Machine learning predicts venous thromboembolism events in traumatic brain injury

Abstract

Abstract Objective: Our study aimed to explore the clinical predict factors and establish a simple and concise model to early identify patients at high risk of venous thromboembolism (VTE) after traumatic brain injury (TBI). Methods: We performed a retrospective study of 685 patients with TBI admitted to our trauma center from April 2018 to April 2021. The experimental group were consisted of the patients which were diagnosed with VTE during hospital, the control group were randomly selected from the rest patients at the ratio of 1:1. We performed a statistical analysis of the clinical predictors of VTE in TBI patients, and established a prediction model of VTE through the logistics and least absolute shrinkage and selection operator (LASSO) regression. Results: Among the 685 included TBI patients, the incidence rate of VTE was 14.74% (101/685). Age, LOS, hemoglobin on admission, and anticoagulant therapy were the common predictors. The prediction model based on the LASSO regression was established and showed a satisfactory AUC value of 0.94 (95% confidential interval: 0.85-0.98) and an excellent calibration ability. The nomogram of the model was also given to help the clinicians identify the targeted patients efficiently. Conclusion: We identified several risk factors for predicting VTE events in TBI patients. The prediction model based on the LASSO regression shows excellent forecasting performance both in the training and validation set.