Purpose: While the C-EOS system helps organize and classify Early Onset Scoliosis (EOS) pathology, it is not data-driven and does not help achieve consensus for surgical treatment. The current study aims to create an automated method to cluster EOS patients based on pre-operative clinical indices.
Methods: A total of 1114 EOS patients were used for the study, with the following distribution by etiology: congenital (240), idiopathic (217), neuromuscular (417), syndromic (240). Pre-operative clinical indices used for clustering were age, major curve (Cobb) angle, kyphosis, number of levels involved in a major curve (Cobb angle) and kyphosis along with deformity index (defined as the ratio of major Cobb angle and kyphosis). Fuzzy C-means clustering was performed for each etiology individually, with one-way ANOVA performed to assess statistical significance (p < 0.05).
Results: The automated clustering method resulted in three clusters per etiology as the optimal number based on the highest average membership values. Statistical analyses showed that the clusters were significantly different for all the clinical indices within and between etiologies. Link to the ACT-EOS web application: https://biomed.drexel.edu/labs/obl/toolkits/act-eos-application .
Conclusion: An automated method to cluster EOS patients based on pre-operative clinical indices was developed identifying three unique, data-driven subgroups for each C-EOS etiology category. Adoption of such an automated clustering framework can help improve the standardization of clinical decision-making for EOS.
Keywords: Automated; Classification; Clustering; Early onset scolosis; Fuzzy C-means.
© 2023. The Author(s).