Scholarlink

Background <p>Identifying patients who are unlikely to achieve significant improvement after anterior cruciate ligament reconstruction (ACLR) facilitates more effective perioperative shared decision-making (SDM). This study aimed to develop and validate machine learning (ML) models for the preoperative prediction of insufficient clinical improvement 1 year after ACLR using the Lysholm Knee Score Scale, and to evaluate important predictive variables.</p> Methods <p>A single-center retrospective study design was adopted. A total of 892 patients who underwent primary ACLR and completed at least 1 year of complete follow-up in the Department of Orthopedics of a tertiary hospital between January 2014 and December 2024 were enrolled. The minimal clinically important difference (MCID) of the Lysholm score for the study population was determined to be 8.5 points using an anchoring method combined with an effect size method. Hyperparameter tuning was performed using 5‑fold stratified cross‑validation combined with grid search. Five models, namely ExtraTrees, random forest (RF), XGBoost, logistic regression (LR), and K‑nearest neighbors (KNN), were evaluated and interpreted using the area under the receiver operating characteristic curve (AUC), calibration curves, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP) analysis.</p> Results <p>Among the five models, the ExtraTrees model achieved the best performance, with a validation set AUC of 0.887 (95% confidence interval [CI] 0.852–0.917), a sensitivity of 0.719, a specificity of 0.838, and a Brier score of 0.091. DCA showed that the ExtraTrees model yielded the highest net clinical benefit over a wide range of threshold probabilities. SHAP analysis revealed that bilateral meniscal injury was the most important predictor of insufficient improvement, followed by higher preoperative baseline Lysholm score, grade III–IV chondral injury, lateral meniscal injury, smoking history, and hypertension.</p> Conclusion <p>Based on the Lysholm score, machine learning models perform well in the preoperative prediction of insufficient clinical improvement 1 year after ACLR. The ExtraTrees model is promising for assisting perioperative shared decision‑making and personalized patient management for ACLR by preoperatively identifying approximately 71.9% of patients at risk of insufficient clinical improvement and providing explanations for relevant contributing factors.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of machine learning models for preoperative prediction of insufficient clinical improvement after anterior cruciate ligament reconstruction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月18日

查看全文
  • Huan Wang,
  • Hujun Wang,
  • Nan Li,
  • Shuyan Qie,
  • Jie Dong,
  • Hanming Wang

摘要

Background

Identifying patients who are unlikely to achieve significant improvement after anterior cruciate ligament reconstruction (ACLR) facilitates more effective perioperative shared decision-making (SDM). This study aimed to develop and validate machine learning (ML) models for the preoperative prediction of insufficient clinical improvement 1 year after ACLR using the Lysholm Knee Score Scale, and to evaluate important predictive variables.

Methods

A single-center retrospective study design was adopted. A total of 892 patients who underwent primary ACLR and completed at least 1 year of complete follow-up in the Department of Orthopedics of a tertiary hospital between January 2014 and December 2024 were enrolled. The minimal clinically important difference (MCID) of the Lysholm score for the study population was determined to be 8.5 points using an anchoring method combined with an effect size method. Hyperparameter tuning was performed using 5‑fold stratified cross‑validation combined with grid search. Five models, namely ExtraTrees, random forest (RF), XGBoost, logistic regression (LR), and K‑nearest neighbors (KNN), were evaluated and interpreted using the area under the receiver operating characteristic curve (AUC), calibration curves, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP) analysis.

Results

Among the five models, the ExtraTrees model achieved the best performance, with a validation set AUC of 0.887 (95% confidence interval [CI] 0.852–0.917), a sensitivity of 0.719, a specificity of 0.838, and a Brier score of 0.091. DCA showed that the ExtraTrees model yielded the highest net clinical benefit over a wide range of threshold probabilities. SHAP analysis revealed that bilateral meniscal injury was the most important predictor of insufficient improvement, followed by higher preoperative baseline Lysholm score, grade III–IV chondral injury, lateral meniscal injury, smoking history, and hypertension.

Conclusion

Based on the Lysholm score, machine learning models perform well in the preoperative prediction of insufficient clinical improvement 1 year after ACLR. The ExtraTrees model is promising for assisting perioperative shared decision‑making and personalized patient management for ACLR by preoperatively identifying approximately 71.9% of patients at risk of insufficient clinical improvement and providing explanations for relevant contributing factors.