Predicting Surgical Complications in Patients Undergoing Elective Adult Spinal Deformity Procedures Using Machine...

Area of Science:

• Orthopedic Surgery
• Data Science
• Machine Learning

Background:

• Machine learning models, including logistic regression (LR) and artificial neural networks (ANNs), are effective for analyzing complex medical datasets.
• Artificial neural networks (ANNs) have not yet been extensively applied to risk factor analysis in orthopedic surgery.

Purpose of the Study:

• To develop and validate machine learning models for predicting complications after adult spinal deformity (ASD) surgery.
• Identify key risk factors associated with post-operative complications in ASD patients.

Main Methods:

• A cross-sectional study utilizing the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database.
• Trained and evaluated logistic regression (LR) and artificial neural network (ANN) models on 4,073 ASD patients.
• Compared model predictive accuracy against the American Society of Anesthesiologists (ASA) class benchmark using area under the receiver operating characteristic curves (AUC).

Main Results:

• Both ANN and LR models significantly outperformed ASA scoring in predicting all assessed complications (p<.05).
• The ANN model demonstrated superior predictive accuracy compared to LR for cardiac complications, wound complications, and mortality (p<.05).

Conclusions:

• Machine learning algorithms provide superior risk prognostication for individual patients undergoing ASD surgery compared to traditional ASA scoring.
• The continuous learning capability of machine learning makes them valuable tools for improving risk assessment in complex clinical scenarios.