Related Experiment Video
Updated: Aug 7, 2025
Assessment and Evaluation of the High Risk Neonate: The NICU Network Neurobehavioral Scale
Published on: August 25, 2014
Development and validation of a Bayesian network predicting neurosurgical intervention after injury in children and
Travis M Sullivan1, Genevieve J Sippel, Elizabeth A Matison
1From the Division of Trauma and Burn Surgery (T.M.S., G.J.S., E.A.M., W.V.G.-T., R.S.B.), Children's National Hospital, Washington, DC; Division of Electrical Engineering and Computer Science (D.O.), Massachusetts Institute of Technology, Boston, Massachusetts; Department of Neurological Surgery (C.O.), Children's National Hospital, Washington, DC; Department of Biomedical Informatics (P.E.D., T.D.B.), University of Colorado School of Medicine; Department of Pediatrics (P.E.D., T.D.B, M.A.C.) Children's Hospital of Colorado, Aurora, Colorado.
Insights
A new Bayesian network predicts the need for neurosurgical intervention in children with traumatic brain injury using four key factors available on hospital arrival. This tool aids in early risk stratification for improved outcomes.
Area of Science:
- Pediatric neurosurgery
- Traumatic brain injury management
- Clinical prediction modeling
Background:
- Timely surgical decompression is crucial for children with traumatic brain injury (TBI) and elevated intracranial pressure.
- Existing scoring systems for surgical decompression in TBI have limitations, including data requirements and accuracy issues.
- There is a need for accessible tools to identify pediatric TBI patients requiring neurosurgical intervention.
Purpose of the Study:
- To develop and validate a Bayesian network model for predicting the probability of neurosurgical intervention in pediatric TBI patients.
- To utilize readily available clinical and injury characteristics at hospital arrival for prediction.
- To address limitations of previous scoring systems in pediatric TBI management.
Main Methods:
- Utilized the 2017-2019 Trauma Quality Improvement Project database for patient and injury characteristics.
- Trained and validated a Bayesian network to predict neurosurgical intervention (craniotomy, craniectomy, ICP monitor placement).
- Evaluated model performance using ROC curves, calibration curves, and relative mutual information (RMI) for predictor importance.
Main Results:
- The final Bayesian network model incorporated Glasgow Coma Scale score (31.9% RMI), pupillary response (11.6% RMI), mechanism of injury (5.8% RMI), and prehospital CPR (0.8% RMI).
- The model demonstrated strong predictive performance with an area under the ROC curve of 0.90 (95% CI, 0.89-0.91).
- Calibration analysis showed a slope of 0.77 (95% CI, 0.29-1.26) and y-intercept of 0.05 (95% CI, -0.14 to 0.25).
Conclusions:
- A novel Bayesian network effectively predicts neurosurgical intervention probability in pediatric TBI using four immediate factors.
- This probabilistic model offers a potential advantage over binary threshold models for risk stratification.
- The model facilitates tailored management strategies for injured children based on predicted risk.
Background:
Timely surgical decompression improves functional outcomes and survival among children with traumatic brain injury and increased intracranial pressure. Previous scoring systems for identifying the need for surgical decompression after traumatic brain injury in children and adults have had several barriers to use. These barriers include the inability to generate a score with missing data, a requirement for radiographic imaging that may not be immediately available, and limited accuracy. To address these limitations, we developed a Bayesian network to predict the probability of neurosurgical intervention among injured children and adolescents (aged 1-18 years) using physical examination findings and injury characteristics observable at hospital arrival.
Methods:
We obtained patient, injury, transportation, resuscitation, and procedure characteristics from the 2017 to 2019 Trauma Quality Improvement Project database. We trained and validated a Bayesian network to predict the probability of a neurosurgical intervention, defined as undergoing a craniotomy, craniectomy, or intracranial pressure monitor placement. We evaluated model performance using the area under the receiver operating characteristic and calibration curves. We evaluated the percentage of contribution of each input for predicting neurosurgical intervention using relative mutual information (RMI).
Results:
The final model included four predictor variables, including the Glasgow Coma Scale score (RMI, 31.9%), pupillary response (RMI, 11.6%), mechanism of injury (RMI, 5.8%), and presence of prehospital cardiopulmonary resuscitation (RMI, 0.8%). The model achieved an area under the receiver operating characteristic curve of 0.90 (95% confidence interval [CI], 0.89-0.91) and had a calibration slope of 0.77 (95% CI, 0.29-1.26) with a y intercept of 0.05 (95% CI, -0.14 to 0.25).
Conclusion:
We developed a Bayesian network that predicts neurosurgical intervention for all injured children using four factors immediately available on arrival. Compared with a binary threshold model, this probabilistic model may allow clinicians to stratify management strategies based on risk.
Level Of Evidence:
Prognostic and Epidemiological; Level III.
