Fig. 5: Stability selection and distribution of imaging biomarkers from multivariate analysis of OA incidence and knee replacement outcomes.

This figure presents imaging biomarkers identified through stability selection and multivariate analysis. These features have the potential to inform knee health dynamics and guide further exploration into biomarkers relevant to digital twin development. a OA incidence stability selection—overall top 20 features: This horizontal bar chart displays the top 20 significant variables associated with OA incidence, ranked by mean coefficients derived from bootstrap elastic net analysis over 1000 iterations. Positive effects extend to the right, while negative effects extend to the left, with 95% confidence intervals indicated by horizontal lines. Detailed results are available in Supplementary Data Table 17 and visualized by imaging biomarker category in Supplementary Fig. 14a–d. b Knee replacement stability selection—overall top 20 features: Presents a horizontal bar chart of the top 20 significant variables for knee replacement outcomes, also ranked by mean coefficients from bootstrap elastic net analysis with 1000 iterations. Positive effects extend to the right, negative effects to the left, with 95% confidence intervals indicated. Additional details are in Supplementary Data Table 20 and categorized views are shown in Supplementary Fig. 14e–h. c Distribution plots of PC modes for OA incidence: Features violin plots of significant PC modes from the final GLM logistic regression, comparing OA incidence (light teal) and Control groups (dark teal). The violin plots are superimposed with dot plots representing Kellgren-Lawrence (KL) grades, a radiographic measure of OA severity (KL grade ≥2 indicates radiographic OA). Dot plots reflect ultimate knee outcomes over time in the OAI study. Statistical details are provided in Supplementary Data Table 18 and further visualized in Supplementary Fig. 15. d Distribution plots of PC modes for surgical intervention: Violin plots display significant PC modes from the final GLM logistic regression, comparing Control (dark purple) and Knee Replacement (light purple) groups, further stratified by non-replacement, partial replacement, and total knee replacement. These dot plots reflect ultimate knee outcomes over time in the OAI study. Additional statistical details can be found in Supplementary Data Table 21 and further visualized in Supplementary Fig. 16.