Explainable Machine Learning Approach to Prediction of Prolonged Intensive Care Unit Stay in Adult Spinal Deformity...

Area of Science:

• Spine Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Prolonged intensive care unit (ICU) stays are associated with increased costs and poorer outcomes in Adult Spinal Deformity (ASD) patients.
• Machine learning (ML) offers potential for pre-operative risk prediction but often lacks transparency in its decision-making processes.

Purpose of the Study:

• To develop and validate ML models for predicting prolonged ICU stay in ASD patients.
• To demonstrate ML models can match or exceed traditional statistical methods in predictive power.
• To ensure ML model decisions align with established clinical reasoning using SHAP values.

Main Methods:

• Retrospective cohort study including 535 ASD patients undergoing posterior or combined fusion.
• Training and validation of five ML models: Decision Tree, Random Forest, Support Vector Classifier, GradBoost, and CNN.
• Utilizing Shapley Additive Explanation (SHAP) values to interpret model predictions and identify key predictive factors.

Main Results:

• ML models achieved an Area Under the Receiver Operating Curve (AUROC) between 0.67 and 0.83.
• The Random Forest model demonstrated the highest predictive performance.
• Key predictors identified by SHAP analysis included surgery duration, complications, and estimated blood loss.

Conclusions:

• Developed an effective ML model for predicting prolonged ICU stay in ASD patients.
• SHAP analysis confirmed that the ML model's predictions align with traditional clinical judgment.
• ML models show significant promise for enhancing risk stratification and optimizing care efficiency in ASD surgery.