Osteoarthritis (OA) is a debilitating chronic multijoint disease of global proportions. OA presence and severity is usually documented by x-ray imaging but whole body imaging is impractical due to radiation exposure, time and cost. Systemic (serum or urine) biomarkers offer a potential alternative method of quantifying total body burden of disease but no OA-related biomarker has ever been stringently qualified to determine the feasibility of this approach. The goal of this study was to evaluate the ability of three OA-related biomarkers to predict various forms or subspecies of OA and total body burden of disease.
Female participants (461) with clinical hand OA underwent radiography of hands, hips, knees and lumbar spine; x-rays were comprehensively scored for OA features of osteophyte and joint space narrowing. Three OA-related biomarkers, serum hyaluronan (sHA), cartilage oligomeric matrix protein (sCOMP), and urinary C-telopeptide of type II collagen (uCTX2), were measured by ELISA. sHA, sCOMP and uCTX2 correlated positively with total osteophyte burden in models accounting for demographics (age, weight, height): R2 = 0.60, R2 = 0.47, R2 = 0.51 (all p<10−6); sCOMP correlated negatively with total joint space narrowing burden: R2 = 0.69 (p<10−6). Biomarkers and demographics predicted 35–38% of variance in total burden of OA (total joint space narrowing or osteophyte). Joint size did not determine the contribution to the systemic biomarker concentration. Biomarker correlation with disease in the lumbar spine resembled that in the rest of the skeleton.
We have suspected that the correlation of systemic biomarkers with disease has been hampered by the inability to fully phenotype the burden of OA in a patient. These results confirm the hypothesis, revealed upon adequate patient phenotyping, that systemic joint tissue concentrations of several biomarkers can be quantitative indicators of specific subspecies of OA and of total body burden of disease.
Citation: Kraus VB, Kepler TB, Stabler T, Renner J, Jordan J (2010) First Qualification Study of Serum Biomarkers as Indicators of Total Body Burden of Osteoarthritis. PLoS ONE5(3): e9739. https://doi.org/10.1371/journal.pone.0009739
Editor: William Taylor, University of Otago, New Zealand
Received: July 29, 2009; Accepted: February 3, 2010; Published: March 17, 2010
Copyright: © 2010 Kraus et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The research was supported by National Institutes of Health/National Institute of Arthritis, Musculoskeletal and Skin Diseases UO1 AR050898; National Institutes of Health/National Institute of Aging Claude D. Pepper Older Americans Independence Centers 2P60 AG11268; MO1-RR-30, National Center for Research Resources, General Clinical Research Centers Program, National Institutes of Health where this study was conducted; and GlaxoSmithKline. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: This is an investigator initiated study funded by GlaxoSmithKline and the National Institutes of Health. The authors have no competing interests with respect to the commercial funder, including no employment, consultancy, patents, products in development or marketed products related to this work. The authors confirm that this does not alter their adherence to all the PLoS ONE policies on sharing data and materials.
Osteoarthritis (OA) is a debilitating chronic multijoint disease with major global impact . OA presence and severity is usually documented by x-ray imaging but whole body imaging is impractical due to radiation exposure, time and cost. Systemic (serum or urine) biomarkers offer a potential alternative method of quantifying total body burden of disease but no OA-related biomarker has ever been stringently validated to determine the feasibility of this approach. Systemic biomarkers in OA are usually only assessed relative to a particular joint site under study. The advent of disease status indicators in the serum or urine would provide badly needed objective means of evaluating and monitoring the disease, could provide quantitative traits for genetic studies, and could contribute to identifying individuals at greatest risk of progression, and therefore, in greatest need of prevention strategies.
The inference of the relevant relationships between serum or urine biomarker concentrations and disease of specific joints is complicated by the fact that many or most OA patients have disease in multiple joints. Moreover, OA is a complex disease increasingly understood to be a mixture of distinct clinical and genetic entities , , , . OA of the knee may or may not share the same etiology as OA of the spine or hands. Biomarkers may be useful as indicators of specific subspecies of OA, or of OA in specific joints. The distinct features of radiographic OA of joint space loss and osteophyte represent different disease processes that may also differ in their associations with biomarkers. These features are highly correlated so it is necessary to adjust for this correlation in order to test the independent association of biomarkers with particular features of OA. Any study aimed at elucidating these relationships will need to collect extensive radiological data to survey all joints potentially involved in OA. Only a small fraction of the patients analyzed will have OA isolated to one set of joints . To focus on this small fraction of the total data would be wasteful and unsatisfying. Instead, we have chosen to analyze the full dataset, relying on statistical analysis, model selection, and model averaging to disentangle the web of correlations among radiological disease, OA risk factors, and biomarkers.
Qualification is the evidentiary process of linking a biomarker with biological processes and clinical end points . In this work, we focus on the process of biomarker qualification for structural endpoints of OA, such as osteophyte formation and radiographic joint space narrowing, as opposed to symptomatic endpoints such as joint pain. We chose to analyze three biomarkers in this study based on the strength of previous evidence showing associations with OA: serum hyaluronan (sHA), serum cartilage oligomeric matrix protein (sCOMP), and urinary C-terminal propeptide epitope of collagen II (uCTX2). Each of these biomarkers has data to support its classification  in at least two categories of the BIPED  biomarker classification scheme (Burden of disease, Investigational, Prognostic, Efficacy of intervention, and Diagnostic biomarkers). The categories to which each of these biomarkers corresponds for structural features of OA are as follows: HA–categories D, B, P, E , , , , , , ; COMP–categories D, B, P , , ; CTX2–categories D, B, P E , , , , , , .
In this work we describe which radiological features are associated with each systemic biomarker and the extent to which these biomarkers reflect burden of disease of the various forms and features of OA. Here we have qualified these three biomarkers as strong quantitative traits of various features of OA. Uncoupling these features or processes of OA provided insights into the origin of these biomarkers. We observed that the canonical radiographic features of OA, namely osteophyte and joint space narrowing, correlated with one another, often in a complex manner with regard to their effects on a biomarker. We further observed that small joints did not necessarily contribute in a small way to serum biomarker concentrations. Moreover, with respect to the impact on biomarkers, the lumbar spine radiographic features behaved similarly to radiographic features in other joints, arguing for a similar disease process ongoing in the lumbar spine as in the rest of the skeleton. Finally, systemic biomarkers were more powerful predictors of multijoint disease than of disease in any single joint when added to a statistical model containing risk factor data (age, height, and weight). A notable exception was the knee in which biomarkers doubled the power of these factors alone for predicting radiographic features of OA.
A total of 461 female participants were studied. Their mean (SD) age was 67 (10) years old. Their mean (SD) weight, height and body mass index were 74 (17) kg, 161 (7) cm, and 28 (6)_(kg/m2), respectively. Just over two-thirds of the participants were evaluated through the Duke site (n = 322) and the remainder at the UNC site (n = 139).
Correlations and bilateral symmetry among radiographic features of OA in the various joint systems
There were seven distinct joint systems considered: distal (DIP) and proximal (PIP) interphalangeal finger joints, metacarpophalangeal (MCP or knuckle) hand joints, carpometacarpal (CMC or base of thumb) joints, knees, hips and lumbar spine. Figure S1A-B shows the age-related increased prevalence of affected joint systems based on any osteophyte or any joint space narrowing scores; Figure S1C shows the numbers of subjects by decade of age with with the OA affected joint system based on the Kellgren Lawrence grade >1 grade (Fig. S1C). age-related increased prevalence of affected joint systems based on the traditional Kellgren Lawrence (KL) grading system  (quantified in Table S1).
Figure 1 shows the correlation matrix among joints affected by osteophyte (OST), and by joint space narrowing (JSN), a surrogate for cartilage loss. The tapestries are complex, reflecting the inherent complexity of OA, but the departures from randomness apparent in their patchiness and symmetries provide important insights into the relationships among the manifestations of disease in the various joints. Two interesting features are discernible in this figure. First, joint faces (medial, lateral, superior and inferior joint margins) within the same joint tended all to be affected by OST, as indicated by the appearance of square patches in the resulting tapestry of this matrix (Figure 1A), particularly in the knees, hands and lumbar spine. Such correlation within joints was less evident for JSN (Figure 1B) but still apparent in the hands, and to a lesser extent, the lumbar spine. Second, there was a high degree of bilateral symmetry in the distribution of OA in these study participants as indicated by the mirror-image symmetry of both panels of Figure 1 across the diagonal. In the JSN tapestry (Figure 1B), this bilateral symmetry extended even to the preferential correlation of the same specific joint faces in contralateral joints, as was especially apparent, for example, in the knees and hips (shown by the strong backward diagonal line). In addition, there was greater collateral involvement of neighboring joints on the right hand, based on JSN, than the left hand (top right versus bottom left corner of Figure 1B). This further highlights the striking multi-joint OA involvement of individuals recruited on the basis of hand OA alone.
The correlation for left and right joints are shown for A) osteophyte score in each pair of joint faces, B) joint space narrowing score for each pair of faces. Each cell indicates the correlation between two joint faces as indicated in the color legend. The correlation matrix is symmetric by definition across the main diagonal. Left and right sides of the body are depicted from bottom to top on the y-axis and from left to right on the x-axis.
For each participant, we summed the total number of joint faces with OST >0 and the total number of joints with JSN >0, designating these sums nOST and nJSN, respectively. We also summed the total number of joints with OA based on KL scoring (KL>1), designating this sum nKL. Among participants, nOST and nJSN had a correlation coefficient of r = 0.66, which was very highly significant (p<10−6).
Associations of biomarkers and radiological features of OA
The OA-related biomarkers, serum hyaluronan (sHA) and urinary C-telopeptide fragment of type II collagen (uCTX2), were strongly positively correlated with the number of affected joint systems based on KL scoring, with their concentrations having significantly positive coefficients in a linear model with nKL as a predictor (Figure S2). Quite surprisingly, serum cartilage oligomeric matrix protein (sCOMP) showed a statistically significant (p = 0.024) concentration decrease as nKL increased. Note that these models were fit with the following risk factors: , ,  age, log height, log weight for sHA; age for sCOMP; and for uCTX2 no other covariate was used. These covariates were tested based on their significance in previous work , , , , ,and those used for these analyses were based on their significant association with OA in models that included the biomarkers.
The KL scoring system is a composite variable, consisting of some OST and some JSN criteria. To assess these features uncoupled, we evaluated the association of the biomarkers to OST and JSN in models fit with the risk factors noted above and with nOST and nJSN (Figure 2), i.e. joint faces were counted without regard to which joint they belonged to and we accounted for correlations among features. Accounting for correlation with JSN, all three biomarkers showed a positive dose-dependent relationship with total OST burden (p<10−6). In contrast, although sHA and uCTX2 showed a strong positive dose-dependent relationship with total JSN burden before adjustment for OST (Figure S3), there was no significant association of sHA or uCTX2 with JSN after accounting for correlation with OST (Figure 2). Surprisingly perhaps, and in contrast to sHA and uCTX2, sCOMP showed a negative dose-dependence with total JSN-affected faces, even after accounting for OST (Figure 2). Some of the most striking associations between individual biomarkers and specific radiographic features are shown in Figure S4: namely sHA and knee OST and CMC JSN; sCOMP and DIP JSN (inverse correlation); and uCTX2 and hip OST.
The data are restricted to female subjects with complete radiographic and biomarker data. The data were fit to a mixed-effects model, with family as the grouping variable for the random effect. The fixed effects are the numbers of joint faces affected by osteophytes and joint-space narrowing, respectively. Point estimates are plotted for mean serum log sHA (top), log sCOMP (middle), and log uCTX2 (bottom) by number of joint faces affected by any osteophyte (grade >0, left panels), and number of joint faces OA affected by any joint space narrowing (grade >0, right panels). Joint faces are counted without regard to which joint they belong. JSN point estimates account for OST and OST point estimates account for JSN. In addition, sHA was adjusted for age, log height, log weight; sCOMP was adjusted for age; and CTX2 did not require adjustment. The R2 values indicated give the proportion of the inter-radiographic class variation explained by the regression. The respective generalized R2 values for the variance explained by the radiographic features are 4.4%, 6.2%, and 10.2%. All p-values for the weighted regressions are less than 10−6, except for the pairing of sHA and JSN (p = 0.18), and uCTX2 and JSN (p = 0.51). The mean (standard deviation) and range of biomarker concentrations prior to logarithmic transformation were: sHA mean 50.7(56.5) ng/ml, range: 1.1–499 ng/ml; sCOMP mean 1130(620) ng/ml, range: 203–4903 ng/ml; uCTX2 mean 372(806) ng/mmol, range: 19.6–12,171 ng/mmol.
We next computed the regression coefficients for each biomarker (dependent variable) for each joint site and radiographic feature (independent variables) in a mixed-effects model with all 14 distinct radiographic features (OST and JSN for each of 7 joint systems) included as predictors. This “saturated” model was not optimally predictive (we developed an optimized model below), but it did allow us to effectively decouple the correlations among the independent variables that would otherwise confound our estimation of the relationships between them and the biomarkers, and to evaluate the contributions of each in its natural context. In Figure 3, a visual comparison is shown of these coefficients considered jointly, through simultaneous fits to all radiographic features; Figure S5 shows the same coefficients estimated by fitting to single radiographic features individually, i.e. not accounting for correlations among features. In both cases, the horizontal reference lines are located at the common coefficient value obtained by considering all joints equivalent, ie, constraining all joints to have identical coefficients. Coefficients significantly above these lines identify features of a particular joint that contribute disproportionately (higher than expected assuming equivalence among joint systems) to the respective biomarker concentrations–while coefficients below the reference lines identify features of a particular joint that provide less than the average contribution to the respective biomarker concentration. Noteworthy findings illustrated in Figure 3 include the fact that CMC JSN contributed disproportionately (positively) to all three biomarker concentrations, while CMC OST had the opposite tendency (disproportionate negative contribution). Knee OST and knee JSN contributed greater than average levels to both sHA and uCTX2. These plots again demonstrate the positive association of sCOMP with OST, but a negative association, with the exception of JSN of the CMC, of sCOMP with JSN.
The linear coefficient was derived for the association of each biomarker with OST and JSN for each joint site. The mean linear coefficient from all joint sites is represented by the red line (for OST) and the blue line (for JSN) depicting the strength of the association. The coefficients are shown for serum log sHA (top), log sCOMP (middle), and log uCTX2 (bottom) by number of joint faces affected by any osteophyte (grade >0), and number of joint faces affected based on any joint space narrowing (grade >0). All features were included in each model to assess the contribution of each feature in the presence of the others, i.e. we accounted for correlations among features. Error bars represent 95% confidence intervals.
Due to the interrelationship of JSN and OST, (illustrated below for the CMC joint and an issue for most other joints as well), misleading conclusions can be drawn by examining given radiographic features in isolation from the rest. Figure S5 (lower panel) shows the results of such an unadjusted analysis, where OST or JSN in each joint was used as the sole predictor for uCTX2. There are some notable differences between these results and those shown in the uCTX2 panel of Figure 3 (adjusted). In both the unadjusted and adjusted analyses, uCTX2 is associated with CMC JSN. In contrast, uCTX2 was associated with CMC OST in the unadjusted analyses but when OST and JSN are accounted for in all joints, the association of uCTX2 and OST not only declined, it became nominally negative, with borderline statistical significance. Figure S6 illustrates the interaction of OST and JSN as contributors to uCTX2 for the CMC joint. It shows that when CMC joint faces affected by OST are measured alone they indicate the likelihood of joint faces affected by JSN, which in turn is predictive of increased levels of uCTX2. If both CMC JSN and OST are measured, greater OST involvement predicts a lower uCTX2 concentration.
Optimal prediction of biomarker concentration
Because radiographic features have the potential to confound each other, we set out to determine the best combination of features for the prediction of biomarker concentrations. We considered all linear mixed-effects models with family, age, height, and weight as well as JSN and OST of each of the seven joint systems as possible linear predictors. For some analyses, we combined PIP and DIP finger joints into a single class, interphalangeal joints (IP). Since each of these terms can be either included or excluded in any given model, there are 217 = 131,072 models to evaluate, which we did exhaustively. We used the Akaike Information Criterion (AIC) ,  to evaluate each model. The resulting optimal models are shown in Table 1. Notably, uCTX2, distinct from the other two biomarkers, had no significant relationship to age, height or weight after accounting for the radiographic features of OA. IP JSN was negatively predictive of sCOMP concentrations. In addition, we computed the Akaike importance weights to examine the relative importance of each joint system apart from their role in the single optimal model (Figure 4). The Akaike weight for a given model is proportional to exp(-AIC/2). The weights are normalized to sum to one over all models in the class under consideration. The Akaike importance weight for a predictor is the sum of Akaike weights over all models in which that predictor is included. The Akaike importance weights (0–1) are intended to provide a measure of the importance of a predictor with respect to a whole class of models, rather than to any individual model, and are thus robust against model mis-specification . Shown in Table 1 (taking into account all joints): sHA reports on MCP JSN, CMC JSN, hip OST and knee OST; sCOMP reports on IP JSN (inverse correlation), MCP OST, MCP JSN (inverse correlation), lumbar spine OST, and knee OST; and uCTX2 reports on CMC OST (inverse correlation), CMC JSN, lumbar spine OST, hip OST and knee OST.
All first-order linear models were fit, and the Akaike weight computed for each one. The bar height is the Akaike importance weight, the sum of the Akaike weights over all models: the greater the height, the more predictive of biomarker concentration.
Use of serum biomarkers to predict OA
Ultimately, we are interested in the information about disease state that measurement of these serum biomarkers can convey. Moreover, because of the wide distribution of predictive covariates, it is plausible that the use of all three systemic biomarkers could suffice to identify which joint-system is affected. To take steps in this direction we sought to determine which covariates, including risk factors (age, height, and weight) and biomarkers, are most useful for predicting the OA burden in each of the joint systems studied. For this purpose we used exhaustive testing of all 26 = 64 linear regression models with interactions among predictors excluded for each of the 2×6 outcomes of interest. The outcome variables used were the percent maximum osteophyte burden and percent maximum joint-space narrowing burden, defined as the sum of the radiographic osteophyte or joint-space narrowing scores, respectively, over all joints within each of the six joint systems, or over all joints in total, divided by the maximum of the relevant sum (maximum sum scores for radiographic features provided in Table S2), and expressed as a percentage. Model selection, as above, was based on the AIC. To determine the proportion of the total variance attributable to the biomarker component of the model, we recomputed the regression with the biomarkers removed. The results of these analyses for total body burden, including Akaike importance weights, are presented in Table 2; complete results for all joint systems are presented in Supplementary Table S3. In all cases, namely individual joint systems and total body burden of disease, the addition of biomarkers to traditional covariates increased the proportion of variance in OA explained. A total of 35–38% of the variance in total body burden of OA (total JSN or total OST) was accounted for by the full model with all three biomarkers, a much smaller proportion of the variance was accounted for in the individual joint systems (Table S3). Total body burden of OST could be optimally predicted with sHA and uCTX2 alone, while total body burden of JSN was best predicted by the use of all three biomarkers.
We have suspected for some time that the correlation of systemic biomarkers with disease has been hampered by the inability to fully phenotype the burden of OA in a patient, and in a sense, giving systemic biomarkers a ‘bad name’. The results presented here illustrate the strong correlation of total body burden of disease and several biomarkers, revealed upon adequate patient phenotyping. These results clearly show that all three biomarkers, sHA, sCOMP and uCTX2 are quantitative traits of the radiographic feature of OST while sCOMP is a negative indicator of joint space narrowing affected joint faces in the body. This paper may serve some in the relevant medical communities as an introduction to some novel methods for assessing biomarkers in combination, joint systems in combination, and OA features in combination. For instance, evaluated in combination using multimodel inference with linear mixed-effects regression, it becomes evident that sCOMP adds little information for predicting total OST burden when sHA and uCTX2 are available, although sCOMP is a marker of osteophytes.
There have been several previous studies that used OA-related biomarkers in combination to quantify and characterize the OA process , , , , , . Four notable examples , , ,  used principal components analysis (PCA). Meulenbelt et. al. identified three components that reflected radiographic OA at different joint sites. The components were comprised of 1) structural markers of cartilage (uCTX2, uTINE, uGlc-Gal-PYD) and bone turnover (uCTX-1 and total sOsteocalcin) associated with hip radiographic OA; 2) a marker of inflammation (serum hsCRP) associated with knee OA, high Western Ontario and McMaster Universities (WOMAC) scores and BMI; and 3) markers of cartilage turnover (sPIIANP, sCOMP) associated with radiographic OA of the hands and lumbar spine as well as age. PCA by Davis et. al. has also identified several factors based on biomarkers that correlated with OA features: 1) a factor representing osteophytes that overlapped, as expected, with a factor that correlated with Kellgren Lawrence score; 2) a separate factor that correlated with subchondral bone mineral density; and 3) a factor that correlated with joint space width that overlapped with biomarkers associated with both osteophytes and bone mineral density . In work by Garnero et al , 10 markers segregated into 5 factors: Factor 1 comprised markers of bone (S-PINP, U-CTX-I) and cartilage (U-CTX2) turnover; factor 2 comprised of SCOMP, S-HA, and S-PIIINP; factor 3 comprised of markers of systemic inflammation S-CRP and S-YKL-40; and factors 4 and 5 comprised of MMP-1 and MMP-3 respectively. Prior work by Otterness et al  showed that 14 biomarkers segregated into 5 rational groups based on inflammation, bone turnover, cartilage anabolism, cartilage catabolism and transforming growth factor beta.
These studies support our finding that different biomarkers report on different aspects of joint pathology. Specifically, different biomarkers may reflect different molecular pathobiologic mechanisms and different joint contributions to the systemic biomarker concentrations. The interpretation of biomarker levels may be further complicated by the complex biology represented by a biomarker. For instance, decreased levels may reflect reduced matrix degradation, decreased synthesis, or impaired release from the tissue or origin . Osteophyte formation can be considered an anabolic phenomenon and all three biomarkers reported independently on this feature of joint pathology. In contrast, joint space narrowing can be considered a phenomenon of catabolism in excess of anabolism, or a net failure of repair. The negative correlation of COMP and joint space narrowing could represent the failed repair phenomenon (waning of an anabolic epitope) with increased disease severity, or depletion of a catabolic epitope with increased disease severity. In this cohort, neither HA nor CTX2 reported independently on JSN. These findings only came to light with comprehensive phenotyping that demonstrated the utility of total body measures of OA when qualifying a systemic OA-related biomarker for a specific purpose. Moreover, these results also highlight the importance of accounting for both OST and JSN. Failure to account for both is problematic and may lead to spurious conclusions as demonstrated here through comparing and contrasting the results obtained from adjusted versus unadjusted analyses (Figure 2 versus Figure S3, and Figure 3 versus Figure S5).
The biomarker uCTX2 was independent of age while sHA and sCOMP increased with age; uCTX2 was also independent of height and weight, as was sCOMP while sHA increased with weight and decreased with height. The biomarkers increased the variance explained over simple demographic factors for predicting total body burden of disease. The different proportions of variance explained by biomarkers for the different joint systems may be related to genetic variation or to measurement variation, i.e. adequacy of phenotyping. In this regard, it is interesting that biomarkers added most to the prediction of knee OA, for which a standardized radiographic phenotyping procedure was used. This specialized x-ray may more accurately represent the disease. Thus, the biomarkers may not be so marginal but rather, true correlations of systemic biomarkers with disease burden may be higher, and only demonstrable with better gold standards for measuring the disease.
The coefficients of a linear model estimate the average contribution per affected joint to the serum concentration of a biomarker. OST and JSN had very similar coefficients for sHA and uCTX2, however, there were cases where the difference among OST and JSN were significant and may have real consequences for the use of these biomarkers. One example of the potential complex interrelationship of OST and JSN was demonstrated by the increase of uCTX2 with increasing number of CMC joint faces affected by JSN. When the number of joint faces affected by JSN was held constant, however, uCTX2 declined with increasing number of CMC joint faces affected by OST. Another point, convincingly demonstrated herein by the large contribution of CMC-OST to both sHA and uCTX2 concentrations, is that joint size does not necessarily determine the contribution of a joint to the systemic biomarker concentration. One could truly say, based on these data that the CMC “sticks out like a sore thumb”. We do not know why the CMC joint, despite its small size, disproportionally impacted the concentration of all three systemic biomarkers. It is possible that the stage of disease accounted for this phenomenon. For instance, relative to other joints, increasing JSN in the CMC represented increasingly active matrix turnover while increasing CMC osteophyte severity, an anabolic phenomenon, represented waning activity with respect to biomarker production. Another possible explanation relates to relative turnover and clearance of the biomarkers from the CMC joint. Strong evidence exists for differences in joint tissue turnover for different joint systems, best exemplified by differences between knees and ankles .
Radiographic features of OST and JSN are interrelated and this correlation can prove to be a potential confounder in attempts to characterize the independent contributors to the systemic level of a biomarker. There are just two examples of OST and JSN features from a single joint system contributing independently to the concentration of a systemic biomarker: MCP on sCOMP, and CMC on uCTX2 (Table 1). In both cases, JSN and OST features had opposing effects on biomarker concentration. This study was limited to women, so these results will need to be validated in other cohorts, including men, selected by other means. Further, because the study participants were selected on the basis of familial hand OA, they may have had greater correlation of OA among joint sites than individuals without so clearly a familial etiology of OA. For this reason, the generalizability of these results would need to be evaluated. Finally, these analyses were limited to the association of a discrete set of biomarkers and structural changes. The field would benefit from similar analyses of a broader range of biomarkers, and qualification studies related to clinical and symptomatic patient-reported outcomes to complement those presented here related to structural endpoints.
Materials and Methods
Written informed consent was obtained from all participants and the study was approved by the institutional review boards of both Duke University and the University of North Carolina at Chapel Hill. The investigation was conducted in accordance with the principles expressed in the Declaration of Helsinki.
The participants for these analyses were enrolled between 1999 and 2002 in the GOGO (Genetics of Generalized Osteoarthritis) study and evaluated at two sites (Duke University and the University of North Carolina at Chapel Hill) . A qualifying family consisted of at least 2 siblings with self-reported Western European descent who fulfilled clinical GOGO hand OA criteria: bony enlargement of ≥3 joints distributed bilaterally, including bony enlargement of at least one DIP joint, and no more than 3 swollen MCP joints. Recruitment was independent of clinical symptoms. Once the sibling pair was positively identified as being clinically affected, the nuclear family was invited to participate, together with potential affected or unaffected siblings beyond the required two affecteds. The cohort consisted of 1060 individuals, 840 women and 220 men. Serum and urine biomarker analyses were conducted on the 461 women from this cohort for whom full radiographic data were available. For this study we focused on the 461 women because OA patterns may vary between sexes.
Weight and height were measured. Participants underwent radiographic evaluation of both hands, hips, knees and lumbar spine as previously reported . Radiographs were scored for Kellgren Lawrence grade (0–4 scale) , and individual radiographic features of OA, osteophyte (OST) and joint space narrowing (JSN), on a 0–3 scale using a standard photographic radiological atlas . All joints, including the lumbar spine, were scored for OST. JSN was scored for DIP, PIP and CMC finger joints, MCP hand joints, knees and hips. The lumbar spine was graded for disc space narrowing which, for purposes of these analyses, was considered JSN. Multiple compartments (or levels in the case of lumbar spine) were graded for these features. The scoring system and possible scores for each joint system are summarized in Table S2 and the number of affected joint systems by KL grade is provided in Table S1. Of the phenotypic data collected, age, and measured weight and height were used in these analyses.
Blood for sera, and an aliquot of unspun urine, were collected and stored at −80°C for biomarker analyses. Serum samples were analyzed in duplicate for sHA using a commercially available kit (Corgenix, Westminster, CO), and sCOMP using an in-house enzyme-linked immunosorbent assay (ELISA) as previously described . The minimum detectable concentrations of sHA and sCOMP were 10 ng/ml and 120 ng/ml, and intra-and inter-assay coefficients of variation (CVs) were <5% and <7% for sHA, and 2.1% and 14.2% for sCOMP. Urinary CTX2 (Nordic Bioscience Diagnostics CartiLaps®, Herlev, Denmark) was measured per the manufacturer's instructions by competitive ELISA to detect the degradation product, C-terminal telopeptides of type II collagen (uCTX2) in urine samples. The concentration of creatinine was measured using a commercially available colorimetric kit (METRA™ QUIDEL Corp., San Diego, CA), and uCTX2 values were corrected for urine creatinine concentration (mmol/L). The minimum detectable concentration of uCTX2 is reported as 0.20 ng/ml by the manufacturer. Intra-and inter-assay CVs were 5.9% and 9.9%, respectively. There is strong evidence for association with OA at specific joint sites for each of these biomarkers , , , , ,  but association with total body burden of disease has not been attempted previously.
All statistical analyses were performed using S-Plus 7.0 (Insightful Corp). For each of the three proteins (sHA, sCOMP and uCTX2), we fit linear mixed-effects models using family as the random factor. These models allowed us to account for the possibility that there was an inheritable component of constitutive levels of expression for these proteins given that most subjects had one or more relatives in the database. Model estimation was done using Maximum Likelihood (rather than Restricted Maximum Likelihood) to permit comparisons among models using Akaike's Information Criterion (AIC). AIC provides a score that represents the balance between model fit and the number of parameters (model complexity) . Lower values for the AIC indicate more suitable models. Predictor selection was performed using AIC and exhaustive enumeration of all first-order linear models over both mixed-effects models and fixed-effects models (the latter of which exclude family as a predictive covariate). R2 values reported for mixed-effects models are computed using Nagelkerke's generalized R2 . Model averaging was performed by computing the Akaike weight for each model as where C is a normalizing constant, and summing over all first-order linear models . Adjustments for the risk factors age, height, and weight were made individually for each biomarker by finding the combination of these factors that give the lowest AIC. This procedure provided the “base model” against which all further comparisons were made.
The S-language scripts used to run these analyses are available upon request.
Classification of participants (N = 461) by radiographic osteoarthritis status.
(0.03 MB DOC)
Scoring system for radiographic features of osteoarthritis.
(0.04 MB DOC)
Predictors and their coefficients comprising the minimum-AIC models for percentage of maximum osteoarthritis burden.
(0.06 MB DOC)
Age-related prevalence of OA based on radiographic features. Prevalence (total number of cases in the age group, divided by the number of individuals in the age group) of OA of affected joint systems based on any osteophyte (Fig. S1A), or any joint space narrowing (Fig. S1B) by decade of age. The error bars show exact binomial 95% confidence intervals. The numbers of subjects with OA of affected joint systems based on the Kellgren Lawrence grade >1 grade (Fig. S1C) by decade of age. The joint subtypes evaluated included the interphalangeal (IP) finger joints (combination of distal and proximal interphalangeal joints), metacarpophalangeal (MCP or knuckle) hand joints, carpometacarpal (CMC or base of thumb) joint, lumbar spine, hip, and knee joints.
(0.25 MB PDF)
Biomarkers and number of affected joint systems based on traditional Kellgren Lawrence (KL) scoring. Mean log sHA (top), log sCOMP (middle), and log uCTX2 (bottom), by number of affected joint systems with OA based on Kellgren Lawrence (KL) grade >1. These models were fit with the following risk factors: age, log height, log weight for sHA; age for sCOMP; and for uCTX2 no other covariate was used. The respective R2 values, giving the proportion of the among-KL grade variation explained by the linear regression are shown. Accompanying p-values are 3.6×10−7, 0.024, and 3.6×10−5.
(0.66 MB TIF)
Biomarkers and total burden of OA based on radiographic features (unadjusted). Point estimates are plotted for mean serum log sHA (top), log sCOMP (middle), and log uCTX2 (bottom) by number of joint faces affected by any osteophyte (grade >0, left panels), and number of joint faces OA affected by any joint space narrowing (grade >0, right panels). These unadjusted data are a companion to the adjusted data shown in Figure 2.
(3.00 MB TIF)
Examples of biomarker concentrations and radiographic features of OA in specific joint systems (adjusted). Point estimates, 95% confidence intervals, and mixed-model regression lines for mean serum log sHA (top), log sCOMP (middle), and log uCTX2 (bottom) by number of joint faces affected by any osteophyte (grade >0, left panels), and number of joint faces affected by any joint space narrowing (grade >0, right panels). Each panel represents an analysis that starts with the optimal mixed-effects model for each biomarker as developed in the text, and supplements it with the predictor of interest if it is not already in the optimal model. The line in each plot represents the prediction under the supplemented mixed-effects model. R2 values are indicated giving the among-radiography class explained variation. Corresponding p-values are (left to right by row): 1.9×10−4, 0.039, 0.11, 3.3×10−4, 1.3×10−3, 0.18.
(1.98 MB TIF)
Linear coefficients of biomarker associations with radiographic features of OA (unadjusted). In this example, features are fit independently, so that correlations among features are not accounted for; the relationships between features and biomarkers in this analysis are complicated by these correlations (compare with Figure 3). Linear coefficients for each joint group and radiographic feature are shown.
(1.71 MB TIF)
Example of complex interaction of OST and JSN on a biomarker concentration. Mean log uCTX2 concentrations varied positively with JSN but negatively with OST. For the depicted component of the mixed-effects model, R2 = 0.977, p = 5.4×10−4.
(1.05 MB TIF)
Conceived and designed the experiments: VK JBR JMJ. Performed the experiments: TS JBR. Analyzed the data: TBK. Contributed reagents/materials/analysis tools: VK TBK JMJ. Wrote the paper: VK TBK. Supervised performance of the experiments: VK. Reviewed the paper: TS JBR JMJ.
- 1. March LM, Bachmeier CJ (1997) Economics of osteoarthritis: a global perspective. Baillieres Clin Rheumatol 11: 817–834.LM MarchCJ Bachmeier1997Economics of osteoarthritis: a global perspective.Baillieres Clin Rheumatol11817834
- 2. Zhang Y, Jordan JM (2008) Epidemiology of osteoarthritis. Rheum Dis Clin North Am 34: 515–529.Y. ZhangJM Jordan2008Epidemiology of osteoarthritis.Rheum Dis Clin North Am34515529
- 3. Arden N, Nevitt MC (2006) Osteoarthritis: epidemiology. Best Pract Res Clin Rheumatol 20: 3–25.N. ArdenMC Nevitt2006Osteoarthritis: epidemiology.Best Pract Res Clin Rheumatol20325
- 4. Nelson A, Kraus V, Sullivan P, Hochberg M, Jordan J (2007) Genetics of osteoarthritis. Curr Prim Care Rep 2: 109–116.A. NelsonV. KrausP. SullivanM. HochbergJ. Jordan2007Genetics of osteoarthritis.Curr Prim Care Rep2109116
- 5. Valdes AM, Spector TD, Doherty S, Wheeler M, Hart DJ, et al. (2008) Association of the DVWA and GDF5 polymorphisms with osteoarthritis in UK populations. AM ValdesTD SpectorS. DohertyM. WheelerDJ Hart2008Association of the DVWA and GDF5 polymorphisms with osteoarthritis in UK populations.Ann Rheum Dis: Dec 3 [Epub ahead of print]. Ann Rheum Dis: Dec 3 [Epub ahead of print].
- 6. Kraus VB, Jordan JM, Doherty M, Wilson AG, Moskowitz R, et al. (2007) The Genetics of Generalized Osteoarthritis (GOGO) study: study design and evaluation of osteoarthritis phenotypes. Osteoarthritis Cartilage 15: 120–127.VB KrausJM JordanM. DohertyAG WilsonR. Moskowitz2007The Genetics of Generalized Osteoarthritis (GOGO) study: study design and evaluation of osteoarthritis phenotypes.Osteoarthritis Cartilage15120127
- 7. Wagner JA, Williams SA, Webster CJ (2007) Biomarkers and surrogate end points for fit-for-purpose development and regulatory evaluation of new drugs. Clin Pharmacol Ther 81: 104–107.JA WagnerSA WilliamsCJ Webster2007Biomarkers and surrogate end points for fit-for-purpose development and regulatory evaluation of new drugs.Clin Pharmacol Ther81104107
- 8. Rousseau JC, Delmas PD (2007) Biological markers in osteoarthritis. Nat Clin Pract Rheumatol 3: 346–356.JC RousseauPD Delmas2007Biological markers in osteoarthritis.Nat Clin Pract Rheumatol3346356
- 9. Bauer DC, Hunter DJ, Abramson SB, Attur M, Corr M, et al. (2006) Classification of osteoarthritis biomarkers: a proposed approach. Osteoarthritis Cartilage 14: 723–727.DC BauerDJ HunterSB AbramsonM. AtturM. Corr2006Classification of osteoarthritis biomarkers: a proposed approach.Osteoarthritis Cartilage14723727
- 10. Garnero P, Piperno M, Gineyts E, Christgau S, Delmas PD, et al. (2001) Cross sectional evaluation of biochemical markers of bone, cartilage, and synovial tissue metabolism in patients with knee osteoarthritis: relations with disease activity and joint damage. Annals Rheum Dis 60: 619–626.P. GarneroM. PipernoE. GineytsS. ChristgauPD Delmas2001Cross sectional evaluation of biochemical markers of bone, cartilage, and synovial tissue metabolism in patients with knee osteoarthritis: relations with disease activity and joint damage.Annals Rheum Dis60619626
- 11. Pavelka K, Forejtova S, Olejarova M, Gatterova J, Senolt L, et al. (2004) Hyaluronic acid levels may have predictive value for the progression of knee osteoarthritis. Osteoarthritis & Cartilage 12: 277–283.K. PavelkaS. ForejtovaM. OlejarovaJ. GatterovaL. Senolt2004Hyaluronic acid levels may have predictive value for the progression of knee osteoarthritis.Osteoarthritis & Cartilage12277283
- 12. Elliott AL, Kraus VB, Luta G, Stabler T, Renner JB, et al. (2005) Serum hyaluronan levels and radiographic knee and hip osteoarthritis in African Americans and Caucasians in the Johnston County Osteoarthritis Project. Arthritis Rheum 52: 105–111.AL ElliottVB KrausG. LutaT. StablerJB Renner2005Serum hyaluronan levels and radiographic knee and hip osteoarthritis in African Americans and Caucasians in the Johnston County Osteoarthritis Project.Arthritis Rheum52105111
- 13. Chua SD Jr, Messier SP, Legault C, Lenz ME, Thonar EJ, et al. (2008) Effect of an exercise and dietary intervention on serum biomarkers in overweight and obese adults with osteoarthritis of the knee. Osteoarthritis Cartilage. SD Chua JrSP MessierC. LegaultME LenzEJ Thonar2008Effect of an exercise and dietary intervention on serum biomarkers in overweight and obese adults with osteoarthritis of the knee.Osteoarthritis Cartilage
- 14. Bruyere O, Collette J, Kothari M, Zaim S, White D, et al. (2006) Osteoarthritis, magnetic resonance imaging, and biochemical markers: a one year prospective study. Ann Rheum Dis 65: 1050–1054.O. BruyereJ. ColletteM. KothariS. ZaimD. White2006Osteoarthritis, magnetic resonance imaging, and biochemical markers: a one year prospective study.Ann Rheum Dis6510501054
- 15. Manicourt DH, Azria M, Mindeholm L, Thonar EJ, Devogelaer JP (2006) Oral salmon calcitonin reduces Lequesne's algofunctional index scores and decreases urinary and serum levels of biomarkers of joint metabolism in knee osteoarthritis. Arthritis Rheum 54: 3205–3211.DH ManicourtM. AzriaL. MindeholmEJ ThonarJP Devogelaer2006Oral salmon calcitonin reduces Lequesne's algofunctional index scores and decreases urinary and serum levels of biomarkers of joint metabolism in knee osteoarthritis.Arthritis Rheum5432053211
- 16. Davis CR, Karl J, Granell R, Kirwan JR, Fasham J, et al. (2007) Can biochemical markers serve as surrogates for imaging in knee osteoarthritis? Arthritis Rheum 56: 4038–4047.CR DavisJ. KarlR. GranellJR KirwanJ. Fasham2007Can biochemical markers serve as surrogates for imaging in knee osteoarthritis?Arthritis Rheum5640384047
- 17. Clark AG, Jordan JM, Vilim V, Renner JB, Dragomir AD, et al. (1999) Serum cartilage oligomeric matrix protein reflects osteoarthritis presence and severity: the Johnston County Osteoarthritis Project. Arthritis & Rheumatism 42: 2356–2364.AG ClarkJM JordanV. VilimJB RennerAD Dragomir1999Serum cartilage oligomeric matrix protein reflects osteoarthritis presence and severity: the Johnston County Osteoarthritis Project.Arthritis & Rheumatism4223562364
- 18. Hunter DJ, Li J, LaValley M, Bauer DC, Nevitt M, et al. (2007) Cartilage markers and their association with cartilage loss on magnetic resonance imaging in knee osteoarthritis: the Boston Osteoarthritis Knee Study. Arthritis Res Ther 9: R108.DJ HunterJ. LiM. LaValleyDC BauerM. Nevitt2007Cartilage markers and their association with cartilage loss on magnetic resonance imaging in knee osteoarthritis: the Boston Osteoarthritis Knee Study.Arthritis Res Ther9R108
- 19. Garnero P, Ayral X, Rousseau JC, Christgau S, Sandell LJ, et al. (2002) Uncoupling of type II collagen synthesis and degradation predicts progression of joint damage in patients with knee osteoarthritis. Arthritis & Rheumatism 46: 2613–2624.P. GarneroX. AyralJC RousseauS. ChristgauLJ Sandell2002Uncoupling of type II collagen synthesis and degradation predicts progression of joint damage in patients with knee osteoarthritis.Arthritis & Rheumatism4626132624
- 20. Christgau S, Tanko LB, Cloos PA, Mouritzen U, Christiansen C, et al. (2004) Suppression of elevated cartilage turnover in postmenopausal women and in ovariectomized rats by estrogen and a selective estrogen-receptor modulator (SERM). Menopause 11: 508–518.S. ChristgauLB TankoPA CloosU. MouritzenC. Christiansen2004Suppression of elevated cartilage turnover in postmenopausal women and in ovariectomized rats by estrogen and a selective estrogen-receptor modulator (SERM).Menopause11508518
- 21. Christgau S, Henrotin Y, Tanko LB, Rovati LC, Collette J, et al. (2004) Osteoarthritic patients with high cartilage turnover show increased responsiveness to the cartilage protecting effects of glucosamine sulphate. Clinical & Experimental Rheumatology 22: 36–42.S. ChristgauY. HenrotinLB TankoLC RovatiJ. Collette2004Osteoarthritic patients with high cartilage turnover show increased responsiveness to the cartilage protecting effects of glucosamine sulphate.Clinical & Experimental Rheumatology223642
- 22. Meulenbelt I, Kloppenburg M, Kroon HM, Houwing-Duistermaat JJ, Garnero P, et al. (2006) Urinary CTX-II levels are associated with radiographic subtypes of osteoarthritis in hip, knee, hand, and facet joints in subject with familial osteoarthritis at multiple sites: the GARP study. Ann Rheum Dis 65: 360–365.I. MeulenbeltM. KloppenburgHM KroonJJ Houwing-DuistermaatP. Garnero2006Urinary CTX-II levels are associated with radiographic subtypes of osteoarthritis in hip, knee, hand, and facet joints in subject with familial osteoarthritis at multiple sites: the GARP study.Ann Rheum Dis65360365
- 23. Garnero P, Aronstein WS, Cohen SB, Conaghan PG, Cline GA, et al. (2008) Relationships between biochemical markers of bone and cartilage degradation with radiological progression in patients with knee osteoarthritis receiving risedronate: the Knee Osteoarthritis Structural Arthritis randomized clinical trial. Osteoarthritis Cartilage 16: 660–666.P. GarneroWS AronsteinSB CohenPG ConaghanGA Cline2008Relationships between biochemical markers of bone and cartilage degradation with radiological progression in patients with knee osteoarthritis receiving risedronate: the Knee Osteoarthritis Structural Arthritis randomized clinical trial.Osteoarthritis Cartilage16660666
- 24. Kellgren JH, Lawrence JS (1957) Radiological assessment of osteo-arthrosis. Ann Rheum Dis 16: 494–502.JH KellgrenJS Lawrence1957Radiological assessment of osteo-arthrosis.Ann Rheum Dis16494502
- 25. Gordon CD, Stabler TV, Kraus VB (2008) Variation in osteoarthritis biomarkers from activity not food consumption. Clin Chim Acta 398: 21–26.CD GordonTV StablerVB Kraus2008Variation in osteoarthritis biomarkers from activity not food consumption.Clin Chim Acta3982126
- 26. Meulenbelt I, Kloppenburg M, Kroon HM, Houwing-Duistermaat JJ, Garnero P, et al. (2007) Clusters of biochemical markers are associated with radiographic subtypes of osteoarthritis (OA) in subject with familial OA at multiple sites. The GARP study. Osteoarthritis Cartilage 15: 379–385.I. MeulenbeltM. KloppenburgHM KroonJJ Houwing-DuistermaatP. Garnero2007Clusters of biochemical markers are associated with radiographic subtypes of osteoarthritis (OA) in subject with familial OA at multiple sites. The GARP study.Osteoarthritis Cartilage15379385
- 27. Garnero P, Sornay-Rendu E, Arlot M, Christiansen C, Delmas PD (2004) Association between spine disc degeneration and type II collagen degradation in postmenopausal women: the OFELY study. Arthritis & Rheumatism 50: 3137–3144.P. GarneroE. Sornay-RenduM. ArlotC. ChristiansenPD Delmas2004Association between spine disc degeneration and type II collagen degradation in postmenopausal women: the OFELY study.Arthritis & Rheumatism5031373144
- 28. Akaike H (1974) A new look at the statistical model identification. Automatic Control, IEEE Transactions 19: 716–723.H. Akaike1974A new look at the statistical model identification.Automatic Control, IEEE Transactions19716723
- 29. Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information-theoretic approach: Springer. KP BurnhamDR Anderson2002Model selection and multimodel inference: a practical information-theoretic approach: Springer.
- 30. Dam EB, Loog M, Christiansen C, Byrjalsen I, Folkesson J, et al. (2009) Identification of progressors in osteoarthritis by combining biochemical and MRI-based markers. Arthritis Res Ther 11: R115.EB DamM. LoogC. ChristiansenI. ByrjalsenJ. Folkesson2009Identification of progressors in osteoarthritis by combining biochemical and MRI-based markers.Arthritis Res Ther11R115
- 31. Garnero P, Mazieres B, Gueguen A, Abbal M, Berdah L, et al. (2005) Cross-sectional association of 10 molecular markers of bone, cartilage, and synovium with disease activity and radiological joint damage in patients with hip osteoarthritis: the ECHODIAH cohort. J Rheumatol 32: 697–703.P. GarneroB. MazieresA. GueguenM. AbbalL. Berdah2005Cross-sectional association of 10 molecular markers of bone, cartilage, and synovium with disease activity and radiological joint damage in patients with hip osteoarthritis: the ECHODIAH cohort.J Rheumatol32697703
- 32. Otterness IG, Swindell AC, Zimmerer RO, Poole AR, Ionescu M, et al. (2000) An analysis of 14 molecular markers for monitoring osteoarthritis: segregation of the markers into clusters and distinguishing osteoarthritis at baseline. Osteoarthritis Cartilage 8: 180–185.IG OtternessAC SwindellRO ZimmererAR PooleM. Ionescu2000An analysis of 14 molecular markers for monitoring osteoarthritis: segregation of the markers into clusters and distinguishing osteoarthritis at baseline.Osteoarthritis Cartilage8180185
- 33. Young-Min SA, Cawston TE, Griffiths ID (2001) Markers of joint destruction: principles, problems, and potential. Ann Rheum Dis 60: 545–548.SA Young-MinTE CawstonID Griffiths2001Markers of joint destruction: principles, problems, and potential.Ann Rheum Dis60545548
- 34. Aurich M, Squires GR, Reiner A, Mollenhauer JA, Kuettner KE, et al. (2005) Differential matrix degradation and turnover in early cartilage lesions of human knee and ankle joints. Arthritis Rheum 52: 112–119.M. AurichGR SquiresA. ReinerJA MollenhauerKE Kuettner2005Differential matrix degradation and turnover in early cartilage lesions of human knee and ankle joints.Arthritis Rheum52112119
- 35. Burnett S, Hart DJ, Cooper C, Spector TD (1994) A Radiographic Atlas of Osteoarthritis. London: Springer-Verlag. S. BurnettDJ HartC. CooperTD Spector1994A Radiographic Atlas of Osteoarthritis.LondonSpringer-Verlag
- 36. Vilim V, Olejarova M, Machacek S, Gatterova J, Kraus V, et al. (2002) Serum levels of cartilage oligomeric matrix protein (COMP) correlate with radiographic progression of knee osteoarthritis. Osteoarthritis Cartilage 10: 707–713.V. VilimM. OlejarovaS. MachacekJ. GatterovaV. Kraus2002Serum levels of cartilage oligomeric matrix protein (COMP) correlate with radiographic progression of knee osteoarthritis.Osteoarthritis Cartilage10707713
- 37. Charni-Ben Tabassi N, Garnero P (2007) Monitoring cartilage turnover. Curr Rheumatol Rep 9: 16–24.N. Charni-Ben TabassiP. Garnero2007Monitoring cartilage turnover.Curr Rheumatol Rep91624
- 38. Birmingham J, Vilim V, Kraus V (2006) Collagen biomarkers for arthritis applications. Biomarker Insights 2: 61–76.J. BirminghamV. VilimV. Kraus2006Collagen biomarkers for arthritis applications.Biomarker Insights26176
- 39. Garnero P, Delmas PD (2003) Biomarkers in osteoarthritis. Current Opinion in Rheumatology 15: 641–646.P. GarneroPD Delmas2003Biomarkers in osteoarthritis.Current Opinion in Rheumatology15641646
- 40. Nagelkerke N (2002) A note on a general definition of the coefficient of determination. Biometrika 78: 691–692.N. Nagelkerke2002A note on a general definition of the coefficient of determination.Biometrika78691692