Functional evaluation in Adult Spinal Deformity can be reduced to three key kinematic parameters that can sufficiently predict health-related quality of life outcomes: an observational prospective study.

Authors

Abstract

Recently, functional evaluation using 3D gait analysis (3DGA) proved to predict health-related quality-of-life (HRQOL) scores better than static radiographic evaluation in adult spinal deformity (ASD). However, 3DGA provides multiple parameters that can be a burden to interpret by non-experts. A recent study showed that the dynamic pelvic tilt (dPT), the forward projection of the head and thorax (dODHA) and walking step length (SL) are the most representative gait kinematics in ASD patients. To determine whether reducing kinematic parameters to only these 3 key parameters would still predict HRQOL outcomes in ASD based on machine learning (ML) random forest regression model. Single-center prospective study. 197 patients with ASD and 57 control subjects OUTCOME MEASURES: Self-report measures: SF36 with the physical and mental components (PCS & MCS), Oswetry Disability Index (ODI), Beck's depression inventory (BDI) and Visual analogue scale (VAS) for pain. Physiologic measures: low-dose full-body biplanar Xrays with 3D skeletal reconstructions. Functional measures: full-body 3D gait analysis during walking. Prediction accuracy: random forest regression ML model. All subjects underwent low-dose full-body biplanar Xrays with 3D skeletal reconstructions (with the calculation of spino-pelvic and global alignment parameters), full-body 3DGA during walking (with the calculation of full-body joint kinematic parameters), and completed HRQOL questionnaires: SF36 with the physical and mental components (PCS&MCS), ODI, BDI and VAS for pain. A random forest regression machine learning model was used to predict HRQOL scores in 4 simulations: (Sim-1) X-ray parameters (spinopelvic and global alignment); (Sim-2) Key-kinematic parameters (dPT, dODHA and SL); (Sim-3) X-ray parameters and dPT, dODHA and SL; (Sim-4) All-kinematic parameters. The prediction accuracy and root mean squared error (RMSE) were evaluated using a 10-fold cross-validation and compared between simulations. The same methodology was applied on a subset of 30 ASD patients followed (6 months to 2 years) after medical, orthopedic and surgical treatment. Simulations 1, 2, 3 and 4 had a median accuracy of 82, 85, 86 and 86%, respectively. Simulations 2, 3 and 4 had comparable accuracies of prediction for all HRQOL scores and higher predictions compared to Simulation 1 (i.e., accuracy for PCS=86±3 vs 90±2, 91±3% and 91±3% for simulations 1, 2, 3 and 4 respectively, p<0.05). Similar results were obtained for the 30 follwed-up ASD patients. Head and pelvis kinematics and step length are sufficient to predict HRQOL scores, even postoperatively, with higher accuracies than classic spinopelvic and global alignment parameters. While the latter play an integrating role in the surgical planning of ASD patients, coupling radiographic to only 3 key functional parameters would be optimal to provide a complete assessment and postoperative follow-up. Future technologies should focus on capturing these 3 parameters alone to allow surgeons to easily access functional assessment, bypassing the complexity of the complete gait analysis process.

Tags

Topics