Ethnic and sex differences in serum levels of cartilage oligomeric matrix protein: The Johnston county osteoarthritis project




Previous descriptions of potential biomarkers of osteoarthritis (OA) have been limited to Caucasians. In the present study, we examined associations between serum levels of cartilage oligomeric matrix protein (COMP) and ethnicity (African American or Caucasian) and sex in the Johnston County Osteoarthritis Project, a population-based study of OA in rural North Carolina.


All African Americans and a randomly selected sample of Caucasians who had available sera and either no radiographic evidence of knee or hip OA according to the Kellgren/Lawrence (K/L) system (K/L grade 0) or radiographic evidence of knee OA (K/L grade 2 or higher) were included. Serum COMP levels were quantified by sandwich enzyme-linked immunosorbent assay, using monoclonal antibodies 16-F12 and 17-C10. Linear regression models were used to assess relationships between serum levels of natural log–transformed COMP (ln COMP) and ethnicity and sex, controlling for age, height, body mass index (BMI), radiographic OA, and the presence of other symptomatic joints. Radiographic OA was defined in separate models as the presence, severity, and laterality of radiographic knee OA, the co-occurrence of radiographic knee and hip OA, and the number of knees and hips with radiographic OA.


The 769 subjects in the study sample had a mean ± SD age of 62 ± 10.3 years. Levels of ln COMP were associated with age, BMI, and all definitions of radiographic OA (P = 0.0001), and varied by ethnicity and sex. In adjusted models, ln COMP was higher in African American women than in Caucasian women (P = 0.003) and higher in Caucasian men than Caucasian women (P = 0.0001). There were no statistically significant differences in serum ln COMP levels between African American men and women.


Serum COMP levels vary by ethnicity and sex. These factors should be considered in the derivation of standards using this, and possibly other, potential biomarkers of OA.

Cartilage oligomeric matrix protein (COMP) is a 524-kd pentameric glycoprotein that is found predominantly in cartilage, but is also present in ligament, tendon, meniscus, and synovium (1–3). Elevated serum levels of COMP have been found in persons with knee osteoarthritis (OA) compared with healthy individuals (2), in persons with OA-related type II collagen gene mutations (4), in those with knee OA with synovitis compared with those with knee OA without synovitis (3), and in those with systemic inflammatory arthropathy due to rheumatoid arthritis (5, 6). Using monoclonal antibody 17-C10 in an inhibition enzyme-linked immunosorbent assay (ELISA), we previously performed the largest population-based analysis of serum COMP in Caucasian cases with radiographically defined knee OA and in controls without radiographic knee or hip OA (7). We documented elevations in serum COMP concentrations with age and with the presence and severity of radiographic knee OA, bilateral radiographic knee OA, concomitant radiographic hip and knee OA, and the number of knees and hips with radiographic evidence of OA (7).

All previous studies of serum COMP, and indeed of any OA-related biomarker, have been limited to Caucasians. The present study is the first to examine ethnic and sex differences in serum COMP levels in African Americans and Caucasians with and without radiographic knee and hip OA from a single, large, population-based sample, using a new sandwich ELISA based on two monoclonal antibodies, 16-F12 and 17-C10. Based on ethnic variability in biomarkers for other conditions (8, 9), we hypothesized that there might be ethnic differences in serum COMP levels, but we expected that these levels would be associated with the presence and severity of OA in African Americans, as was previously noted in Caucasians. We found ethnic and sex differences in serum COMP levels and noted similar associations with serum COMP and radiographic OA in African Americans as in Caucasians.


Study participants.

The Johnston County Osteoarthritis Project is an ongoing, population-based, prospective study of the occurrence and progression of knee and hip OA in a rural, multi-ethnic population of North Carolina. Details of the study design and protocol have been reported elsewhere (10). Briefly, between May 1991 and December 1997, a total of 3,187 African American or Caucasian individuals ages ≥45 years were recruited into the project from the community, using probability sampling, without regard to knee or hip pain or radiographic OA status. All participants underwent clinical evaluation, radiographic assessment of the knees and hips, and serum sampling (10). The present study utilized a selected subset of data from the baseline evaluation.

From the 3,187 study participants, 388 African Americans and 820 Caucasians had an available serum sample that had been drawn at the time of radiography plus either Kellgren/Lawrence (K/L) radiographic grade 0 findings in both knees and both hips (radiographic OA unaffected) or K/L radiographic grade ≥2 changes in at least 1 knee (radiographic knee OA affected). Of these subjects, we selected all 388 African Americans and randomly selected 394 Caucasians, in approximately equal numbers from each of the 12 cells obtained from the cross-classification of age group (ages 45–54 years, 55–64 years, ≥65 years), sex, and OA status (affected, unaffected) for evaluation of serum COMP levels.

One outlier subject and subjects with questionable or missing data for other variables of interest were eliminated. Thus, 379 African Americans and 390 Caucasians were included in the present analysis.

Radiographic and clinical evaluation.

All participants underwent bilateral anteroposterior radiography of the knee with weight bearing; women over the age of 50 years and all men also underwent supine anteroposterior pelvic radiography. Radiographs were read by a single musculoskeletal radiologist (JBR) using the K/L radiographic atlas for overall knee and hip radiographic grades (11), as defined below. Interrater reliability (assessed with another trained radiologist) and intrarater reliability for the radiologist were high (weighted kappa for interrater reliability 0.859; kappa for intrarater reliability 0.886), as described previously (10).

K/L grade 0 (normal findings) was defined as the absence of radiographic features of OA. K/L grade 1 (questionable) included a minute radiographic osteophyte of doubtful pathologic significance. Radiographs showing an osteophyte but no joint space narrowing were assigned a K/L grade of 2 (mild); moderate diminution of the joint space was assigned a K/L grade of 3 (moderate). K/L grade 4 (severe) was defined as severe joint space narrowing with subchondral bone sclerosis (11).

Height without shoes was measured (expressed in centimeters), and weight was measured using a balance beam scale (expressed in kilograms). The presence of other symptomatic joints was determined by subject's report of joint symptoms, using a homunculus (shoulders, neck, upper back/mid-back, lower back, elbows, wrists, ankles, hands, and feet).


Blood was collected, and sera were separated and immediately stored on ice. Sera were frozen at –20°C within 8 hours of collection and were then transferred to a –86°C environment for long-term storage. COMP was quantified by a sandwich ELISA as previously described (12), with a slight modification. Alkaline phosphatase–conjugated avidin was used rather than peroxidase, followed by detection at an absorbance of 405 nm with p-nitrophenyl phosphate as the substrate.

Sera from two normal subjects without disease were pooled, and aliquots of this pool were frozen for use as a normal control. Sera from 4 subjects with OA previously determined to have high COMP levels were pooled, and aliquots of this pool were frozen for use as a high-COMP control. These controls were run with each assay. The results were used to determine the precision of the assay and to establish an acceptable control range for the assay. Interassay (between-run) variability was 9.7% for the normal control (n = 82 assays) and 8.7% for the high-COMP control (n = 82). Intraassay (within-run) variability was 5.8% for the normal control (n = 23) and 6.6% for the high-COMP control (n = 23). An acceptable control range was defined as the mean ± 2SD, as determined from the interassay results. Any assay in which either control fell outside this range was excluded, and that assay was repeated.

Definitions of variables for analysis.

Ethnic group was self-defined as African American or Caucasian. Age and body mass index (BMI) were used as continuous variables in all analyses. The presence and severity of radiographic OA were defined in 5 ways for evaluation in 5 separate analyses, as follows: 1) the presence of grade 2–4 radiographic knee OA in at least 1 knee (yes/no); 2) the severity of radiographic knee OA, defined by K/L grades as a 4-level ordinal variable (normal, mild, moderate, severe; coded 0, 1, 2, 3, respectively); 3) the laterality of radiographic knee OA, defined as a 3-level ordinal variable (none, unilateral, bilateral; coded 0, 1, 2, respectively); 4) the co-occurrence of radiographic knee and hip OA of at least K/L grade 2 in at least 1 knee and 1 hip (none, radiographic knee OA only, radiographic knee and hip OA; coded 0, 1, 2, respectively); and 5) the number of knee and hip joints with radiographic OA, defined as a count from 0 to 4.

Statistical analysis.

All statistical computations were performed using SAS version 8.2 software (SAS Institute, Cary, NC). Comparisons between African Americans and Caucasians with respect to demographic and clinical variables of interest were performed with Student's t-test and Pearson's chi-square test. Since the distribution of serum COMP was skewed, natural log transformation was performed. The transformed data did not violate the assumptions of linearity, normality, and equal variances (homoscedasticity), underlying the use of linear regression model methods. Descriptive statistics for serum COMP and serum ln COMP were calculated, and all analyses were done using the ln COMP data. Associations between ln COMP and categorical variables were assessed using Student's t-test and tests for linear trend (i.e., tests for a linear relationship with ordinal variables with equally spaced values). To evaluate the linear relationship of ln COMP with continuous variables, Pearson's correlation coefficients were calculated and tests for correlation performed.

Separate multivariable linear regression models included terms for ethnic group, sex, age, BMI, and each definition of radiographic OA. Tests for interaction were performed, and effect modification was determined to be present if the P value of the interaction term was less than or equal to 0.10. From each model, the differences between adjusted least squares means (LSMs) and corresponding 95% confidence intervals (95% CIs) for these differences were calculated for categories of interest.


The total sample for analysis included 258 African American women, 121 African American men, 190 Caucasian women, and 200 Caucasian men. The African American sample was slightly older, had a higher proportion of women, and had a higher mean BMI than the Caucasian sample (Table 1). Overall, the mean ln COMP level increased linearly with age (r = 0.32, r2 = 0.10, P = 0.0001), weight, and BMI (r = 0.14, r2 = 0.02, P = 0.0001 for both weight and BMI), but was not associated with height (r = 0.01, r2 = 0.0001, P = 0.78). The linear relationships between age, weight, and BMI, respectively, and the ln COMP level were further supported by a nonsignificant quadratic effect of each variable on the ln COMP. The expected increase in the mean ln COMP level for a 10-year increase in age was 0.125, and the expected increase in the mean ln COMP level for a 10-unit increase in BMI was 0.086.

Table 1. Selected characteristics of the study sample
 African American subjects (n = 379)Caucasian subjects (n = 390)
  • *

    P < 0.001 by Student's t-test or Pearson's chi-square test.

  • P < 0.01 by Pearson's chi-square test.

Age, mean ± SD years*63.3 ± 10.860.6 ± 9.6
Age group, %  
 45–54 years26.731.8
 55–64 years26.933.8
 ≥65 years46.434.4
Sex, % female*68.148.7
Body mass index, mean ± SD kg/m2*31.5 ± 7.428.9 ± 6.1

The mean ln COMP level was higher in men than in women (difference in LSMs 0.06 [95% CI –0.01, 0.12]; P = 0.05) and higher in African Americans than Caucasians (difference in LSMs 0.12 [95% CI 0.06, 0.18]; P = 0.0001). The mean ln COMP level was higher in those with radiographic knee OA compared with those without radiographic knee or hip OA (P = 0.0001 for all radiographic OA definitions) (Table 2). There were no statistically significant differences in mean ln COMP levels according to symptoms in other joint groups of the homunculus; therefore, height and symptomatic joints were not included in multivariable models.

Table 2. Serum levels of COMP and ln COMP according to the presence, severity, and laterality of radiographic knee OA, the co-occurrence of radiographic knee and hip OA, and the number of knees and hips with radiographic OA*
GroupNo. of subjectsCOMP level, median (range) ng/mlln COMP level, mean ± SD
  • *

    Osteoarthritis (OA) was defined as grade 2–4 radiographic changes according to the Kellgren/Lawrence (K/L) scale. P = 0.0001 for all comparisons, by Student's t-test or by test for linear trend for natural log–transformed (ln) COMP.

Presence of radiographic knee OA   
 K/L grade 0 knee OA and grade 0 hip OA302805.6 (170.3–2,131.1)6.68 ± 0.40
 K/L grade 2–4 knee OA467943.2 (196.0–4,017.6)6.83 ± 0.40
Severity of radiographic knee OA   
 K/L grade 0 knee OA and grade 0 hip OA302805.6 (170.3–2,131.1)6.68 ± 0.40
 K/L grade 2 knee OA313890.2 (196.0–3,338.0)6.79 ± 0.40
 K/L grade 3 knee OA110971.9 (381.7–2,299.6)6.87 ± 0.34
 K/L grade 4 knee OA441,102.0 (239.1–4,017.6)7.00 ± 0.48
Laterality of radiographic knee OA   
 K/L grade 0 knee OA and grade 0 hip OA302805.6 (170.3–2,131.1)6.68 ± 0.40
 Unilateral K/L grade 2–4 knee OA223860.4 (196.0–3,338.0)6.77 ± 0.42
 Bilateral K/L grade 2–4 knee OA244982.8 (239.1–4,017.6)6.88 ± 0.38
Co-occurrence of radiographic knee and hip OA   
 K/L grade 0 knee OA and grade 0 hip OA302805.6 (170.3–2,131.1)6.68 ± 0.40
 K/L grade 2–4 knee OA and grade 0 hip OA255928.2 (316.5–2,312.9)6.82 ± 0.36
 K/L grade 2–4 knee OA and grade 2–4 hip OA152971.5 (239.1–4,017.6)6.87 ± 0.41
No. of knees and hips with radiographic OA   
 0 joints affected302805.6 (170.3–2,131.1)6.68 ± 0.40
 1 joint affected131831.6 (316.5–2,312.9)6.74 ± 0.37
 2 joints affected158985.8 (381.7–2,299.6)6.88 ± 0.35
 3 joints affected66971.5 (239.1–2,162.4)6.87 ± 0.43
 4 joints affected521,057.9 (488.1–4,017.6)6.93 ± 0.40

The serum COMP concentration ranged from 170.3 to 4,017.6 ng/ml in the African Americans and from 239.1 to 2,299.6 ng/ml in the Caucasians. Table 3 shows the medians and ranges of serum COMP levels in African American and Caucasian men and women according to the presence of radiographic knee OA in each age group.

Table 3. Serum COMP levels according to radiographic knee OA status and ethnic, sex, and age group*
Radiographic OA status and ethnic, sex, and age groupNo. of subjectsCOMP level, median (range) ng/ml
  • *

    Osteoarthritis (OA) was defined as grade 2–4 radiographic changes according to the Kellgren/Lawrence (K/L) scale. COMP = cartilage oligomeric matrix protein.

K/L grade 0 knee and hip OA  
 African American  
   45–54 years14725.2 (307.8–978.3)
   55–64 years29888.8 (170.3–1,571.5)
   ≥65 years15979.7 (305.6–1,702.8)
   45–54 years24743.3 (372.0–1,398.4)
   55–64 years10791.1 (502.8–1,378.7)
   ≥65 years141,031.8 (427.2–1,741.4)
   45–54 years25655.6 (305.6–1,140.1)
   55–64 years34631.9 (280.9–1,141.7)
   ≥65 years34915.5 (323.4–2,131.1)
   45–54 years35816.9 (293.4–1,958.3)
   55–64 years34813.5 (451.3–1,524.9)
   ≥65 years34962.5 (487.1–1,842.2)
K/L grade 2–4 knee OA  
 African American  
   45–54 years48883.7 (196.0–4,017.6)
   55–64 years45960.8 (392.6–1,654.2)
   ≥65 years1071,075.4 (364.6–3,338.0)
   45–54 years15831.6 (453.0–2,245.3)
   55–64 years18948.0 (316.5–1,411.3)
   ≥65 years401,066.5 (392.1–2,162.4)
   45–54 years31744.1 (279.3–1,703.4)
   55–64 years33764.7 (349.2–1,275.3)
   ≥65 years33891.3 (522.6–2,299.6)
   45–54 years33790.4 (463.9–1,667.7)
   55–64 years31940.8 (372.8–1,529.7)
   ≥65 years331,042.7 (239.1–2,041.9)

The association between the ln COMP level and ethnic group varied by sex, with African American women having higher levels than Caucasian women had (P < 0.0001), but no ethnic difference in these levels was noted in the men (P = 0.470). Caucasian men had higher ln COMP values than did Caucasian women (P < 0.0001), but there was no comparable sex difference in these levels in African Americans (P = 0.848).

Figure 1 shows the mean serum ln COMP levels in African American and Caucasian men and women by age group and OA status. In all subgroups, serum ln COMP levels increased linearly with age and were higher in those with radiographic OA than in those without radiographic OA. Figure 1 also demonstrates that the association between serum COMP levels and sex may vary by ethnicity.

Figure 1.

Mean serum levels of natural log–transformed cartilage oligomeric matrix protein (ln COMP) in A, women and B, men, by radiographic knee osteoarthritis (OA) status (according to the Kellgren/Lawrence [K/L] scale), ethnicity, and age group (years). • = unaffected (K/L grade 0 knee and hip OA) African Americans; ▓ = unaffected Caucasians; ⧫ = affected (K/L grade 2–4 knee OA) African Americans; ▴ = affected Caucasians.

Of all interactions tested, only the ethnic-by-sex interaction was statistically significant in all multivariable models (P < 0.06). Table 4 shows the differences in LSMs and their 95% CIs for the 4 ethnic-by-sex groups, adjusted for age, BMI, and the presence of radiographic knee OA. After adjustment, African American women had significantly higher mean ln COMP levels than did Caucasian women (P = 0.003), and Caucasian men had higher levels than did Caucasian women (P < 0.0001). There was no statistically significant difference in adjusted LSMs between the African American and Caucasian men (P = 0.970). Similar results were seen in models utilizing other OA definitions (data not shown).

Table 4. Differences between LSMs for ln COMP levels and corresponding 95% CIs for comparisons of groups of interest*
GroupDifference between LSM (95% CI)
  • *

    Data were adjusted for age, body mass index, and the presence of Kellgren/Lawrence grade 2–4 radiographic knee osteoarthritis. LSMs = least squares means; ln COMP = natural log–transformed cartilage oligomeric matrix protein; 95% CIs = 95% confidence intervals.

African American women vs. Caucasian women0.111 (0.039, 0.184)
African American men vs. Caucasian men0.002 (−0.083, 0.086)
African American men vs. African American women0.071 (−0.012, 0.154)
Caucasian men vs. Caucasian women0.181 (0.107, 0.255)

Associations between ln COMP levels and age, BMI, and radiographic OA were similar in the 4 ethnic-by-sex groups (P > 0.273 for all second-order interactions with ethnic group and sex). Age and BMI each remained highly statistically significantly associated with ln COMP in all adjusted models (P = 0.0001).


This is the first report of any serum biomarker of OA in African American subjects. The sample is population-based, utilizes strict radiographic criteria for defining affected and unaffected OA status, and is the largest sample thus far in which serum COMP levels were determined using a sandwich ELISA based on 2 monoclonal antibodies, 16-F12 and 17-C10. Our data show ethnic and sex differences that cannot be explained by age, skeletal size as approximated by height, BMI, the presence or severity of radiographic knee or hip OA, or the presence of other symptomatic joints. Importantly, associations between serum COMP levels and age, BMI, and radiographic OA were not dissimilar in African American and Caucasian men and women.

What might underlie these observations? One potential explanation for the ethnic and sex differences in serum COMP levels might be differences in radiographic OA in other joints that were not evaluated in this study. While this is possible, we believe that it is unlikely to be the sole explanation. Recognizing that symptoms do not always reflect radiographic disease, it is nonetheless notable that the serum COMP level was not strongly associated with symptoms in any other joint group we examined. Further, while symptoms in hands were more common in women than in men, they did not differ by ethnic group (data not shown).

Serum COMP levels have been shown to be associated with radiographic OA in large joints, such as the knees and hips (3, 7, 13). Although not previously formally evaluated, one might, by analogy, conjecture that symptoms in the large joint groups of the lumbosacral and cervical spine, if associated with radiographic disease, might also be associated with higher serum COMP levels. If this were true, then we would expect that subjects in our study who had evidence of lumbosacral or cervical spine disease might have higher serum COMP levels than those without such involvement. However, we found that symptoms in these areas were actually more common in Caucasian women than in African American women (data not shown); we would expect, then, to see higher levels of serum COMP in Caucasian women, quite the opposite of what we observed. We cannot rule out the possibility of confounding by occult asymptomatic disc disease and OA in the spine or other joints not evaluated in this study. Whether ethnic and/or sex differences in the experience of OA or in the reporting of symptoms might influence these issues is also open to conjecture.

Possible differences in bone density and metabolism, body composition, skeletal and joint size, cartilage and tendon mass, or COMP levels in other organ systems could also be proposed as potential explanations for the ethnic and sex differences we observed. The recent report of the presence of COMP in osteoblasts (14) suggests that differences in osteoblastic activity and expression of COMP could possibly account for the ethnic and sex effects in COMP levels which we observed. A growing body of literature suggests that the rate of bone turnover is important in OA pathogenesis (15–19) and that cartilage and bone metabolism are linked (20, 21). We have recently observed an inverse association between serum ln COMP levels and the use of hormone replacement therapy in postmenopausal women (22). Whether this observation is mediated through an effect of hormone therapy on bone and/or cartilage metabolism (23) or whether the observed ethnic and sex associations with COMP levels might be explained through such a mechanism is unclear and will be the subject of further study. We suspect that the serum COMP level probably reflects both a component of bone metabolism and a component of cartilage metabolism, each potentially influenced, singly or together, by sex and ethnicity.

We cannot exclude the possibility that some of the ethnic and sex differences in serum ln COMP levels are related to differences in skeletal or joint size or to the total body content of cartilage, meniscus, and tendon. Magnetic resonance imaging has been used to quantify cartilage volume in the knee (24, 25), but few other joints have been studied in this way, and normative data are few (24). There are currently no standard ways to extrapolate measures from one joint to estimate cartilage content for the entire body. Several investigators have noted associations between height and cartilage volume of the knee (24–26), but determinants of cartilage volume in different joints may not be identical (24). In our study, serum ln COMP levels were not associated with height, which we considered a surrogate measure for skeletal size.

Genetic differences in genes associated with COMP might also exist between the ethnic groups. To date, more than 50 unique COMP mutations have been identified as causing two skeletal dysplasias, pseudoachondroplasia and multiple epiphyseal dysplasia (27). Whether variations in COMP genes, COMP-associated proteins, or COMP regulation related to COMP gene expression explain ethnic and sex differences in serum COMP has not been explored.

As in our earlier study of serum COMP levels using an inhibition ELISA with 1 monoclonal antibody against human COMP (17-C10) (7), this analysis, using a sandwich assay with 2 monoclonal antibodies (16-F12 and 17-C10) has similarly demonstrated associations of serum ln COMP levels with age. The strength of the statistical association between ln COMP levels and age was not affected by adjustment for any definition of radiographic OA, BMI, height, ethnic group, sex, or other symptomatic joints, suggesting that other factors not previously examined must be responsible for the association. As above, a contribution to serum COMP by a greater burden of asymptomatic joint disease with aging cannot be ruled out. Although found predominantly in cartilage and tendon, COMP has also been shown to be expressed in the vascular smooth muscle of arteries (28). It is intriguing to speculate whether differences in cardiovascular disease or other conditions associated with aging in the men and women in the two ethnic groups might explain some of the ethnic and sex differences in serum COMP levels that we observed. Unfortunately, it is not possible at this time to identify the tissue origin of this biomarker.

In contrast to our previous analysis with the inhibition ELISA, ln COMP levels in the present analysis were positively associated with BMI. This relationship was based upon the association of the ln COMP level with weight but not with height. This statistical association was unchanged by controlling for the other covariates. For reasons similar to those cited above, we doubt that this association is solely the result of asymptomatic radiographic OA in sites not examined in this study, but this possibility cannot be excluded. BMI has been noted to be independently associated with the C-propeptide of type II collagen in the synovial fluid of patients with primary knee OA (29). Whether serum COMP is associated with body composition, glucose metabolism, hormone status, bone density, and other factors associated with BMI has not been examined.

In our previous study utilizing the inhibition ELISA with a single monoclonal antibody, 17-C10, ln COMP levels were not associated with obesity, defined as a BMI ≥30 kg/m2 (7), or with BMI. Measurements of serum COMP with the sandwich ELISA utilized in the current study (using monoclonal antibodies 16-F12 and 17-C10) have been shown to correlate strongly with those obtained with the inhibition ELISA using 17-C10 (30). Further, this new sandwich assay has been shown to be highly reproducible, with low day-to-day and diurnal variations (30). One possible reason why our previous 17-C10 analysis failed to show a relationship between BMI and serum ln COMP is that there was a smaller sample size than in the current study; moreover, the previous study did not include African American subjects. Other potential explanations for the different results with the two assays include potential differences in the avidity of COMP to the different monoclonal antibodies, interactions of COMP with other serum components, or other technical differences in the assays (30). Finally, it should be noted, that while the association of BMI with serum ln COMP levels in the current study was highly statistically significant, the amount of the variability in serum ln COMP levels explained by BMI was small.

These results have implications as the significance of biomarkers for OA continues to grow. Importantly, associations with COMP levels and radiographic OA were not significantly different in the two ethnic groups. These results emphasize the need to obtain normative data and suggest that separate consideration be given to men and women and different ethnic groups in the development of standards for this biomarker, and perhaps other biomarkers as well. Despite the variation in this marker with age, sex, and ethnicity, serum COMP levels may prove to be useful in the study of OA, particularly when used in conjunction with other data and potential markers under investigation.