Fig. 4: Evaluating matching quality and identifying significant biomarkers in OA incidence and knee replacement cohorts.

This figure evaluates the quality of cohort matching and accentuates significant differences in imaging biomarkers between the Control and Outcome groups for OA Incidence and Knee Replacement analyses. a OA incidence matching quality: Standardized Mean Difference (SMD) analysis compares numerical covariates before (teal) and after (purple) matching, visualizing improved balance (see Supplementary Data Table 8). b Knee replacement matching quality: The SMD assessment for numerical covariates illustrates the effectiveness of the matching process. Effect sizes are depicted in teal for pre-match and purple for post-match conditions, demonstrating improved equivalence between cohorts (see Supplementary Data Table 10). c OA incidence covariate balance: Bar plots depict the distribution of categorical covariates between Control (dark teal) and OA incidence (light teal) groups. This analysis highlights the improvement in balance for categorical variables, such as Sex (see Supplementary Data Table 9). d Knee replacement covariate balance: Bar plots illustrate the distributions of categorical covariates for Control (dark purple) versus Knee Replacement (light purple) groups. These plots show the success of the matching process in equalizing group characteristics (see Supplementary Data Table 11). e Significant features in OA incidence cohort study: Box plots display the distribution of significant PC mode features between the Control and OA incidence groups. Box plots display the median, interquartile range, and whiskers extending to 1.5 times the IQR. Statistical significance was determined using Paired Wilcoxon Rank Sum Tests with Benjamini-Hochberg correction for multiple comparisons (p-values adjusted using the Hochberg method). Significance levels are denoted as follows: single asterisk for p < 0.05, double asterisks for p < 0.01, and triple asterisks for p < 0.001, aligning with results in Supplementary Data Table 14. f Significant features in knee replacement cohort study: Box plots, as described in panel e, compare significant PC mode features between Control and Knee Replacement groups. Statistical significance was determined using Paired Wilcoxon Rank Sum Tests with Hochberg correction. Annotated p-values and significance thresholds are provided in Supplementary Data Table 15. The remaining significant PC mode plots not shown are in Supplementary Figs. 12 and 13.