Identifying high healthcare utilizers following cervical spine surgery using comprehensive predictive modeling...

Area of Science:

• Spine Surgery
• Health Services Research
• Machine Learning in Healthcare

Background:

• Rising healthcare expenditures for chronic spinal conditions and surgery necessitate predictive models.
• Identifying patients at high risk for postoperative healthcare utilization is crucial for resource allocation and care optimization.

Purpose of the Study:

• To develop machine learning (ML) models for predicting postoperative healthcare utilization in cervical spine fusion patients.
• To leverage comprehensive preoperative patient data, including medical, surgical, and social histories, for granular prediction.

Main Methods:

• Analysis of a cohort of 4480 anterior and posterior cervical decompression and fusion surgeries (2002-2022).
• Systematic extraction of patient and operative characteristics.
• Application and optimization of various ML algorithms to predict high healthcare utilizers within 90 days post-surgery.
• Computation of SHAP feature importance for the top-performing model.

Main Results:

• 12% of patients were identified as high healthcare utilizers.
• All ML models surpassed the American Society of Anesthesia benchmark.
• The Balanced Random Forest model achieved the highest discriminative performance (AUC: 0.772 ± 0.007).
• Top predictive features included increased operating room duration, and 90-day preoperative neuromodulator and opioid usage.

Conclusions:

• Successfully developed prognostic ML models to predict high healthcare utilization within 90 days of cervical spine surgery.
• These validated models show potential to assist spine surgeons in clinical workflows.
• Accurate prediction aids in optimizing patient care and managing healthcare resources effectively.