Cone-Beam Computed Tomography-Based Three-Dimensional Phenotypes of Skeletal Class II Malocclusion in Yemeni Adults: A Principal Component and Cluster Analysis

Introduction and aims: Skeletal Class II malocclusion is heterogeneous, and conventional two-dimensional cephalometry may not fully capture relevant three-dimensional (3D) craniofacial variation. This study aimed to identify 3D skeletal phenotypes of Class II malocclusion in Yemeni adults using cone-beam computed tomography (CBCT) and multivariate analysis. Methods: This retrospective observational study included CBCT scans of 120 Yemeni adults with skeletal Class II malocclusion (56 males, 64 females; age range 16–30 years; mean 23.1 ± 3.5 years) retrieved from the archives of the Department of Orthodontics, Faculty of Dentistry, Sana’a University. CBCT-derived cephalometric measurements were obtained from lateral reconstructions using Dolphin Imaging (version 12.0 Premium). Twenty prespecified skeletal variables were standardised and analysed by principal component analysis (PCA). Seven components were retained and entered into Ward’s hierarchical cluster analysis. The optimal cluster solution was selected using dendrogram inspection, agglomeration coefficients, average silhouette, and clinical interpretability. Canonical discriminant analysis with leave-one-out classification was used for internal validation, and cluster distribution by sex was assessed using chi-square testing. Results: The seven retained principal components explained 64.4% of the variance in the 20 skeletal variables. Cluster analysis identified five skeletal Class II phenotypes (Cluster 1, n = 34; Cluster 2, n = 24; Cluster 3, n = 32; Cluster 4, n = 24; Cluster 5, n = 6) representing different combinations of maxillary position, mandibular retrusion, mandibular plane angulation, and facial height pattern. The five-cluster solution showed acceptable internal separation (average silhouette = 0.253). Internal validation demonstrated robust discrimination, with an overall leave-one-out correct classification rate of 99.2%. Cluster distribution did not differ significantly by sex (chi-square P = .921). Conclusion: CBCT-based multivariate analysis identified five distinct skeletal phenotypes among Yemeni adults with skeletal Class II malocclusion, highlighting the heterogeneity of this condition and the limitations of relying only on sagittal descriptors such as ANB and Wits appraisal. Phenotype-specific 3D characterisation may support more individualised treatment planning and may provide clinically interpretable labels for future machine-learning studies. Clinical relevance: Phenotype-specific 3D characterisation of skeletal Class II malocclusion may support individualised treatment planning (eg, growth modification, camouflage, or orthognathic approaches). The resulting cluster labels may also serve as clinically interpretable phenotypic descriptors for future machine-learning studies, but no AI modelling was performed in the present work.