Predicting Adolescent Idiopathic Scoliosis among Chinese Children and Adolescents

Abstract

Objective. Adolescent idiopathic scoliosis (AIS) affects 1%-4% of adolescents in the early stages of puberty, but there is still no effective prediction method. This study aimed to establish a prediction model and validated the accuracy and efficacy of this model in predicting the occurrence of AIS. Methods. Data was collected from a population-based school scoliosis screening program for AIS in China. A sample of 884 children and adolescents with the radiological lateral Cobb angle≥10° was classified as an AIS case, and 895 non-AIS subjects with a Cobb angle<10° were randomly selected from the screening system. All selected subjects were screened by visual inspection of clinical signs, the Adam’s forward-bending test (FBT), and the measurement of angle of trunk rotation (ATR). LR and receiver operating characteristic (ROC) curves were used to preliminarily screen the influential factors, and LR models with different adjusted weights were established to predict the occurrence of AIS. Results. Multivariate LR and ROC curves indicated that angle of thoracic rotation (adjusted odds ratios AOR=5.18−10.06), angle of thoracolumbar rotation (AOR=4.67−7.22), angle of lumbar rotation (AOR=6.97−8.09), scapular tilt (area under the curve AUC=0.77, 95% CI: 0.75-0.80), shoulder-height difference, lumbar concave, and pelvic tilt were the risk predictors for AIS. LR models with different adjusted weights (by AOR, AUC, and AOR+AUC) performed similarly in predicting the occurrence of AIS compared with multivariate LR. The sensitivity (82.55%-83.27%), specificity (82.59%-83.33%), Youden’s index (0.65-0.67), positive predictive value (82.85%-83.58%), negative predictive value (82.29%-83.03%), and total accuracy (82.57%-83.30%) manifested that LR could accurately identify patients with AIS. Conclusions. LR model is a relatively high accurate and feasible method for predicting AIS. Increased performance of LR models using clinically relevant variables offers the potential to early identify high-risk groups of AIS.

Development and validation of machine learning-based models for prediction of adolescent idiopathic scoliosis: A retrospective study

“…Fortunately, a few studies have made efforts to explore the early warning of AIS risk. For example, Yan et al built a prediction model for AIS based on binary linear regression algorithm, and the prediction accuracy can reach 82.57% to 83.30% [ 12 ] ; Xu et al developed a genetic prediction model that can be used to assess the discrimination between AIS children and adolescents and normal control groups. [ 28 ] Nault et al established a prediction model using a general linear method, taking 3D spinal parameters and clinical parameters as predictors, and found that the positive predictive value of the model was 79% (provided that the Cobb angle was 35° as the critical point).…”

“…In this study, the candidate variables screened by various machine learning algorithms showed that ROSHTSH, AOLR, ST, SHD, LC, PT, and AOTR were the most valuable candidate variables for AIS prediction model, which was consistent with previous research results. [ 12 ] Interestingly, previous studies have shown that the height of children and adolescents with scoliosis increases rapidly, with the average height growth rate of female AIS children and adolescents peaking at 8.1 cm/year, while the average height growth rate of normal female children and adolescents of the same age peaking at 7.1 cm/year. [ 1 , 30 ] Therefore, it is believed that the growth and development of AIS children and adolescents at the peak of growth and development is faster than that of normal children and adolescents of the same age.…”

“…As for AOLR, ST, SHD, LC, PT, and AOTR, some researchers speculate that changes in the scapula and lumbar spine may be considered as adaptive compensation or muscle activation strategies for AIS children and adolescents, which is consistent with the results of this study. [ 12 , 31 , 32 ] We are encouraged that these follow-up variables are relatively easy to obtain and have no negative effects such as radiation, so we can often have more advantages in the process of building the AIS early warning model.…”

“…The postural evaluation of participants mainly comes from visual inspection, Adam FBT and angle of trunk rotation to obtain correlation coefficients, namely: Ratio of sitting height to standing height (ROSHTSH), angle of lump rate (AOLR), scapular tilt (ST), shoulder height difference (SHD), lump construct (LC), pelvic tile (PT), angle of thoracolumbar rotation (AOTR), thoracic kyphosis, flat back (FB), and lumbar kyphosis. [ 12 ] Among them, for the evaluation and measurement of the above indicators, in order to ensure the quality and reliability of the data obtained, all subjects were independently evaluated by 2 experts, and then the data were verified by a third expert. If there was data bias or error, the 3 experts jointly supervised the repeated testing, and finally the data measured by the 3 experts were included in the final study.…”

Sex differences in musculoskeletal injury and disease risks across the lifespan: Are there unique subsets of females at higher risk than males for these conditions at distinct stages of the life cycle?

“…A third category of MSK conditions that exhibit sex differences in incidence/prevalence are those not associated with any overt linkage to immune dysfunction or obvious injury event. An example of such a condition is adolescent idiopathic scoliosis, a condition affecting 1%–4% of adolescents ( Penha et al, 2018 ; Yan et al, 2020 ), and a condition resulting in spinal deformities of varying extent, with onset in the post-puberty environment of growth and maturation ( Chung et al, 2020 ; Liang et al, 2021 ). It occurs mostly in young females ( Chung et al, 2020 ; Yan et al, 2020 ; Liang et al, 2021 ), with a F/M ratio of up to 10/1 for those with curvatures of more than 30° ( Raggio, 2006 ).…”