Machine learning-based prediction of perioperative complications in spine surgery: a large-scale model development...

Area of Science:

• Spine Surgery Outcomes
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Spinal surgery poses significant perioperative risks, necessitating early identification of high-risk patients.
• Traditional risk scores have limitations; machine learning (ML) offers enhanced predictive accuracy by analyzing complex clinical data.
• Existing ML models often lack large, diverse patient cohorts for robust validation.

Purpose of the Study:

• To develop and validate ML models for predicting perioperative complications in spine surgery.
• To assess the fairness and equity of these predictive models across various patient subgroups.
• To improve patient outcomes and surgical safety through accurate risk prediction.

Main Methods:

• A retrospective cohort study of 5,060 adult patients from the SpineReg registry (2015-2023) was conducted.
• 160 preoperative features including demographics, clinical data, imaging, and patient-reported outcomes were analyzed.
• Six ML algorithms were trained and validated, with random forest (RF) showing superior performance.

Main Results:

• The RF model achieved an AUC of 0.87, balanced accuracy of 0.80, and PPV of 0.71.
• Key predictors of complications included sagittal imbalance and multilevel surgery; degenerative pathology was protective.
• The model demonstrated robust performance across diverse patient and surgical subgroups, ensuring fairness.

Conclusions:

• RF-based ML models accurately and equitably predict spine surgery complications in varied contexts.
• These models support personalized patient counseling, targeted monitoring, and optimized resource allocation.
• Clinical integration of these ML tools can enhance surgical safety and efficiency.