Machine learning model successfully identifies important clinical features for predicting outpatients with rotator
Cheng Li1, Yamuhanmode Alike1, Jingyi Hou1
1Department of Orthopedics, Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University, 107 Yan Jiang Road West, Guangzhou, 510120, Guangdong, China.
Summary
Machine learning models accurately predict rotator cuff tears (RCTs) using clinical data. Explainable AI identified key predictors like the Jobe test, Bear hug test, and age, aiding prompt treatment.
Area of Science:
- Orthopedics
- Medical Informatics
- Machine Learning
Background:
- Rotator cuff tears (RCTs) are a common cause of shoulder pain and dysfunction.
- Accurate and efficient prediction of RCTs is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To develop and validate machine learning (ML) models for predicting rotator cuff tears (RCTs) in outpatient settings.
- To utilize explainable artificial intelligence (XAI) to identify key clinical features associated with RCTs.
Main Methods:
- Retrospective analysis of a clinical registry dataset (2019-2022) including patients with shoulder pain.
- Development and comparison of six ML algorithms (XGBoost, Random Forest, etc.) for RCT prediction.
- Evaluation of model performance using AUC and Brier scores; interpretability assessed with SHAP values.
Main Results:
- The Extreme Gradient Boost (XGBoost) model demonstrated superior performance with an AUC of 0.92 and accuracy of 0.85.
- Key predictors for RCTs identified by SHAP analysis included the Jobe test (1.458), Bear hug test (0.950), and patient age (0.790).
- The study included 1684 patients, with 417 diagnosed with RCTs via arthroscopy.
Conclusions:
- ML models, particularly XGBoost, effectively predict rotator cuff tears using clinical data.
- XAI revealed significant clinical predictors, enabling a better understanding of RCTs.
- The developed ML tool, integrated into an application, can assist in outpatient RCT prediction and management.
