Sensitivity to change and association of three-dimensional meniscal measures with radiographic joint space width loss in rapid clinical progression of knee osteoarthritis

The Osteoarthritis Initiative (OAI) is a multicentre, prospective observational cohort study of 4,796 subjects, designed to identify biomarkers and risk factors for KOA incidence and progression [18]. The OAI was approved by the institutional review board of the University of California, San Francisco, as well as each OAI Clinical Center, all patients provided informed consent [18]. OAI participants were 45-79 years of age at enrolment and had (or were at risk of) symptomatic KOA in at least one knee. Clinical data, 3Tesla MR images (Siemens Magnetom Trio, Erlangen, Germany; quadrature transmit-receive knee coils from USA Instruments, Aurora, OH, USA) and radiographs of the knee were acquired at annual visits [18‐20].

To be eligible as a case for the current study, total or unicompartmental medial KR had to be confirmed by radiography or by hospital records at the 36-month (M), 48M, or 60M follow-up visit. Further, 3T MR images for which quantitative cartilage analysis [21, 22] had been performed, had to be available for the annual visit before KR (T₀) and the annual visit two years before T₀ (T_-2) (Fig. 1). KRs detected at the 12M or 24M follow-up visit were not included, due to an insufficient longitudinal observation period before KR. If both knees of one participant were reported as replaced at the same, or different time points, both knees were included in the analyses. Only knees with baseline Kellgren-Lawrence grade (KLG) ≤2 were included to ensure rapid and clinically relevant progression of KOA [21, 22]. Knees with lateral Osteoarthritis Society International JSN grades >0 were excluded from the analyses, since lateral compartment JSN increases the risk of lateral progression, potentially masking progression in the medial compartment [12].

Weight-bearing, posterior-anterior fixed-flexion (10°) radiographs were acquired from OAI participants at each annual visit [18, 23]. Semi-automated radiographic JSW measurements were performed for the majority of these knees [9, 10] (Fig. 2). In the current study, we analysed medial compartment mJSW and medial compartment fixed-location JSW at 22.5% of the mediolateral width of the distal femur [medial fixed-location (medf)JSW, Fig. 2] [9] as this measure has been previously shown to display greater sensitivity to change in KOA [10, 24] and a stronger relationship with subsequent KR than mJSW [2].

MR images from a sagittal double echo steady state sequence with water excitation (DESSwe, 0.7mm slice thickness; interpolated in-plane resolution 0.37mm × 0.37mm) [19] were used for segmenting the femorotibial cartilages at the T_-2 and T₀ visits [22] (Fig. 3a). Total subchondral bone area and the cartilage joint surface area of the medial tibia (MT) and the central (weight-bearing) medial femoral condyle (cMF) were manually traced by experienced readers, as described previously [22]. The analysis, conducted with blinding to acquisition order, relied on custom software (Chondrometrics GmbH, Ainring, Germany). Quality control readings were performed by an expert reader. The mean cartilage thickness was computed for MT and cMF, for the total medial femorotibial compartment (MFTC=MT+cMF), and for a combined central femorotibial subregion (cMFTC) (Fig. 3b) [25, 26].

Coronal multi-planar reconstructions (1.5mm slice thickness; interpolated in-plane resolution 0.37mm × 0.37mm) were derived from the near isotropic sagittal DESSwe MR images. The medial tibial plateau area (i.e. the area of cartilage surface including denuded areas of subchondral bone) and the surfaces of the medial meniscus (tibial, femoral and external area) were segmented at T_-2 and T₀ from the coronally reconstructed images obtained, which were previously used for the cartilage thickness measurements (with blinding to acquisition order, Fig. 4a). The segmentations were performed by a PhD (M.R.), who was first formally trained in quantitative meniscus analysis; all segmentations done in this study were quality controlled by an expert reader (K.E.) with >5 years of experience in quantitative meniscus analysis. Each case (two time points, two menisci) required approximately 2.5 h, including quality control readings and potential corrections following quality control. Measures of meniscal morphology included mean and maximal meniscal thickness and width, and meniscal volume (Fig. 4b-c) [13]. Measures of meniscal position relative to the tibial plateau encompassed the tibial plateau area covered by the meniscus (absolute/percent), the tibial meniscal surface area not covering the tibial plateau (absolute/percent), mean and maximal extrusion distance and overlap distance between the meniscus and the medial tibial plateau area (Fig. 4b) [13].

Paired t-tests were used to determine whether significant changes in JSW, cartilage and meniscal parameters occurred between T_-2 and T₀. The bootstrap method [27] (1000 replications, BCa method, simple sampling), that renders P values relatively robust against non-normally distributed data, was applied to use the same statistical approach for all variables, even in the case of non-normal distribution. The Wilcoxon signed-rank test was additionally used to confirm the results for non-normally distributed variables. Sensitivity to change of the parameters was assessed using the standardized response mean (SRM=mean change/standard deviation of change).

Multiple linear regression analyses were performed to examine the association of meniscal and cartilage measures with JSW change based on a hierarchical approach, the variables being predefined manually by the study team: Step 1 forced cMFTC cartilage thickness (Fig. 3b) into the model, because this cartilage parameter has previously displayed high sensitivity to change in medial KOA [21]. Please note that in correlation analyses cMFTC cartilage thickness also displayed the strongest coefficient between JSW change amongst the cartilage measures (Table 1). Step 2 allowed the model to include one positional (tibial meniscal surface area not covering the tibial plateau; Fig. 4b) and two morphological (maximal thickness, mean width; Fig. 4b-c) meniscal measures in a stepwise fashion. The specific meniscal measures were chosen based on their high correlation with JSW parameters and their responsiveness (SRM), and so that different meniscal properties were represented in the model. In step 3, the model was allowed to include age (at T_-2) and body weight change (from T_-2 to T₀) in a stepwise manner to compensate for possible confounding. The Akaike information criterion (AIC) permitted comparison of the models at the different stages. For statistical comparison of the model steps, analysis-of-variance was performed. For exploratory purposes the models were also calculated based on a second approach, which exchanged step 1 and 2 in their order. p values of < 0.05 were considered significant, no correction was performed for multiple testing due to the exploratory nature of this study.

All statistical analyses, except for the AIC (Stata V14.2, StataCorp, TX, USA), were done using SPSS 23 (IBM Corporation, Armonk, NY, USA).

Thirty-five knees of 33 OAI participants (23 women; baseline age 64.7±7.1, body mass index 30.0±4.0 kg/m², KLG0/1/2: 5/8/22, 33 total/two medial unicompartmental KRs) received a KR between the 36M and 60M follow-up visit (13 at 36M, seven at 48M, and 15 at 60M; Fig. 1). Descriptive statistics are presented in Table 1.

All meniscal measures, except for meniscal volume and mean thickness, showed significant changes over the two-year observation period before KR. The position of the medial meniscus relative to the tibial plateau changed significantly, with the positional parameters showing a loss in coverage of the tibial plateau by the meniscus. Meniscal width displayed a reduction over time, whereas maximal meniscal thickness increased (Table 1). For non-normally distributed parameters the Wilcoxon signed-rank test displayed the same results (data not shown).

The percent area of the tibial plateau covered by the meniscus and the mean overlap distance were observed to be the most sensitive meniscal measures (SRM =  − 0.76/0.75). The majority of the positional measures reached an SRM ≥ |0.62|, which was the lowest absolute SRM observed for cartilage measures. Somewhat lower absolute SRMs were observed for morphological meniscal measures. As a reference, the SRM of mJSW change was −0.79 and −1.00 for medfJSW change. The most responsive cartilage measure was MT cartilage thickness change (SRM = 0.83 ) (Table 1).

The correlation coefficients of mJSW change with change in meniscal position were generally greater than those with change in meniscal morphology (Table 1). Among meniscal measures, the highest correlation with mJSW change was observed for change in tibial meniscal surface area not covering the tibial plateau (r=-0.66; p<0.01); this association was similar in magnitude to that between mJSW change and change in cartilage thickness (0.63≤r≤0.70; p<0.01) (Table 1).

For medfJSW, the correlations with change in meniscal measures were observed to be similar for positional and morphological parameters (Table 1). In contrast to mJSW, the strongest correlation among the meniscal parameters with medfJSW change was noted for change in maximal meniscal thickness (r=-0.62; p<0.01). The correlations of medfJSW change with change in cartilage thickness measures were observed to be somewhat greater (0.64≤r≤0.73; p<0.01) than those of medfJSW change with change in meniscal measures (Table 1).

In the multiple regression models, 47% of the variance in 2-year mJSW change before KR was explained by change in cMFTC cartilage thickness. Change in meniscal position relative to the tibial plateau added 9% to the explained variance, and baseline age another 5%. This resulted in a total of 61% of explained variance in mJSW change (Table 2).

Change in cMFTC cartilage thickness explained 52% of the variance in medfJSW change. An additional 20% of the variance of medfJSW change was explained by meniscal morphology changes (thickness, width), amounting to a total of 72% variance explained. Unlike for mJSW, the regression model did not include a positional meniscal or demographic measure (Table 2).

Explorative regression analyses, which first included meniscal measures and then cartilage thickness, yielded the same explanatory parameters and the same total variance explained for mJSW (61% = 41% meniscal position change + 15% cMFTC cartilage thickness change + 5% age), as well as for medfJSW (72% = 48% meniscal morphology changes + 24% cMFTC cartilage thickness change).