Machine learning prediction of post-traumatic osteoarthritis based on three-dimensional printing-derived joint...

Post-traumatic osteoarthritis (PTOA) develops in 20-40% of patients following ankle fracture fixation despite anatomic reduction. This study aimed to develop a machine learning model incorporating three-dimensional (3D) printing-derived joint congruence biomechanics for individualized PTOA prediction.

This retrospective cohort study included 263 patients (January 2020-January 2024) who underwent preoperative 3D-printed model-assisted surgical planning with minimum 24-month follow-up. Finite element analysis quantified joint congruence parameters including peak contact pressure, pressure inhomogeneity index, and contact center offset. Four machine learning algorithms were developed using training data (n = 158, 60%) and validated temporally (n = 105, 40%). The primary outcome was PTOA defined by Kellgren-Lawrence grade ≥ II with clinical symptoms.

PTOA developed in 102 patients (38.8%) during mean 36.4-month follow-up. Patients with PTOA demonstrated significantly higher peak contact pressure (15.2 ± 3.8 vs. 9.6 ± 2.4 MPa, P < 0.001) and pressure inhomogeneity index (2.8 ± 0.6 vs. 2.3 ± 0.4, P < 0.001). In the validation set, the XGBoost full model achieved an area under the curve (AUC) of 0.83 (95% CI 0.75-0.90), significantly outperforming the restricted model without biomechanical variables (AUC 0.74, 95% CI 0.65-0.82; P = 0.008). Among 198 patients with anatomic reduction, the full model maintained superior discrimination (AUC 0.81 vs. 0.68, P = 0.01), identifying a high-risk subgroup (peak pressure > 13 MPa) with 58.3% PTOA incidence versus 18.2% in low-pressure patients (P < 0.001).

Joint congruence parameters from 3D printing-based finite element analysis significantly improve machine learning prediction of PTOA following ankle fracture, identifying high-risk patients even after anatomic reduction.