Dr. Eckstein has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Novo Nordisk, Inc., BioSyntech, Novartis, and Wyeth, and (more than $10,000) from Pfizer, Inc. and Merck, and owns stock and/or holds stock options in Chondrometrics GmbH.
To propose a novel strategy for more efficiently measuring changes in cartilage thickness in osteoarthritis (OA) using magnetic resonance imaging, and to hypothesize that determining the magnitude of thickness change independent of the anatomic location provides improved discrimination between healthy subjects and OA participants longitudinally.
A total of 148 women were imaged; 90 were Kellgren/Lawrence (K/L) grade 0, 30 were K/L grade 2, and 28 were K/L grade 3. Magnetic resonance images (3T) were acquired at baseline and at 24 months. Changes in femorotibial cartilage thickness were determined in 5 tibial and 3 femoral medial and lateral subregions, respectively (conventional approach). The new strategy provided ordered values of subregional change in each compartment, ranked according to the direction and magnitude of change.
Using the new ordered values approach, the minimal P value for the differences in 2-year change in medial cartilage thickness of K/L grade 3 and K/L grade 0 participants was 0.001 (Wilcoxon test), with 4 ordered medial subregions differing significantly between both groups. With the conventional approach, only 1 medial subregion differed significantly between K/L grade 3 and K/L grade 0 (P = 0.037). Cartilage thickening was significantly greater in K/L grade 2 versus K/L grade 0 participants in 1 medial subregion using the conventional approach (P = 0.016), and in 2 medial subregions (minimal P = 0.007) using the ordered values approach.
The novel ordered values approach is more sensitive in detecting cartilage thinning in K/L grade 3 and cartilage thickening in K/L grade 2 versus K/L grade 0 participants. The new method may be particularly useful in the context of other comparisons, e.g., a group treated with a disease-modifying OA drug versus one treated with a placebo.
Changes in femorotibial cartilage are of high interest in the study of the structural progression of knee osteoarthritis (OA) (1). Magnetic resonance imaging (MRI) can provide quantitative data on cartilage morphology and its changes over time (2–4). Because observational studies have reported that cartilage lesions occur and progress more frequently in some parts of the joint than in others (5), computational algorithms have been applied to study cartilage morphology metrics in defined anatomic subregions of the femorotibial joint (6–8). Recent longitudinal studies have confirmed that thickness changes tend to be localized, with certain subregions showing greater average changes than others (8–11).
Although studies that focus on the average progression rates in certain anatomic subregions can provide an understanding of the spatial pattern of disease progression in a given group of patients (e.g., varus and valgus versus normal knees) (10), they may not necessarily provide a more sensitive measure of disease progression than the analysis of total cartilage plates because of the spatial heterogeneity in progression (defined as the change in cartilage thickness) between subjects (8–11). When changes within specific subregions are averaged across several subjects, it may appear that the sensitivity to change is still unsatisfactory because only a few participants showed progression, whereas others progressed more strongly in other subregions. The spatial heterogeneity of progression may therefore result in a relatively low mean change, a relatively large SD of change, and hence in an unfavorable standardized response mean (mean/SD of change).
In the current work, we propose a strategy for a more efficient measure of change in cartilage thickness in a joint based on the measurement of subregional changes. The strategy is designed to remove the link between the spatial location and the magnitude/direction of the change in cartilage thickness. It focuses on the magnitude/direction of the change alone, independent of where in the joint (specifically in which subregion) the maximal or minimal change occurred in each participant. The strategy is not intended to provide unbiased estimates of the magnitude of change that could be used to determine whether and where changes have occurred, but to provide the opportunity to more efficiently compare the magnitude of change between 2 populations than conventional approaches (e.g., healthy subjects versus OA patients or treatment versus placebo groups in a disease-modifying OA drug study). To test this hypothesis, the new strategy was applied to a longitudinal multicenter trial involving both healthy participants and patients with symptomatic, radiographic knee OA (9, 12–14). We hypothesized that an algorithm that relies on so-called ordered values of subregional cartilage change in the femorotibial joint can more effectively discriminate rates of change in cartilage thickness between healthy and OA participants than conventional strategies of averaged (subregional) analyses that are based on spatial location.
SUBJECTS AND METHODS
The study included 180 women, of which 152 (mean ± SD age 56.7 ± 8.6 years) completed the baseline and the month 24 visits. Inclusion criteria for OA participants were frequent symptoms, mild to moderate radiographic OA (Kellgren/Lawrence [K/L] grade 2 or 3) in the medial femorotibial compartment in conventional weight bearing, extended anteroposterior radiographs, joint space width in the medial femorotibial compartment less than or equal to the lateral femorotibial compartment, a body mass index (BMI) of ≥30 kg/m2, and a medial tibiofemoral joint space width (JSW) of ≥2 mm in modified Lyon schuss radiographs (12, 15). Healthy control participants had to have infrequent knee pain, no sign of radiographic knee OA (K/L grade 0), and a BMI of ≤28 kg/m2. The study was conducted in compliance with the ethical principles derived from the Declaration of Helsinki and in compliance with the local Institutional Review Board, informed consent regulations, and the International Conference on Harmonization Good Clinical Practices Guidelines.
The subjects were imaged at 7 clinical centers. The K/L grade was first scored at the imaging sites and was then reassessed by a single experienced reader after enrollment was completed. If the first 2 readers disagreed, the K/L grade was adjudicated by a third reader. There were 90 K/L grade 0 participants (mean ± SD BMI 24.7 ± 4.0 kg/m2), 30 K/L grade 2 participants (mean ± SD BMI 35.5 ± 4.7 kg/m2), and 28 K/L grade 3 participants (mean ± SD BMI 38.0 ± 5.8 kg/m2); 4 K/L grade 1 participants were excluded from further analysis because the group was too small to provide a useful statistical analysis.
MRI was performed using 3T scanners: 3 of the 7 sites used Siemens Magnetom Trio magnets (Siemens, Erlangen, Germany) and 4 sites used Signa Excite magnets (GE Healthcare Technologies, Waukesha, WI) (14). The precision and stability of the measurements in this multicenter study have been described previously (14). Previously validated (16, 17) double oblique coronal MRI acquisitions were obtained at baseline and at 24 months with water excitation spoiled gradient-echo sequences at a 0.31 mm × 0.31 mm × 1.0 mm resolution (13, 14). Segmentation of the femorotibial cartilages (medial tibia, lateral tibia, weight-bearing medial femoral condyle, and weight-bearing lateral femoral condyle) (Figures 1 and 2) was performed by 7 technicians with formal training and more than 3 years of experience each in cartilage segmentation, using custom software (Chondrometrics, Ainring, Germany) (14). The readers segmented the bone interface and the articular surface of the tibial and femoral cartilages manually based on visual assessment. The images were read in pairs (24 months versus baseline), with the readers blinded to the order of acquisition. Quality control of all segmentations was performed by a single reader (FE).
As a cartilage morphology metric, we used mean cartilage thickness over the total area of subchondral bone, as defined in a previous proposal for a nomenclature for MRI-based measures of articular cartilage in OA by an international group of experts (14, 18). The computation included denuded areas of subchondral bone with 0-mm cartilage thickness, but did not include osteophytes and cartilage covering the osteophytes (14, 18). Cartilage thickness over the total area of the subchondral bone was computed for total cartilage plates (medial tibia, weight-bearing medial femoral condyle, lateral tibia, and weight-bearing lateral femoral condyle) and for 5 tibial subregions (central, external, internal, anterior, and posterior) and 3 femoral subregions (central, external, internal) (Figures 1 and 2), as previously described in detail (7). The borders of the subregions within each cartilage plate were determined by a software-based, fully automatic algorithm, with the central tibial subregions (medial tibia, lateral tibia) occupying 20% of the total tibial and the central femoral subregions (weight-bearing medial femoral condyle and weight-bearing lateral femoral condyle) occupying 33% of the total weight-bearing femoral subchondral bone area (7) (Figure 1).
The size of the subregions has been shown to be highly consistent under conditions of repositioning (SD 0.0–0.3%), with the precision errors for regional cartilage thickness measurements ranging from 19 μm (1.5%) to 84 μm (4.7%). Precision errors (test–retest repeatability under repositioning) for total tibial and femoral cartilage plates have been shown to range from 1.8% to 2.3% (17).
The new strategy for estimating change in cartilage thickness considered the following set of metrics: the rates of change (mm/year) found in the 8 medial subregions (5 tibial, 3 femoral) were sorted in each subject and these ordered values were viewed as new variables of change in the medial and lateral femorotibial compartments, respectively. Therefore, instead of the 8 values for rate of change being associated with specific anatomic locations, the values were associated with new variables, labeled here as C(1) or minimum, C(2), C(3), C(4), C(5), C(6), C(7), and C(8) or maximum, where the subscripted numbers represent the rank of the observed rate of change relative to other rates of change in the same subject and compartment. The value of subregional change with the most negative rate of change in each subject (and compartment), minimum or C(1), represents maximal thinning, whereas the most positive rate of change, maximum or C(8), represents maximal thickening. For example, suppose a subject had the following rates of change in their 8 medial subregions: central weight-bearing medial femoral condyle = −0.22 mm/year, external weight-bearing medial femoral condyle = −0.43 mm/year, internal weight-bearing medial femoral condyle = 0.11 mm/year, central medial tibia = −0.32 mm/year, external medial tibia = −0.17 mm/year, internal medial tibia = 0.25 mm/year, anterior medial tibia = 0.05 mm/year, and posterior medial tibia = −0.13 mm/year; the ordered values would be: minimum or C(1) = −0.43 mm/year, C(2) = −0.32 mm/year, C(3) = −0.22 mm/year, C(4) = −0.17 mm/year, C(5) = −0.13 mm/year, C(6) = 0.05 mm/year, C(7) = 0.11 mm/year, and maximum or C(8) = 0.25 mm/year. This exercise was performed independently in each subject in the study and was repeated for lateral compartment subregions.
Summary statistics were computed for each K/L grade group based on organizing the data within each subject by spatial location, i.e., the average in each cartilage plate and subregion (conventional approach), and based on organizing the data by the ordered values (the new approach, i.e., averaging the rates of change over the subregions with maximum thinning, maximal thickening, and magnitudes of change in between, i.e., C(1) through C(8), independent of the specific anatomic location in the joint). Nonparametric Wilcoxon tests were carried out to compare K/L grade 2 versus K/L grade 0 and K/L grade 3 versus K/L grade 0 participants, using both the conventional approach and the new ordered values approach in order to directly compare the sensitivity to change of the 2 approaches. Wilcoxon tests were applied because the ordered values were not expected to be normally distributed given the smaller sample sizes in this study.
The results of the comparative tests for change over time in the medial femorotibial compartment for K/L grade 2 versus K/L grade 0 and for K/L grade 3 versus K/L grade 0 participants are shown in Tables 1 and 2, and for the lateral compartment are shown in Tables 3 and 4, respectively.
Table 1. Estimates of change in cartilage thickness between baseline and 2-year followup (in mm/year) in K/L grade 2 versus K/L grade 0 and in K/L grade 3 versus K/L grade 0 participants in the medial femorotibial compartment using the conventional subregion approach (anatomic subregions)*
Table 2. Estimates of change in cartilage thickness between baseline and 2-year followup (in mm/year) in Kellgren/Lawrence (K/L) grade 2 versus K/L grade 0 and in K/L grade 3 versus K/L grade 0 participants in the medial femorotibial compartment using the ordered values approach
Table 3. Estimates of change in cartilage thickness between baseline and 2-year followup (in mm/year) in K/L grade 2 versus K/L grade 0 and in K/L grade 3 versus K/L grade 0 participants in the lateral femorotibial compartment using the conventional subregion approach (anatomic subregions)*
Table 4. Estimates of change in cartilage thickness between baseline and 2-year followup (in mm/year) in Kellgren/Lawrence (K/L) grade 2 versus K/L grade 0 and in K/L grade 3 versus K/L grade 0 participants in the lateral femorotibial compartment using the ordered values approach
K/L grade 0
K/L grade 2
K/L grade 2 vs. K/L grade 0
K/L grade 3
K/L grade 3 vs. K/L grade 0
The mean difference in the rate of change between K/L grade 2 and K/L grade 0 participants for the medial subregions ranged from −0.004 mm/year to 0.014 mm/year (Table 1), whereas the range for the ordered values was nearly the same (−0.005 mm/year to 0.014 mm/year) (Table 2). Using the conventional approach, only in the external subregion of the weight-bearing medial femur did the rate of change significantly differ between K/L grade 2 and K/L grade 0 participants (P = 0.016); the K/L grade 2 participants displayed a significantly greater increase in cartilage thickness over time than the K/L grade 0 participants (Table 1). No significant difference between K/L grade 2 versus K/L grade 0 was observed for the total cartilage plates (medial tibia or medial femur). Using the ordered values approach, 2 ranked subregions showed significantly greater cartilage thickening in K/L grade 2 than in K/L grade 0 participants (P = 0.0071 and 0.012, respectively) (Table 2). Both ordered variables had lower P values than the external subregion of the central weight-bearing medial femoral condyle with the conventional approach, suggesting that the novel method was more sensitive in detecting the greater increase in cartilage thickness in K/L grade 2 versus K/L grade 0 participants in the medial femorotibial compartment.
Using the conventional approach, the range of differences in the rate of change between K/L grade 3 versus K/L grade 0 was from −0.042 mm/year to 0.001 mm/year (Table 1), whereas when using the ordered values, the differences ranged from −0.061 mm/year to 0.008 mm/year (Table 2). Therefore, the largest difference in the mean change of the ordered values approach (−0.061 mm/year) showed a much larger decrease in cartilage thickness than the largest decline observed using the conventional subregion approach (−0.042 mm/year). Using the conventional subregion approach, cartilage thinning was significantly (but only marginally) greater in K/L grade 3 than in K/L grade 0 participants in 1 subregion, specifically the external medial tibia (P = 0.037). In contrast, using the novel approach, 4 of the 8 ordered values (ranks) had significant P values (for differences in change between K/L grade 3 and K/L grade 0). The minimal P value was 0.001 and 3 of the ordered values displayed P values less than the smallest P values obtained with the conventional subregional approach (external medial tibia). Therefore, 4 of the 8 medial subregions displayed a significantly greater decrease in cartilage thickness in K/L grade 3 versus K/L grade 0 participants, but the specific regions involved varied between the OA participants.
In the lateral compartment (Tables 3 and 4), K/L grade 3 participants displayed a larger increase in cartilage thickness in the external weight-bearing lateral femur than the K/L grade 0 participants using the conventional subregion approach (Table 3). None of the ordered values variables, however, were significantly different between K/L grade 2 versus K/L grade 0 or between K/L grade 3 versus K/L grade 0 participants (Table 4).
We propose here a novel strategy (the ordered values approach) for providing a more statistically efficient measure of change in cartilage thickness, based on the measurement of various anatomic subregions. This approach removes the link between the spatial location and the magnitude/direction of change in individual study participants. Although this strategy will not be valid for testing whether the change observed was significantly different from zero, it provides the potential to more efficiently compare the magnitude of change between 2 or more populations (i.e., OA participants versus healthy controls or participants treated with a disease-modifying OA drug versus a placebo).
The new strategy was applied to participants in a longitudinal multicenter trial with different grades of medial radiographic knee OA (9, 12–14). The results suggest that the ordered values of subregional cartilage change provide a more sensitive measure of discerning structural progression in OA participants versus healthy controls than conventional approaches, where changes in the same cartilage plates or subregions are averaged over all participants. The higher sensitivity of the ordered values approach was most evident in the medial femorotibial compartment of the K/L grade 3 participants. This was not surprising, given that the OA participants were selected to have predominantly medial disease, and given that no significant cartilage loss was observed laterally.
Interestingly, the K/L grade 2 participants did not show significant cartilage thinning in either the medial or the lateral femorotibial compartment, but showed cartilage thickening in the medial compartment. Also, the cartilage thickening in the K/L grade 2 participants became more clearly apparent using the ordered values approach versus the conventional approach. Given that the ranges of differences between the K/L grade 2 and K/L grade 0 participants were similar for the conventional and ordered values approaches, the increased sensitivity must have been due to reduced variability between subjects within each K/L group (rather than an increased difference between groups) using the ordered values approach.
The increase in sensitivity by the ordered values approach can be evaluated in several ways. First, the number of ordered values variables that showed a significant change was more than the number of subregions in the medial compartment that showed significant change. Second, the P values were generally smaller in the ordered values variables compared with the anatomic subregions. Comparing the most sensitive test results, i.e., the smallest P value, from both approaches indicates that a study using the ordered values approach would require only 33% of the sample size used in the subregion approach to attain the same significance in tests comparing K/L grade 3 and K/L grade 0 participants in the medial compartment, and only 75% of the sample size of the subregion approach in tests comparing the K/L grade 2 and K/L grade 0 participants. This increased sensitivity did not carry over to the lateral compartment, where no tests were significant using the ordered values approach and only 1 test was significant for the anatomic subregion approach. Given that change was not expected in the lateral compartment because the participants had medial disease, this suggests that the increased sensitivity in the medial compartment seen in the ordered values approach was not due to a test bias.
A limitation of the study is that the results are specific to the choice of size and location of the subregions used here, which was to some extent arbitrary, and therefore improvements could vary with different definitions of subregions. A 20% central subregion was used in the tibia because 5 anatomic subregions were used, and a 33% central subregion was used in the femur because 3 anatomic subregions were used. However, recent studies found these subregions useful to compare the spatial pattern of cartilage change in participants with varus and valgus malalignment versus those with normal knee alignment (10), and another study showed that the rates of change and the standardized response mean did not differ relevantly when differently sized central subregions (range 10–66% of the subchondral bone area) were used (11). No anterior or posterior subregion was defined in the weight-bearing femur (central weight-bearing medial femoral condyle and central weight-bearing lateral femoral condyle) because the region already provides a subregion of the total femur, with the trochlea being located anteriorly and the posterior aspects of the femoral condyles being located posteriorly (7). Also, as mentioned before, the proposed technique has the limitation that it cannot be used to evaluate whether the changes observed are significantly different from zero, but solely provides a tool to more efficiently compare the magnitude of change between populations.
It is important to note that the ordered values approach of the subregional analysis provided a much more sensitive measure of the difference in longitudinal change between OA participants and healthy subjects than the analysis of total cartilage plates (medial tibia and central weight-bearing medial femoral condyle). The reasons for this are that the decoupling of the direction and magnitude of the subregional change and the actual location where the subregional change occurs removes the noise introduced by the spatial heterogeneity of the cartilage loss between subjects and thus allows one to exploit the full benefit and potential of subregional cartilage analysis.
In addition to providing a more sensitive measure of change, there are several other important advantages of the novel approach. Organizing subjects by the magnitude of change provides a natural mechanism for organizing subjects by potential differences in disease stage. There is evidence that subjects may encounter significant increases in cartilage thickness at some stage of the disease, while they are likely to see significant decreases at another stage of the disease. This evidence comes from animal studies, where cartilage thickening (e.g., hypertrophy/swelling) was observed at the early phase of OA (19–23) and where cartilage hypertrophy has been shown to precede cartilage breakdown. In humans, evidence from cross-sectional comparisons in JSW in radiographs showed a greater JSW in K/L grade 2 participants than in healthy knees, and a smaller JSW in K/L grade 3 participants than in healthy knees at baseline in the same cohort as described here (12). Also, an increase in cartilage thickness over time has recently been observed using quantitative MRI in patients after acute injury of the anterior cruciate ligament (24). The results from our current study provide further evidence that participants with radiographic OA do not only display cartilage thinning, but also cartilage thickening. Although some subregions display changes more frequently than other regions, the specific subregions where most of the change is observed varied substantially between participants. This observation led to the development of a new metric for measuring change by examining the ordered values of change within each individual. The current methodology may thus provide a more sensitive tool to identify subjects with regional cartilage thickening and thinning and to better classify subjects in different disease stages than currently possible using radiography.
At this time, it is very challenging to identify a priori at what stage in the disease a subject may be when enrolled in a study, and where in the joint cartilage thickening and/or thinning may occur in the future. Use of spatial location (mean in subregions) to assess change over time lumps together subjects at different disease stages and may cancel out important differences in these individuals because cartilage thickening and thinning may occur at the same time. In contrast, the use of the ordered values approach naturally separates subjects as to whether they show significant cartilage thinning or thickening by assigning results to the smallest and largest ordered values, respectively. As evidenced in this study, this can increase the observed signal (sensitivity to change) and reduce the variability within the groups that are compared. Moreover, the ordered values approach can provide an approximate measure of the size of the area that displays significant change. As the number of significant ordered subregional values increases, this indicates that the area with significant changes is also increasing. In this context, the ordered values approach allows the specific locations being considered for change to vary between subjects, potentially providing a better opportunity to see the full spatial extent of the change, regardless of location, in a group of OA participants.
In the current study, medial and lateral subregions were treated separately because the study population was restricted to subjects with medial femorotibial radiographic OA. When examining cohorts with either medial or lateral disease, however, the methodology proposed here may be used to provide ordered values (ranks) based on medial and lateral subregions together in order to eliminate the variability in change that originates from participants that progress only medially or laterally. This may provide an opportunity to study the effect of a disease-modifying OA drug versus placebo more effectively in participants with either medial or lateral femorotibial OA than the conventional approach that is based on anatomic regions or subregions.
In conclusion, we propose a new metric of ordered values (ranks) of individual subregional cartilage change to provide a more sensitive measure than conventional approaches that are restricted to mean differences in total cartilage plates or within given subregions. We show that the novel approach is more sensitive in detecting cartilage thinning over time in K/L grade 3 participants versus healthy participants, and cartilage thickening in K/L grade 2 participants versus healthy controls. This novel approach has the benefit of naturally separating subjects that may be at different stages of the disease (encountering cartilage thinning or thickening) and of providing an estimate of the size of the region with significant change in a group of patients. The new method may also be potentially superior in the context of other comparisons, e.g., a group treated with a disease-modifying OA drug versus a group treated with a placebo.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Buck had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Buck, Wyman, Hellio Le Graverand, Eckstein.
Acquisition of data. Buck, Wyman, Hellio Le Graverand, Hudelmaier, Wirth.
Analysis and interpretation of data. Buck, Wyman, Hellio Le Graverand, Hudelmaier, Wirth, Eckstein.
We are grateful to the dedicated group of study coordinators whose skills were essential in assuring the successful conduct of this study: Sandra Chapman, Emily Brown, Kristen Fredley, Donna Gilmore, Joyce Goggins, Norine Hall, Thelma Munoz, and Kim Tally. We would like to thank all of the site investigators: Deb Burstein, Julia Crim, David Hunter, Gary Hutchins, Chris Jackson, Virginia Byers Kraus, Nancy Lane, Thomas M. Link, Sharmila Majumdar, Steve Mazzuca, Prasad Pottumarthi, Thomas Schnitzer, Mihra Taljanovic, Berchman Vaz, and associates. We would also like to thank Cecil Charles, the Duke Image Analysis Laboratory staff, the dedicated MRI technologists, and the Pfizer A9001140 Team, Peggy Coyle and Charles Packard, for their invaluable efforts in conducting this study. Also, we would like to thank Gudrun Goldmann, Linda Jakobi, Manuela Kunz, Sabine Mühlsimer, Annette Thebis, and Dr. Barbara Wehr (Chondrometrics GmbH) for data segmentation, Steve Mazzuca and Kenneth Brandt for the central reading and adjudication of the K/L grade scores, and John Kotyk and Jeff Evelhoch for their advice on the study design and setup.