Machine Learning for Individualized Risk Estimation in Anterior Lumbar Interbody Fusion.

Abstract

Background and objectives: Although the anterior approach to the spine for anterior lumbar interbody fusion (ALIF) has been shown to be an effective procedure, there are different surgical risks compared with conventional posterior fusion. ALIF patients could potentially receive more personalized care plans that minimize the risk of negative outcomes by forecasting short-term postoperative results before the surgical procedure. The objective of this research was to evaluate the performance of machine learning (ML) algorithms in predicting short-term unfavorable postoperative outcomes after ALIF and to develop an easy-to-use and readily available instrument for this purpose.

Methods: Using the American College of Surgeons National Surgical Quality Improvement Program database, we identified ALIF patients and used 6 ML algorithms to build models predicting postoperative outcomes. These models were then incorporated into an open-access web application.

Results: The analysis included 8304 ALIF patients. The LightGBM models achieved area under the receiver operating characteristic scores of 0.735 for prolonged length of stay and 0.814 for nonhome discharges. The random forest models achieved area under the receiver operating characteristics of 0.707 for 30-day readmissions and 0.701 for major complications. These top-performing models were integrated into a web application for individualized patient predictions.

Conclusion: ML techniques show promise in predicting postoperative outcomes for ALIF surgeries. As data in spinal surgery expand, these predictive models could significantly improve risk assessment and prognosis. We present an accessible predictive tool for ALIF surgeries to achieve the goals mentioned above.