APPROACHES TO OPTIMIZE ANALYSES OF MULTIDIMENSIONAL ORDINAL MRI DATA IN OSTEOARTHRITIS RESEARCH AND CLINICAL TRIALS

INTRODUCTION

Knee OA is a disease of the whole joint involving multiple tissues. MRI-based semi-quantitative (SQ) scoring of knee OA is a method based on ordinal grading to perform multi-tissue joint assessment. SQ scoring is used to measure severity of structural disease on a tissue level and allows evaluation of disease progression. While guidance is available to describe how SQ scoring may be applied and can be used for clinical trial enrichment, less information is available on how these parameters should be used to assess outcomes in research and clinical trial contexts.

OBJECTIVE

Here we describe how SQ scoring can optimally be used to quantify longitudinal change in knee OA and highlight its potential as an outcome measure in research and clinical trials.

METHODS

The two most widely used SQ scoring systems for knee OA, MOAKS and WORMS, rely on standard MRI acquisitions (usually intermediate-weighted fat-suppressed sequences in three orthogonal planes) and ordinal ratings of knee features by expert readers. Key pathoanatomic features of the joint are assessed, including cartilage damage (both in surface area extent and in full-thickness loss), meniscus damage, osteophytes, BM lesions, synovitis, and others. The knee joint is divided into subregions (SRs) (e.g., MOAKS scores cartilage damage across 14 SRs) or locations (e.g., osteophytes) and each SR or location is scored for a given feature.

RESULTS

The following approaches may be considered to assess longitudinal change. Worsening across SRs is quantified by the count of the number of SRs with a worse (higher) score at follow-up vs. baseline and by the change in the number of SRs affected (score=0 at baseline and >0 at follow-up). Improvement across SRs is quantified as the number of SRs with improvement from baseline to follow-up (i.e., lower score at follow-up vs. baseline). The delta-SR approach considers worsening and improvement simultaneously and is calculated as the number of SRs with worsening minus the number of SRs with improvement (particularly relevant for fluctuating features such as BM lesions). The delta-sum approach considers the ordinal score in each SR: the sum of ordinal scores across all SRs is computed and change is quantified by the difference in total score. Finally, maximum grade change is the maximum change across all SRs. Within-grade changes are changes that do not fulfill the definition of a full-grade change but do represent definite SQ visual change. Including such changes in SQ assessment of longitudinal change increases sensitivity to change. Examples are shown in Table 1.

The various approaches to quantifying longitudinal change may result in variables that are counts, ordered categories, binary categories, or continuous parameters. Count data may be analyzed with Poisson regression, binary data with log-binomial or logistic regression, and continuous data with linear regression. Special attention must be paid when analyzing repeated measures or clustered data, random-effect, mixed-effect, or marginal models may be considered when analyzing clustered data, for example, when conducting analyses at the compartment level vs. whole knee level, or when including one vs. two knees per participant. Reliability needs to be determined using different measures depending on the nature of data and number of readers (e.g. weighted kappa, ICC etc).

CONCLUSION

To date no disease-modifying therapies have been approved for knee OA. SQ imaging assessment of longitudinal change provides an opportunity to better understand disease progression across multiple tissue types, which may allow future trials to match outcome to patient phenotype or to treatment mechanism of action. In addition, safety signals can be assessed that are not necessarily visualized using dedicated (e.g. cartilage-focused) imaging protocols.