Predicting Spinal Cord Injury Prognosis Using Machine Learning: Systematic Review and Meta-Analysis

Area of Science:

• Medical Informatics
• Computational Biology
• Rehabilitation Medicine

Background:

• Spinal cord injury (SCI) presents complex challenges in patient prognosis.
• Machine learning (ML) techniques are emerging as powerful tools for predicting SCI outcomes.

Purpose of the Study:

• To evaluate the efficacy and caliber of ML models in forecasting SCI consequences.
• To compare the performance of various ML algorithms for SCI prognosis.

Main Methods:

• Comprehensive literature searches across multiple databases (PubMed, Web of Science, Embase, etc.).
• Meta-analysis of the area under the receiver operating characteristic curve (AUC) for ML models.
• Inclusion of 13 eligible studies from 1254 retrieved articles.

Main Results:

• XGBoost algorithm achieved the highest AUC (0.867) for spinal cord function prognosis.
• Other ML models showed varying performance for predicting complications, independent living, and walking ability.
• Random forest and logistic regression also demonstrated significant predictive power for different outcomes.

Conclusions:

• ML models accurately predict SCI outcomes, particularly spinal cord function prognosis.
• XGBoost algorithm is the top-performing model for SCI prognosis.
• Advancements in ML and large datasets will further enhance predictive capabilities for clinicians.