Abstract:
Background Pelvic fractures have high incidence in all blunt trauma cases presenting to the emergency department. However, associated urethral injuries were frequently ignored.
Objective To investigate the risk factors associated with urethral injury in cases of pelvic fracture and establish a predictive model so as to give valuable guidance for clinicians in the diagnosis and treatment of urethral injuries resulting from pelvic fractures.
Methods Clinical data about 1 358 patients with pelvic fractures hospitalized in the First Medical Center of Chinese PLA General Hospital from 2000 to 2020 were retrospectively analyzed. Rsik factors were analyzed using univariate and multivariate logistic regression. The data were divided into model training set and validation set according to the ratio of 7∶3. Urethral injury risk prediction models were developed by several machine learning models (including Gradient Boosting Machine, Support Vector Machine, Random Forest, XGBoost). The effectiveness of the model was evaluated using the AUC of the subject's work characteristic curve (ROC curve).
Results To tally1 358 patients with pelvic fractures were included, with median age of 39(IQR: 26 —52)years, and 64.4% of them were male. The incidence of pelvic fracture urethral injury was 12.0%. Multivariate logistic regression showed that gender, department of consultation, Tiles staging of pelvic fracture, cause of injury, bipulmonary breath sounds, and neutrophils were independent risk factors for the occurrence of urethral injuries in patients with pelvic fracture(P＜0.05). Random forest model had the highest AUC of 0.955, and Gradient Boosting Machine, Support Vector Machine, and XGBoost model had AUCs of 0.936, 0.940, and 0.953, respectively.
Conclusion The occurrence of urethral injury among patients with pelvic fracture is associated with a variety of factors, including gender, department of consultation, Tiles typing of pelvic fracture, cause of injury, et al. Based on these independent risk factors, an early prediction model for urethral injury in patients with pelvic fracture has been developed and established using the random forest machine learning method with good predictive efficacy, which is helpful in clinically predicting and preventing the occurrence of combined urethral injury in patients with pelvic fracture at an early stage.
下载: