The Value of Artificial Intelligence in Predicting Perioperative Complications and Key Quality Metrics Following Elective Degenerative Spine Surgery: A Narrative Review of the Current Landscape.

Abstract

Narrative review. To perform a review discussing current applications of artificial intelligence (AI) and machine learning (ML) in the prediction of quality metrics in elective degenerative spinal surgeries. A major barrier to the long-term effectiveness of elective spine surgery is the high incidence of perioperative complications. The use of AI and ML in the preoperative evaluation of elective spine surgery has been limited. A literature review was performed to identify adults who underwent surgery for elective lumbar degenerative spinal pathology or deformity. Studies were stratified by cohort size, type of complications that the AI/ML models predicted, type of machine learning algorithm, and performance of AI/ML models. Included for analysis were 46 studies. The median study sample size to build and validate predictive models was 4538 (109-279,135). Models and associated AUCs are as follows: AUC 0.57-0.95 [extended length of stay (LOS), discharge disposition, and costs], AUC 0.59-0.95 (medical complications), and AUC 0.64-0.87 (short-term readmissions/reoperations). The AI/ML models favored supervised learning algorithms and opened the possibility for additional development of unsupervised and reinforcement-based algorithms. Future models should utilize additional granular predictors such as the surgical invasiveness index, social support, and socioeconomic variables to enhance predictive capabilities.