Diagnostic model for radiographic instability of L4-5 degenerative lumbar spondylolisthesis based on propensity score matching and LASSO regression Author’s List

Abstract

Abstract Background: This study aims to establish a diagnostic model for radiographic instability of L4-5 degenerative lumbar spondylolisthesis (DLS) based on propensity score matching (PSM) and least absolute shrinkage and selection operator (LASSO) regression. Method: This retrospective study included 163 patients diagnosed with DLS at the L4-5 level. Radiographic instability was defined as a relative translation of >8% and an intervertebral angulation > 10° on standing lateral flexion-extension radiographs. 62 pairs of individuals with stable and unstable DLS were matched by PSM to minimize the influence of confounding baseline characteristics. LASSO regression was performed to select the optimal combination of features. Finally, a diagnostic model for radiographic instability was constructed using multifactor binary logistic regression. The model's efficiency was evaluated by the area under the curve (AUC) of the receiver operating characteristic curve (ROC). Result: Patients in unstable group had a higher fat fraction (FF) of multifidus (MF) and erector spinae (ES) muscles and a thicker ligamentum flavum (LF). There was no significant difference between the two groups for MF and ES relative cross-sectional areas, total endplate scores and modified Pfirrmann grades. As for T2 histogram features of paraspinal muscle (PM), significant differences between the two groups were found for mean, variance, skewness, kurtosis, 10th, 50th and 90th percentiles. There was a linear correlation between relative translation and MF FF (r=0.517), ES FF (r=0.456). 58 features were extracted from MRI images and T2 histograms of PM, and five features including MF FF, sum ES variance, left MF kurtosis, left MF skewness and left MF variance were selected by LASSO regression. MF fat fraction (OR=1.394, P<0.001), ES variance sum (OR=1.001, P<0.05) and thickened LF(Y/N) (OR=4.892, P<0.05) were potential risk factors for unstable DLS, whereas left MF variance (OR=0.998, P<0.01) was protective feature for stable DLS. The AUC, sensitivity and specificity of the diagnostic model were 0.972, 86.46% (95%IC 81.72%-91.20%) and 95.19% (95%IC 92.23%-98.16%) respectively. Conclusion: FF and T2 histogram features of PM and LF morphology are valuable for lumbar dynamic instability. A diagnostic model based on these features in conventional MRI images and T2 histograms can evaluate radiological segmental stability of DLS.