Derivation and Validation of a Machine Learning Algorithm for Predicting Venous Thromboembolism in Injured Children

Abstract

BackgroundVenous thromboembolism (VTE) causes significant morbidity in pediatric trauma patients. We applied machine learning algorithms to the Trauma Quality Improvement Program (TQIP) database to develop and validate a risk prediction model for VTE in injured children. MethodsPatients ≤18 years were identified from TQIP (2017–2019, n = 383,814). Those administered VTE prophylaxis ≤24 h and missing the outcome (VTE) were removed (n = 347,576). Feature selection identified 15 predictors: intubation, need for supplemental oxygen, spinal injury, pelvic fractures, multiple long bone fractures, major surgery (neurosurgery, thoracic, orthopedic, vascular), age, transfusion requirement, intracranial pressure monitor or external ventricular drain placement, and low Glasgow Coma Scale score. Data was split into training (n = 251,409) and testing (n = 118,175) subsets. Machine learning algorithms were trained, tested, and compared. ResultsLow-risk prediction: For the testing subset, all models outperformed the baseline rate of VTE (0.15%) with a predicted rate of 0.01–0.02% (p < 2.2e−16). 88.4–89.4% of patients were classified as low risk by the models. High-risk predictionAll models outperformed baseline with a predicted rate of VTE ranging from 1.13 to 1.32% (p < 2.2e−16). The performance of the 3 models was not significantly different. ConclusionWe developed a predictive model that differentiates injured children for development of VTE with high discrimination and can guide prophylaxis use. Level of EvidencePrognostic, Level II. Type of StudyRetrospective, Cross-sectional.