Developing a Siamese Network for UTIs Risk Prediction in Immobile Patients Undergoing Stroke

Area of Science:

• * Clinical Informatics
• * Artificial Intelligence in Medicine
• * Rehabilitation Medicine

Background:

• * Immobility post-stroke elevates complication risks, including urinary tract infections (UTIs).
• * UTIs are linked to poor prognosis in stroke survivors.
• * Existing datasets show a low prevalence (4%) of new-onset UTIs during hospitalization, creating imbalanced data challenges for predictive modeling.

Purpose of the Study:

• * To develop an effective prediction model for identifying UTIs risk in immobile stroke patients.
• * To compare the performance of a novel Siamese Network model against traditional machine learning approaches for UTI prediction.

Main Methods:

• * Development of a Siamese Network model utilizing common clinical features.
• * Model derivation and validation using a nationwide dataset of 3982 Chinese stroke patients.
• * Comparative analysis against traditional machine learning models on imbalanced data.

Main Results:

• * The Siamese Network demonstrated superior performance in predicting UTIs in imbalanced datasets.
• * Achieved a sensitivity of 0.810 and an Area Under the Curve (AUC) of 0.828.
• * Outperformed traditional machine learning models in identifying patients at risk for UTIs.

Conclusions:

• * The Siamese Network offers a promising solution for accurate UTI risk prediction in immobile stroke patients, even with imbalanced data.
• * This model can aid clinicians in early identification and intervention for UTIs, potentially improving patient outcomes.
• * The study highlights the potential of advanced machine learning techniques in addressing challenges in clinical prediction modeling for post-stroke complications.