Answers are grounded in this paper and cite their sources.
A clinical dataset of patients with degenerative scoliosis (DS) was constructed based on clinical data, and the current status and related factors of DS in this dataset were analyzed. Multiple machine learning methods were used to establish candidate risk assessment models for DS and perform internal performance evaluation. Finally, corresponding nomograms were developed as preliminary visualization tools for future validation. This retrospective study enrolled patients with chronic low back pain treated at Zhongda Hospital, Southeast University, from January 2023 to January 2025. Cross-sectional area (CSA) and lean CSA/gross CSA (LCSA/GCSA) of the multifidus and erector spinae were measured on T2-weighted MRI at L3–S1 mid-disc levels using ImageJ; Surgimap was used to measure facet joint angle (FJA), facet overhang (FO) length, and facet joint space width (FJSW). Baseline and laboratory data were collected from the Hospital Information System and Laboratory Information System. After excluding highly redundant variables (ρ > 0.90) and those with variance inflation factor > 5, LASSO regression and feature importance analysis were applied to identify candidate predictors. Multiple machine learning algorithms, including Firth logistic regression and random forest, were employed to construct models for assessing DS risk and severity. Candidate models were compared by internal resampling and selected based on discrimination, calibration, optimism, net benefit, parsimony, and interpretability. Performance measures included accuracy, precision, recall, F1, and AUC. Models for nomogram visualization were selected by considering predictive performance together with interpretability and feasibility of graphical implementation, rather than the highest performance estimate alone. A total of 462 patients (300 DS, 162 non-DS) were enrolled; the DS group was further stratified into mild ( n = 144), moderate ( n = 112), and severe ( n = 44) scoliosis. Between-group comparison revealed significant differences in 36 parameters ( P < 0.05): DS patients were older, more often female, had lower bone mineral density, and exhibited more severe paraspinal muscle atrophy, facet joint and disc degeneration, and spinopelvic malalignment. Cobb angle showed strong positive correlation with FJA, moderate positive correlations with FO length, FJSW, and Facet Tropism (FT), and moderate negative correlations with LCSA/GCSA, CSA, and Sagittal Range of Motion (sROM). DS severity was associated with advancing age, diabetes, and progressive declines in nutritional and hematological markers, alongside worsening imaging phenotypes and sagittal imbalance. Significant concave–convex differences in muscle and facet parameters were present in mild-to-moderate DS but largely disappeared in severe DS. Multivariable analyses showed that reduced paraspinal muscle CSA at L5–S1 was independently associated with DS status, whereas greater L4–L5 facet joint angle, preoperative facet tropism, T1 pelvic angle, female sex, segmental concave–convex degeneration parameters, and reduced preoperative sagittal range of motion were independently associated with DS severity. Among the candidate classifiers, Firth logistic regression showed the most favorable overall internal performance for assessing prevalent DS status (AUC = 0.931; bootstrap-corrected AUC = 0.923; sensitivity = 0.900; specificity = 0.864). For DS severity, the parsimonious linear regression model (R² = 0.771, RMSE = 3.96, MAE = 3.06) was selected for nomogram development over more complex algorithms with higher development-set fit (SVM R² = 0.886; XGBoost R² = 0.997), in order to prioritize interpretability, robustness against overfitting, and clinical usability. Based on these two selected models, two nomogram visualization tools were developed. This study demonstrated that, compared with non-DS patients, middle-aged and older DS patients exhibit more severe paraspinal muscle atrophy, facet joint and disc degeneration, sagittal imbalance, and lower bone mineral density, all of which worsen with disease progression. The constructed multimodal assessment models for DS status and severity showed good discriminability in the development dataset; after external validation, the derived nomograms may serve as practical clinical tools.
No takes yet. Share an insight, caveat, or question.
Ji et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: