248 - Development of a predictive model identifying infants undergoing physical abuse evaluations at high risk for intracranial injury: A multicenter study

Publication Number: 2239.248

M Katherine Henry, Children's Hospital of Philadelphia/UPenn, Philadelphia, PA, United States; Jing Huang, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA, United States; Jan Leonard, CHOP, Philadelphia, PA, United States; Jingya Yu, Children's Hospital of Philadelphia, Philadelphia, PA, United States; Chris Feudtner, Children's Hospital of Philadelphia, Philadelpia, PA, United States; Rachel Berger, UPMC Childrens Hospital of Pittsburgh, Pittsburgh, PA, United States; Daniel M. Lindberg, University of Colorado Anschutz Medical Campus, Denver, CO, United States; Kristine A. Campbell, University of Utah School of Medicine, Salt Lake City, UT, United States; James Anderst, Children's Mercy Hospitals and Clinics, Kansas City, MO, United States; Angela Bachim, Baylor College of Medicine, Houston, TX, United States; Farah W.. Brink, Brink, Dublin, OH, United States; Nancy S. Harper, University of Minnesota Masonic Children's Hospital, Minneapolis, MN, United States; Natalie Laub, University of California, San Diego School of Medicine, San Diego, CA, United States; John Melville, University of South Carolina School of Medicine, Char, SC, United States; Joanne Wood, Children's Hospital of Philadelphia, Philadelphia, PA, United States

Background: Neurologically asymptomatic infants with concern for physical abuse are at risk for intracranial injuries (ICI), yet the decision to obtain neuroimaging can be challenging. A predictive model could help identify which infants are at highest risk for asymptomatic ICI and inform neuroimaging decisions.

Objective: To develop a model predicting ICI in infants undergoing subspeciality abuse evaluations.

Design/Methods: This was a retrospective, cross-sectional study of infants 30-364 days who underwent subspecialty child physical abuse evaluations at 9 CAPNET sites from 2/2021-12/2022. All infants underwent both a skeletal survey (which would identify most skull fractures) and neuroimaging (CT, MRI, fast MRI). Infants with ICI prompting abuse evaluation, seizures, altered mental status, cardiorespiratory collapse, clinical need for intubation, or GCS < 13, and those undergoing screening exams due to family violence were excluded. The outcome was ICI, defined as intracranial hemorrhage, parenchymal injury, or other possible trauma-related finding. The dataset was divided into 80% training and 20% test sets. A total of 348 predictors related to history provided, laboratory values, physical exam, injuries identified, and social risk factors were considered. We used random forest models to identify the 20 most predictive variables which were then used to generate a single classification and regression tree. Missing data in hemoglobin, platelets, and head circumference (HC) percentile were imputed with medians of those values in the training population. Area under the ROC curve, sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) were assessed in the test set.

Results: Of 1108 infants, 142 (13%; 95% CI: 11%, 15%) had ICI. Tree-based methods identified the following predictor variables in the training set (N=886): skull fracture (number, unilateral vs bilateral, and presence), scalp injury (reported symptom or exam), appendicular fracture (classic metaphyseal lesion [CML] and non-CML), mobility, fall history (presence, type, height), age, hemoglobin, platelets, HC percentile, musculoskeletal finding (symptom or exam), number of non-skull fractures, history of CPS involvement (child, caregiver), bruising characteristics (number, patterned). In the test set (N=222), area under the ROC curve was 0.776 (95% CI 0.680, 0.873). Sensitivity, specificity, PPV, NPV were 92%, 38%, 17% and 97%, respectively.

Conclusion(s): A complex tree-based model can identify infants at high risk for ICI. Multiple predictors overlap with those in the Pittsburgh Infant Brain Injury Score.