Purpose: Understanding patterns of symptomatic and structural progression has been a major research focus in knee osteoarthritis (OA). Latent class growth analysis is a statistical approach that aims to identify clusters of subjects with similar progression trajectories. This approach has been used recently in knee OA to uncover longitudinal trajectories of both structural (e.g., quantitative MRI, joint space width (JSW), bone shape) and symptom (e.g., WOMAC Pain) progression. In longitudinal studies, it is common for subjects to drop out before the scheduled end of follow-up. Previous studies have shown that subjects undergoing total knee replacement (TKR) progress more quickly in the years leading up to TKR than subjects not undergoing the procedure. By not accounting for dropout due to TKR in estimating disease trajectories, we may fail to identify a trajectory of rapid progressors. Our objective was to evaluate the impact of incorporating TKR on estimates of structure (JSW) and symptom (WOMAC Pain) trajectories in knee OA. Methods: We used data from the Osteoarthritis Initiative (OAI), a multicenter, longitudinal, observational study of knee OA. We selected knees with radiographic, symptomatic OA at baseline (KL 2+, WOMAC Pain greater than 0). Eligible subjects had at least one follow-up visit. To assess joint structure we used serial measures of fixed JSW at x=0.25 in the medial tibiofemoral compartment, measured annually from baseline to year 4, and at years 6 and 8. WOMAC Pain was used to assess symptoms (0 - 100, 100 worst), assessed annually through year 9. After selecting patients with baseline pain, we modeled trajectories starting at year 1 to avoid modeling regression to the mean. We used latent class growth analysis (LGCA) to identify distinct subgroups, separately for JSW and pain. While LCGA can handle missing data under the missing at random assumption, these models on their own do not accommodate informative dropout. Thus, we used an extension of the LCGA framework to jointly model longitudinal outcome and time to TKR using joint latent class mixed models. We used the BIC, posterior group membership probability, and sufficient sample size per group (n=50) to select the final models. Results: We used data from 1,578 subjects with radiographic, symptomatic knee OA at baseline. 157 (10%) of subjects underwent TKR, with median time from baseline to TKR 5.5 years. On average, subjects undergoing TKR increased in pain by 2.3 points per year (19 points over 8 years) while subjects not undergoing TKR increased by 0.21 points per year (1.7 points over 8 years). Subjects undergoing TKR decreased in JSW by 0.31mm per year (2.5mm over 8 years) while subjects not undergoing TKR decreased in JSW by 0.09mm per year (0.75mm over 8 years). We found three distinct JSW trajectories. In our final model the majority (84%) of the cohort was in a stable trajectory, 5.9% of the cohort was in a late progression trajectory, and 8.4% of the cohort was an early progression trajectory (Figure 1). While the number and shape of trajectories was largely unchanged when we moved from the initial JSW model to the final joint JSW and TKR model, 32% of TKRs were re-classified, most from the stable to a progressing trajectory. In our initial model of pain trajectories we found 4 distinct trajectories: stable low pain (70%), rapidly increasing pain (13%), rapidly decreasing pain (5%) and increasing high pain (12%) (Figure 2a). After incorporating TKR our model identified a fifth trajectory. The increasing high pain trajectory (year 1 WOMAC pain 40) was split into a moderate increasing pain trajectory (year 1 WOMAC pain 28) and a high pain trajectory (year 1 WOMAC pain 52) (Figure 2b). 68% of TKRs were re-classified, most from the stable to increasing pain trajectory, stable to moderate pain, or the high to moderate pain. In the final model, only 2% of TKRs are in the stable pain trajectory, compared to 42% in the initial model. Conclusions: In longitudinal studies restricting analysis to completers may lead to underestimation of disease progression. In analyses seeking to uncover latent subgroups this could result in subjects being misclassified or entire trajectories being missed. Investigators studying disease progression in OA should consider the potential impact of dropout due to TKR, particularly when a large proportion of subjects undergo the procedure.View Large Image Figure ViewerDownload Hi-res image Download (PPT)View Large Image Figure ViewerDownload Hi-res image Download (PPT)