Utilizing Machine Learning to Predict Total Charges for Primary Admission Following Lumbar Disc Arthroplasty

Introduction: Lumbar degenerative disc disease (LDDD) affects millions worldwide, leading to chronic low back pain and reduced mobility. Traditional fusion methods have limitations such as long recovery times and potential adjacent segment disease. Lumbar disc arthroplasty (LDA) offers an alternative, aiming to preserve motion and reduce adjacent segment disease risks. The aims of this study were to evaluate the impact of 1) hospital size, 2) geographic region, and 3) patient-specific variables on charges associated with the primary admission period following LDA.

Methods: A retrospective analysis was conducted using machine learning models to predict and analyze factors associated with total charges for LDA. Data from the National Inpatient Sample (NIS) database, covering the years 2016 to 2020, was queried, focusing on patients who underwent LDA. The primary outcome of interest was the total charge associated with the primary admission for LDA, which was analyzed in relation to patient demographics, hospital characteristics, and local/regional economic conditions. Multivariate linear regression and machine learning algorithms, including logistic regression, random forest, and gradient boosting trees, were employed to assess their predictive power on charge outcomes. Statistical significance was set at a P value of < 0.05.

Results: The analysis included 568 eligible LDA cases, comprising 526 single-level and 42 multi-level procedures. The mean total charge for all operative admissions was $124,946, with high-charge cases defined by a cutoff of $155,770. The average length of stay (LOS) was 2.3 days. Major charge predictors included LOS, large hospital size, and private investor-owned hospital type. The random forest model demonstrated the highest accuracy (AUC = 0.836), followed by gradient boosting (AUC = 0.826) and logistic regression (AUC = 0.822).

Conclusion : The findings suggest that incorporating machine learning models into clinical practice can enhance economic decision-making in spine surgery. Given the significant impact of hospital characteristics and patient-specific factors on costs, future strategies should prioritize optimizing resource allocation and promoting LDA adoption in suitable clinical settings to improve patient outcomes and ensure cost-effective healthcare delivery.