# Preoperative prediction of residual rotational instability after ACL reconstruction using a machine learning model

## TL;DR

This study developed a machine learning model to predict residual knee instability after ACL surgery, using preoperative clinical and MRI data.

## Contribution

A novel machine learning model was developed and validated to predict postoperative rotational instability after ACL reconstruction.

## Key findings

- XGBoost achieved the best performance with an AUC of 0.87 in predicting residual pivot-shift.
- Posterior tibial slope and lateral meniscal extrusion were the strongest predictors in the model.
- The model showed consistent results in external validation with an AUC of 0.84.

## Abstract

Residual rotational instability persists in 15%–30% of patients after anterior cruciate ligament (ACL) reconstruction and is associated with subjective instability, reduced return‐to‐sport rates and increased graft failure risk. Accurate preoperative prediction of residual pivot‐shift could improve surgical planning and guide selective anterolateral reinforcement. This study aimed to develop and validate a machine‐learning model to predict postoperative rotational instability (pivot‐shift ≥2) using routinely available clinical and magnetic resonance imaging (MRI)‐derived variables.

A multicenter retrospective cohort of patients undergoing primary ACL reconstruction was screened (n = 312), of whom 246 met inclusion criteria and were analysed, including 79 patients with postoperative pivot‐shift ≥2. Variables included demographic factors, clinical laxity, posterior tibial slope, lateral meniscal extrusion, graft type and anterolateral reinforcement. Three algorithms—Random Forest, extreme gradient boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO) logistic regression—were trained (70%) and internally validated (30%) using five‐fold cross‐validation. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), calibration and decision‐curve analysis. External validation was performed in an independent cohort (n = 60).

XGBoost showed the best discriminative performance (AUC 0.87; sensitivity 0.83; specificity 0.79), with consistent results in external validation (AUC 0.84). Posterior tibial slope and lateral meniscal extrusion were the strongest predictors. Decision‐curve analysis demonstrated superior net clinical benefit compared with rule‐based approaches using International Knee Documentation Committee (IKDC) or Lachman thresholds.

A machine‐learning model based on routine preoperative clinical and MRI variables accurately predicts residual mechanical pivot‐shift after ACL reconstruction. This tool may support individualized surgical planning and selective indications for anterolateral reinforcement. Prospective preoperative evaluation of its impact on clinical decision‐making is warranted.

Level II, retrospective diagnostic‐predictive study.