Harnessing machine learning to predict and prevent proximal junctional kyphosis and failure in adult spinal deformity surgery: A systematic review

Introduction

Adult spinal deformity (ASD) surgery involves high costs and risks, with Proximal Junctional Kyphosis (PJK) and Proximal Junctional Failure (PJF) being major concerns. Artificial intelligence (AI) and machine learning (ML) offer potential in predicting and preventing these complications. This review examines the role of AI in predicting PJK/PJF, its effectiveness, and future research needs.

Research question

Can AI-based models accurately predict PJK/PJF after ASD surgery, and what factors affect their performance?

Material and methods

A systematic review was conducted following PRISMA guidelines, analyzing Medline, Scopus, Embase, and Cochrane Library databases up to December 2024. Keywords included “Adult Spinal Deformity,” “PJK,” “PJF,” “AI,” and “ML.” Data extracted included study characteristics, patient demographics, surgical details, AI model parameters, and performance metrics. Bias risk was assessed using the MINORS score.

Results

Among 164 studies, 7 met inclusion criteria (n = 2179 patients). Mean age was 63.2 ± 3.7 years, BMI 26.1 ± 2.4 kg/m², and fusion levels 9.82 ± 1.8. PJK/PJF occurred in 41.1 %. AI models (Random Forest, supervised learning) had accuracy from 72.5 % to 100 % (AUC up to 1.0). Key predictors included age, BMD, spinal alignment, and implant type.

Discussion and conclusions

AI and ML models show promise in predicting PJK/PJF after ASD surgery. However, larger multicenter studies with standardized definitions, BMD assessments, and preoperative MRI integration are needed for broader clinical application and validation.