Mylène Jansen

MRI cartilage thickness up to ten years after KJD 257 13 Statistical analyses Whole-joint analyses MATLAB R2020a and the SurfStat MATLAB package (https://www.math.mcgill.ca/keith/ surfstat/, modified for this specific application by Graham Treece of the University of Cambridge) were used for whole-joint, vertex-wise data analysis and visualization. The average cartilage thickness was displayed for each time point separately by averaging data of all available patients at each specific time point. Statistical parametric mapping (SPM) was used for analysis of changes over time. SPM uses all subject values at each vertex for testing between time points and delivers p- values corrected for multiple comparisons. 18 For differences at each follow-up moment compared to baseline, SPM with linear mixed models was used. The influence of baseline patient characteristics on the changes over time was also analyzed with SPM, using a separate linear regression model for each different patient characteristic and its influence on short-term (2-year) and long-term (ten-year) changes. In all cases, a threshold p- value <0.05 was considered statistically significant. Since KJD has previously shown significant results mostly in the patients’ most affected compartment (MAC), patients were separated in 2 groups based on whether their MAC was the medial or lateral compartment. Compartmental analyses For each time point, the average cartilage thickness was calculated for the medial and lateral femur and tibia. Instead of analyzing changes over time for the medial and lateral side areas, changes over time were analyzed for the MAC (either medial of lateral) and least affected compartment (LAC; either lateral or medial). As such, the 4 different compartments analyzed at each time point were the MAC and LAC femur and tibia. Compartmental statistical analyses were performed in IBM SPSS Statistics 25. In case of missing data over the entire 10 years, for the statistical compartmental analyses (not for the whole-joint surface-based analyses) multiple imputation was performed for each compartment separately for all patients; missing data was replaced by the average of 5 imputations considering the data available before loss of follow-up data. This was considered valid, as previous data have shown that those patients that underwent arthroplasty after several years within the 10-year follow-up period had no significant change in clinical or structural radiographic outcome shortly before arthroplasty. 15 As a sensitivity analysis for imputation validity, patients with complete data sets were analyzed separately. Changes over time were analyzed using repeated measures ANOVA. Additionally, as patients filled out the Western Ontario McMaster Osteoarthritis Index (WOMAC) at the same time points MRI scans were performed, the influence of compartmental cartilage thickness changes over time on the change in total WOMAC over time was analyzed using linear mixed models, with total WOMAC as outcome variable, a random intercept at patient level and fixed effects of time and compartmental cartilage thickness. In case the cartilage thickness change in a compartment had a statistically significant influence, its influence on the change in WOMAC