Mastering Resource Allocation: Machine Learning Predictions for Lumbar Spinal Fusion Patients

Introduction: Machine learning models have demonstrated their effectiveness in making reliable predictions and assisting in clinical decision-making regarding various outcomes such as readmission, discharge location, blood transfusion, and postoperative complications. Applying Machine learning models in patients undergoing lumbar spinal fusion, especially when using larger databases.

Methods: The study utilized the National Inpatient Sample datasets from 2009 to 2020 to analyze patients who underwent dorsal and lumbar fusion for nondeformity. The variables used included age, sex, race, ethnicity, admission type, patient refined risk of mortality, severity of illness, and clinical classification software diagnosis codes. To predict payments, length of stay, and discharge disposition, a Naïve Bayes machine learning model was employed, with the aim of developing a payment scheme stratified by risk for dorsal and lumbar fusion.

Results: An adaptive boosting machine learning algorithm was used to analyze national inpatient sample datasets from 2009 to 2020, focusing on dorsal and lumbar fusion for nondeformity. The algorithm showed high reliability, with area under the receiver and operating characteristic curve of 0.875, 0.938, and 0.904 for Cost, Length of stay, and discharge disposition, respectively. The model found that patients with increased mortality risk or illness severity had higher costs and payments, reflecting the higher risk associated with caring for these patients. The study also revealed a wide range in expected payments due to preoperative comorbidities, which suggests a need for a risk-based model for these patients.

Conclusion : The Naïve Bayes machine learning model demonstrated good to excellent reliability and responsiveness in predicting cost, length of stay, and discharge disposition. The study showed a significant difference in costs between the lowest and highest risk strata based on mortality risk and illness severity. The results suggest that institutions face higher financial risk in a flat bundled payment model for patients with higher comorbidity rates.