Prediction of primary admission total charges following lumbar disc arthroplasty utilizing machine learning.

Background: Lumbar degenerative disc disease (LDDD) is a widespread condition contributing to chronic lower back pain and impaired mobility. While spinal fusion has been the conventional treatment, it poses drawbacks including extended recovery periods and the risk of adjacent segment degeneration. Lumbar disc arthroplasty (LDA) has emerged as a motion-preserving alternative with the potential to mitigate these risks. This study aimed to assess how factors such as hospital size, regional location, and patient characteristics influence hospitalization charges during the initial admission for LDA. Methods: This retrospective study utilized the National Inpatient Sample (NIS) database from 2016 to 2020 to identify patients who underwent LDA. The primary endpoint was total hospitalization charges for the initial surgical admission. Variables analyzed included demographic data, hospital attributes, and economic indicators at the regional level. Both multivariate linear regression and machine learning (ML) techniques-logistic regression, random forest, and gradient boosting-were applied to evaluate predictive factors for cost. A significance threshold was set at P<0.05. Results: A total of 568 patients met the inclusion criteria, consisting of 526 single-level and 42 multi-level LDA procedures. The average admission charge was $124,946, with high-cost admissions defined as those exceeding $155,770. The mean hospital stay was 2.3 days. Key predictors of increased charges included longer length of stay, treatment at large hospitals, and for-profit hospital ownership. Among the models tested, the random forest algorithm yielded the highest predictive accuracy [area under the receiver operating characteristic curve (AUC) =0.836], followed by gradient boosting (AUC =0.826) and logistic regression (AUC =0.822). Conclusions: Charges associated with LDA are significantly influenced by institutional and patient-level factors. ML models effectively predicted cost variability and hold promise for informing cost-effective strategies in spine surgery. Integrating these models into clinical workflows may enhance both financial planning and patient care.