Can unsupervised cluster analysis identify patterns of complex adult spinal deformity with distinct perioperative outcomes?

On behalf of The International Spine Study Group, Renaud Lafage, Mitchell S. Fourman, Justin S. Smith, Shay Bess, Christopher I. Shaffrey, Han Jo Kim, Khaled M. Kebaish, Douglas C. Burton, Richard Hostin, Peter G. Passias, Themistocles S. Protopsaltis, Alan H. Daniels, Eric O. Klineberg, Munish C. Gupta, Michael P. Kelly, Lawrence G. Lenke, Frank J. Schwab, Virginie Lafage

Research output: Contribution to journalArticlepeer-review


OBJECTIVE The objective of this study was to use an unsupervised cluster approach to identify patterns of operative adult spinal deformity (ASD) and compare the perioperative outcomes of these groups. METHODS A multicenter data set included patients with complex surgical ASD, including those with severe deformities, significant surgical complexity, or advanced age who underwent a multilevel fusion. An unsupervised cluster analysis allowing for 10% outliers was used to identify different deformity patterns. The perioperative outcomes of these clusters were then compared using ANOVA, Kruskal-Wallis, and chi-square tests, with p values < 0.05 considered significant. RESULTS Two hundred eighty-six patients were classified into four clusters of deformity patterns: hyper-thoracic kyphosis (hyper-TK), severe coronal, severe sagittal, and moderate sagittal. Hyper-TK patients had the lowest disability (mean Oswestry Disability Index [ODI] 32.9 ± 17.1) and pain scores (median numeric rating scale [NRS] back score 6, leg score 1). The severe coronal cluster had moderate functional impairment (mean physical component score 34.4 ± 12.3) and pain (median NRS back score 7, leg score 4) scores. The severe sagittal cluster had the highest levels of disability (mean ODI 49.3 ± 15.6) and low appearance scores (mean 2.3 ± 0.7). The moderate cluster (mean 68.8 ± 7.8 years) had the highest pain interference subscores on the Patient-Reported Outcomes Measurement Information System (mean 65.2 ± 5.8). Overall 30-day adverse events were equivalent among the four groups. Fusion to the pelvis was most common in the moderate sagittal (89.4%) and severe sagittal (97.5%) clusters. The severe coronal cluster had more osteotomies per case (median 11, IQR 6.5-14) and a higher rate of 30-day implant-related complications (5.5%). The severe sagittal and hyper-TK clusters had more three-column osteotomies (43% and 32.3%, respectively). Hyper-TK patients had shorter hospital stays. CONCLUSIONS This cohort of patients with complex ASD surgeries contained four natural clusters of deformity, each with distinct perioperative outcomes.

Original languageEnglish
Pages (from-to)547-557
Number of pages11
JournalJournal of Neurosurgery: Spine
Issue number5
StatePublished - May 2023


  • adult spine deformity
  • artificial intelligence
  • cluster analysis
  • multicenter database
  • perioperative outcomes


Dive into the research topics of 'Can unsupervised cluster analysis identify patterns of complex adult spinal deformity with distinct perioperative outcomes?'. Together they form a unique fingerprint.

Cite this