Algorithmic fairness reporting in artificial intelligence for spinal deformity: a systematic review

Artificial intelligence (AI) and machine learning (ML) are increasingly studied for diagnosis, surgical planning, risk stratification, and outcome prediction in spinal deformity. However, these models may reproduce inequities when datasets, outcome definitions, or validation strategies incompletely represent diverse populations. Given the rapid growth of AI in spinal deformity, there is a need to evaluate how algorithmic fairness is reported and assessed.

A systematic review was performed in accordance with PRISMA guidelines. PubMed, Embase, Cochrane, Scopus, and Web of Science were searched in May 2026 for studies applying AI or ML to spinal deformity care. Studies were evaluated for application domain, population, data modality, model type, validation strategy, demographic reporting, subgroup performance, calibration, and fairness audit or mitigation.

Fifty-eight studies were included: 33 image-based diagnosis or measurement studies (57%) and 25 clinical prediction or decision-support studies (43%). Age and sex or gender were commonly reported but rarely used for subgroup evaluation. Race, ethnicity, or skin tone were reported in 8 studies (14%), and socioeconomic or insurance-related variables in 1 study (2%). Fairness-aware evaluation was uncommon: 2 studies (3%) reported subgroup performance analyses, none evaluated subgroup calibration, and 2 studies (3%) described formal fairness audit or mitigation. Despite increasing technical sophistication, evaluation of demographic representativeness, subgroup reliability, and algorithmic equity remained limited.

AI applications in spinal deformity show promise across diagnosis, measurement, surgical planning, prognosis, and postoperative prediction. Before clinical implementation, studies should improve demographic reporting, test subgroup performance and calibration, and evaluate whether model outputs perform equitably across clinically relevant populations.