Abstract
Objective
To use software-based magnetic resonance imaging (MRI) measures of multiple features of knee osteoarthritis (KOA) to predict radiographic and pain progression in persons with KOA, and compare to a study that used primarily semi-quantitative (SQ) scoring.
Design
Data from the Foundation for the National Institutes of Health Osteoarthritis Biomarkers Consortium (FNIH) nested case-control study (600 subjects divided into case and control groups based on knee pain and/or radiographic progression) were used. The MRI Osteoarthritis Software Scoring (MOSS) was used to quantitatively assess medial femoral cartilage, bone marrow lesions, osteophyte volume, effusion-synovitis volume, and a measure of Hoffa's synovitis at baseline and 24-months using readers with diverse levels of expertise. Association between baseline and baseline to 24-month change with progressor status was examined and discriminative ability assessed using the c-statistic (AUC) computed under 10-fold cross validation.
Results
AUC values ranged from 0.690 to 0.726 to predict combined pain/radiographic progression and from 0.709 to 0.804 to predict radiographic progression alone. Bone marrow lesions and osteophyte volume played a role in all analyses. Medial femoral cartilage was significant for all but the cross-sectional analysis involving pain progression. Comparison to results from a separate publication showed that MOSS offered similar discrimination to a published model that primarily used SQ scoring.
Conclusions
We found a high level of discrimination particularly for radiographic progression analysis. Use of fast automated software and readers with varied prior experience make MOSS a useful tool for enriching future clinical trials and for other large studies of KOA.
Keywords: Osteoarthritis, Magnetic resonance imaging, Quantitative imaging, Knee, Software, Synovitis
1. Introduction
Osteoarthritis (OA) is the most common joint disorder in the United States, affecting more than 32.5 million adults [1]. Currently, 23 % of adults over 40 years of age have knee osteoarthritis (KOA) worldwide [2]. Arthritis is also the leading cause of disability in the United States, with the overall economic burden associated with OA in the United States estimated at $136.8 billion in annual costs due to work disability, medical care, and lost earnings [1].
To date, there is no US Food and Drug Administration approved treatment to prevent progression or cure osteoarthritis (OA) [3]. However, multiple disease modifying osteoarthritis drugs are in development aiming to modify OA pathophysiology with some of these drugs having been found to modify structural progression [4,5] Longitudinal clinical trials using both radiography and magnetic resonance imaging (MRI) to evaluate these drugs have led to the increased need for accurate, sensitive, efficient, and reliable MRI evaluation of the structural components and anatomic biomarkers involved in OA.
MRI is currently the most sensitive imaging modality for the evaluation of osteoarthritis structural markers. KOA MR imaging biomarkers, including cartilage, bone marrow lesions (BMLs), meniscal damage, osteophytes, and effusion/synovitis (ES) are well-visualized on MRI. Synovial joint tissues are known to be involved in the development of OA, with destruction of articular cartilage and bone involvement leading to decreased joint function [6]. BMLs or subchondral inflammation have shown the strongest association with knee pain in comparison to other structures, with a tendency to wax and wane [7]. There is currently great interest in therapies targeted at the above structures, in particular towards decreasing inflammation implicated in synovial thickening, edema or effusion (the latter 2 of which demonstrate high signal intensity on fluid sensitive MRI sequences).
Current methods of MRI measurement include the following: manual evaluation (mouse tracing), semi-quantitative (SQ) scoring, quantitative, semiautomated and more recently, fully automated, with deep learning (DL) or artificial intelligence (AI) gaining traction [[8], [9], [10]]. AI or DL offer the potential for substantial improvements in efficiency as we have demonstrated for BML volume measurement [11]. To date, SQ MR scoring methods have been used in longitudinal trials to assess OA progression; these scores include the Whole Organ Magnetic Resonance Imaging Score (WORMS), Boston Leeds Osteoarthritis Knee Score (BLOKS) and MRI Osteoarthritis Knee Score (MOAKS) [[12], [13], [14]]. Several studies have proven the utility of these methods in the Osteoarthritis Initiative (OAI) and the Foundation for the National Institutes of Health OA Biomarkers Consortium (FNIH) cohort [[15], [16], [17], [18]].
While MOAKS has been considered the gold standard, it involves subjective evaluation of knee joint parameters and requires scoring by expert radiologists. This may be expensive and time consuming. This has led to focused development of more objective evaluation techniques that can be performed by non-expert readers and thus may be more feasible to implement in a DMOAD clinical trial setting or large observational studies such as MOST or OAI where large numbers of images may go unscored due to resource constraints. A recent publication by Hunter et al. stated that “there is need for further refinement and improvement of measures of joint structural change based on imaging in order to identify individuals likely to have progression of disease” [19].
MOSS consists of several semi-automated software tools used to quantify various structural measurements related to KOA. Each component combines low-level C code with a graphical user interface that allows a reader to check the results and make any necessary edits. The goal of our study was to analyze multiple (5) quantitative MRI-based structural measures from MOSS in a case-control model to understand whether these features accurately predict radiographic and pain progression in KOA, and to make a comparison to results from a previously published study that used SQ scoring to a large extent and used the same data set [19].
2. Method
2.1. Study design and participants
Our study used data and images from the FNIH study (https://nds.nih.gov.oai), a nested case-control study within the OAI, a longitudinal cohort study of 4796 participants with or at risk for Knee Osteoarthritis (KOA). Approval was obtained from the institutional review boards at each OAI site. The OAI enrolled men and women ages 45–79 with or at high risk of developing knee OA at four centers in the United States between 2004 and 2006. The nested FNIH case-control study included 600 patients, who were divided into 4 groups based on pain and/or radiographic progression. Knees were frequency matched on KL grade and body mass index. Radiographic progression was defined by medial tibiofemoral radiographic joint space loss ≥0.7 mm from BL to 24, 36, or 48 months [20]. Pain progression was defined as a persistent increase on the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain scale between 24 and 60 months. Persistence required an increase in pain at two or more time points. A pre-determined number of knees were selected into the following groups [19]:
Group 1: radiographic and pain progressors (n = 194),
Group 2: radiographic-only progressors (n = 103),
Group 3: pain-only progressors (n = 103),
Group 4: no radiographic or pain progressors (n = 200).
These four case-control groups were selected to enable separate evaluation of biomarkers for structural and symptom progression. Further details of the FNIH cohort can be found in separate publications [19,21,22]. To facilitate comparisons to prior work, we used the following case definitions in our analyses:
Primary analysis: For the primary analysis, cases were defined as having both radiographic and pain progression (Group 1) and controls consisted of subjects from Groups 2, 3, and 4.
Secondary analysis: In secondary analyses focusing on radiographic progression, we also investigated comparisons of all subjects with radiographic progression (Group 1 and 2) to subjects without radiographic progression (Groups 3 and 4).
2.2. Knee MRI acquisition
Non-contrast MRI acquisition was performed at 4 OAI clinical centers using a 3 T MRI system (Trio, Siemens Healthcare, Erlangen, Germany). Additional parameters of the full OAI pulse sequence protocol and sequence parameters have been published in detail elsewhere [23]. Sagittal, coronal and axial images were analyzed as described below. Details of these MR sequences are provided in Supplemental Table S1.
2.3. Image analysis methods
MOSS consists of several semi-automated software tools used to quantify various structural measurements related to KOA. Each component combines low-level C code with a graphical user interface that allows a reader to check the results and make any necessary edits. We used five different components of MOSS to assess medial femoral cartilage, BMLs, osteophytes, effusion synovitis (ES), and Hoffa's synovitis (HS); each is briefly described below. Each measurement method was developed and independently validated in a published study. Supplemental Table S2 summarizes the automated and semi-automated steps required to make each measurement. The total reader time for all five methods was under 40 min per knee. The reader time and intra/inter-reader precision for each measure is provided in Supplemental Table S3.
All measurements were made with the readers blinded to case/control status and time point; for cartilage and osteophyte volume measurements, the readers were able to view images from both time points simultaneously to ensure consistency. The readers possessed a variety of skills and experience. Those without prior experience were trained by an MSK fellowship trained radiologist on a subset of cases with agreement by consensus before the official study began. For BMLs and osteophytes, the readers (MS and MY) were recent undergraduates with no formal medical training. A radiologist (SS or LS), performed the reader training and a quality assurance check after the measurements were made. For cartilage, the reader (LS) was a radiology resident/research fellow at the time of the study. For ES, the readers (LL and SB) were both board certified MSK radiology clinical fellows. Finally for HS, the reader (SS) was a board-certified MSK fellowship trained attending radiologist and member of the medical school faculty with 20 years’ experience. Generally MOSS can be performed by readers without extensive formal training as long as they undergo training by a more skilled individual [24].
2.4. Local area cartilage segmentation (LACS) method
The LACS method uses a cylindrical coordinate system linked to anatomical landmarks to objectively and reproducibly define a region in the central weight-bearing portion of the medial femur (MF) for cartilage volume measurement. LACS has an inherent flexibility whereby the regions can be placed at any location, but for our study we chose a region that roughly matched the central weight-bearing location of a commonly used method [20]. Furthermore, since the coordinate system is linked to anatomical landmarks, the regions are normalized to knee size. LACS was applied to sagittal double echo steady state (DESS) 3D MR images. This has been described and validated in detail [25].
2.5. BML volume measurement
To measure BML volume (regions of subchondral inflammation demonstrating bright signal intensity on fluid sensitive MRI), we used a threshold-based software algorithm to highlight regions of hyperintense signal intensity on sagittal turbo spin echo fat-suppressed (TSE FS) intermediate-weighted MR images. The software produced BML volume measurements of the medial and lateral femur (MF and LF), medial and lateral tibia (MT and LT), and medial and lateral patella (MP and LP). This technique was validated and published previously [26].
2.6. ES volume measurement
As with the BML software, ES volume was measured using a thresholding technique [27] on axial reformatted DESS MR images and was assessed for all slices where the patella was present. The reader followed the guidelines set by the MOAKS scoring system [14], assessing ES volume only in the anterior regions of the knee within the confines of the suprapatellar bursa extending from the superior suprapatellar bursa to the level of the inferior patella using the validated technique described by Smith et al. [27].
2.7. Osteophyte volume measurement
Marginal osteophyte volume in the MF, LF, MT, and LT regions was measured using an edge-detection software algorithm on coronal reformatted DESS MR image in a central portion of the knee [28]. The software determined the outer margins between each osteophyte and the neighboring soft tissue, while the reader indicated the boundary between the normal bone and the osteophyte.
2.8. HS assessment
HS was measured in 5 central slices on the sagittal TSE FS MRI image using software that analyzed manually drawn regions of interest (ROIs). This technique produced a dimensionless metric that reflected the level of synovitis in the ROIs and is normalized to the fat pad ROI size [27].
2.9. Statistical methods
We examined the unadjusted association between biomarkers and outcome using descriptive statistics and unadjusted logistic regression. Using Spearman correlation, we computed the correlation between biomarkers, separately for baseline biomarkers and 24 month change in biomarkers.
To investigate potential non-linear associations between each biomarker and case status we first modeled the association between continuous biomarker and outcome using restricted cubic splines, a flexible modeling strategy which uses cubic functions to model the association between predictor and outcome between knots [29]. For cartilage, osteophytes, ES, and HS, we used five knots (5th, 27.5th 50th, 72.5th and 95th percentiles). Median BML size at baseline was zero for all compartments, so for BML we used knots at the 80th, 90th, and 95th percentiles. To incorporate these non-linear associations in the multivariable regression models we then created categorical variables for each predictor, using two (any vs. none) or three levels for baseline BMLs, 3 levels (improve vs. no change vs. worsen) for change in BMLs, and quartiles for all other predictors. Categorical variables were included as nominal variables to allow for modeling of non-linear associations. This approach aided in interpretation of parameter estimates, facilitating comparison with prior FNIH analyses.
We used logistic regression to examine the association between each biomarker and case status for the primary and secondary analyses, calculating odds ratios with associated 95 % confidence intervals. Models were adjusted for sex, race, age, body mass index (BMI), Kellgren and Lawrence grade (KLG), WOMAC pain, pain medication use, and baseline JSW. Models were fit separately for baseline and 24-month change in biomarkers.
To facilitate comparison with prior work, modeling building was undertaken in a similar manner as the primary FNIH phase I analyses [19]. Three different stepwise selection methods were used to determine the best subset of predictors: 1) Akaike Information Criterion (AIC); 2) Schwarz Bayesian Criterion (SBC); and 3) p-value (p = 0.2 for entry/0.1 for retention). We additionally included (least absolute shrinkage and selection operator) LASSO selection due to suspected correlation between biomarkers. The discriminative ability was assessed using the c-statistic (AUC). AUC was computed under 10-fold cross-validation and averaged over 20 iterations.
Data analyses were conducted using Statistical Analysis Software (SAS), version 9.4.
3. Results
Baseline demographic and clinical characteristics are provided in Table 1. The average age was 62 years, 59 % female/41 % male, with KLG percentages as follows: 13 % KL 1, 51 % KL2, 37 % KL3. Fig. 1 shows the results of restricted cubic spline analyses to investigate the relationship between each biomarker and the magnitude of the outcome for baseline cartilage volume, MF osteophytes, MF BMLs, and effusion-synovitis. Most figures demonstrate non-linear relationships. For example, the estimated odds of becoming a composite case decreases in a fairly linear fashion for baseline cartilage volumes from approximately 250 mm3–600 mm3, and then flattens for cartilage volumes > ∼600 mm3 (Fig. 1A). Results for the remaining biomarkers are shown in Fig. 2 (baseline) and 3 (24 month change). These results support the strategy to categorize biomarkers.
Table 1.
Cohort Characteristics. Presented in cells: mean (SD) or n (%).
| Composite Case |
|||
|---|---|---|---|
| Variable | Overall | Yes (n = 194) | No (n = 406) |
| Baseline age | 61.5 (8.9) | 62.0 (8.8) | 61.3 (8.9) |
| Sex | |||
| Male | 247 (41 %) | 84 (43 %) | 163 (40 %) |
| Female | 353 (59 %) | 110 (57 %) | 243 (60 %) |
| Baseline BMI | 30.7 (4.8) | 30.7 (4.8) | 30.7 (4.8) |
| Race: White | |||
| No | 125 (21 %) | 39 (20 %) | 86 (21 %) |
| Yes | 475 (79 %) | 155 (80 %) | 320 (79 %) |
| Baseline pain medication use | |||
| No | 423 (71 %) | 131 (68 %) | 292 (72 %) |
| Yes | 177 (30 %) | 63 (32 %) | 114 (28 %) |
| Baseline WOMAC pain score (0–100, 100 worst) | 12.1 (15.6) | 10.2 (13.0) | 13.0 (16.7) |
| Baseline JSW | 3.8 (1.2) | 3.8 (1.4) | 3.9 (1.1) |
| Baseline KLG | |||
| 1 | 75 (13 %) | 24 (12 %) | 51 (13 %) |
| 2 | 306 (51 %) | 84 (43 %) | 222 (55 %) |
| 3 | 219 (37 %) | 86 (44 %) | 133 (33 %) |
Fig. 1.
Association between MF baseline biomarkers and composite case (JSN + pain) status using the restricted cubic spline method.
Legend: The biomarker is along the x-axis and the predicted probability of being a composite case is along the y-axis. The predicted probability of being a case for each value of biomarker is shown by each dot, while the band represents the 95 % confidence interval. A) Cartilage Volume Baseline (BL); B) Osteophytes Medial Femur (MF); C) BMLs Medial Femur (MF); D) Effusion-Synovitis. Baseline (BL), Medial Femur (MF).
Fig. 2.
Association between baseline biomarkers and composite case (JSN + pain) status using the restricted cubic spline method.
Legend: The biomarker is along the x-axis and the predicted probability of being a composite case is along the y-axis. The predicted probability of being a case for each value of biomarker is shown by each dot, while the band represents the 95 % confidence interval. Osteophytes are shown in panels A through C; BML volume in panels D through H. A) Osteophytes Lateral Femur (LF); B) Osteophytes Lateral Tibia (LT); C) Osteophytes Medial Tibia (MT); D) BMLs Medial Patella (MP); E) BMLs Lateral Patella (LP); F) BMLs Medial Tibia (MT); G) BMLs Lateral Tibia (LT); H) BMLs Lateral Femur (LF); I) Effusion-synovitis.
3.1. Primary analysis
Models predicting composite pain and radiographic progression from baseline biomarkers are shown in Table 2. The table shows each biomarker selected by the various selection strategies (AIC, SBC, p-value, LASSO) along with the odds of progression for each selected biomarker estimated from the multivariable logistic regression model.
Table 2.
Multivariable models of baseline biomarkers to predict radiographic and pain progression in the primary analysis.
| MODEL |
M1 |
M2 |
M3 |
M4 |
|---|---|---|---|---|
| Selection Method | Stepwise, AIC | Stepwise, SBC | Stepwise, P-value | LASSO |
| Model characteristics | ||||
| Biomarkers included | Ost-LT, BMLs-MP, BMLs-MT, BMLs-MF, BMLs-LF | BMLs-MP, BMLs-MT | ||
| AUC (unadjusted) | 0.706 | 0.655 | Same as M1 | Same as M2 |
| AUC (adjusteda) | 0.737 | 0.702 | ||
| AUC (adjusteda, 10 fold cross-validation) | 0.690 | 0.664 | ||
| Adjusted odds ratios (95 % CIs) | ||||
| Osteophytes-LT quartile (range) | ||||
| 0 (0, 2.4) | reference | |||
| 1 (2.5, 33) | 1.4 (0.82, 2.5) | |||
| 2 (34, 104) | 1.6 (0.92, 2.9) | |||
| 3 (105, 919) | 3.1 (1.7, 5.6) | |||
| BMLs-MP category (range) | ||||
| 0 (0) | reference | reference | ||
| 1 (8, 1529) | 1.8 (1.2, 2.8) | 2.1 (1.4, 3.2) | ||
| BMLs-MT category (range) | ||||
| 0 (0) | reference | reference | ||
| 1 (0.4, 5815) | 4.0 (2.4, 6.7) | 4.2 (2.6, 6.9) | ||
| BMLs-MF category (range) | ||||
| 0 (0) | reference | |||
| 1 (1.9, 134) | 1.4 (0.85, 2.3) | |||
| 2 (135, 5931) | 2.5 (1.5, 4.1) | |||
| BMLs-LF category (range) | ||||
| 0 (0) | reference | |||
| 1 (2.3, 7576) | 0.57 (0.36, 0.90) | |||
OST=Osteophyte, BML=Bone marrow lesion, LF=Lateral femur, LT = Lateral tibia, MF = Medial femur, MP = Medial patella, MT = Medial tibia, AIC = Akaike Information Criterion; SBC=Schwarz Bayesian Criterion; LASSO = least absolute shrinkage and selection operator.
Adjusted for sex, race, age, BMI, KLG, WOMAC pain, pain medication use, and JSW.
AIC and p-value based selection resulted in the same set of biomarkers: osteophytes in the LT, BMLs in the MP, MT, MF, and LF (Table 2, M1, M3). The cross-validated AUC was 0.690. SBC and LASSO selection resulted in the same set of biomarkers: BMLs in the MP and MT. The cross-validated AUC was 0.664. In both models, higher BML volume in the MP and MT was associated with increased odds of being a case. In M1, higher osteophyte volume in the LT was associated with higher odds of becoming a composite case, while higher BML volume in the LF was associated with lower odds of becoming a composite case.
Over 24 months, composite cases had, on average, greater loss of cartilage volume compared to controls and greater increases in osteophyte volume. On average, cases had an increase in effusion-synovitis while controls demonstrated a decrease. Controls were more likely to have no change in BML volume (versus increasing or decreasing BML volume).
Models predicting composite pain and radiographic progression from 24 month change in biomarkers are shown in Table 3. AIC (M1) and p-value (M3) based selection resulted in the same model and included cartilage, effusion-synovitis, osteophytes in the LF and MF, and BMLs in the MP, LP, MT, and LT. The cross-validated AUC was 0.725. LASSO selection resulted in a very similar model, included BMLs in the MF, and did not include BMLs in the LP or osteophytes in the LF. The cross-validated AUC was 0.726. SBC based selection resulted in a model with only cartilage and osteophytes in the MF and had a cross-validated AUC of 0.699. In all models, the higher cartilage volume group (i.e., less cartilage loss) was associated with lower odds of being a composite case. In M1, those with the least cartilage loss (quartile 4, increase in cartilage volume ≥8.8 mm3) had 0.3 times the odds of being a composite case compared to those with the most cartilage loss (quartile 1, cartilage loss between 74 and 569.2 mm3). Increasing osteophyte volume in the MF was associated with increased odds of being a case, while increasing cartilage volume in the LF was associated with decreased odds of being a case.
Table 3.
Multivariable models of 24 month change in biomarkers to predict radiographic and pain progression in the primary analysis.
| MODEL |
M1 |
M2 |
M3 |
M4 |
|---|---|---|---|---|
| Selection Method | Stepwise, AIC | Stepwise, SBC | Stepwise, P-value | LASSO |
| Model characteristics | ||||
| Biomarkers included | Cartilage, effusion-synovitis, Ost-LF, Ost-MF, BMLs-MP, BMLs-LP, BMLs-MT, BMLs-LT | Cartilage, Osteophyte-MF | Cartilage, effusion-synovitis, Ost-MF, BMLs-MP, BMLs-LT, BMLs-MT, BMLs-MF | |
| AUC (unadjusted) | 0.766 | 0.711 | Same as M1 | 0.759 |
| AUC (adjusteda) | 0.787 | 0.740 | 0.781 | |
| AUC (adjusteda, 10 fold cross-validation) | 0.725 | 0.699 | 0.726 | |
| Adjusted odds ratios (95 % CIs) | ||||
| Cartilage volume quartile (range) | ||||
| 0 (−569.2, −74.0) | reference | reference | reference | |
| 1 (−74.0, −24.2) | 0.50 (0.28, 0.87) | 0.42 (0.25, 0.69) | 0.52 (0.30, 0.89) | |
| 2 (−24.2, 8.7) | 0.45 (0.25, 0.80) | 0.40 (0.23, 0.67) | 0.51 (0.29, 0.90) | |
| 3 (8.8, 154.3) | 0.28 (0.15, 0.51) | 0.23 (0.13, 0.41) | 0.30 (0.16, 0.54) | |
| Effusion-synovitis quartile (range) | ||||
| 0 (−12906.2, −955.1) | reference | reference | ||
| 1 (−934.2, −146.0) | 1.5 (0.81, 2.6) | 1.6 (0.86, 2.8) | ||
| 2 (−140.3, 488.9) | 1.3 (0.69, 2.3) | 1.4 (0.78, 2.6) | ||
| 3 (504.5, 11016.3) | 2.3 (1.3, 4.0) | 2.3 (1.3, 4.0) | ||
| Osteophytes-LF quartile (range) | ||||
| 0 (−212.07, −2.05) | reference | |||
| 1 (−2.05, 8.21) | 0.71 (0.39, 1.3) | |||
| 2 (8.39, 46.07) | 0.96 (0.54, 1.7) | |||
| 3 (46.25, 425.07) | 0.56 (0.30, 1.0) | |||
| Osteophytes-MF quartile (range) | ||||
| 0 (−148.46, −0.18) | reference | reference | reference | |
| 1 (0.00, 8.02) | 0.46 (0.25, 0.86) | 0.45 (0.25, 0.81) | 0.47 (0.26, 0.87) | |
| 2 (8.21, 40.10) | 1.1 (0.60, 1.9) | 0.99 (0.57, 1.7) | 0.99 (0.56, 1.8) | |
| 3 (40.47, 448.56) | 2.6 (1.4, 4.9) | 2.4 (1.4, 4.1) | 2.1 (1.2, 3.8) | |
| BMLs-MP category (range) | ||||
| 0 (−1200.20, −1.90) | reference | reference | ||
| 1 (0.00, 0.00) | 0.49 (0.28, 0.85) | 0.56 (0.33, 0.93) | ||
| 2 (5.40, 1050.70) | 0.71 (0.36, 1.4) | 0.62 (0.31, 1.2) | ||
| BMLs-MT category (range) | ||||
| 0 (−4199.70, −0.40) | reference | reference | ||
| 1 (0.00, 0.00) | 0.34 (0.18, 0.65) | 0.35 (0.18, 0.67) | ||
| 2 (0.80, 6476.20) | 0.64 (0.33, 1.2) | 0.56 (0.29, 1.1) | ||
| BMLs-LP category (range) | ||||
| 0 (−2129.70, −2.30) | reference | |||
| 1 (0.00, 0.00) | 0.85 (0.49, 1.5) | |||
| 2 (2.30, 2120.50) | 0.45 (0.26, 0.81) | |||
| BMLs-LT category (range) | ||||
| 0 (−1190.60, −0.80) | reference | reference | ||
| 1 (0.00, 0.00) | 0.92 (0.36, 2.3) | 0.91 (0.37, 2.3) | ||
| 2 (8.40, 2728.40) | 2.1 (0.69, 6.5) | 2.1 (0.70, 6.5) | ||
| BMLs-MF category (range) | ||||
| 0 (−4253.60, −1.50) | reference | |||
| 1 (0.00, 0.00) | 0.59 (0.34, 1.0) | |||
| 2 (0.40, 6334.30) | 1.1 (0.68, 1.8) | |||
OST=Osteophyte, BML=Bone marrow lesion, LF=Lateral femur, LT = Lateral tibia, LP=Lateral patella, MF = Medial femur, MP = Medial patella, MT = Medial tibia, AIC = Akaike Information Criterion; SBC=Schwarz Bayesian Criterion; LASSO = least absolute shrinkage and selection operator.
Adjusted for sex, race, age, BMI, KLG, WOMAC pain, pain medication use, and JSW.
3.2. Secondary analysis
Figs. S3, S4, and S5 show the results of restricted cubic spline analyses to investigate the relationship between each biomarker and the estimated probability of being a JSN case. Models predicting JSN case status from baseline biomarkers are shown in Table 4. Cross-validated AUCs ranged from 0.685 to 0.709. BMLs in the medial femur and medial tibia were selected by all methods. Cartilage volume, osteophytes in the medial femur, and effusion synovitis were selected by the AIC and p-value selection methods.
Table 4.
Multivariable models of baseline biomarkers to predict radiographic progression in the secondary analysis.
| MODEL |
M1 |
M2 |
M3 |
M4 |
|---|---|---|---|---|
| Selection Method | Stepwise, AIC | Stepwise, SBC | Stepwise, P-value | LASSO |
| Model characteristics | ||||
| Biomarkers included | Cartilage, effusion-synovitis Ost-MF, BMLs-MF, BMLs-MP, BMLs-MT, BMLs-LF | BMLs-MF, BMLs-MT | Cartilage, effusion-synovitis Ost-MF, BMLs-MF, BMLs-MT, BMLs-LF | Same as M2 |
| AUC (unadjusted) | 0.720 | 0.663 | 0.716 | |
| AUC (adjusteda) | 0.758 | 0.714 | 0.755 | |
| AUC (adjusteda, 10 fold cross-validation) | 0.709 | 0.685 | 0.708 | |
| Adjusted odds ratios (95 % CIs) | ||||
| Cartilage volume quartile (range) | ||||
| 0 (228.9, 523.4) | reference | reference | ||
| 1 (524.0, 644.2) | 0.47 (0.27, 0.82) | 0.48 (0.27, 0.83) | ||
| 2 (644.3, 819.2) | 0.68 (0.37, 1.3) | 0.71 (0.39, 1.3) | ||
| 3(819.6, 1468.3) | 0.47 (0.21, 1.1) | 0.50 (0.22, 1.1) | ||
| Effusion-synovitis quartile (range) | ||||
| 0 (16.0, 699.8) | reference | reference | ||
| 1 (702.2, 1551.4) | 0.96 (0.58, 1.6) | 0.96 (0.58, 1.6) | ||
| 2 (1551.6, 3038.2) | 0.85 (0.51, 1.4) | 0.86 (0.51, 1.4) | ||
| 3 (3057.1, 15906.9) | 1.8 (1.0, 3.1) | 1.8 (1.0, 3.0) | ||
| Osteophyte-MF category (range) | ||||
| 0 (0.0, 3.7) | reference | reference | ||
| 1 (4.1, 32.1) | 1.1 (0.66, 1.9) | 1.1 (0.66, 1.9) | ||
| 2 (32.6, 111.5) | 1.9 (1.1, 3.3) | 1.9 (1.1, 3.3) | ||
| 3 (111.9, 1316.2) | 2.4 (1.3, 4.4) | 2.4 (1.3, 4.3) | ||
| BMLs-MF category (range) | ||||
| 0 (0) | reference | reference | reference | |
| 1 (1.9, 134) | 1.5 (0.95, 2.4) | 1.5 (0.94, 2.3) | 1.6 (0.99, 2.5) | |
| 2 (135, 5931) | 2.9 (1.7, 4.7) | 3.0 (1.9, 4.8) | 3.1 (1.9, 5.1) | |
| BMLs-MP category (range) | ||||
| 0 (0) | reference | |||
| 1 (8, 1529) | 1.4 (0.93, 2.2) | |||
| BMLs-MT category (range) | ||||
| 0 (0) | reference | reference | reference | |
| 1 (0.4, 5815) | 4.1 (2.4, 6.9) | 4.2 (2.5, 7.0) | 4.2 (2.5, 7.1) | |
| BMLS-LF category (range) | ||||
| 0 (0) | reference | reference | ||
| 1 (2.3, 7576) | 0.51 (0.33, 0.79) | 0.52 (0.34, 0.80) | ||
OST=Osteophyte, BML=Bone marrow lesion, LF=Lateral femur, LT=Lateral tibia, LP=Lateral patella, MF=Medial femur, MP=Medial patella, M=Medial tibia, AIC=Akaike Information Criterion; SBC=Schwarz Bayesian Criterion; LASSO = least absolute shrinkage and selection operator.
Adjusted for sex, race, age, BMI, KLG, WOMAC pain, pain medication use, and JSW.
Models predicting JSN case status from 24 month change in biomarkers are shown in Table 5. Cross-validated AUCs ranged from 0.788 to 0.804. All selection strategies selected change in cartilage volume, osteophytes in the medial femur, and BMLs in the medial tibia.
Table 5.
Multivariable models of 24 month change in biomarkers to predict radiographic progression in the secondary analysis.
| MODEL |
M1 |
M2 |
M3 |
M4 |
|---|---|---|---|---|
| Selection Method | Stepwise, AIC | Stepwise, SBC | Stepwise, P-value | LASSO |
| Model characteristics | ||||
| Biomarkers included | Cartilage, Ost-MT, Ost-MF, BMLs-MF, BMLs-MT, BMLs-LP | Cartilage, Ost-MF, BMLs-MT | Cartilage, effusion-synovitis, Ost-MT, Ost-MF, BMLs-MF, BMLs-MT, BMLs-LP | |
| AUC (unadjusted) | 0.799 | 0.780 | Same as M1 | 0.819 |
| AUC (adjusteda) | 0.834 | 0.814 | 0.843 | |
| AUC (adjusteda, 10 fold cross-validation) | 0.798 | 0.788 | 0.804 | |
| Adjusted odds ratios (95 % CIs) | ||||
| Cartilage volume quartile (range) | ||||
| 0 (−569.2, −74.0) | reference | reference | reference | |
| 1 (−74.0, −24.2) | 0.39 (0.22, 0.72) | 0.35 (0.20, 0.63) | 0.41 (0.22, 0.77) | |
| 2 (−24.2, 8.7) | 0.24 (0.13, 0.44) | 0.23 (0.13, 0.41) | 0.26 (0.14, 0.49) | |
| 3 (8.8, 154.3) | 0.14 (0.07, 0.26) | 0.14 (0.08, 0.25) | 0.14 (0.07, 0.27) | |
| Effusion-synovitis quartile (range) | ||||
| 0 (−12906.2, −955.1) | reference | |||
| 1 (−934.2, −146.0) | 1.4 (0.79, 2.5) | |||
| 2 (−140.3, 488.9) | 0.95 (0.53, 1.7) | |||
| 3 (504.5, 11016.3) | 2.6 (1.4, 4.8) | |||
| Osteophytes-MT quartile (range) | ||||
| 0 (−212.07, −2.05) | reference | reference | ||
| 1 (−2.05, 8.21) | 0.79 (0.45, 1.4) | 0.77 (0.43, 1.4) | ||
| 2 (8.39, 46.07) | 1.2 (0.69, 2.1) | 1.1 (0.63, 1.9) | ||
| 3 (46.25, 425.07) | 2.1 (1.1, 3.9) | 2.1 (1.1, 4.0) | ||
| Osteophytes-MF quartile (range) | ||||
| 0 (−148.46, −0.18) | reference | reference | reference | |
| 1 (0.00, 8.02) | 0.67 (0.38, 1.2) | 0.61 (0.35, 1.1) | 0.69 (0.38, 1.2) | |
| 2 (8.21, 40.10) | 1.1 (0.64, 2.0) | 1.2 (0.67, 2.0) | 1.2 (0.68, 2.2) | |
| 3 (40.47, 448.56) | 3.1 (1.6, 6.0) | 4.0 (2.1, 7.3) | 3.2 (1.6, 6.3) | |
| BMLs-MT category (range) | ||||
| 0 (−4199.70, −0.40) | reference | reference | reference | |
| 1 (0.00, 0.00) | 0.23 (0.11, 0.47) | 0.21 (0.10, 0.42) | 0.23 (0.11, 0.49) | |
| 2 (0.80, 6476.20) | 1.1 (0.51, 2.3) | 1.0 (0.51, 2.1) | 1.1 (0.51, 2.3) | |
| BMLs-LP category (range) | ||||
| 0 (−2129.70, −2.30) | reference | reference | ||
| 1 (0.00, 0.00) | 0.79 (0.47, 1.3) | 0.82 (0.48, 1.4) | ||
| 2 (2.30, 2120.50) | 0.39 (0.22, 0.70) | 0.39 (0.21, 0.70) | ||
| BMLs-MF category (range) | ||||
| 0 (−4253.60, −1.50) | reference | reference | ||
| 1 (0.00, 0.00) | 0.55 (0.33, 0.93) | 0.61 (0.36, 1.0) | ||
| 2 (0.40, 6334.30) | 1.3 (0.78, 2.2) | 1.4 (0.84, 2.5) | ||
OST=Osteophyte, BML=Bone marrow lesion, LF=Lateral femur, LT=Lateral tibia, LP=Lateral patella, MF=Medial femur, MP=Medial patella, MT=Medial tibia, AIC=Akaike Information Criterion; SBC=Schwarz Bayesian Criterion; LASSO = least absolute shrinkage and selection operator.
Adjusted for sex, race, age, BMI, KLG, WOMAC pain, pain medication use, and JSW.
Correlation between biomarkers is shown in Supplemental Figs. S1 and S2. Correlation between biomarkers was generally higher at baseline when compared to 24 months. For both sets of biomarkers, correlation was highest between osteophyte sub-regions. Descriptive statistics of biomarkers and biomarker categories by composite case status were calculated and composite radiographic and pain cases had less cartilage volume at baseline, and higher osteophyte and BML volume in most compartments.
4. Discussion
In the primary analysis, the baseline MOSS measures were predictive of case control status with AUC values (adjusted with 10-fold cross validation) of 0.690 (Model M1) and 0.664 (Model M2). For both AIC (M1) and SBC (M2) based selection, we found that medial-compartment BML volume was a main contributor, but additional BML sub-regions and LT osteophyte volume added significantly to the models.
Not surprisingly, the AUC values were higher for the BL to 24 month change (Table 3) given that the radiographic component of the case/control definitions were based on structural progression, namely medial compartment joint space width change. For M1, several new quantities added significantly, including MF cartilage, ES, and additional regions of the osteophyte and BML volume.
Given that our study used imaging parameters, it was not surprising that the secondary analyses where radiographic progression defined the case/control status performed better with cross validated AUC values of 0.709 (BL) and 0.804 (24 month change). Unlike for the primary analysis, both cartilage and ES were significant contributors to the model that only used BL data. The model using BL to 24 month change (Table 5) had cross-validated AUC values as high as 0.804 based on additional significant contributors compared to the BL model.
Interestingly, the metric reflecting HS did not significantly contribute to any model. It is worth noting that, unlike the others, the HS variable was not a direct measurement of structure rather it was a dimensionless quantity that was optimized based on a comparison with MOAKS scoring. This measurement was found to have lower concurrent validity compared to ES in an initial validation study [27]; it is possible that a re-optimized HS metric, based on a larger number of SQ scored knees than used previously, will perform better.
The spline plots (Fig. 1, Fig. 2, Fig. 3) offer a visual way to understand how each variable contributes to the case/control probability. The high level of non-linearity of these plots demonstrates the complicated relationships between the measures and the probabilities, and suggests that a more sophisticated statistical model, such as generalized additive models, fractional polynomials, or perhaps incorporating new machine learning techniques such as random forest or artificial neural networks, may perform better; investigators should be cautious in assuming a linear relationship between quantitative biomarkers and progression outcomes. This will require further study.
Fig. 3.
Association between 24 month change in biomarkers and composite case (JSN + pain) Status using the restricted cubic spline method.
Legend: The biomarker is along the x-axis and the predicted probability of being a composite case is along the y-axis. The predicted probability of being a case for each value of biomarker is shown by each dot, while the band represents the 95 % confidence interval. Osteophytes are shown in panels B through E; BML volume in panels F through K. A) Cartilage Volume; B) Osteophytes Lateral Femur (LF); C) Osteophytes Lateral Tibia (LT); D) Osteophytes Medial Tibia (MT); E) Osteophytes Medial Femur (MF); F) BML's Medial Patella (MP); G). BML volume lateral patella (LP); H) BML volume medial tibia (MT); I) BML volume lateral tibia (LT); J) BML volume medial femur (MF); K) BML volume lateral femur (LF); L) Effusion-synovitis; M) Hoffa-synovitis. Baseline (BL).
In Hunter et al. [19] the investigators used the same FNIH data and MRI-based measurements of cartilage, BMLs, osteophytes, ES, and HS. Hunter et al. generally used SQ MOAKS scoring for image assessment. They also assessed several quantitative measurements not used in our study, including cartilage thickness, any measure of cartilage in the lateral femur or tibia, the meniscus, trabecular bone texture, bone shape, as well as a set of biochemical markers. Despite using substantially fewer inputs, our models demonstrated similar discrimination with cross-validated AUC values of 0.690 for BL predictors (M1, Table 2) and 0.725 for the BL-24 month predictors (Table 3). This can be compared to AUCs of 0.694 (Hunter et al. M1, Supplementary Table 5) and 0.700 (Hunter et al. M1, Supplementary Table 6) in the study referenced above by Hunter et al. Comparisons are made to the supplementary material of the Hunter study since their main analysis did not use all 600 FNIH subjects. We hypothesize that the similar discrimination found despite using fewer inputs is due to the use of fully quantitative objective measurements that may offer improved accuracy. Further study is needed to confirm this hypothesis.
While SQ scoring is widely used and proven to be a useful tool, we believe it has several limitations. It provides only ordinal scores, generally requires expert readers and has the potential to be time consuming although reader times are generally not reported. MOSS has several advantages over MOAKS. Unlike the ordinal output typical of SQ scoring, MOSS provides floating-point numbers that are a direct reflection of the size of the KOA-related structures and are also more convenient for statistical analysis. We have shown that MOSS can be operated by a diverse group of readers after sufficient training, including individuals without formal medical education. The use of MOSS by these readers provided measurements that reflected radiographic and pain progression with similar performance to a study that used primarily SQ scoring performed by highly trained readers. Improvements to efficiency for MOSS are highly feasible particularly in light of advances in machine learning algorithms. We have already made progress in this area [11]. While we have shown that models developed with MOSS parameters offer similar discriminative performance as models including SQ parameters, future work should consider a direct comparison to models including only SQ parameters. This validation study is one step toward a goal of a nearly fully automated MOSS capable of analyzing very large data sets in a small amount of time, To our knowledge, this is first study to use multiple exclusively fully quantitative software MRI-based measures for assessment of KOA. In addition to offering further insight into the relationships between structural parameters and the progression of KOA, it further validates the MOSS method.
Our study does have some limitations. To facilitate interpretation of parameter estimates and comparison with prior FNIH analyses using semi-quantitative scoring and because we noted non-linear associations between biomarkers and outcomes, we created categories for predictor variables. However, it is recognized that categorizing continuous variables can lead to inefficiency and loss of statistical power. Future work will investigate model building using models for non-linear data, such as generalized additive models, and will investigate cut-points in biomarkers that best predict progression. Other than cartilage volume and HS, the measurements are not normalized knee size, which could create a bias. Knees with both radiographic and pain progression by 12 months were excluded from the FNIH cohort, as were participants with knee or hip replacement between baseline and 24 months. This may have excluded a small number of cases with very fast progression. In addition, participants had to have complete imaging, biochemical biomarker, and clinical data at both baseline and 24 months in order to be considered for this nested case-control study.
Not all MOAKS features were assessed with the MOSS tool, and the Hunter et al. analysis included other features, in addition to MOAKS. For example, we did not evaluate the menisci or cartilage outside the MF compartment, both of which are MOAKS features. Meniscal damage has been shown to be an important risk factor in the incidence and progression of KOA [30]. We also did not evaluate bone shape, which is not a MOAKS feature but was evaluated in the Hunter paper. Hunter et al. described a longitudinal validation of periarticular bone area and 3D shape as biomarkers for KOA progression also using the FNIH as Biomarkers Consortium data [22].
The current MOSS tool is not fully automated and presents a substantial reader burden for studies involving large numbers of scans and, unlike the more unified MOAKS system, each measurement method is made using an independent software tool. Future work will include incorporating the MOSS components into a single software tool, adding the lateral femur and tibia cartilage, and making a direct comparison with MOAKS. As described above, efforts are underway to apply AI methods to substantially decrease the reader time, which will mitigate this limitation. Applying DL was able to reduce the reader time for the BML component 10-fold [11]. We expect similar improved performance after applying DL to all the MOSS measures, lowering the total reader time to less than 5 min.
5. Conclusions
To our knowledge, this is the first study to assess multiple exclusively quantitative MRI-based measurements of KOA for predicting radiographic and/or pain progression. Our results suggest that MOSS may be more useful than semi-quantitative scoring for enriching clinical trials and for other large studies of KOA, as a “one-stop shop” technique for consistent, reliable, precise, and time-sensitive structural evaluation, particularly with further advancement of the technique using AI.
Author contributions
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. Drs. Duryea, Smith and Collins had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study conception and design: Smith, Duryea, Acquisition of Data: Smith, Duryea, Collins, Lo, Sury, Bahouth, Yin, and Schaefer. Analysis and Interpretation of data: Smith, Duryea, Collins.
Ethics approval and consent to participate
The OAI study included participants recruited from the University of Maryland School of Medicine and Johns Hopkins University (Baltimore, MD), Ohio State University (Columbus, OH), University of Pittsburgh (Pittsburgh, PA), and Memorial Hospital of Rhode Island (Pawtucket, RI). The OAI Coordinating Center was located at the University of California, San Francisco (UCSF). Approval was obtained from the institutional review boards at each site.
Availability of data and materials
The datasets used/and or analyzed during the current study are available in the OAI repository available at: https://nda.nih.gov/oai/.
Role of the funding source
Research reported in this publication was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases of the National Institutes of Health under Award Number R011AR956664 (PI: Duryea) and K01AR075879 (PI: Collins). The Osteoarthritis Initiative is a public-private partnership comprised of five contracts (N01-AR [1]24 2–2258; N01-AR-2-2259; N01-AR-2-2260; N01-AR-2-2261; N01-AR-2-2262) funded by the National Institutes of Health, a branch of the Department of Health and Human Services, and conducted by the Osteoarthritis Initiative Study Investigators. Private funding partners include Merck Research Laboratories; Novartis Pharmaceuticals Corporation, GlaxoSmithKline; and Pfizer, Inc. Private sector funding for the Osteoarthritis Initiative is managed by the Foundation for the National Institutes of Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Arthritis and Musculoskeletal and Skin Diseases, or the National 32 Institutes of Health. The private funding partners had no role in the study design or in the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by the private funding partners.
Declaration of competing interest
Drs. Duryea and Collins report grants from NIH – National Institute of Arthritis and Musculoskeletal and Skin Diseases as noted above. Drs. Smith, Low, Sury, Bahouth, Yi, and Schaefer report no financial disclosures or conflicts of interest.
Acknowledgements
The authors would like to thank Ms. Quinley Maio from the Quantitative Musculoskeletal Imaging Group (QMIG), Department of Radiology, Brigham and Women's Hospital, for her technical assistance with this project.
Handling Editor: Professor H Madry
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.ocarto.2025.100638.
Contributor Information
Stacy E. Smith, Email: ssmith@bwh.harvard.edu.
Lawrence Lo, Email: Lawrence.yh.lo@gmail.com.
Meera Sury, Email: msury@partners.org.
Sara M. Bahouth, Email: sbahouth@mgh.harvard.edu.
Ming Yin, Email: ming.yin02@gmail.com.
Lena F. Schaefer, Email: lenafranziskaschaefer@yahoo.com.
Jamie E. Collins, Email: jcollins13@bwh.harvard.edu.
Jeffrey Duryea, Email: jduryea@bwh.harvard.edu.
Appendix A. Supplementary data
The following is/are the supplementary data to this article:
References
- 1.CDC.gov/arthritis/basic/osteoarthritis.htm. (Checked January, 2025).
- 2.Cui A., Li H., Wang D., Zhong J., Chen Y., Lu H. Global, regional prevalence, incidence and risk factors of knee osteoarthritis in population-based studies. eClinicalMedicine. 2020;29 doi: 10.1016/j.eclinm.2020.100587. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Cho Y., Jeong S., Kim H., Kang D., Lee J., Kang S.-B., et al. Disease-modifying therapeutic strategies in osteoarthritis: current status and future directions. Exp. Mol. Med. 2021;53:1689–1696. doi: 10.1038/s12276-021-00710-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Conaghan P.G., Bowes M.A., Kingsbury S.R., Brett A., Guillard G., Rizoska B., et al. Disease-modifying effects of a novel cathepsin K inhibitor in osteoarthritis: a randomized controlled trial. Ann. Intern. Med. 2020;172:86–95. doi: 10.7326/M19-0675. [DOI] [PubMed] [Google Scholar]
- 5.Hochberg M.C., Guermazi A., Guehring H., Aydemir A., Wax S., Fleuranceau-Morel P., et al. Effect of intra-articular sprifermin vs placebo on femorotibial joint cartilage thickness in patients with osteoarthritis: the FORWARD randomized clinical trial. JAMA. 2019;322:1360–1370. doi: 10.1001/jama.2019.14735. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Eckstein F., Guermazi A., Gold G., Duryea J., Le Graverand M.-P.H., Wirth W., et al. Imaging of cartilage and bone: promises and pitfalls in clinical trials of osteoarthritis. Osteoarthr. Cartil. 2014;22:1516–1532. doi: 10.1016/j.joca.2014.06.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Felson D.T., Niu J., Guermazi A., Roemer F., Aliabadi P., Clancy M., et al. Correlation of the development of knee pain with enlarging bone marrow lesions on magnetic resonance imaging. Arthritis Rheum. 2007;56:2986–2992. doi: 10.1002/art.22851. [DOI] [PubMed] [Google Scholar]
- 8.Gandhamal A., Talbar S., Gajre S., Razak R., Hani A.F.M., Kumar D. Fully automated subchondral bone segmentation from knee MR images: data from the Osteoarthritis Initiative. Comput. Biol. Med. 2017;88:110–125. doi: 10.1016/j.compbiomed.2017.07.008. [DOI] [PubMed] [Google Scholar]
- 9.Fripp J., Crozier S., Warfield S.K., Ourselin S. Automatic segmentation and quantitative analysis of the articular cartilages from magnetic resonance images of the knee. IEEE Trans. Med. Imag. 2009;29:55–64. doi: 10.1109/TMI.2009.2024743. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Cigdem O., Deniz C.M. Artificial intelligence in knee osteoarthritis: a comprehensive review for 2022. Osteoarthr. Imag. 2023;3 doi: 10.1016/j.ostima.2023.100161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Preiswerk P., Sury M., Wortman J., Duryea J. Deep learning-based bone marrow lesion segmentation on knee MRI for the assessment of osteoarthritis. Osteoarthr. Cartil. Open. 2019;4:1–6. doi: 10.1016/j.ocarto.2022.100234. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Peterfy C.G., Guermazi A., Zaim S., Tirman P.F., Miaux Y., White D., et al. Whole-organ magnetic resonance imaging score (WORMS) of the knee in osteoarthritis. Osteoarthr. Cartil. 2004;12:177–190. doi: 10.1016/j.joca.2003.11.003. [DOI] [PubMed] [Google Scholar]
- 13.Hunter D.J., Lo G.H., Gale D., Grainger A.J., Guermazi A., Conaghan P.G. The reliability of a new scoring system for knee osteoarthritis MRI and the validity of bone marrow lesion assessment: BLOKS (Boston Leeds Osteoarthritis Knee Score) Ann. Rheum. Dis. 2008;67:206–211. doi: 10.1136/ard.2006.066183. [DOI] [PubMed] [Google Scholar]
- 14.Hunter D.J., Guermazi A., Lo G.H., Grainger A.J., Conaghan P.G., Boudreau R.M., et al. Evolution of semi-quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score) Osteoarthr. Cartil. 2011;19:990–1002. doi: 10.1016/j.joca.2011.05.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Felson D.T., Lynch J., Guermazi A., Roemer F.W., Niu J., McAlindon T., et al. Comparison of BLOKS and WORMS scoring systems part II. Longitudinal assessment of knee MRIs for osteoarthritis and suggested approach based on their performance: data from the Osteoarthritis Initiative. Osteoarthr. Cartil. 2010;18:1402–1407. doi: 10.1016/j.joca.2010.06.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Collins J.E., Losina E., Nevitt M.C., Roemer F.W., Guermazi A., Lynch J.A., et al. Semi-quantitative imaging biomarkers of knee osteoarthritis progression: data from the FNIH OA biomarkers consortium. Arthritis Rheumatol. 2016:68 2422–2431. doi: 10.1002/art.39731. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Roemer F.W., Guermazi A., Collins J.E., Losina E., Nevitt M.C., Lynch J.A., et al. Semi-quantitative MRI biomarkers of knee osteoarthritis progression in the FNIH biomarkers consortium cohort− Methodologic aspects and definition of change. BMC Muscoskelet. Disord. 2016;17:1–17. doi: 10.1186/s12891-016-1310-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Guermazi A., Hayashi D., Roemer F., Felson D.T., Wang K., Lynch J., et al. Severe radiographic knee osteoarthritis–does Kellgren and Lawrence grade 4 represent end stage disease?–the MOST study. Osteoarthr. Cartil. 2015;23:1499–1505. doi: 10.1016/j.joca.2015.04.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Hunter D.J., Deveza L.A., Collins J.E., Losina E., Katz J.N., Nevitt M.C., et al. Multivariable modeling of biomarker data from the phase I foundation for the national institutes of health osteoarthritis biomarkers consortium. Arthritis Care Res. 2022;74:1142–1153. doi: 10.1002/acr.24557. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Eckstein F., Collins J., Nevitt M., Lynch J., Kraus V., Katz J., et al. Cartilage thickness change as an imaging biomarker of knee osteoarthritis progression–data from the fnih OA biomarkers consortium. Arthritis Rheumatol. 2015;67:3184. doi: 10.1002/art.39324. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hunter D.J., Nevitt M., Losina E., Kraus V. Biomarkers for osteoarthritis: current position and steps towards further validation. Best Pract. Res. Clin. Rheumatol. 2014;28:61–71. doi: 10.1016/j.berh.2014.01.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Hunter D., Nevitt M., Lynch J., Kraus V.B., Katz J.N., Collins J.E., et al. Longitudinal validation of periarticular bone area and 3D shape as biomarkers for knee OA progression? Data from the FNIH OA Biomarkers Consortium. Ann. Rheum. Dis. 2016;75:1607–1614. doi: 10.1136/annrheumdis-2015-207602. [DOI] [PubMed] [Google Scholar]
- 23.Peterfy C.G., Schneider E., Nevitt M. The osteoarthritis initiative: report on the design rationale for the magnetic resonance imaging protocol for the knee. Osteoarthr. Cartil. 2008;16:1433–1441. doi: 10.1016/j.joca.2008.06.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Amesbury R., Ragati-Haghi H., Mathiessen A., Laffaye T., Stein R.I., Collins J.E., et al. Gender differences in patterns of cartilage loss: data from the Osteoarthritis Initiative. Osteoarthr. Cartil. 2024;32:1149–1153. doi: 10.1016/j.joca.2024.04.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Duryea J., Iranpour-Boroujeni T., Collins J.E., Vanwynngaarden C., Guermazi A., Katz J.N., et al. Local area cartilage segmentation: a semiautomated novel method of measuring cartilage loss in knee osteoarthritis. Arthritis Care Res. 2014;66:1560–1565. doi: 10.1002/acr.22332. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Ratzlaff C., Guermazi A., Collins J., Katz J.N., Losina E., Vanwyngaarden C., et al. A rapid, novel method of volumetric assessment of MRI-detected subchondral bone marrow lesions in knee osteoarthritis. Osteoarthr. Cartil. 2013;21:806–814. doi: 10.1016/j.joca.2013.03.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Smith S.E., Hosseinzadeh S., Maetani T., Shilpa P., Collins J.E., Kwoh C.K., et al. Association of quantitative measures of effusion-synovitis and hoffa-synovitis with radiographic and pain progression: data from the FNIH OA biomarkers consortium. Osteoarthr. Cartil. Open. 2021;3 doi: 10.1016/j.ocarto.2021.100138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Hakky M., Jarraya M., Ratzlaff C., Guermazi A., Duryea J. Validity and responsiveness of a new measure of knee osteophytes for osteoarthritis studies: data from the osteoarthritis initiative. Osteoarthr. Cartil. 2015;23:2199–2205. doi: 10.1016/j.joca.2015.07.003. [DOI] [PubMed] [Google Scholar]
- 29.Desquilbet L., Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat. Med. 2010;29:1037–1057. doi: 10.1002/sim.3841. [DOI] [PubMed] [Google Scholar]
- 30.Berthiaume M.-J., Raynauld J.-P., Martel-Pelletier J., Labonté F., Beaudoin G., Bloch D.A., et al. Meniscal tear and extrusion are strongly associated with progression of symptomatic knee osteoarthritis as assessed by quantitative magnetic resonance imaging. Ann. Rheum. Dis. 2005;64:556–563. doi: 10.1136/ard.2004.023796. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets used/and or analyzed during the current study are available in the OAI repository available at: https://nda.nih.gov/oai/.