Methods

A total of 180 women, age ⩾40 years, were recruited at 7 clinical centres to participate in this non-interventional, observational method study (Pfizer trial registration number A9001140).3 23 In all, 35 participants were not included in the analysis because not all visits were completed (n = 24), because of protocol violations or MRI artifacts (n = 5), because radiographs were missing (n = 2), or because the Kellgren–Lawrence Grade (KLG) was grade 1 in the weight-bearing extended anterior–posterior (AP) radiographs (n = 4; see below). In total, 145 subjects had complete MRI scans at baseline, 6, 12 and 24 months, and weight-bearing LS and FF radiographs at baseline, 12 and 24 months.

Inclusion criteria for the OA participants were frequent knee pain, KLG2 or KLG3 radiographic OA, severity of medial JSN greater than (or equal to that) in the lateral compartment, BMI ⩾30 and medial femorotibial minimum joint space width (mJSW) ⩾2 mm in LS radiographs22 23 (table 1). The controls (healthy reference population) were asymptomatic with no evidence of radiographic OA (KLG0) in either knee, had a BMI ⩽28, and were age-matched to within 5 years of the OA participants (table 1); the dominant leg was selected for analysis. Subjects with a history of intra-articular fracture, arthroplasty, meniscectomy, crystalline diseases, knee infection, and avascular necrosis were not included. While cases with anterior cruciate ligament tears were not excluded, a review revealed no cases of injury and/or reconstruction.

An experienced central reader reassessed KLG after enrolment was completed (SAM), and differences between readers were adjudicated by a third reader (KB).3 Of the 145 participants, 86 belonged to the reference population (KLG0) and 55 to the OA population (27 KLG2, 28 KLG3, table 1). The study was conducted in compliance with the ethical principles derived from the Declaration of Helsinki and with the local Institutional Review Board, informed consent regulations and International Conference on Harmonization Good Clinical Practices Guidelines.

Three of the seven sites used Magnetom Trio magnets (Siemens AG, Erlangen, Germany), and four Signa Excite magnets (GE Healthcare Technologies, Waukesha, Wisconsin, USA).3 Previously validated16 27 double oblique coronal spoiled gradient recalled acquisition at steady state (SPGR) sequences with selective water excitation (we), 1 mm slice thickness and 0.31 mm in-plane resolution were acquired as described previously (fig 1).3 16 Analysis of the femorotibial cartilages was performed by seven technicians with thorough experience in cartilage segmentation, using proprietary software (Chondrometrics GmbH, Ainring, Germany).3 The bone interface and the cartilage surface of the medial tibia (MT), lateral tibia (LT), central (weight-bearing) medial femoral condyle (cMF) and lateral femoral condyle (cLF; fig 1) were traced manually.3 28 Images were read in pairs (6, 12 and 24 months versus baseline, respectively) with blinding of the readers to the order of acquisition. Quality control of all segmentations was performed by a single reader (FE). The mean cartilage thickness over the total area of subchondral bone (ThCtAB)3 29 was computed in five tibial (central, external, internal, anterior and posterior)20 and three femoral subregions (central, external, internal20; fig 2), as described previously.20 21 The central tibial subregions (cMT, cLT) occupied 20%, and the central femoral subregions (ccMF and ccLF) 33% of the total subchondral bone area20 21 (fig 2). The cartilage thickness in MT and cMF, LT and cLF, cMT and ccMF and cLT and ccLF was added, in order to obtain measures of mean cartilage thickness in the medial and lateral femorotibial compartment (MFTC and LFTC) and in the central MFTC and LFTC (cMFTC, cLFTC), respectively20 30 (figs 1 and 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Double oblique coronal spoiled gradient recalled acquisition at steady state (SPGR) sequence with selective water excitation (we), 1 mm slice thickness and 0.31 mm in-plane resolution. cLF, central weight-bearing lateral femoral condyle; cMF, central weight-bearing medial femoral condyle; LFTC, lateral femorotibial compartment (LT and cLF added); LT, lateral tibia; MFTC , medial femorotibial compartment (MT and cMF added); MT, medial tibia.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Images showing the femorotibial subregions analysed using quantitative MRI. (A) Posterior view of femorotibial subchondral bone areas (tibia at the bottom, weight-bearing femur at the top), with subregions displayed by different grey values. (B) Superior view of the tibial subchondral bone area, with subregions labelled. (C) Inferior view of the femoral subchondral bone area, with subregions labelled. aMT, anterior MT; ccMF, central cMF; cMF, central (weight-bearing) medial femoral condyle; cMT, central MT; ecMF, external cMF; eMT, external MT; icMF, internal cMF; iMT, internal MT; MT, medial tibia; pMT, posterior MT.

FF and LS views22 23 31 were acquired at baseline, 12 months and 24 months. Independent quantitative measurements of the mJSW were obtained in the medial femorotibial compartment of all radiographs by an experienced observer, using digitised image analysis software (Holy’s software, UCLB, Lyon, France).23 32 The qMRI and radiographic data were used to compute the annualised change in mJSW and in ThCtAB (in mm), negative changes indicating cartilage thinning (or JSN) and positive changes, cartilage thickening (or joint space widening). The mean and standard deviation (SD) of the change per year, and the standardised response means (SRM  =  mean/SD of change) were computed for each anatomical subregion, KLG grade and time point. Percentage mean change was expressed in relation to the mean baseline value observed across all subjects. Statistical significance of the change between baseline and follow-up was evaluated by a paired t test. Differences in changes (between baseline and follow-up) of qMRI and radiographic end points in the participants with radiographic OA and the non-arthritic controls were evaluated using mixed effects models, with “KLG” being treated as fixed effect, “participants” as random effect and each KLG being allowed to have its own variance.

Results

Follow-up at 6 months

An −0.8% (annualised) change in cartilage thickness (ThCtAB) by qMRI was observed in the total medial femorotibial compartment (MFTC) of the control group (SRM  = −0.18) at 6 months (table 2) and a −1.8% change (SRM  = −0.31; p<0.01 between baseline and follow-up) in the central MFTC (cMFTC). In the KLG3 participants, the reduction was somewhat higher in MFTC (−1.2%; SRM  = −0.14) and cMFTC (−3.7%; SRM  = −0.33), but the changes did not differ significantly from that in the control group. Given the smaller number of KLG3 subjects than KLG0 controls, the change between baseline and follow-up was not significant (table 2). In the KLG2 participants, a 2.2% (annualised) increase in ThCtAB was observed in MFTC and a 2.7% increase in cMFTC (table 2).

In the control group, the greatest reduction in cartilage thickness among cartilage plates and subregions was observed in the cMT and ccMF (table 2). In the KLG3 participants, the greatest changes were found in cMT (−3.4%; SRM −0.29), eMT (−3.9%; SRM −0.19) and ccMF (−4.0%; SRM −0.30). In the KLG2 subjects, the greatest increase in ThCtAB among MT subregions was observed in cMT and among the cMF subregions in icMF (table 2).

In the LFTC, ThCtAB tended to increase (range +1.0 to +2.0% per annum) in all participants (table 2). Among cartilage plates and subregions (table 2), the lateral weight-bearing femur (cLF) showed a greater increase than the LT. In KLG2 participants, the increase in ThCtAB in cLF (+5.6%; p<0.01) was contrasted by a decrease in ThCtAB of LT (−1.1%), in particular in iLT (−4.0%; p<0.001).

Follow-up at 12 months

Over 12 months, no relevant changes (⩽0.2%) in ThCtAB of MFTC or mJSW were detected in the controls using qMRI or in FF or LS radiographs (table 3). In the KLG3 participants, the SRMs for qMRI were similar to those in the 6-month follow-up, but the annualised rates of change were lower (−0.9% for MFTC and −2.0% for cMFTC; table 3). The change in mJSW in the LS radiographs (−7.4%; SRM −0.68; p<0.01) exceeded that of ThCtAB in qMRI (table 3) and was significantly different (p<0.001) from that in healthy controls. The FF radiographs, in contrast, did not show a significant change in mJSW between baseline and follow-up and the change was smaller than in qMRI (−1.3%; SRM −0.13 (table 3). In the KLG2 participants, a small (but non-significant) reduction in ThCtAB (−0.7%; SRM −0.14) was observed for cMFTC with qMRI, and some reduction of mJSW in LS radiographs (−2.2%; SRM −0.30). In contrast, a small increase in mJSW was noted in the FF radiographs (+0.1%; SRM +0.17) (table 3). Among plates and subregions, the greatest rate of cartilage thinning in KLG3 participants (qMRI) was observed in eMT (−2.4%; SRM −0.23), ccMF (−3.7%; SRM −0.39) and ecMF (−2.4%; SRM −0.24) (table 3).

In LFTC, ThCtAB tended to increase in all participants (except for cLFTC in KLG2 participants), but the annualised rates were lower than at 6 months (⩽+0.8%; table 3). As at 6 months, consistent increases in ThCtAB of cLF and its subregions were contrasted by only small changes in LT, except for a significant increase in eLT of KLG2 participants (table 3).

Follow-up at 24 months

Over 24 months, only small annualised changes (⩽0.5%) in ThCtAB of the MFTC and mJSW were found in healthy controls with qMRI, LS and FF radiographs, the qMRI changes reaching statistical significance in the medial compartment (MFTC: p<0.05; SRM −0.25; cMFTC: p<0.01; SRM −0.32) (table 4). In the KLG3 participants, the SRMs for qMRI were higher at 24 than at 6 and 12 months, and the changes between the baseline and follow-up images reached statistical significance (cMFTC: p<0.05, −2.2%, SRM −0.48; table 4). The annualised rate of change (−5.6%; p<0.01) and SRM (−0.62) of the LS radiographs was lower than at 12 months, but was still higher than for qMRI (table 4). The FF radiographs, in contrast, found only small changes (−1.7%; SRM  = −0.20), which were not statistically significant. In the KLG2 participants, a small change in ThCtAB (cMFTC: −0.3%; SRM −0.15) was observed with qMRI, and a somewhat greater change of mJSW with LS radiography (−1.9%; SRM −0.32), but neither was significant. The FF radiographs also showed no statistically significant change between baseline and follow-up (+0.7%; SRM +0.11) (table 4). Among cartilage plates and subregions, the greatest annualised rate of change in the KLG3 participants (qMRI) was observed in eMT (−3.4%; SRM −0.57; p<0.01) and cMT (−2.0%; SRM −0.53; p<0.01), the changes in eMT and MT significantly exceeding those in control participants (table 4). The changes in cMF and its subregions (up to −3.0% for ecMF; SRM −0.34) did not reach statistical significance (between baseline and follow-up) and were not significantly different from controls (table 4).

In contrast to the data from 6 and 12 months, no trend towards an increased cartilage thickness of LFTC was observed at 24 months, but small decreases in controls and KLG2 participants (table 4). Small increases (<2.0%) in ThCtAB of cLF and its subregions (consistent with findings at 6 and 12 months, tables 2–4) were contrasted by a mild decrease in LT of KLG3 participants, and by a significant decrease of ThCtAB in LT (−1.1%; SRM −0.57; p<0.01) and its subregions in KLG2 participants (table 4).

Discussion

In this study, qMRI and LS radiographs detected a significant reduction in medial cartilage thickness/medial minimum JSN over time in participants with symptomatic KLG3 OA, but no significant reduction in participants with symptomatic KLG2 OA and only small changes in the controls. The SRMs observed with LS radiographs (−0.62) at 24 months exceeded those of qMRI in the central medial femorotibial compartment (−0.48) and in other subregions of the medial tibia and weight-bearing medial femur (up to −0.57). FF radiographs, in contrast, did not identify a significant reduction in mJSW over 24 months (SRM −0.20). In the lateral femorotibial compartment, qMRI revealed an increase of cartilage thickness across all participants (including controls) at 6 months (particularly in the femur), but at later time points the increase in femoral cartilage thickness tended to be accompanied by a decrease in lateral tibial cartilage thickness.

A limitation of this study is the relatively small number of participants with radiographic OA, particularly in comparison with the relatively large number of regions of interest tested. Results on the relative performance of the different subregions and outcome measures should therefore be viewed as descriptive, providing a basis for power calculations for future trials. A strength of the study, however, is the relatively large control group and the relatively uniform inclusion criteria of the participants with radiographic OA (women with BMI >30, medial mJSW ⩾2 mm and medial ⩽ lateral JSW). The focus on obese women in this study was driven by the desire to maximise risk of progression in this relatively small cohort but not to marginalise men and the non-obese, as female sex and obesity have been reported to be risk factors of OA progression.33 34 35 As such, the homogeneity of the sample may also represent a limitation with respect to generalisability, but we expect these results to generalise, in principle, to the OA population as a whole.

In contrast with the KLG3 participants, those with KLG2 did not display significant change over time in either radiography or qMRI. A reason for this may be that these participants were still at an early stage of the disease, when cartilage swelling/hypertrophy may occur, as observed in animal models of OA.36 37 38 39 40 This is also supported by the finding that, in the same cohort,23 JSW at baseline was significantly greater in obese KLG2 subjects than in non-arthritic controls. The rate of change and SRM in medial femorotibial cartilage thickness in 3.0 T qMRI in this study was similar to that observed over 12 months in participants of the Osteoarthritis Initiative (OAI) progression subcohort studied also with 3.0 T qMRI.17 18 In contrast to the OAI data, where the medial femur displayed a higher SRM than the medial tibia, this was not the case (at 24 months) in the current study. Some studies that were previously performed at 1.5 T reported higher rates of change (and higher SRMs) than the current study,4 6 7 8 9 but others found smaller rates and SRMs.5 10

The higher SRMs for LS radiographs compared with 3.0 T qMRI may appear surprising; however, there are several potential explanations: (a) previous9 21 and the current qMRI study revealed that the most prominent changes occur in the central and external aspects of the medial femorotibial compartment, locations similar to where minimum medial radiographic JSW is usually measured in LS views; (b) in contrast to the MRIs the knee x rays were acquired under weight-bearing conditions, where the mechanical compression of cartilage may have been of relevance. If cartilage swelling or hypertrophy occurs and the matrix composition and biomechanical properties are compromised,36 37 38 39 40 the cartilage may be thickened in qMRI, but may be compressed in weight-bearing x rays; (c) x rays display a very sharp contrast at the bone cartilage interface and have a higher in-plane resolution (0.18 mm on average in the present study, range 0.12–0.30 mm) than qMRI (here 0.31 mm); d) mJSW in x rays is dependent on cartilage thickness and on meniscal extrusion/subluxation.41 42 43 Worsening of meniscal pathology with progression of OA may thus affect mJSW, but not cartilage thickness measured with qMRI. Note, however, that the x ray beam needs to be appropriately aligned with the medial tibial plateau to be able to measure change,44 as achieved here with the LS technique.22 23 Although FF radiographs yield reproducible data on mJSW,24 25 no significant reduction in mJSW in this study was observed using this x ray protocol. A previous report found FF radiographs to be sensitive to change over 24 months (SRM −0.52), when the tibial plateau in the baseline and follow-up acquisition were appropriately aligned,26 but this was only the case in 42% of the film pairs.26 The current data confirm the relative performance of LS and FF radiographs over 24 months, as previously reported based on 12-month data from the same study.23 An SRM of −0.68 was also reported in another 12-month study using the LS technique,45 in subjects in whom medial tibial plateau alignment was reproducible. LS was also superior to conventional extended weight-bearing AP radiographs46 in detecting JSN.

Despite the apparent superiority of LS radiography over 3.0 T qMRI in detecting medial femorotibial change over time in this study, it should be noted that: (a) these results are specific to LS radiographs, but not to other radiographic techniques frequently used in clinical trials; (b) reproducible and accurate radioanatomic positioning of the subject remains a challenge in multicentre clinical trials, including those that employ the LS technique; (c) qMRI can reveal the spatial pattern of cartilage change in the medial femorotibial compartment and thus provide insights into the pathophysiology of the disease that radiography may not; (d) qMRI can additionally provide quantitative data for the lateral femorotibial compartment or other compartments of the knee that are less accessible to radiography; and (e) the results may be specific to this cohort of women with BMI >30, mJSW >2 mm and medial ⩽ lateral mJSW at baseline, but may not necessarily apply to other cohorts with more advanced disease.

In conclusion, qMRI and LS radiographs detected significant change in joint space width/cartilage thickness over time in symptomatic subjects with KLG3 that were at high risk of progression, but not in KLG2 participants, and only small changes in healthy controls. At 12 and 24 months, LS displayed greater, and FF radiography less, sensitivity to change in KLG3 participants than qMRI.