Development of a population-anchored Z-score for MRI-based knee osteoarthritis disease activity: Data from the Osteoarthritis Initiative

Objective: We previously developed a novel imaging biomarker, reflecting whole-knee bone marrow lesion and effusion-synovitis volumes, to measure disease activity. This study describes the process used to develop a z-score and its percentile to improve interpretability and comparability across studies. Design: We used data from several Osteoarthritis Initiative projects (n = 1084 participants with 6876 observations). For 76% of observations, participants had radiographic knee osteoarthritis (Kellgren-Lawrence grade (KL) ≥ 2). We targeted a synthetic reference population with an overall prevalence of 25% for radiographic knee osteoarthritis. A Box-Cox transformation was used to derive a standard normal distribution for a composite metric. A within-cluster resampling method was used to summarize the reference population and estimate model parameters. The new z-score was associated with categories of WOMAC knee pain to evaluate construct validity (cross-sectionally and longitudinally). Results: The probability of being a normal distribution for a Box-Cox transformed composite metric was 99.5%, and the probability that the newly developed disease activity z-score follows the standard normal distribution was 95.7%. Because increasing age was associated with an upward shift in the mean z-score distribution, we also developed an age-adjusted z-score. We found 94% consistency between the new z-score and the initial composite metric of disease activity and construct validity with WOMAC knee pain. Conclusion: The unadjusted and age-adjusted z-score for disease activity developed using a synthetic reference population improves both the statistical and clinical usefulness of the disease activity metric. This new composite metric improves comparability across studies, clinical interpretability, and statistical modeling utility.