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


This study identified new predictors of progression for patients with mild to moderate AIS. Using only the information available at first visit, the prediction model was able to explain 64.3% of the variance in the final Cobb angle at skeletal maturity. As in previous reports, initial Cobb angle, skeletal maturation, and type of curvature were expected determinants of the final Cobb angle. The addition of the plane of maximal curvature and of two specific 3D disk wedging levels (T3-T4, T8-T9) further refined the overall prediction. This new predictive model should help assess patients with AIS and tailor their treatment plan accordingly.

Using the model for a new patient would require the clinicians to upload standing full spine x-rays (EOS or standard x-rays with a calibration object) and standard clinical data in the shared decision-making tool currently being developed and validated. The program will compute all relevant 3D geometrical indices before processing these values into the implemented model for an accurate prediction of Cobb at maturity. For example, if a 10-year-old presents with a 20° curve, the model would propose a final Cobb angle, in one scenario, the model could propose a 3D Cobb of 22° at maturity. In this case, the clinician may conclude that the curve is stable and does not require bracing or active treatment, thus decreasing the radiation currently imposed on the patient. In an alternative scenario, the model could predict progression to a final deformity of 55°, prompting the clinician to discuss the benefits of bracing and may propose a less-invasive and more growth-friendly approach prior to skeletal maturity. This would lead to patients benefiting from a more personalized treatment approach, with less inadequate use of braces and less invasive fusion procedures. An example of the clinical application is presented in Figure 3, in which two skeletally immature patients were seen for the first time at the clinic with a similar curve pattern and comparable Cobb angle (26° and 29°). At skeletal maturity, the Cobb angle in Patient 2 remained stable (29°), whereas Patient 1 progressed to 50°. The predictive model led to excellent predictions of 29° for Patient 2 and 48° for Patient 1.

Figure 3.

Radiographs at first visit and 35 months later. Patients had similar initial. The model accurately predicted a final Cobb angle of 29° and 48°, respectively, using the 3D descriptors available at first visit. Patients' characteristics at first visit: Patient 1 (Risser 0, triradiate closed, thoracic rib hump of 12°, positive family history, brace treatment and referral for surgery) and Patient 2 (Risser 0, triradiate closed, thoracic rib hump of 5°, no family history, brace treatment).

When performing a multivariate analysis on our data using the classic risk factors[3] only, an R2 of 0.476 is obtained. This value is remarkably lower than that of the final prediction model (R2 = 0.643) presented in our study. In terms of clinical usefulness, this improvement in quality of adjustment in the model leads to better discrimination of patients in need of orthopedic management and more confidence in keeping nonprogressive curves under observation only. Indeed, classification of patients using the classic risk factors reduces the sensitivity to 70% and the positive predictive value to 68% (Cobb>35°).

The angle of the plane of maximal curvature is an important 3D parameter. Villemure et al[12] found a tendency for the angle to increase with increasing Cobb angle in the frontal plane, but this is the first study to demonstrate that this parameter is a risk factor for AIS progression. It is associated with rotation of the curve and may be more sensitive for detecting AIS progression than the traditional Cobb angle.

Until now, apical vertebral body wedging had been associated with Cobb angle progression.[12,16] It is noteworthy, however, that apical vertebral body wedging was not identified as a predictor in this study. The fact that disc wedging at levels T3-T4 and T8-T9 was predictive supports the hypothesis that wedging begins in the disks and then, following the Hueter–Volkmann principle, appears in the vertebral body as the deformation and skeletal maturation progress with asymmetrical loading.[17]

The prospective study managed by a tertiary care scoliosis referral center provided a large sample of participants and a case mix representative of what we would expect in specialized clinics, favoring generalizability of results. Our target sample size was determined considering both the recruitment potential in our institution and general criteria on the number of study units compared to the number of degrees of freedom corresponding to the potential predictors to be tested in the model. In addition, with classical predictors we expected a R2 around 0.45 and we expected that the addition of our 3D potential additional predictors with interaction term would add about 0.15 to the R2 (effect size = 0.27). The inclusion of 195 patients with 20% expected lost to follow-up would have ensured a power over 0.95 in the refined linear model, with a nominal significance level of 0.05.

One limitation of this study is the inclusion of all curve types in the model. Further research could assign AIS patients to subgroups and build a GLM models accordingly. Sample size in this study precluded subgroup analysis. Furthermore, this predictive model includes all patients, whether they received orthotic treatment or not. No adjustment in the analyses was made for brace wear, which could alter the distribution of final Cobb angle.[18]

One of the limitations in applying the findings of the current study is the availability of 3D reconstruction of the spine in the clinical setting. However, recent developments allow for more rapid 3D spine reconstruction with low radiation radiographs and high-quality images with improved clinical acceptance worldwide. Furthermore, AIS pathology is more and more focused on 3D definition and classification as well as 3D surgical treatment correction. It seems logical that 3D-based evaluation will be part of the clinical setting in the near future, even though the precision is not perfect; it is a better prediction tool to help patients' management. The model will help tailor the follow-up interval for each patient based on its own progression risk. The physician will still have the liberty to monitor the patient's progression at regular intervals to limit the risk of a false negative.

The next step is to assess the external validity of the model by testing it on an international prospective AIS cohort. The findings of the current study are a major advancement for AIS patients and families. The addition of 3D spine parameters to the prediction model has the potential to change the management of patients with AIS by identifying patients at risk of significant progression and those at low risk of progression. Having a better understanding of the deformity at maturity will help modulate treatment strategies and follow up intervals adapted to the risk of progression.