Comparison of two clustering methods on surgical patients with adult spinal deformity: Importance of the variable choice on the obtained results and their interpretation

Larrieu D., Baroncini A., Bourghli A., Pizones J., Kleinstueck F. S., ALANAY A., ...Daha Fazla

Brain and Spine, cilt.6, 2026 (ESCI, Scopus) identifier identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 6
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.bas.2025.105904
  • Dergi Adı: Brain and Spine
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), Scopus, BIOSIS, EMBASE, Directory of Open Access Journals
  • Anahtar Kelimeler: Adult spine deformity, Clustering, Health-related quality of life, Machine learning, Spine surgery
  • Acıbadem Mehmet Ali Aydınlar Üniversitesi Adresli: Evet

Özet

Introduction: Clustering techniques can reveal patterns in complex datasets and enable further statistical analysis, but outcomes may vary based on variable selection. Research question: Does the choice of input variables affect clustering results in patients undergoing surgery for adult spinal deformity (ASD)? Materials and methods: Hierarchical clustering was applied using two variable sets: C16 (16 variables including demographic, radiographic and quality-of-life metrics) and C12 (12 primarily radiographic variables). Results: Data from 784 patients were analyzed. Both C16 and C12 identified three clusters. In Cluster 1, C16 included younger idiopathic scoliosis patients (age 29.42 ± 11.69 years), while C12 grouped slightly older patients (35.77 ± 15.44 years) with similar sagittal alignment and Cobb angles, but C16 had better quality of life (inverse ODI: 82.24 ± 11.90 vs 74.50 ± 16.90). Cluster 2 included patients with sagittal malalignment and moderate disability, showing similar demographics and ODI, but differing in radiographic features such as Cobb angle (41.39° vs 36.40°), coronal balance (22.12 mm vs 18.66 mm), and lumbar lordosis index (0.77 vs 0.71). Cluster 3 captured patients with severe sagittal malalignment and greater disability. Here, C12 showed more pronounced malalignment (global tilt: 47.45° vs 39.81°), but better quality of life (inverse ODI: 45.94 vs 41.41). The PCA revealed that clustering was driven by quality-of-life metrics in C16 and by radiological parameters in C12. Discussion and conclusion: both algorithms identified similar cluster numbers and profiles, but the dominant clustering variables differed, highlighting the need to align variable selection with specific study goals.