DOI: 10.1097/pxr.0000000000000582 ISSN: 1746-1553

Machine learning–based prediction model for initial correction effect of Cheneau brace in adolescent idiopathic scoliosis

Kai-Hua Li, Li-Ning Zhang, Hui-Ling Xiong, Bao-Sheng Chen, Li-Sha Yi
Show PDF Cite

Background:

Predicting the initial correction rate (ICR) of Cheneau brace treatment for adolescent idiopathic scoliosis (AIS) is critical for personalized therapeutic planning but remains underexplored. This study aims to develop a clinical prediction model for ICR evaluation.

Methods:

A retrospective analysis included 391 spinal curves from 310 patients with AIS (since some patients have double curves) from 4 orthotic centers. Key variables encompassed demographics, axial trunk rotation (ATR), Risser sign, curve type (C-shaped/S-shaped), apical location (thoracic/lumbar), and prebrace Cobb angle. ICR was calculated as [(prebrace Cobb − postbrace Cobb)/prebrace Cobb] × 100%, with satisfactory correction defined as ICR ≥50%. Logistic regression and random forest models were trained (70% data) and validated (30%). Feature importance was assessed through mean decrease accuracy (MDA) and Gini (MDG). A nomogram integrated significant predictors.

Results:

Multivariate analysis identified Risser sign (OR = 0.033, P < 0.001), ATR (OR = 0.853, P < 0.001), curve type (OR = 0.485, P = 0.036), apical location (OR = 5.827, P < 0.001), and BMI (OR = 0.897, P = 0.041) as independent predictors. Random forest confirmed Risser sign as the most influential variable (MDA = 28.07%, MDG = 24.35%), followed by ATR (MDA = 21.99%, MDG = 19.50%). Both models demonstrated high predictive accuracy: logistic regression (AUC = 0.951, accuracy = 0.879) and random forest (AUC = 0.959, accuracy = 0.897). Lumbar curves and C-shaped curves exhibited superior correctability.

Conclusion:

Machine learning models effectively predict Cheneau brace ICR, with Risser sign and ATR angle as primary determinants. The developed nomogram provides a practical tool for pretreatment outcome anticipation, enhancing clinical decision-making in AIS management.

More from our Archive