Predictive Models for Length of Stay and Discharge Disposition in Elective Spine Surgery: Development, Validation,...

Area of Science:

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

Background:

• A retrospective study analyzed 3678 adult patients undergoing elective spine surgery from 2014-2019.
• Data was sourced from electronic health records at a single academic institution.

Purpose of the Study:

• To develop and validate machine learning models for predicting hospital length of stay (LOS) and discharge disposition after elective spine surgery.
• To compare the performance of these machine learning models against the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) prediction calculator.

Main Methods:

• Patients were stratified into cervical degenerative, lumbar degenerative, and adult spinal deformity cohorts.
• Predictive models were built using regression, classification trees, and LASSO, incorporating demographics, BMI, surgical details, and comorbidities.
• Model performance was validated using AUROC, sensitivity, specificity, and correlation, with direct comparison to the ACS NSQIP calculator.

Main Results:

• Machine learning models, particularly Poisson regression and LASSO, showed significantly higher correlation with observed LOS (R²=0.29) compared to NSQIP (R²=0.16).
• Logistic regression achieved an AUROC of 0.79 for predicting discharge location, statistically equivalent to the NSQIP calculator's AUROC of 0.75.
• The study included 3678 patients, with an average LOS of 3.66 days; 78% were discharged home, and 22% to rehabilitation.

Conclusions:

• Machine learning models provide accurate preoperative estimation of LOS and rehabilitation discharge risk for elective spine surgery.
• The developed models demonstrate superior performance in predicting LOS and comparable performance in predicting discharge location versus the ACS NSQIP calculator.