Patient-reported outcomes, predictive modeling, and artificial intelligence in spine surgery: current limitations and future directions

Chronic back pain is a leading cause of disability worldwide, contributing to reduced productivity and diminished quality of life for many patients. Spine surgery, particularly lumbar decompression and fusion, serves as an intensive intervention after conservative measures fail. Despite advances in surgical technique and instrumentation, outcomes remain highly variable, with up to half of patients experiencing little or no meaningful improvement. This variability has prompted growing interest in predictive models that forecast which patients are most likely to benefit from surgery.

This review examines the current landscape of predictive modeling for patient-reported outcomes (PROs) in spine surgery. We evaluated the use of PRO instruments, which have become central to assessing surgical success. We further explored existing predictive models that incorporate demographic, clinical, and psychosocial variables, as well as emerging applications of artificial intelligence (AI) and machine-learning techniques aimed at improving diagnostic accuracy and prognostic performance.

Current predictive models demonstrate variable performance across populations and remain constrained by incomplete datasets, inconsistencies in definitions, and limited integration of psychosocial factors. While most models report moderate discrimination for categorical outcomes (AUC 0.62-0.84), the subjective nature of PROs and the complexity of postoperative recovery pose challenges to accurate prediction. Machine-learning approaches, including neural networks, show promise in identifying complex relationships among diverse predictors of surgical outcomes. Future advancements will rely on leveraging multimodal data sources and integrating them through AI-driven frameworks. Transparency and external validation remain essential for clinical implementation.

While PROs remain essential for understanding patient experiences and benchmarking surgical success, they often fall short as standalone predictive tools. The next generation of modeling should integrate multidimensional data and utilize AI's capacity to detect clinically meaningful patterns. When implemented, these models have the potential to enhance preoperative counseling, guide surgical decision-making, and advance personalized spine care.