Mortality prediction and long-term outcomes for civilian cerebral gunshot wounds: A decision-tree algorithm based on a single trauma center.
2020
Abstract Gunshot wounds (GSW) are one of the most lethal forms of head trauma. The lack of clear guidelines for civilian GSW complicates surgical management. We aimed to develop a decision-tree algorithm for mortality prediction and report long-term outcomes on survivors based on 15-year data from our level 1 trauma center. We retrospectively reviewed 96 consecutive patients who presented with cerebral GSWs between 2003 and 2018. Clinical information from our trauma database, EMR, and relevant imaging scans was reviewed. A decision-tree model was constructed based on variables showing significant differences between survivors and non-survivors. After excluding patients who died at arrival, 54 patients with radiologically confirmed intracranial injury were included. Compared to survivors (51.9%), non-survivors (48.1%) were significantly more likely to have perforating (entry and exit wound), as opposed to penetrating (entry wound only), injuries. Bi-hemispheric and posterior fossa involvement, cerebral herniation, and intraventricular hemorrhage were more commonly present in non-survivors. Based on the decision-tree, Glasgow Coma Scale (GCS) > 8 and penetrating, uni-hemispheric injury predicted survival. Among patients with GCS ≤ 8 and normal pupillary response, lack of 1) posterior fossa involvement, 2) cerebral herniation, 3) bi-hemispheric injury, and 4) intraventricular hemorrhage, were associated with survival. Favorable long-term outcomes (mean follow-up 34.4 months) were possible for survivors who required neurosurgery and stable patients who were conservatively managed. We applied clinical and radiological characteristics that predicted survival to construct a decision-tree to facilitate surgical decision-making for GSW. Further validation of the algorithm in a large patient setting is recommended.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
38
References
5
Citations
NaN
KQI