Machine learning prediction of multidrug-resistant urinary tract infections in brain and spinal cord injury patients: a dual-center validation study

Abstract

Background and objective: Multidrug-resistant urinary tract infections (MDR UTIs) are a growing concern in patients with brain and spinal cord injuries due to their high susceptibility to UTIs and repeated antibiotic exposure. This study aimed to develop and externally validate a machine learning model for predicting MDR UTIs in patients with brain and spinal cord injuries.

Methods: We retrospectively analyzed 849 patients with brain or spinal cord injuries and culture-confirmed UTIs. Data from a single institution were used for model training, and the other served as an external test set. Five machine learning models and ensemble models were trained and evaluated. Model performance was evaluated using accuracy, sensitivity, specificity, positive/negative predictive value, and area under the receiver operating characteristic curve.

Results: The finally selected ensemble model, combining Naive Bayes and Random Forest, achieved an area under the receiver operating characteristic curve of 0.8425, with an accuracy of 0.7688, sensitivity of 0.7647, specificity of 0.7697, positive predictive value of 0.4262, and negative predictive value of 0.9360. Important predictors included recent antibiotic use, total Functional Independence Measure score, neutrophil-to-platelet ratio, neutrophil-to-lymphocyte ratio, erythrocyte sedimentation rate, age, hematocrit, hemoglobin, serum protein, serum albumin, and urethral indwelling catheter.

Conclusion: Externally validated, the model showed consistent performance across institutions. It may support early MDR risk stratification in patients with brain and spinal cord injuries. Early identification of at-risk individuals may guide antibiotic stewardship in neurorehabilitation settings.