Repeatability and discrimination validity of cartilage imaging biomarkers for experimental medicine studies of knee osteoarthritis

Abstract

PURPOSE The utility of magnetic resonance imaging (MRI) quantitative imaging biomarkers (QIBs) of cartilage in experimental medicine studies of potential disease modifying osteoarthritis treatments (DMOATs) is uncertain due to the small sample sizes and short follow-up involved. To be relevant to such studies, candidate QIBs must demonstrate acceptable repeatability, discriminative ability and responsiveness. The purpose of this study was to evaluate test-retest repeatability and discriminative ability of 4 cartilage QIBs at the knee (thickness, T1rho, T2 and delayed gadolinium enhanced MRI of cartilage [dGEMRIC]) to assess their utility in experimental medicine studies. METHODS We imaged 9 participants with mild-moderate knee osteoarthritis (OA), characterised by radiographs with medial tibiofemoral predominant disease and Kellgren-Lawrence grades 2-3, and 4 healthy volunteers (HVs) matched for age, sex and body mass index. Participants were imaged at baseline and 1 month. MR studies were performed on a 3T system (MR 750, GE Healthcare) and were split into 2 sessions. The first MR session included a 3-dimensional spoiled gradient echo (3D SPGR) sequence and T1rho and T2 mapping sequences. At the end of this session, we administered an intravenous gadolinium based contrast agent (Dotarem, Guerbet LLC) at double dose (0.2 mmol/kg). Participants then performed 10 minutes of exercise on a stationary cycle to facilitate contrast penetration into the joint for dGEMRIC. The second MR session began 90 minutes following injection of contrast agent and consisted of T1 mapping using a variable flip angle technique. MR pulse sequence details are provided in table 1. We performed surface based analysis of cartilage parameters using Stradwin software (http://mi.eng.cam.ac.uk/~rwp/stradwin). In brief, this consists of semi-automated thickness measurement on 3D SPGR images, with generation of accurate inner and outer cartilage surfaces. Following rigid registration of the compositional (T1rho, T2, dGEMRIC) data to the 3D SPGR images, the same surfaces were used to sample the compositional data. This analysis pipeline allows each parameter to be measured at ~5,000 surface vertices per participant. We assessed overall (pooled OA and HV data) test-retest repeatability by calculating the root-mean-square coefficient of variation (RMS-CV) and the smallest detectable change (SDC) for each parameter. We assessed overall discriminative validity by estimating effect sizes for the difference in the mean value of each parameter in OA participants and HVs. We also mapped the test-retest errors and mean difference in each parameter between groups onto a representative surface for each cartilage region. RESULTS Participant characteristics are provided in table 2. RMS-CVs, SDCs and effect size estimates are provided in table 3. Overall test-retest RMS-CV values ranged from 2.5% (T1rho) to 10.3% (dGEMRIC). The largest overall effect size was demonstrated for dGEMRIC (-1.3, 90% confidence interval [CI] -2.3 to -0.2), followed by thickness (-0.67, 90% CI -1.5 to 0.3), T1rho and T2 (both 0.61, 90% CI -0.4 to 1.5). Mapping of test-retest errors and between-group differences to a representative surface revealed substantial spatial heterogeneity in these metrics for all parameters (figure 1). This suggests that surface based approaches may offer improved sensitivity over traditional region of interest based measurements. CONCLUSION This study demonstrates that cartilage QIBs may be useful in experimental medicine studies of knee osteoarthritis. Statistical approaches reflecting the spatial heterogeneity of changes in cartilage should be used to provide an indication of where significant results are located, not just whether an average value is significantly different. Future work will evaluate responsiveness of these cartilage QIBs.