Answers are grounded in this paper and cite their sources.
Study Design. A retrospective observational study. Objective. This study aimed to apply machine learning-based clustering to identify distinct phenotypes of DLS patients with diverse clinical profiles and recovery trajectories. Summary of Background Data. Patients with degenerative lumbar scoliosis (DLS) exhibit substantial heterogeneity in pathology, baseline impairments, frailty, and surgical expectations. However, existing classification systems primarily rely on radiographic parameters and fail to integrate multidimensional patient characteristics, limiting their ability to capture patient heterogeneity and guide individualized surgical strategies. Methods. Latent profile analysis (LPA) was performed to determine possible patient phenotypes based on a set of screened preoperative demographic, clinical, and radiographic variables. Goodness-of-fit indices were used to determine the optimal cluster number. Baseline and outcome differences among phenotypes were evaluated using analysis of variance and subsequent post hoc Tukey’s test. A phenotype classification tool was then established and deployed as a web-based application for clinical use. Results. A total of 257 DLS patients were included. Three distinct phenotypes were identified: (1) mild type (n=98, 38.1%), characterized by younger age, minimal deformity, and the most favorable recovery with the lowest dissatisfaction; (2) frailty-dominant type (n=95, 37.0%), defined by advanced age, higher frailty burden, slower recovery, and persistent postoperative pain; and (3) deformity-dominant type (n=64, 24.9%), marked by severe coronal and sagittal malalignment, greater surgical complexity, and the highest rate of postoperative residual coronal malalignment. Across all phenotypes, combined lateral-posterior approaches reduced blood loss and instrumented levels compared with posterior-only approaches. Notably, in frailty-dominant patients, combined approaches further improved early ambulation (46.9% vs. 17.5%, P =0.002) and reduced dissatisfaction (21.9% vs. 44.4%, P =0.031). The phenotype classification tool demonstrated good agreement with the LPA-derived phenotype assignments (F1-score: 0.866). Conclusions. Machine learning-based clustering identified three distinct phenotypes of DLS patients. This classification captures the multidimensional heterogeneity of patients with DLS and highlights the importance of phenotype-specific surgical strategies. The accompanying web-based classification tool offers a practical means to integrate these findings into routine clinical practice. Level of Evidence. III
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: