthritis or noninflammatory joint disease. They found bone loss to be associated with RF and mean CRP level during followup; however, their study did not look for an association between bone loss and future erosive disease.
Many studies have examined other potential predictors of joint damage with varying and sometimes conflicting results. Our findings in relation to other predictors of erosive disease are broadly consistent with published studies. For example, Liao et al () reported that age is associated with susceptibility to erosive disease, and Weinblatt et al () reported that CRP level after 12 weeks of methotrexate therapy was predictive of radiographic progression at 1 year. Previous studies of anti-CCP have perhaps yielded the most contradictory results, and it is here that our observation that factors affecting erosion susceptibility and severity are different may be the most pertinent. For example, Kaltenhauser et al () reported that anti-CCP positivity was associated with severe joint destruction (i.e., upper third of the study population after a 6-year followup), whereas Liao et al () reported that anti-CCP was not associated with susceptibility to erosive disease. Both of these studies are consistent with our results.
Our finding that BMD loss at 6 months is a predictor of erosive disease at 12 months is a new and potentially useful finding. Three types of bone loss have been described in RA: focal articular bone loss (erosions), periarticular bone loss adjacent to inflamed joints (periarticular osteopenia), and generalized systemic bone loss affecting the axial and appendicular skeleton and leading to increased fracture risk (osteoporosis) (). In animal models of inflammatory arthritis, there is a strong association between inflammation and focal bone erosions, but this has been more difficult to demonstrate in humans with RA, probably due to disease heterogeneity and differing treatments (). Inflammatory mediators such as tumor necrosis factor α, interleukin-1 (IL-1), IL-6, and IL-17 produced in the synovium and pannus have been shown to play a role in erosive joint damage (). Focal bone destruction (erosions) due to osteoclast overactivity occurs due to increased production of RANKL () and decreased production of its inhibitor osteoprotegerin within the RA joint (). To date, there have been no animal studies specifically looking at the mechanisms leading to periarticular bone loss in inflammatory arthritis.
It has previously been suggested that the association between BMD loss and the development of focal erosions implies a common pathway in the pathogenesis of these 2 types of bone loss (). The findings of our study, which employed a hurdle model to identify separate risk factors for erosion susceptibility and erosion progression (severity) suggest an alternative, albeit not mutually exclusive, hypothesis. The risk factors for erosion severity were RA disease specific but the risk factors for erosion susceptibility were not, suggesting that changes in bone associated with aging and bone loss may render it less resistant to attack by erosive processes, the severity of which is determined by disease activity and anti-CCP status. However, whether there is a direct causal relationship between BMD loss and later erosive disease has not yet been established.
For a measurement to have utility as a predictor of disease progression it needs to be inexpensive, readily available, and quantifiable. Hand DXA fulfills these criteria. An additional criterion is a lack of redundancy relative to simpler measures. In the present study, the predictive value of DXA for erosive disease needs to be considered within the context of ready availability of a patient's age and anti-CCP antibody status, which is now a routine component of diagnostic assessment. For example, the predictive value of age ≥50 years plus anti-CCP status may be sufficient to justify an intensive treatment approach in patients without erosive disease, with change in hand BMD being most useful when one of these criteria is not met. Further, it is also possible that higher degrees of BMD loss may have stronger predictive value than lesser degrees of BMD loss. Larger studies are needed to develop and validate predictive algorithms for patients at risk of erosive disease.
A limitation of longitudinal BMD measurements is the difficulty in defining patients with BMD loss. This requires a statistical approach and some inevitable uncertainty and misclassification. Earlier studies have determined the mean change in BMD and divided patients according to which side of this value they fell. In this study, we used a clustering algorithm to optimally partition the patients into BMD loss and no loss groups. This approach was validated by the subsequent finding that the no loss group was associated with expected risk factors for bone loss, such as smoking, age, and female sex. An alternative method is to use the LSC, which is derived from BMD precision measurements. The LSC identifies patients for whom there is a 95% probability that the BMD score has changed, which is necessary when interpreting ΔBMD for an individual patient in a clinical setting. However, in a population sample, this method will have a high specificity, yet a poorer sensitivity, for identifying patients with bone loss. In our study, 82% of subjects classified as having BMD loss met the LSC criteria.
The main limitations of this study are its size (small number of patients with erosive disease) and short duration of followup. However, this is the first study to date that examines in a prospective design the relationship between bone loss and erosive disease in DMARD-naive patients with very recent–onset RA, coupled with a treat-to-target treatment program. Further, other studies have used a specialized technique, DXR, to measure BMD, whereas our study utilized DXA, which is a simple, reproducible, and readily available technique.
In conclusion, whereas previous studies have shown an association between BMD loss in the first 12 months and long-term joint damage, this the first study to demonstrate that early BMD loss measured by DXA in the first 6 months of DMARD therapy for recent-onset RA is a predictor for erosive disease at 12 months. Early BMD loss at 6 months may be a means of identifying patients who require additional therapy despite an apparently adequate clinical response to treatment. Conversely, the absence of early bone loss may allow treatment to be less intensive without unacceptable risk of disease progression. The potential predictive value of BMD loss warrants further investigation. Ultimately, the utility of hand DXA will need to be evaluated prospectively in randomized controlled trials by incorporating it into a treatment algorithm whereby those with high BMD loss receive additional DMARD therapy at 6 months, irrespective of disease activity.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Proudman had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Black, Chatterton, Cleland, Hill, Proudman.
Acquisition of data. Black, Spargo, Schultz, Chatterton, Cleland, Proudman.
Analysis and interpretation of data. Black, Spargo, Schultz, Cleland, Lester, Hill, Proudman.
The authors wish to thank Ms Leah McWilliams, BNg, for metrological assistance, and Ms Cindy Hall, AssDipMedLabSci, for help with data management.