Machine Learning-Based Prognostic Models for Functional Outcomes in Spinal Cord Injury: Systematic Review

Area of Science:

• Medical Informatics
• Neurology
• Biostatistics

Background:

• Machine learning (ML) is increasingly utilized for prognostic prediction models in spinal cord injury (SCI).
• Existing SCI prognostic models exhibit significant heterogeneity in outcome measures, predictors, modeling strategies, and validation methods.
• Systematic evaluation of reporting quality, risk of bias, and clinical applicability of these ML models is lacking.

Purpose of the Study:

• To assess the reporting quality and risk of bias of ML-based prognostic models for SCI.
• To evaluate the clinical applicability, model features, validation, and implementation barriers of these models.

Main Methods:

• A comprehensive literature search was conducted across multiple databases up to May 14, 2025.
• Two independent investigators screened studies, extracted data, and assessed risk of bias.
• The Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement and the Prediction Model Risk of Bias Assessment Tool (PROBAST) were used for evaluation.

Main Results:

• Nineteen cohort studies were included, with TRIPOD adherence ranging from 54.8% to 81.1%.
• All 19 studies exhibited a high risk of bias, primarily due to analytical limitations.
• External validation was present in only one study; most relied on internal validation or model development only, with issues in sample size justification, missing data handling, and calibration.

Conclusions:

• ML holds potential for SCI prognostic modeling, particularly with complex data.
• However, current evidence is hampered by incomplete reporting, high risk of bias, heterogeneity, and insufficient external validation.
• Robust, larger studies with standardized outcomes and external validation are essential for clinical implementation.