Background/Purpose: Establishing a threshold of clinically important cartilage loss is of great interest for the design and interpretation of knee OA trials of structure-modifying treatments. Our objective was to obtain predictive probabilities of knee replacement (KR) based on the combination of femorotibial joint (FTJ) cartilage thickness (CtTh) loss over 2 years, and CtTh at baseline.

Methods: Knees with symptomatic OA, defined by definite osteophyte (OARSI atlas grade 1-3) and participant-reported frequent knee symptoms at baseline, were selected prospectively from the Osteoarthritis Initiative (Project 09B). FTJ cartilage thickness at baseline and year 2 were measured quantitatively with MRI (3T; sagittal DESS sequence) and calculated as total  femorotibial cartilage volume divided by the cartilage surface area. KRs reported up to 7 years following the 2-year imaging window were self-reported and confirmed. We used a Bayesian discrete time logistic survival model with intercepts (α_k) for each year of follow-up: logit[p_k(x)] = α_k + β′x. Diffuse reference prior distributions for the model parameters, α_k, β_j ~ N(0, 10⁴) were specified. Posterior densities of the regression coefficients were estimated with 10,000 Markov Chain Monte Carlo samples generated with the R package R2OpenBUGS. Models were compared using deviance information criteria. Sensitivity of the results to the prior specification was considered. 

Results: Among 582 knees (one knee per participant), 95 underwent KR up to 7 years following the initial 2-year imaging window (median follow-up 6 years). Mean baseline CtTh was 1.85 mm (SD 0.29), while mean CtTh loss over 2 years was 0.04 mm (SD 0.07). Greater CtTh at baseline (standardized) was associated with lower odds of KR (OR=0.73 [95% credible interval (CI): 0.59, 0.90]; Pr(OR<1 | data) = 0.9995). Greater 2-year loss of CtTh (standardized) was associated with higher odds of KR (OR=1.71 [95%CI: 1.41, 2.05]; Pr(OR>1| data) =1.0). The figure presents three representative curves of predictive probabilities of KR as a function of 2-year CtTh loss corresponding to three baseline CtTh values. For example, a knee with baseline CtTh of 1.85 mm and 2 year loss of 0.17 mm has a probability of KR within 5 years of 0.27. A knee with less baseline CtTh, 1.56 mm (-1SD), has a probability of KR of 0.34 with the same measured loss, while a knee with greater CtTh at baseline, 2.14 mm (+1SD), has probability of KR of 0.20.

Conclusion: The same rate of 2-year FTJ cartilage thickness loss has varying predictive probability of future KR, depending on the baseline cartilage thickness, highlighting the difficulty of defining minimum clinically important differences in a tissue that eventually disappears entirely at the end-stage disease process. Predictive probabilities of clinically relevant outcomes provide a more interpretable way to communicate the clinical relevance of imaging biomarkers beyond traditionally reported odds ratios.