Early Prediction of Positive Hospital-Acquired MRSA Screening using Deep Learning on Vital Signs Time Series

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Infectious Disease Diagnostics

Background:

• Methicillin-Resistant Staphylococcus Aureus (MRSA) is a significant cause of hospital-acquired infections (HAIs), leading to increased patient morbidity, mortality, and healthcare expenses.
• Current MRSA diagnostic methods require laboratory testing, often resulting in delays in treatment initiation and effective infection control.
• There is a critical need for rapid, non-invasive methods for early MRSA detection to enable timely clinical intervention.

Purpose of the Study:

• To develop and evaluate a deep learning model for the early prediction of positive MRSA screening outcomes.
• To utilize time-series vital signs data for predicting MRSA status, aiming to provide a predictive capability before laboratory confirmation.
• To assess the model's performance across different prediction windows and lookback periods for optimizing early detection.

Main Methods:

• A Bidirectional Long Short-Term Memory (BiLSTM) deep learning model was developed using the MIMIC-IV database.
• The model was trained on time-series data of six vital signs: body temperature, heart rate, systolic and diastolic blood pressure, respiratory rate, and oxygen saturation.
• Model performance was evaluated using metrics such as F1-score and Area Under the Receiver Operating Characteristic (AUROC) curve across various prediction and lookback windows.

Main Results:

• A 1-day prediction window achieved a high F1-score of 89.19% and an AUROC of 0.96.
• The 1-day prediction window significantly outperformed a 3-day window, which yielded an F1-score of 82.67% and an AUROC of 0.93.
• The developed deep learning approach demonstrated superior performance compared to recent studies using the same dataset, indicating the predictive value of vital sign fluctuations.

Conclusions:

• Time-series vital signs data possess significant predictive value for the early detection of MRSA.
• The AI-driven approach offers a non-invasive, real-time solution for proactive MRSA risk mitigation in healthcare settings.
• This predictive model has the potential to enhance infection control, reduce HAIs, improve patient safety, and optimize resource allocation.