Determining individual trajectories of joint space loss: improved statistical methods for monitoring knee osteoarthritis disease progression.

Summary Objectives Knee osteoarthritis (KOA) progression is frequently monitored by calculating the change in knee joint space width (JSW) measurements. Such differences are small and sensitive to measurement error. We aimed to assess the utility of two alternative statistical modelling methods for monitoring KOA. Material and methods We used JSW on radiographs from both the control arm of the Strontium Ranelate Efficacy in Knee Osteoarthritis trial (SEKOIA), a 3-year multicentre, double-blind, placebo-controlled phase three trial, and the Osteoarthritis Initiative (OAI), an open-access longitudinal dataset from the USA comprising participants followed over 8 years. Individual estimates of annualised change obtained from frequentist linear mixed effect (LME) and Bayesian hierarchical modelling, were compared with annualised crude change, and the association of these parameters with change in WOMAC pain was examined. Results Mean annualised JSW changes were comparable for all estimates, a reduction of around 0.14 mm/y in SEKOIA and 0.08 mm/y in OAI. The standard deviation (SD) of change estimates was lower with LME and Bayesian modelling than crude change (SEKOIA SD = 0.12, 0.12 and 0.21 respectively; OAI SD = 0.08, 0.08 and 0.11 respectively). Estimates from LME and Bayesian modelling were statistically significant predictors of change in pain in SEKOIA (LME P-value = 0.04, Bayes P-value = 0.04), while crude change did not predict change in pain (P-value = 0.10). Conclusions Implementation of LME or Bayesian modelling in clinical trials and epidemiological studies, would reduce sample sizes by enabling all study participants to be included in analysis regardless of incomplete follow up, and precision of change estimates would improve. They provide increased power to detect associations with other measures.