Using Machine Learning to Predict Surgical Site Infection After Lumbar Spine Surgery

Area of Science:

• Neurosurgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Surgical site infection (SSI) is a significant complication following posterior lumbar spinal surgery.
• Accurate prediction of SSI risk is crucial for effective patient management and prevention strategies.

Purpose of the Study:

• To employ machine learning techniques to analyze perioperative factors and identify blood glucose levels predictive of SSI.
• To develop a dynamic prediction model for SSI following posterior lumbar spinal surgery.

Main Methods:

• Utilized logistic regression, Lasso regression, support vector machine, and random forest to screen variables in a cohort of 4019 patients.
• Intersected variables identified by four methods to construct a dynamic prediction model.
• Assessed model performance using ROC and calibration curves, with internal validation.

Main Results:

• Identified four key predictors for SSI: Modic changes, sebum thickness, hemoglobin, and glucose.
• The prediction model demonstrated high accuracy with an Area Under the ROC Curve (AUC) of 0.988 in the test group and 0.987 in the validation group.
• The model showed favorable consistency between predicted and real measurements, with C-indices of 0.986 and 0.982 for the test and validation groups, respectively.

Conclusions:

• Logistic regression and machine learning successfully identified Modic changes, sebum thickness, hemoglobin, and glucose as significant risk factors for SSI.
• A dynamic prediction model was constructed to assist clinicians in simplifying SSI monitoring and prevention.