Article Text
PDF
Abstract
Objective To summarise the relationship between joint damage and functional disability in rheumatoid arthritis (RA) patients.
Methods A systematic review of the literature from 1990 to 2008 was conducted using MEDLINE and EMBASE databases. The search strategy focused on RA, joint damage and disability. Only longitudinal studies or randomised clinical trials with 1 year or more of follow-up containing data correlating joint damage and disability were included. The comparisons were categorised in four ways: baseline damage versus disability at end of follow-up (correlation A); damage versus disability measured cross-sectionally at each of several time points (correlation B); changes in damage versus final disability (correlation C) and changes in damage versus changes in disability (correlation D).
Results From a total of 1902 abstracts, 42 studies met the inclusion/exclusion criteria. More than 50% of the studies that measured baseline damage to later disability (A) reported a statistically significant association. Correlation was significant when measured at multiple time points over time (B; 16/19 studies). Statistically significant associations between changes in damage and either disability at end of follow-up or changes in disability were also found (C and D; 11/13 studies).
Conclusions While many of the studies did not include multivariate analysis with confounder adjustment, the published evidence indicates a link between joint damage and functional disability and that an increase in joint damage is associated with an increase in disability over time. Treatments to limit progressive joint damage may lead to better joint function and improved patient outcome with less disability.
Statistics from Altmetric.com
Joint destruction often begins within 3 months of symptom onset in rheumatoid arthritis (RA),1 2 and most patients develop some joint damage within 2 years.3 4 Joint damage comprises bone destruction and cartilage breakdown, visualised radiographically as erosions and joint space narrowing (JSN). Radiographic scoring commonly involves either global scoring (eg, Steinbrocker et al5 or Larsen scores,6 or a total score derived from the sum of separate erosion and JSN (Sharp) scores. Both score types have undergone further modifications.7,–,9 MRI scoring focuses on erosions.10 Some MRI sequences also assess synovium and ‘bone oedema’ or osteitis. Bone and cartilage destruction over time in RA and the effects of RA therapeutics can be measured using radiographic and functional outcome measures.
Intuitively, inhibition of progressive damage leads to preservation of structure and presumably function, and ultimately to better patient outcomes as assessed by patient-reported outcome measures such as the health assessment questionnaire (HAQ) disability index. This argument would be best confirmed if progressive joint damage (increasing radiographic changes) were longitudinally associated with reduced physical function. However, physical function is mostly a consequence of disease activity (acute inflammation) and chronic joint damage.11 Disability measures are complex instruments, affected by both acute and chronic components. Acute inflammatory disease activity is reversible, leading to potential difficulty in assessing the disability caused over time by irreversible joint damage.
Although it is well known that joint damage and functional disability increase with disease duration, their relationship is still uncertain. Review articles assessing potential predictors of functional disability12,–,15 generally concluded that irreversible joint damage accounts for much chronic disability and that the correlation is strong in later RA. However, these reviews relied on a wide spectrum of study designs, and evidence was often reported selectively rather than systematically. This study aims to improve our understanding through a systematic review assessing the temporal relationship between joint damage and disability.
Methods
Literature search
A systematic review of the published literature was conducted using MEDLINE and EMBASE databases. The search was limited to studies published in English between 1990 and 2008 focused on three components: RA, joint damage and disability. Joint damage was searched using components (eg JSN, erosion, joint destruction) and scores of joint damage (eg, Sharp score, Larsen score). Disability was identified using instruments to measure physical disability (eg, HAQ disability index, arthritis impact measurement scale, grip strength, short-form 36 (SF-36)), work disability and quality of life (eg, loss of productivity, absenteeism, activities of daily living). The results from these three components were combined to obtain abstracts and titles, which were then limited to meta-analyses, systematic reviews, longitudinal studies and randomised controlled trials (RCT). The listing and order of the search terms are provided in supplementary table S1 (available online only). Using this search strategy, 1902 unique citations were retrieved.
Screening strategy
All citations were screened by two independent reviewers (MB and TG) for:
Inclusion criteria
Adult RA
Joint damage, measured on x-ray with Larsen, Sharp or van der Heijde/Sharp score or on MRI
Disability measured with one or more of the following: HAQ, arthritis impact measurement scale, RA quality of life, work disability, absenteeism, presenteeism, SF-36
Longitudinal studies (including RCT with patient follow-up ≥1 year)
Exclusion criteria
Correlation between damage and disability estimated cross-sectionally
Patient follow-up less than 1 year
Methodological studies or narrative papers
Reviews
No measurement of damage/disability association
Of 1902 citations, 188 fulfilled all inclusion criteria. After exclusion criteria were applied, 118 citations were eligible for full review. Citations with cross-sectional correlation coefficients (exclusion 1) or follow-up less than 1 year (exclusion 2) were held as potential sources of additional information in case conclusions from the other studies were unclear. Some studies were excluded because, despite damage and disability measurements, correlations were not measured or reported. Bibliographies of included articles were scanned as a final step to identify studies that might have been missed.
Forty-three studies were identified using the search strategy. Two additional articles were found by scanning bibliographies. The final sample of studies for data extraction included 40 longitudinal studies and five RCT. Supplementary table S2 (available online only) lists studies excluded with the reasons for exclusion.
Data abstraction
A data abstraction form was developed to summarise the studies (see supplementary Figure S1, available online only). Patient samples were closely examined to avoid duplication. If more than one study reported results using the same sample, the study with the most complete data was included in the data abstraction form.
To address the study objective, four kinds of comparisons were examined:
Comparison A: The correlation between radiographic damage at the time of study entry and HAQ at the end of follow-up; assesses the ability of radiographic score at baseline to predict future disability.
Comparison B: The correlation between joint damage and functional disability estimated at sequential time points during the follow-up (eg, correlation of the two at baseline, 3 years, 6 years etc.); assesses whether the strength of association between radiographic damage and disability increases or decreases over time.
Comparison C: The correlation between change in radiographic progression and mean HAQ score during either follow-up or at the end of follow-up; investigates whether progression in joint damage is associated with mean HAQ or final patient disability, whichever was available.
Comparison D: Correlation between changes in radiographic progression and changes in HAQ score during follow-up; estimates whether joint damage progression is associated with functional disability change.
Potential confounders likely to impact the damage/disability association were collected, along with the statistical technique in order to document the presence of covariate adjustments and multivariate analyses.
Results
Within the 45 studies identified, three32,–,34 used reported data for the same patients described in earlier studies and were excluded.35 36 Key characteristics of the final 42 studies are described in more detail in the online supplementary data and are summarised in supplementary table S3 (available online only). All studies included an estimate of one or more of the four types of comparisons, including 19 type A, 19 type B, six type C and seven type D comparisons.
Comparison A (correlation between baseline radiographic damage and HAQ at the end of follow-up)
The majority of these studies (9/19) showed statistically significant correlations between baseline radiographic damage and HAQ at the end of follow-up; however, seven studies reported no statistical significance, and three studies were significant based on the measurement instrument used. The studies differed in their analytical approach. Table 1 stratifies the studies based on univariate analysis (standard t test) or multivariate analysis (regression model). Among eight studies that used univariate analysis, only three reached statistical significance; those studies are presented in more detail in the supplementary material (available online only). Among the 11 studies with multivariate analyses including regression models, six found a statistically significant correlation between damage at baseline and disability at the end of follow-up, three found no association, and two had contradicting results depending on the instrument used to measure damage66 or the type of statistical analysis performed.59
Synthesis of results for studies that estimate the correlation between radiographic damage at baseline and HAQ at the end of follow-up (type A comparison)
Of the studies demonstrating correlations, a study in 43 Swedish patients62 found a statistically significant correlation between the baseline Larsen score and the HAQ score at 2 years (Spearman correlation coefficient 0.385; p<0.05). Also, when the HAQ score change over 2 years was used as a dependent variable in a regression model (adjusted for confounders), the baseline Larsen score explained much of this variation. Two RCT conducted in The Netherlands38 75 and an Austrian pooled analysis of RCTs37 reached similar conclusions using the Sharp score as the damage measure and HAQ as the functional disability measure; p<0.001. Both Aletaha et al37 and van der Heijde et al75 developed regression models adjusting for potential confounders to reduce the risk of bias in detecting the association between Sharp and HAQ. A similar approach60 showed a statistically significant longitudinal relationship between joint damage at baseline and HAQ at 9 years (p<0.0001) using a mixed model regression analysis in 115 Dutch patients. Finally, a 74 patient Japanese study54 estimated the factors predicting prognosis by a logistic regression analysis for quality of life score as the endpoint variable, with a statistically significant relationship between initial Steinbrocker's stage and quality of life even after 15 years.
Two Dutch studies46 61 showed no association between baseline damage (total Sharp score) and functional disability (HAQ) at the end of follow-up. In the study by Van der Heijde et al,46 147 patients over 2 years showed no correlation between damage and disability in a multiple regression analysis; however, limited information was reported. Landewe et al61 followed 156 patients up to 5 years; correlation between the baseline Sharp score and HAQ at follow-up was not statistically significant (linear regression, p=0.684). Bansback69 constructed a multivariate logistic regression to predict the 5-year functional outcome in RA (based on the HAQ) in a study of 985 UK patients. In the bootstrap resampling validation (resampling with replacement of patient data from an existing dataset) of the logistic regression, six variables (functional grade II at year 1, functional grade III or IV at 1 year, HAQ at baseline, HAQ at 1 year, haemoglobin level at baseline, Carstairs deprivation index 5) consistently predicted moderate/severe functional outcome, but the baseline Larsen score did not.
Comparison B (correlation between joint damage and functional disability measured cross-sectionally at different time points during the follow-up)
Nineteen studies compared joint damage and functional disability cross-sectionally at different time points (eg, 3, 6, 10 years) during follow-up, providing information on whether associations increase or decrease over time (table 2). Most studies (13/19) conducted only univariate analyses, although six also performed a regression model. Ten studies showed statistically significant correlations between damage and disability at every time point considered, six studies found statistically significant correlation only during the follow-up years, while three studies did not find associations at any time point using the HAQ score as a disability measure. Details are provided in the supplementary material (available online only).
Synthesis of results for studies that estimate the correlation between joint damage and functional disability cross-sectionally at different time points during the follow-up (type B comparison)
Six studies used a regression model in addition to standard univariate analysis; all concluded that correlation between damage and disability was statistically significant, although the relationship was apparent only after years of follow-up in three studies. Drossaers-Bakker et al36 estimated the correlation between the Sharp score and HAQ at different time points (baseline, 3, 6 and 12 years) and showed significant associations at all time points. The correlation strength increased with time: Spearman correlation coefficient of the Sharp score to HAQ was 0.22 at baseline (p<0.05); 0.29 at 3 years (p<0.001); 0.41 at 6 years (p<0.001) and 0.57 at 12 years (p<0.001). In addition, a stepwise linear regression model showed that by putting the variable describing the pooled joint destruction into the model first, 40% of the variation in the HAQ score was explained (p<0.01). Another study57 in this population using multivariate analysis also showed correlations between the Sharp score and HAQ and between the Larsen score and HAQ at baseline and 12 years. Ødegård et al70 came to similar conclusions estimating the association between the Sharp score and HAQ at 1, 2, 5 and 10 years using generalised estimating equations (GEE), a technique for analysing longitudinal and other response data, particularly if responses are binary. Adjusting for potential cofounders, a statistically significant association was found (p=0.02). Welsing et al60 estimated the correlation between the Sharp score and HAQ at baseline, 3, 6 and 9 years. While the correlation was weak and not significant at baseline or 3 years (Pearson correlation coefficient 0.15 and 0.06), it was stronger and significant at 6 and 9 years (Pearson correlation coefficient 0.75 and 0.57, respectively, p<0.05). The Sharp score explained a high proportion of the variation in the HAQ score after adjusting for key confounders. Similarly, in a multivariate analysis, Scott et al12 found that the correlation between the Larsen score and HAQ was not statistically significant until 5 years of follow-up. Finally, Plant et al40 also found no baseline correlation between Larsen and HAQ, but a significant correlation at 5 years (Spearman correlation coefficient 0.29, p<0.01). Moreover, in a multiple linear regression model adjusted for confounders, the Larsen score and HAQ score were significantly associated only at the end of follow-up (p<0.001).
Comparison C (correlation between changes in radiographic damage and mean HAQ score during or at the end of follow-up)
Six studies estimated the correlation between changes in radiographic progression and mean HAQ score during or at the end of follow-up (table 3). Of these, five found significant correlations between damage and disability and one found contradictory results depending on the instrument used to assess disability (HAQ (significant) vs grip strength (NS)). Studies using univariate analyses are described in the supplementary material (available online only). Among the studies using multivariate analysis or regression models, Ødegård et al70 estimated the association between the change in the total Sharp score over 10 years and HAQ using GEE, with a statistically significant association (p<0.019) after adjusting for potential confounders (see also type B studies). Similar results47 used a discriminant analysis to determine potential predictors of disability over approximately 15.2 years, concluding that erosive change during the first 2 years of disease was the most powerful single predictive feature for poor functional outcome (p<0.05).
Synthesis of results for studies that estimate the correlation between changes in radiographic progression and mean HAQ score during the follow-up or HAQ score at the end of follow-up (type C comparison)
Comparison D (correlation between changes in radiographic progression and changes in HAQ score during follow-up)
Seven studies measured the correlation between changes in radiographic progression and changes in the HAQ score (table 4). Five found an association; one did not. The seventh study reported a correlation score of 0.2 but did not report a p value; the correlation was assumed to be non-significant. Three studies12 40 44 conducted only univariate analysis and are again presented in detail in the supplementary material (available online only).
Synthesis of results for studies that estimate the correlation between changes in radiographic progression and changes in HAQ score during follow-up (type D comparison)
González-Alvaro et al42 found a significant association between changes in the Larsen score and changes in the HAQ score over 4 years (p<0.001). van der Heijde et al75 obtained similar findings at 2 years assessing the correlation between changes in the Sharp score and changes in the HAQ score (see also comparison A). A statistically significant association between damage and disability (p<0.0001) was found using a generalised mixed linear model, even when adjusted for potential confounders. Clarke et al43 used a hierarchical model to estimate the strength of relationship between radiographic scores and HAQ to take the variability across patients into account (see also comparison B). Over 10 years, a 25-unit (50%) increase in radiographic score was estimated to result in an increase in baseline disability (HAQ) of approximately 0.46 units (35%). Only radiographic score and baseline HAQ were reported as predictors of HAQ in the hierarchical models. On the contrary, Rupp et al39 did not find an association (p>0.05) between change in the Sharp score and change in the physical component of the SF-36 over 2 years using multiple linear regression.
Discussion
Our systematic review identified studies that assessed the association between joint damage and functional disability in RA patients. Inclusion was restricted to longitudinal studies and RCT with 1 year or more of follow-up in order to make stronger assessments of the causal relationship between damage and disability. Forty-two eligible studies providing a total of 50 correlation results and four types of comparisons were found.
Inconsistency was found for the correlation between joint damage at baseline and functional disability at end of follow-up (comparison A). Of 19 studies, 12 found correlations, while in three, the significance depended on the instrument used to measure damage or disability. Seven studies did not find associations. However, the correlation between baseline damage and disability at the end of follow-up was statistically significant for all four studies that conducted a multivariate analysis with confounder adjustment; the risk of bias for those studies is likely to be low.
Published evidence also suggests that a correlation exists between radiographic progression and disability when estimated cross-sectionally at multiple time points (comparison B). Sixteen of 19 studies showed a correlation at one or more time point during follow-up, with 10 revealing a correlation at every time point; associations between damage and disability increased with disease duration and follow-up. This was particularly evident in two studies,43 60 in which there was no correlation at baseline, but strong correlation was found over longer patient follow-up. Weak correlations between joint damage and disability in early RA may be attributed to several factors: the extent of joint damage at baseline, the effect of acute inflammation on the HAQ, and the progression pattern chronic disability has on the HAQ assessment. Joint damage at baseline in early RA patients differs considerably from one patient to another, resulting in confounding variability that may not allow the detection of correlation with subsequent long-term functional impairment. In contrast, in later RA, more damage may be easier to evaluate for correlations with disability. Another confounding challenge is that HAQ scores increase in early RA, tend to decrease as the initial acute inflammation is treated and then increase again as chronic impairment and joint damage accrues.13 This is likely to be a consequence of inflammation reduction in early disease, when inflammation control will increase function, whereas with long-standing disease, joint damage leads to an increase in irreversible functional impairment,11 so that even when disease activity is quiescent, residual disability is highly correlated with the degree of joint damage and disease duration.37 While radiographic and HAQ scores seem to move in opposite directions in early RA, both tend to increase in later disease. Therefore, in studies of early RA with a short follow-up, a correlation between damage and disability may not be evident; however, these associations become evident as RA progresses.
The published evidence suggests a strong correlation between change in the radiographic score and functional disability at the end of follow-up (correlation C) and between changes in radiographic score and changes in disability over patient follow-up (correlation D). Only two of 13 studies failed to show statistically significant associations.39 56 All four studies employing multivariate analyses with confounder adjustment showed statistically significant correlations. From clinical perspectives, studies reporting correlation types C and D are more informative because they suggest that if treatment is able to inhibit progressive joint damage, it may lead to better joint function and outcome.
Reviews4 12,–,14 have also concluded that correlations between damage and functional disability strengthen as disease duration increases, while damage and disability are often not correlated early in RA.12 13 However, our review sheds additional light on the relationship between joint damage and disability because it is methodologically more robust than the previously published reviews. Of the previous reviews, only Scott et al13 was systematic; all others included both longitudinal and cross-sectional studies. Second, this review stratified results by types of comparisons (A, B, C, D) with the objective of highlighting the different types of links between damage and disability. Third, the type of statistical analysis was highlighted and studies stratified by univariate analysis versus multivariate analysis with adjustment for confounders. Finally, studies were included that measured the association between damage and disability through a correlation score regardless of the instrument or model used.
This study has some limitations. No quality assessment of the studies identified was conducted, and the results were not adjusted for the quality of each study. Second, no specific adjustments were made for the sample size or appropriateness. Only the number of studies with or without a statistically significant association between joint damage and disability were calculated. Therefore, these analyses only allow inference from correlations that damage and function are associated with one another over time. Quantitative estimates to translate Sharp or van der Heide/Sharp score increases into increases in HAQ have recently been suggested.76 Third, although common radiographic scoring methodologies were used, heterogeneity in scoring between studies might have affected the strength of the correlation between damage and disability. Fourth, the studies were performed between 1991 and 2008 (17 years) and changes in treatment patterns over the years may impact both damage and disability progression. Fifth, the type of analyses performed may limit the result interpretation. Fewer than half (23/50) of the comparisons measured were based on a multivariate analysis, and only 12 of these adjusted for potential confounders. The association between damage and disability might thus have been influenced by the effect of various factors, such as disease duration and disease activity14 and, therefore, has not been formally addressed for its actual strength but rather presented as the statistical significance of this association. However, the fact that so many studies have described a statistically significant association in both univariate and more complex analyses implies that this association is not only present but is strong, because it is reproducible in various patient populations across various countries. Sixth, the length of follow-up was relatively short (5 years or less) for 23/42 studies. Most of the studies extending 5 years or longer measured the correlation between damage and disability cross-sectionally at multiple time points (comparison B). Since correlation increases as disease duration increases, a longer follow-up might be needed to detect fully the significance of the association between damage and disability. Finally, the level of information provided in the literature varied greatly among studies. Some studies were not designed with the primary aim of assessing the correlation between damage and disability.
In summary, published evidence indicates a link between joint damage and functional disability (especially as disease duration increases) and that an increase in joint damage is associated with an increase in future disability over time. Future long-term studies (>5 years) with appropriate sample sizes using multivariate analysis with confounder adjustments are needed to corroborate these findings.
Acknowledgments
The authors thank Tripat Gill and Nicki Protopapas from OptumInsight for their assistance in literature review and data abstraction, and Crystal Watson and Jackie Mortan from Amgen Inc. for their assistance with the literature search.
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:
- Web Only 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.
- Web Only 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.
Footnotes
-
Funding Funding for this study was provided by Amgen Inc.
-
Competing interests DM and DJZ are employees of Amgen Inc. VW, AP and MB are paid consultants for Amgen Inc. JSS has received grant support and/or honoraria from Abbott, Amgen, BMS, Centocor, Roche, Sanofi-Aventis, Schering-Plough and Wyeth.
-
Provenance and peer review Not commissioned; externally peer reviewed.