Risk stratification for postoperative hematoma following anterior cervical spine surgery: a machine learning approach

Area of Science:

• Neurosurgery
• Spine Surgery
• Medical Informatics

Background:

• Anterior cervical spine procedures are increasing, with low complication rates.
• Postoperative hematoma is a rare but serious complication, potentially causing respiratory compromise.
• This study aimed to identify risk factors for postoperative hematoma using machine learning.

Purpose of the Study:

• To utilize machine learning algorithms (MLAs) to characterize the clinical risk profile for readmission and reoperation due to postoperative hematoma.
• To identify preoperative variables associated with postoperative hematoma development in patients undergoing anterior cervical spine procedures.

Main Methods:

• Utilized the American College of Surgeons National Surgical Quality Improvement Program database (2012-2018).
• Included adult patients undergoing elective anterior cervical spine procedures with postoperative hematoma requiring readmission/reoperation.
• Employed 1:5 propensity score matching and constructed six MLAs to identify risk factors, deriving feature importance from the top-performing model.

Main Results:

• Of 1,056 matched patients, 16.67% developed postoperative hematoma.
• MLAs achieved an average AUC of 0.824 and accuracy of 87.75%, but low sensitivity (30.2%) due to rarity.
• Significant risk factors included diabetes, smoking history, preoperative WBC, preoperative sodium, and dependent function status.

Conclusions:

• MLAs identified key variables associated with postoperative hematoma after anterior cervical spine surgery.
• Due to low sensitivity, these algorithms are not reliable for individual patient prediction or screening.
• MLAs can serve as adjunctive tools to improve perioperative risk awareness and stratification for these rare events.