Prediction model-based comparison of medial stabilized and cruciate-retaining total knee arthroplasty: A two-phase study.

BACKGROUND: Medial stabilized (MS) inserts have been introduced to improve knee kinematics and mid-flexion stability in total knee arthroplasty (TKA). However, direct comparisons between MS and cruciate-retaining (CR) designs are often confounded by patient-related factors.

PURPOSE: To compare postoperative knee flexion between MS and CR designs using a prediction model-based counterfactual framework.

METHODS: This retrospective two-phase study included patients who underwent primary TKA at a single institution. In Phase 1, a multivariable regression model was developed using 129 CR-TKA cases to predict 1-year postoperative knee flexion based on preoperative and intraoperative variables. Internal validation was performed using 10-fold cross-validation, and model performance and calibration were assessed using out-of-fold predictions. In Phase 2, the validated model was applied to MS-TKA cases to estimate the postoperative flexion expected under a hypothetical CR condition. Predicted CR flexion and observed MS flexion were compared using paired t-tests.

RESULTS: In the CR cohort (n = 129), the mean difference between predicted and observed postoperative flexion was 0.07° (95% confidence interval (CI), -1.5 to 1.7°; P = 0.996). Ten-fold cross-validation demonstrated moderate predictive performance (R2 = 0.218, mean absolute error = 7.78°, root mean squared error = 9.84°). In the MS cohort (n = 24), observed postoperative flexion (120.0 ± 10.5°) did not differ significantly from the predicted CR flexion (122.5 ± 6.5°), with a mean paired difference of -2.43° (95% CI, -6.59 to 1.73°; P = 0.239).

CONCLUSION: The CR-based prediction model demonstrated moderate internal predictive performance within the study cohort. Within the limitations of this exploratory analysis, no statistically significant difference in postoperative knee flexion was detected between observed MS-TKA outcomes and the flexion predicted under hypothetical CR conditions. Larger externally validated studies with adequate statistical power are required before definitive conclusions regarding implant-specific differences can be drawn.