A Preoperative Predictive Model of Lower Lumbar Spine Instability Based on Three‐Dimensional Computed Tomography: A Retrospective Case–Control Pilot Study

摘要

Objective This study aimed to build a predictive model of lower lumbar instability. Methods This retrospective study included 199 patients. Patients were divided into the lower lumbar instability group (LLIG) (n = 98) and lower lumbar stability group (LLSG) (n = 101). All participants of LLIG were recruited over a 2‐year period (2015–2017) from the patients who accept lumbar surgery at the First Hospital of Jilin University. The LLSG was selected from outpatients who had underwent lumbar spine computed tomography (CT) and Flexion and extension radiographs (FER) at the First Hospital of Jilin University from 2015 to 2017. Several lower lumbar parameters were measured, including Lordosis angle (LA), intervertebral height (IH), ratio of anterior height to posterior height (APR), angle between endplate and anterior edge of vertebral body (AEPVa), sagittal slip ratio (SSR), and angle between the upper endplate and z‐axis on sagittal plane (AUEZS). These parameters were keyed into the SPSS software to create a predictive model for classification. Sensitivity, specificity, predictive accuracy, and Kappa value were used to evaluate the predictive model. Results Compared with LLSG, the LA of LLIG decreased by 3.49° (126.54° vs 130.3°). Similarly, the IH of LLIG decreased by 1.23°mm, 1.66°mm, and 0.71°mm at L3‐4, L4‐5, and L5‐S1. Compared with LLSG, the SSR of LLIG is higher at L3‐4, L4‐5, and L5‐S1 (0.54 vs 0.51, 0.57 vs 0.46, and 0.59 vs 0. 47). Moreover, the APR of LLIG is higher than those of LLSG at L3‐4, L4‐5, and L5‐S1 (1.97 vs 1.81, 2.40 vs 1.97, and 2.69 vs 2.26). The LLIG has bigger AEPVa than LLIG at L3‐4, L4‐5, and L5‐S1. Compared with LLSG, the AUEZS of LLIG is bigger at L3‐4 (91.75° vs 90.81°) and smaller at L4‐5 and L5‐S1(84.63° vs 85.85° and 73.27° vs 75.01°). The SSR (L4) show highest predictive accuracy (83%) when every parameter was fed to LDA classifier to generate a univariate model. All parameters represent a statistically significant difference ( P ?0.05) between LLSG and LLIG. The model including LA, APR (L5‐S1), IH (L4‐5), SSR (L5), AUEZS (L5) has highest predictive accuracy of 88.2%. The sensitivity, specificity, and Kappa value are 88.7%, 93.1%, and 0.77. Conclusion The predictive model has good classification performance and can be an auxiliary tool for clinicians to evaluate lumbar instability in preoperative patients with severe pain aggravated by lumbar movement.