1. Statistical learning of blunt cerebrovascular injury risk factors using the elastic net.
- Author
-
Cooper ME, Risk B, Corey A, Fountain AJ, and Allen JW
- Subjects
- Humans, Retrospective Studies, Risk Factors, Vertebral Artery diagnostic imaging, Vertebral Artery injuries, Carotid Artery Injuries, Cerebrovascular Trauma, Wounds, Nonpenetrating diagnostic imaging
- Abstract
Purpose: To compare logistic regression to elastic net for identifying and ranking clinical risk factors for blunt cerebrovascular injury (BCVI)., Materials and Methods: Consecutive trauma patients undergoing screening CTA at a level 1 trauma center over a 2-year period. Each internal carotid artery (ICA) and vertebral artery (VA) was independently graded by 2 neuroradiologists using the Denver grading scale. Unadjusted odds ratios were calculated by univariate and adjusted odds ratios by multiple logistic regression with FDR correction. We applied logistic regression with the elastic net penalty and tenfold cross-validation., Results: Total of 467 patients; 73 patients with BCVI. Maxillofacial fracture, basilar skull fracture, and GCS had significant unadjusted odds ratios (OR) for ICA injury and C-spine fracture, spinal ligamentous injury, and age for VA injury. Only transverse foramen fracture had significant adjusted OR for VA injury, with none for ICA injury, after FDR correction. Using elastic net, ICA injury variables included maxillofacial fracture, basilar skull fracture, GCS, and carotid canal fracture. For VA injury, these included cervical spine transverse foramen fracture, ligamentous injury, C1-C3 fractures, posterior element fracture, and vertebral body fracture., Conclusion: Elastic net statistical learning methods identified additional risk factors and outperformed multiple logistic regression for BCVI. Elastic net allows the study of a large number of variables, and is useful when covariates are correlated., (© 2021. American Society of Emergency Radiology.)
- Published
- 2021
- Full Text
- View/download PDF