Baseline 3D joint space width on weight-bearing CT in comparison with 2D JSW_x on radiographs for predicting 24-month worsening of knee structure in the multicenter osteoarthritis study

Abstract

Objective:

This study evaluated the predictive validity of baseline 3D joint space width (3D-JSW) measured using weight-bearing computed tomography (WBCT) versus 2D-JSW on radiographs for predicting knee structural worsening over 24 months, assessed by radiographs and MRI.

Design:

A prospective observational study using data from the Multicenter Osteoarthritis Study (MOST) included 425 knees. Baseline 3D-JSW was assessed using WBCT acquired in a standing fixed-flexion position, and 2D-JSW was assessed from standing fixed-flexion posteroanterior radiographs. Outcomes included 24-month changes in tibiofemoral Osteoarthritis Research Society International (OARSI) joint space narrowing (JSN) grade and MRI Osteoarthritis Knee Score Cartilage Morphology (MOAKS-CM). Predictive validity was evaluated using logistic regression and comparison of areas under ROC curves.

Results:

The study population (N=302) had a mean age of 63.2 years (standard deviation 8.9) and BMI of 28.2 kg/m² (standard deviation 4.8), with radiographic OA observed in 45.9 %. For medial JSN worsening, 2D-JSW (AUC=0.711) showed significantly better predictive validity than the basic demographic model (AUC=0.626, p=0.032), while 3D-JSW (AUC=0.681) did not outperform 2D-JSW (p=0.184). For lateral JSN worsening, 3D-JSW (AUC=0.900) and 2D-JSW (AUC=0.893) were similar (p=0.814). For medial MOAKS-CM worsening, 3D-JSW (AUC=0.729) and 2D-JSW (AUC=0.742) were similar (p=0.579), as were 3D-JSW (AUC=0.699) and 2D-JSW (AUC=0.692) for lateral worsening (p=0.779).

Conclusions:

3D-JSW did not show superior predictive validity over 2D-JSW in predicting knee structural worsening over 24 months in a cohort predominatly without radiographic knee OA. WBCT remains a promising imaging modality for osteoarthritis assessment. Further studies are needed to determine its role as a diagnostic tool.

Introduction

Osteoarthritis (OA) is a prevalent joint disorder that leads to joint pain and disability, affecting an estimated 595 million people worldwide as of 2020 [1]. The prevalence of OA has been steadily increasing in recent years, contributing to a growing global health burden. The most common weight-bearing joint affected is the knee with over 32.5 million Americans diagnosed in varying levels of severity [2,3]. However, in the early stages, the disease may be more amenable to preventative interventions, highlighting the importance of identifying individuals at high risk of knee structural worsening [2,4,5]. Early identification of these individuals is crucial to prevent or slow the progression of the disease, ultimately improving long-term outcomes for those at risk.

Radiography is the most common imaging examination used to evaluate structural disease in OA clinical trials, with radiographic joint space width (JSW) being a very common measure for assessing the progression of knee OA in clinical trials to date [6]. In longitudinal studies, JSW, the distance between the distal femur and proximal tibia on standing radiographs, is used as a surrogate marker of tibiofemoral cartilage thickness albeit acknowledging that JSW is a composite measure refelcting cartilage damage or integrity, meniscus damage or integrity and meniscal positioning including extrusion and either of these features or a combination may contribute to longitudinal joint space narrowing (JSN) [7]. Magnetic resonance imaging (MRI) is now recommended for clinical trials, particularly for assessing cartilage morphology [6]. MRI semi-quantitative scoring of knee OA has proven to be a valuable method for performing multi-feature joint assessment [8]. However, challenges remain due to the high cost and longer scan times of MRI, which impose a significant burden on patients.

The recent increase in availability of low-dose weight-bearing CT (WBCT) may provide an opportunity to improve both detection and longitudinal monitoring of knee OA structural change. This modality allows the knee to be imaged with the patient in a standing fixed-flexed position, with advantages compared to radiographs due to 3D acquisition overcoming the challenge of bony overlap and compared to non-weight bearing CT and MRI scans[9]. WBCT imaging markers can be acquired at similar costs to radiographs, and much less than MRI, with a scan time that is one-tenth that of MRI, and at a Relative Radiation Level equivalent to radiographs, without contraindications for patients with implanted metal [10,11].

Recent evidence demonstrates that 3D JSW measurements on WBCT are more responsive to temporal changes in the joint than 2D JSW measurements on radiographs [12]. However, a necessary next step toward qualification of 3D JSW as an imaging marker for OA, is to establish the prognostic validity for longitudinal worsening of structural disease. If suitably validated, WBCT imaging markers may allow less expensive and shorter duration clinical trials to develop more timely interventions that may mitigate the long-term effects of knee OA.

Thus, this study aimed at evaluating the predictive validity of baseline 3D JSW obtained from WBCT compared to 2D JSW on radiographs for predicting knee structural worsening over 24-month follow-up.

Methods

Participant selection and study design

This study was conducted ancillary to the Multicenter Osteoarthritis Study (MOST), an NIA-funded investigation, at the University of Iowa (UI) clinic site . UI-MOST followed a cohort of 1503 participants for 12 years (Original Cohort) and, at the 12-year visit, recruited an additional 747 participants with no more than mild OA or mild pain (New Cohort). For the present ancillary study, we included participants from the MOST cohort who underwent WBCT imaging at the 144-month (12-year) follow-up visit. To study the predictive value of 2D and 3D JSW for worsening knee joint structure, this cohort was selected since participants are representative of those who would be most likely to benefit from the development of improved imaging biomarkers. Participants were drawn from the community surrounding the clinic, but selected as likely to have either preexisting knee OA or to be at elevated risk for knee OA while maintaining a distribution of age and sex similar to the U. S. population as previously described [12-14].

To maximize sampling efficiency in assessing and comparing the predictive performance of 3D JSW and 2D JSW for knee structural worsening, participants were recruited through UI-MOST in the following manner. Exclusion criteria for the MOST Study included inability to attend 24-month follow-up, inflammatory arthritis, inability to walk independently, or serious health conditions such as end-stage renal disease, cancer except non-melanoma skin cancer, severe heart failure. Additional exclusion criteria for this ancillary study included inability to safely stand for the 2-minute WBCT image acquisition or tremor that might result in motion artifact. All participants participated in an IRB-approved consent process, culminating in providing written consent.

Image acquisition

Bilateral, standing fixed-flexion posteroanterior (PA) radiographs of the tibiofemoral compartments were acquired at baseline and the 24-month follow-up as previously described [15,16]. The pixel spacing was 0.10mm. Following radiograph acquisition, bilateral fixed-flexion WBCT images were acquired using a standing frame that externally rotated participants’ toes 10°while aligning the great toes, the patellae, and the anterior superior iliac spines in the saggital plane. This custom frame resulted in a knee flexion angle of approximately 20°. WBCT was acquired using a prototype scanner (CurveBeam, Warrington, PA) that reconstructed cone beam images (0.37mm isotropic voxels; FOV 20cm high x 35cm width and depth; 120 kVp, 5 mA) and an effective radiation dose of approximately 0.1 mSv.

2D JSW_x measurements on radiographs

A semi-automated software tool developed and documented by Duryea et al [17,18] was used to delineate the femoral and tibial margins on digital knee radiographs and enable measurements of JSW at fixed locations (JSW_x). In brief, this method involved establishment of a coordinate system referenced to anatomical landmarks, such that each x-location represents the position of the JSW_x measurement along the projected tibiofemoral joint. The variable x is a dimensionless quantity that represents the fractional distance from the medial to the lateral extents of the femur. In this study, following previous reports[12,19,20], radiographs were evaluated for change in 2D JSW_x at 22.5 % (JSW_.225) and 75.0 % (JSW_.750) of this distance (Fig. 1).

Fig. 1.

2D JSW_x measurements on radiographs.

A posterioranterior radiograph of a left knee. The x-axis, which is a line tangent to both femoral epicondyles, was positioned automatically by the software. The x variable indicates the location along the joint surface where the JSW_x measurement was taken. The y-axis is perpendicular to the x-axis and touching the most prominent part of the medial epicondyle. The x=1.0 line was defined as a tangent to the most prominent part of the lateral epicondyle. In this study, we measured the distance at two points: the medial joint space at JSW_.225 and the lateral joint space at JSW_.750. For clarity, tibial margins are not displayed in this schematic. In the actual semi-automated JSW measurement process, both femoral and tibial margins are automatically delineated.

3D JSW measurements by WBCT

Tibiofemoral geometries were analyzed using semi-automated segmentation to generate digital 3D surface models of the femur and tibia (Seg3D Version 2.2.1). These initial segmentations were presented to an expert user along with tools for correcting any inaccurate surface details. For each acquisition, the 3D JSW was determined by calculating the Euclidean distances between thetriangular faces of the opposing femoral and tibial surfaces within eight defined tibiofemoral subregions: central medial and lateral compartments of the femur (CMF/CLF) and tibia (CMT/CLT), as well as the anterior and posterior aspects of tibia (AMT/ALT, PMT/PLT) (Fig. 2A, B). To further investigate the 3D JSW in the anterior, central, and posterior, each mediolateral location was subdivided into three segments along the anteroposterior direction, approximating the locations used for 2D JSW_x meansurments on radiography (Fig. 2C). The mean 3D JSW values obtained from the anterior, central, and posterior tibia was utilized as the measurement values in this study.

Fig. 2.

Method for defining the 8 tibiofemoral subregions for measuring 3D JSW.

A. Anatomical delineation of the central medial femur (CMF) region in the sagittal view. The tibia was divided into anterior (AMT), central (CMT), and posterior (PMT) sub-regions, each equally divided into thirds. B. Lateral compartment sub-regions, including anterior lateral tibia (ALT), central lateral tibia (CLT), posterior lateral tibia (PLT), and central lateral femur (CLF), were defined in a similar manner. C. Division of the tibial plateau into six regions using axial images from 3D CT.

OARSI JSN scoring

The medial and lateral compartments for each knee were scored using the Osteoarthritis Research Society International (OARSI) JSN grading scale [21]. Standing fixed-flexion PA radiographs at baseline and 24-month follow-up were graded by comparing each image with the OARSI reference atlas, assigning a severity grade for JSN ranging from zero to three, where grade 0 indicates normal joint space, grade 1 indicates minimal narrowing, grade 2 indicates moderate narrowing, and grade 3 indicates severe joint space narrowing. OARSI JSN scoring was performed using a standardized multi-reader process. Paired baseline and follow-up radiographs were read independently by two primary readers (Reader #1: rheumatologist; Reader #2: senior musculoskeletal radiologist). When their grades differed, a two-stage adjudication process was implemented. First, Reader #1 reviewed the radiographs again with knowledge of Reader #2’s scores and indicated whether he agreed with the senior reader. If Reader #1 agreed, the senior reader’s score was used. For films with continued disagreement, Reader #2 re-evaluated the images. Persistent discrepancies were adjudicated by a consensus panel including both primary readers and a third senior musculoskeletal radiologist (Reader #3). All readers were blinded to participants’ clinical characteristics throughout the scoring process. The reliability of this OARSI JSN scoring approach has been previously reported; Felson et al. demonstrated inter-reader agreement on half-grade progression of κ = 0.58 and agreement for either half- or full-grade progression of κ = 0.66 [16].OARSI JSN worsening was defined as an increase of at least one grade.

MRI acquisition and scoring

Seated MRI were acquired as part of MOST, using axial and sagittal proton-density weighted fat-suppressed (PDFS) fast spin echo sequences and a coronal short tau inversion recovery (STIR) sequence [22]. The MRI acquisition protocol utilized a 1.5T extremity scanner (GE Optima MR430s, GE Healthcare, Waukesha, WI) with a dedicated knee coil. The detailed acquisition parameters were as follows: axial PDFS images were obtained with repetition time (TR) 4680 ms, echo time (TE) 13 ms, 3 mm slice thickness with no gaps, 20 slices, 288 × 192 matrix, 140mm FOV, and echo train length (ETL) 8; sagittal PDFS images with TR 4800 ms, TE 35 ms, 3 mm slice thickness with no gaps, 32 slices, 288 × 192 matrix, 140 mm FOV, and ETL 8; and coronal STIR images with TR 6650 ms, TE 15 ms, inversion time (TI) 100 ms, 3mm slice thickness with no gaps, 28 slices, 256 × 192 matrix, 140 mm FOV, and ETL 8. Each knee scanned was evaluated for the tibiofemoral compartment in both the medial and lateral regions, using the same subregional definitions applied in the WBCT analysis, with the medial compartment represented by 5 subregions (AMT, CMT, PMT, CMF, PMF) and the lateral compartment represented by 5 subregions (ALT, CLT, PLT, CLF, PLF). Cartilage in each subregion was scored using the MRI Osteoarthritis Knee Score Cartilage Morphology Score (MOAKS-CM) [23]. MOAKS-CM scoring was performed by two evaluators with 15 and 20 years of experience, respectively. Images were assessed with knowledge of their chronological order; however, both evaluators were blinded to participants’ clinical characteristics. The reliability of MOAKS-CM scoring by these readers has been validated previously, with intra-rater weighted kappa values of 0.71–0.95 and inter-rater weighted kappa values of 0.36–0.85 for cartilage area and depth, as well as percent agreement of 70–95 %, as reported by Hunter et al [23]. In brief, MOAKS quantifies the size of cartilage lesions as: 0 (none), 1 (<10 %), 2 (10–75 %), and 3 (>75 %). Additionally, a separate score is assigned to indicate the percentage of the subregion affected by full-thickness cartilage loss: 0 (none), 1 (<10 %), 2 (10–75 %), and 3 (>75 %). For this study, these two scores were combined to create 10 categories ({0.0} (normal), {1.0}, {1.1}, {2.0}, {2.1}, {2.2}, {3.0}, {3.1}, {3.2}, {3.3}), which were used for classification and evaluation (Fig. 3) [24,25]. Worsening was defined as an increase in the MOAKS-CM score between baseline and the 24-month follow-up, and was considered present if at least one of the five subregions within each of the medial or lateral compartments showed progression.:

Fig. 3.

MRI-based cartilage scoring using MOAKS Cartilage Morphology Scores.

The light blue area represents cartilage. The blue line indicates regions with partial cartilage loss, while the red line indicates regions with full-thickness cartilage loss. These categories are numerically identified by combining two separate feature scores into a single number, where the integer portion represents the score for lesion size and the decimal portion represents the score for the amount of full-thickness cartilage loss.

1. Increase in lesion size: This was defined as a change in the integer portion of the score, for example, from 0→1, 1→2, or 2→3.

2. Increase in full-thickness cartilage loss: This was defined as an increase in the decimal portion of the raw score, for example, from 1.0→1.1, 2.0→2.1, 2.0→2.2, 3.0→3.1, 3.0→3.2, 3.0→3.3, 3.1→3.2, 3.1→3.3, or 3.2→3.3.

In addition, even when the baseline and follow-up scores were numerically identical, cases with documented within-grade worsening—i.e., an increase in the extent of full-thickness cartilage loss within the same overall grade—were also considered to represent structural worsening. The degree of MOAKS-CM worsening was classified into three levels: within-grade worsening (reader-identified change without a numerical score increase), full-grade worsening (score increased by one grade), and more than full-grade worsening (score increased by more than one grade).

Statistical analysis

The outcome of knee structural worsening was defined based on two measures respectively: a) OARSI JSN worsening and b) MOAKS-CM worsening. Analyses were conducted separately for the medial and lateral compartments of each knee. Logistic regression was first performed to evaluate the prediction of 24-month knee structural worsening using sex, BMI, and age (Basic Model). Additional logistic regression models were constructed to assess the association of baseline 2D JSW_x and mean 3D JSW with 24-month knee structural worsening, adjusting for sex, BMI, and age. Next, the predictive validity of 2D JSW_x and mean 3D JSW, averaged across the anterior, central, and posterior regions, was assessed using logistic regression and receiver operating characteristic (ROC) curves to evaluate their ability to distinguish between participants with and without knee structure worsening. The area under the ROC curve (AUC) was used to compare the predictive validity of 2D and 3D measurements for structural worsening. To estimate the confidence intervals for the AUCs and assess the stability of the predictive models, bootstrapping was performed. For OARSI JSN worsening, the analysis was conducted at the knee level, and logistic regression with generalized estimating equations (GEE) was used to account for the correlation between knees within a participant. For MOAKS-CM worsening, analyses were conducted at the participant level because MRI was acquired for only one knee per participant. For all analyses, P values less than .05 were considered statistically significant. All statistical analyses were performed using SAS v.9.4 (SAS Institute, Cary, North Carolina).

Results

Participant characteristics

At baseline for this ancillary study, 824 participants who underwent WBCT were enrolled. Of these, 70 participants who did not return for the 24-month follow-up clinic visit were excluded, leaving 754 participants. Among them, 190 participants without collected imaging data and 262 participants who were not selected for 2D JSW parameter processing by the parent study (MOST) were further excluded. This resulted in 302 participants (604 knees) being eligible for analysis (Fig. 4). At the knee level, 21 knees were excluded due to knee arthroplasty, osteonecrosis, or KL grade 4, along with 158 knees excluded due to motion artifacts or poor image quality. The final analytical sample comprised 302 participants, including 123 participants with bilateral knees (246 knees) and 179 participants with unilateral knees (179 knees) included, for a total of 425 knees analyzed for OARSI JSN worsening. For the analysis of MOAKS-CM worsening, 291 participants (291 knees) from this sample who had MRI scans available were included. Table 1 presents the participants’ age and BMI, as well as the characteristics of the knees, including side, compartment more affected by OA and KL grade.

Fig. 4.

Flowchart for inclusion of participants.

Table 1.

Overview of participant demographics (total sample and MRI subsample).

Characteristics	Total Sample (N=302)	MRI Subsample (N=291)
Female / Male	167 / 135	160 / 131
Knees (Left / Right)	157 / 268 (Total: 425)	126 / 165 (Total: 291)
Age (years), mean ± SD	63.2 ± 8.9	63.3 ± 8.9
Female	63.2 ± 8.8	63.2 ± 8.8
Male	63.2 ± 9.1	63.4 ± 9.1
BMI (kg/m2), mean ± SD	28.2 ± 4.8	28.2 ± 4.8
Female	27.6 ± 5.3	27.7 ± 5.3
Male	29.0 ± 4.0	28.9 ± 4.1
KL grade (0 / 1 / 2 / 3)*	230 / 91 / 82 / 22	154 / 64 / 55 / 18

^*Values represent number of knees.

For medial OARSI JSN worsening, 48 knees (11.3 %) showed worsening scores and 377 knees did not through the 24-month follow-up. Supplementary Table 1 shows the distribution of worsening across baseline medial OARSI JSN grades. The AUC of 2D JSW (AUROC=0.711) was 0.085 higher (95 % CI: −0.163 to −0.008, p=0.032) than that of the basic model (AUROC=0.626). However, mean 3D (AUROC=0.681) was not statistically superior (p=0.184) to 2D JSW (Table 2).

Table 2.

Associations between 3D JSW and 2D JSW with medial OARSI JSN worsening.

Medial OARSI JSN worsening (48/425 knees)
Predictor	AUC	Contrast (95 % CI)	p-value
* Basic model	0.626	−0.085 (−0.163, −0.008)	0.032
2D JSW.225	0.711	Reference
Mean 3D JSW	0.681	−0.030 (−0.075, 0.014)	0.184

^*The basic model refers to a logistic regression model adjusted for age, sex, and BMI.

For lateral OARSI JSN worsening, 11 knees (2.6 %) showed worsening scores and 414 knees did not through the 24-month follow-up. Supplementary Table 2 shows the distribution of worsening across baseline medial OARSI JSN grades. The AUC of 2D JSW (AUROC=0.893) was 0.089 higher than that of the basic model (AUROC=0.804), but this difference was not statistically significant (95 % CI: −0.187 to 0.008, p=0.073). Similarly, although the AUC for mean 3D (AUROC=0.900) was 0.007 higher than that for 2D JSW, this difference was also not statistically significant (95 % CI: −0.048 to 0.061, p=0.814) (Table 3).

Table 3.

Associations between 3D JSW and 2D JSW with lateral OARSI JSN worsening.

Lateral OARSI JSN worsening (11/425 knees)
Predictor	AUC	Contrast (95 % CI)	p-value
* Basic model	0.804	−0.089 (−0.187, 0.008)	0.073
2D JSW.750	0.893	Reference
Mean 3D JSW	0.900	0.007 (−0.048, 0.061)	0.814

^*The basic model refers to a logistic regression model adjusted for age, sex, and BMI.

Regarding medial MOAKS-CM worsening, 53 knees (18.2 %) exhibited worsening scores, while 238 knees showed no worsening over the 24-month follow-up. Supplementary Table 3 shows the distribution of medial MOAKS-CM worsening across baseline scores within the five medial tibiofemoral subregions. Among the 53 knees with worsening, 13 showed within-grade worsening, 26 had full-grade worsening, and 14 had more than full-grade worsening. The AUC for 2D JSW (AUROC = 0.742) was 0.080 higher than that of the basic model (AUROC = 0.662), whereas the AUC for mean 3D (AUROC = 0.729) was 0.013 lower than that of 2D JSW (Table 4). However, neither difference was statistically significant (95 % CI: −0.161 to 0.002, p = 0.055, and 95 % CI: −0.058 to 0.032, p = 0.579, respectively).

Table 4.

Associations between 3D JSW and 2D JSW with medial MOAKS-CM worsening.

Medial MOAKS-CM worsening (53/291 knees)
Predictor	AUC	Contrast (95 % CI)	p value
* Basic model	0.662	−0.080 (−0.161, 0.002)	0.055
2D JSW.225	0.742	Reference
Mean 3D JSW	0.729	−0.013 (−0.058, 0.032)	0.579

^*The basic model refers to a logistic regression model adjusted for age, sex, and BMI.

Regarding lateral MOAKS-CM worsening, 34 knees (11.7 %) exhibited worsening scores, while 257 knees showed no worsening over the 24-month follow-up. Supplementary Table 4 shows the distribution of lateral MOAKS-CM worsening across baseline scores within the five lateral tibiofemoral subregions. Among the 34 knees with worsening, 8 showed within-grade worsening, 22 had full-grade worsening, and 4 had more than full-grade worsening. The AUC for 2D JSW (AUROC = 0.692) was 0.031 higher than that of the basic model (AUROC = 0.662), whereas the AUC for mean 3D (AUROC = 0.699) was 0.007 lower than that of 2D JSW (Table 5). However, neither difference was statistically significant (95 % CI: −0.105 to 0.043, p = 0.413, and 95 % CI: −0.040 to 0.054, p = 0.779, respectively).

Table 5.

Associations Between 3D JSW and 2D JSW with lateral MOAKS-CM worsening.

Lateral MOAKS-CM worsening (34/291 knees)
Predictor	AUC	Contrast (95 % CI)	p value
* Basic model	0.662	−0.031 (−0.105, 0.043)	0.413
2D JSW.750	0.692	Reference
Mean 3D JSW	0.699	−0.007 (−0.040, 0.054)	0.779

^*The basic model refers to a logistic regression model adjusted for age, sex, and BMI.

Discussion

The objective of this study was to evaluate the predictive validity of baseline 3D JSW on WBCT in comparison with 2D JSW_x on radiographs to predict worsening of tibiofemoral and cartilage morphology on MRI. In the group of participants studied, 3D JSW on WBCT did not demonstrate superior predictive validity for worsening OARSI JSN scores or MOAKS-CM scores compared to 2D JSW assessed on knee radiographs. Regarding only worsening of medial OARSI JSN, 2D JSW showed statistically significant predictive superiority over the basic model with only demographic factors. Results are likely representative of this sample in which 24.5 % of knees had radiographic OA and radiographic worsening was detected in 11.3 % of medial and 2.6 % of lateral tibiofemoral compartments and MOAKS worsening in 18.2 % and 11.7 % respectively.

Radiography has long been utilized as the traditional imaging technique in the evaluation of knee osteoarthritis. Other imaging modalities, like WBCT, have shown potential due to their superior detection of OA-related features such as osteophytes, subchondral cysts, and meniscal abnormalities [26]. Although radiographs have been the standard imaging modality in the clinical assessment of OA, they are not without limitations. While radiography allows visualization of bony features and offers an indirect measurement of joint space width, its accuracy is limited by its projectional nature. In contrast, WBCT comprehensively visualizes bony structures in 3D and provides improved visualization of OA features. This, in turn, potentially allows for greater sensitivity and responsiveness to change in OA[12]. Moreover, radiography commonly yields poor reproducibility due to factors such as positional variability compounded by the aforementioned visualization challenges. This consequently gives rise to a critical measurement error of JSW, which serves as an essential marker of structural worsening. It has also been demonstrated to be a poor predictor of clinical outcomes in OA [27]. The weight-bearing nature of WBCT marks this modality as an ideal acquisition method in OA-related JSW analysis. The positioning of the joint under functionally loaded conditions is essential to avoid underestimating JSW loss which may occur in non-weight bearing imaging.

In this study, we assessed prediction of tibiofemoral OARSI JSN and MOAKS-CM worsening by baseline 3D JSW on WBCT compared to 2D JSW_x on radiography. We did not find statistically significant differences in the ability of these two modalities to predict worsening of radiographic and MRI-based knee structural worsening. WBCT has been shown to demonstrate significantly greater sensitivity and accuracy than radiographs for the identification of patellofemoral OA [28]. Additionally, previous studies have reported that WBCT provides JSW measurements that correlate more highly with MRI assessments of the presence and severity of medial tibial cartilage lesions than does minimum JSW by plain radiographs [29]. However, this finding reflects concurrent validity, which differs from the predictive validity examined in our study. As for previous reports on predictive validity, Wirth et al. reported that lateral and medial joint space narrowing predicts subsequent cartilage loss in the narrowed compartment but not in the non-narrowed femorotibial compartment [30]. In our study population, 75 % of the participants did not have radiographic knee OA, which may have contributed to the lack of statistically significant findings regarding predictive validity for structural worsening.

It is important to note that 3D imaging techniques such as WBCT provide insight into other disease-specific features while promising similar cost performance, relative radiation exposure and timeliness compared to radiographs [26]. Thus, WBCT may nonetheless demonstrate promise in the future as a routine diagnostic imaging technique for OA. The ability to more accurately follow disease progression is an important goal in both clinical care settings and future research trials. This consideration substantiates the need for imaging biomarkers that provide greater sensitivity and accuracy in the detection of OA-specific features and joint structural worsening. Such imaging biomarkers may allow for earlier detection of mild OA-related abnormalities, such as localized cartilage lesions or early cartilage matrix degeneration, that may go otherwise undetected by the standard radiographic measures [26].

Our study used MOAKS-CM as an MRI-based outcome to evaluate the performance of 3D and 2D JSW measurements for predicting knee structural worsening. This choice aligns with growing evidence that MRI-derived cartilage assessments may capture structural deterioration not visualized on radiographs. Wirth et al. reported that subregional cartilage thinning and thickening on MRI provided discrimination similar to that of radiographic JSW in predicting knee replacement [31]. In our study, both 3D and 2D JSW demonstrated comparable predictive validity for MOAKS-CM worsening, supporting the utility of joint space measurements in identifying knees at risk for early structural changes detectable on MRI. A limitation of this study included the use of a prototype WBCT machine built in 2013 for image acquisition. The image quality of the WBCT unit used in this study, was lower than that of modern WBCT scanners, which have significantly less artifacts and have higher spatial resolution (.37mm in the scanner used for this study and 0.15-0.30mm in current scanners). The pixel size of 0.10mm on radiographs and 0.37mm on WBCT could have provided an advantage for radiographic JSW measurement over those from WBCT. A secondary limitation is that JSW_.225 and JSW_.750 were selected for the medial and lateral compartments on radiographs. There is a possibility that assessing JSW at multiple additional points could have provided a more comprehensive evaluation. Third, due to the relatively low prevalence of structural worsening, the study may have been underpowered to detect modest differences in predictive validity between 2D and 3D JSW, making type II error a key limitation. Therefore, additional studies with larger samples and higher event counts are warranted. Finally, a limiting factor in this study concerns WBCT as a novel imaging approach and the lack of a defined method by which to measure 3D JSW. While we assessed multiple definitions for 3D JSW, it is possible that an alternative approach to defining 3D JSW may have provided different results. Future investigation to validate and standardize 3D JSW measurements on WBCT could advance understanding of WBCT predictive potential.

Despite these limitations, this study was conducted ancillary to MOST, which provided a sufficiently large sample size to allow for multivariable adjustment. Additionally, the imaging acquisition methods used in this study were more comprehensive and precise than those typically employed in routine clinical practice. Furthermore, participants were selected from the general population with age and gender distributions that closely match those of the U.S. population at risk for knee OA, enhancing the generalizability of the results. This study is the first to report the predictive validity of 3D JSW on WBCT in comparison with 2D JSW_x on radiographs to predict knee structural worsening using data from a large-scale epidemiological study.

In conclusion, 3D JSW, as defined in this study, was not found to be a superior to radiographic 2D JSW in predicting worsening OARSI JSN and MOAKS-CM. Although 3D JSW by WBCT performed similarly to that of 2D JSW by radiographs, WBCT may hold several advantages that could support its use as a standard imaging technique in the assessment of OA.

Supplementary Material

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ostima.2025.100389.

Abbreviations:

OA

osteoarthritis

JSW

joint space width

WBCT

weight-bearing computed tomography

MRI

magnetic resonance imaging

OARSI

Osteoarthritis Research Society International

JSN

joint space narrowing

MOAKS-CM

MRI Osteoarthritis Knee Score – Cartilage Morphology

KL grade

Kellgren-Lawrence grade

PDFS

proton density fat-suppressed

STIR

short tau inversion recovery

JSW_x

joint space width at a specific x-location

GEE

generalized estimating equations