Machine learning models for predicting postoperative paraplegia in acute type A aortic dissection patients.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Neurosurgery

Background:

• Acute type A aortic dissection (ATAAD) is a life-threatening emergency requiring immediate surgery.
• Postoperative paraplegia is a severe complication impacting patient prognosis and healthcare costs.
• Identifying risk factors and prediction models for paraplegia is crucial for improving patient outcomes.

Purpose of the Study:

• To explore risk factors for postoperative paraplegia in ATAAD patients.
• To develop and validate a predictive risk model for postoperative paraplegia.
• To enhance surgical decision-making and patient care for ATAAD.

Main Methods:

• Retrospective analysis of 572 ATAAD patients undergoing surgery.
• Utilized Least Absolute Shrinkage and Selection Operator (Lasso) regression for variable selection.
• Developed and validated seven machine learning models, including Neural Networks, using stratified sampling and SHAP analysis.

Main Results:

• 22 patients (3.84%) developed postoperative paraplegia.
• The Neural Networks model exhibited the highest predictive performance (AUC, Brier score, F1, Kappa).
• Significant risk factors identified: pancreatic lipase, left subclavian artery involvement, Sun's procedure, age, pancreatic amylase, hemoglobin, and secondary surgery.

Conclusions:

• The Neural Networks model offers superior predictive capability for postoperative paraplegia in ATAAD.
• Understanding identified risk factors can guide preventative strategies and surgical planning.
• This model can aid clinicians in assessing and mitigating paraplegia risk in ATAAD patients.