Predicting pulmonary function in adolescent idiopathic scoliosis: is a simple 2D radiograph still the winner?

While 3D radiographic parameters are increasingly used to assess adolescent idiopathic scoliosis (AIS), barriers to 3D imaging limit their routine use. 2D-to-3D prediction algorithms have been proposed as a substitute. This study sought to determine if a predicted 3D kyphosis, derived from 2D images, is a more robust predictor of FEV1 in patients with AIS than a traditional 2D analysis.

A retrospective, cross-sectional review of 259 AIS patients with surgical-range thoracic curves (> 40°) was performed. We built two multivariate linear regression models to predict FEV1. Due to structural multicollinearity, the deformity measures were tested separately. Model A included Main Thoracic Cobb, BMI, and Age. Model B included Predicted 3D T5-T12 Kyphosis, BMI, and Age. The models were compared using the Akaike Information Criterion (AIC) and adjusted.

The cohort (87% female, mean age 15.7 ± 3.3 years) presented with severe deformities (mean Main Thoracic Cobb 68.4° ± 16.5°) and widespread restrictive impairment (71%). Multivariate Analysis revealed that Model A, (2D Main Thoracic Cobb) was statistically superior to Model B (Predicted 3D T5 - T12 Kyphosis) with a lower AIC (-292.5 vs. -265.9) and a higher adjusted R² (0.241 vs. 0.159). All factors were significant independent predictors.

The predicted 3D-based model was not superior, while a parsimonious 2D-based multivariate model including Main Thoracic Cobb, BMI, and Age explained a significantly larger proportion of the variance in FEV1. This specific 2D-to-3D prediction algorithm is an imperfect proxy and is not a valid substitute for true 3D imaging in predicting pulmonary risk.