A Predictive Model of Progression for Adolescent Idiopathic Scoliosis Based on 3D Spine Parameters at First Visit

Marie-Lyne Nault, MD, PhD; Marie Beauséjour, PhD; Marjolaine Roy-Beaudry, MSc; Jean-Marc Mac-Thiong, MD, PhD; Jacques de Guise, PhD; Hubert Labelle, MD; Stefan Parent, MD, PhD


Spine. 2020;45(9):605-611. 

In This Article



Of 195 consecutive AIS patients enrolled in the cohort, 21 were lost to follow-up, and two had calibration errors on imaging that precluded 3D reconstruction; leaving 172 patients to be analyzed (88%).

Descriptive characteristics of the cohort at first visit were 20 male (11.6%) and 152 female (88.4%) with mean age of 12.5 ± 1.3 years. Seventy-four patients were Stage 0, Risser 0-triradiate cartilage open (43.0%), and 98 patients with Stage 1, Risser 0, and triradiate closed or Risser 1 (57.0%). The mean initial Cobb angle was 22.6° ± 7.8°. The average follow-up was 3.2 ± 1.2 years and the final Cobb angle was 27.3° ± 14.1°. Forty-nine patients had Right Thoracic curves (28%), 36 Right Thoracic and Left lumbar (21%), 46 left thoracolumbar (27%), and 41 patients with other types (24%). A total of 104 (60%) patients followed a brace treatment.

Pearson's and Spearman's correlations with final Cobb angle were done for 41 spinal descriptors. Of these, 23 parameters resulted in a correlation associated with a P value < 0.1 (Table 1).

On GLM analysis, significant predictors of the final Cobb angle were: skeletal maturation status, curve type, 2D initial Cobb angle, angle of the plane of maximal curvature, disk wedging of T3-T4 and T8-T9 (Table 2). The interaction between skeletal maturation status and curve type was also statistically significant. The R2 of this predictive model was 0.643 (adjusted R2 = 0.618).

The model is parcimonious and it features normal distribution of residuals, less than 5% residuals >|1.96| and Cook's distances << 1.0.

The Bland–Altman plot of differences between the predicted and observed values and their means is shown in Figure 2. The global mean of differences is not statistically different from 0.0 (bias is 0), with a standard deviation of 8.41. The regression test showed a low but statistically significant proportional bias, suggesting that the error of prediction increases with increasing Cobb angle. The mean standard error of prediction is 2.3° ± 0.3° (min = 1.7°, max = 3.9°). For example, to detect a final Cobb angle at maturity over 35° (after removing cases with initial Cobb between 35° and 40°), the model showed in our sample, a sensitivity of 75% and a specificity of 94%, leading to a positive predictive value of 79% and a negative predictive value of 94%, for an overall agreement of 91%.

Figure 2.

Bland–Altman Plot: Differences between observed and predicted values as a function of their means. Reference lines indicate mean of differences and 95% CI of limits of agreement (mean of differences ± 1.96 standard deviation of differences).