Identifying bladder rupture following traumatic pelvic fracture: A machine learning approach

Area of Science:

• Trauma Surgery
• Medical Informatics
• Diagnostic Imaging

Background:

• Bladder rupture is a rare but serious complication of blunt pelvic trauma.
• Predicting bladder injury in these patients can be challenging.
• Machine learning offers a potential tool for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms in predicting bladder rupture in blunt pelvic trauma patients.
• To identify key factors at presentation that correlate with bladder injury.
• To enhance the early detection of bladder rupture.

Main Methods:

• Retrospective analysis of adult patients with blunt pelvic trauma and pelvic fractures.
• Inclusion of patients with available urinalysis, ICD-9 codes, and mechanism of injury data.
• Comparison of patients with and without bladder rupture using various machine learning classifiers.

Main Results:

• Out of 3063 patients, 208 (6.8%) had bladder ruptures.
• Machine learning models, particularly Gaussian Naïve Bayes and Kernel Naïve Bayes, achieved high accuracy (97.8%).
• Excellent performance metrics included 99% specificity, 83% sensitivity, and an AUC of 0.99.

Conclusions:

• Machine learning models can accurately predict bladder rupture in blunt pelvic trauma.
• Utilizing readily available clinical data improves prediction accuracy.
• This approach can optimize patient selection for further imaging and resource allocation.