Measurement-level clustering of ligamentum flavum thickness identifies distinct phenotypes associated with lumbar lordosis in adult spinal deformity: a proof-of-concept study

Study design Retrospective cohort study. Objectives To identify distinct ligamentum flavum (LF) thickness phenotypes in adult spinal deformity (ASD) using measurement-level clustering and to characterize their association with traditional spinopelvic parameters in a proof-of-concept analysis. Methods LF thickness was measured on preoperative MRI in 44 consecutive ASD patients (2018–2023), yielding 1,250 measurements; complete clinical and radiographic data were available in 40. K-means clustering ( k = 3) was performed at the measurement level, treating thickness regimes as exploratory and unsupervised. Patients were assigned to a dominant phenotype, and all inferential comparisons were performed at the patient level. Between-phenotype differences in spinopelvic parameters were assessed with independent-samples t -tests and Mann–Whitney U -tests, with Hedges g effect sizes. Preliminary clinical and radiological outcomes were reported descriptively. Results Three measurement-level thickness regimes were identified: Low LF (2.21 ± 0.51 mm, n = 619 measurements), Medium LF (3.92 ± 0.56 mm, n = 499), and High LF (6.11 ± 0.90 mm, n = 132). High-cluster measurements were distributed across 32 patients, but only one patient had High LF as a dominant phenotype and was excluded from comparisons (43 patients analyzed; 25 Low, 18 Medium). Lumbar lordosis differed significantly between Low and Medium phenotypes (Low: −44.6 ± 12.0° vs. Medium: −32.5 ± 8.9°; p = 0.003; Hedges g = −1.10, 95% CI −1.81 to −0.39), a large effect that was robust to nonparametric testing. Cluster purity was higher in Low LF (61.9% vs. 53.3%, p = 0.027; g = 0.70). Principal component analysis of standard parameters showed that mean LF thickness was moderately correlated with the dominant pelvic-morphology component ( r = −0.44, p = 0.008), indicating that LF thickness is partially related to, rather than independent of, traditional parameters. Intra-rater reliability, assessed by blinded re-measurement of 15 patients, was moderate [ICC(2,1) = 0.632, 95% CI 0.242–0.855; Cohen kappa for phenotype = 0.524], and sensitivity analysis showed that the phenotype-lordosis association does not survive substitution of the re-read values ( p = 0.005 to p = 0.245). In contrast, the continuous rank correlation does (rho = + 0.496, p = 0.0015). Conclusions Measurement-level clustering identifies distinct LF thickness phenotypes in ASD that are associated with lumbar lordosis. These exploratory findings establish a methodological foundation and motivate larger, outcome-based studies with formal measurement-reliability assessment to determine clinical utility.