Use of Artificial Intelligence to Predict Complications in Degenerative Thoracolumbar Spine Surgery: A Systematic Review.

Abstract

Objective: We aim to conduct a systematic review of the literature to evaluate the effectiveness of artificial intelligence prediction models in predicting complications in adult patients undergoing surgery for degenerative thoracolumbar pathology compared with other commonly used prediction techniques.

Methods: A systematic literature review was conducted in Medline/Pubmed, Cochrane Library, and Lilacs/Portal de la BVS to identify machine learning models in predicting complications in patients undergoing surgery for degenerative thoracolumbar spine pathology between January 1, 2000, and May 1, 2023. The risk of bias was assessed using the PROBAST tool. Study characteristics and outcomes focusing on general or specific complications were recorded.

Results: A total of 2,341 titles were identified (763 were duplicates). Screening was performed on 1,578 titles, and 22 were selected for full-text reading, with 18 exclusions and 4 publications selected for the subsequent review. Additionally, 8 publications were included from other sources (Argentine Association of Orthopedics and Traumatology Library; manual citation search). In 5 (41.6%) articles, the effectiveness of artificial intelligence predictive models was compared with conventional techniques. All were globally classified as having a very high risk of bias. Due to heterogeneity in samples, outcomes of interest, and algorithm evaluation metrics, a meta-analysis was not performed.

Conclusion: Although the available evidence is limited and carries a high risk of bias, the studies analysed suggest that these models may achieve promising performance in predicting complications, with area under the curve values mostly ranging from acceptable to excellent.