Personalized machine learning-based prognostic model for ICU-acquired bloodstream infections.

Area of Science:

• Critical Care Medicine
• Infectious Diseases
• Machine Learning in Healthcare

Background:

• Intensive care unit-acquired bloodstream infections (ICU-BSIs) are a significant cause of mortality in intensive care units (ICUs).
• Predicting the prognosis of ICU-BSIs is crucial for effective patient management and resource allocation.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based model for predicting the prognosis of ICU-BSIs.
• To identify key clinical variables associated with 28-day mortality in ICU-BSI patients.

Main Methods:

• Utilized data from adult patients with blood cultures drawn ≥48 hours after ICU admission from two centers (AMU, China and MIMIC-IV, USA).
• Developed and optimized an eXtreme Gradient Boosting (XGBoost) model using routinely collected clinical variables and time-series data.
• Validated the model internally using the AMU dataset and externally using the MIMIC-IV dataset.

Main Results:

• The XGBoost model demonstrated strong predictive performance, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.92 for training and 0.85 for internal validation.
• Key predictors of 28-day mortality included antibiotic duration, platelet count, serum creatinine, invasive mechanical ventilation duration, and Charlson Comorbidity Index (CCI).
• A simplified model using the top 10 variables identified by SHAP analysis maintained good accuracy, with external validation AUROC of 0.71.

Conclusions:

• Developed and externally validated a personalized ML-based prognostic model for ICU-BSIs using multicenter time-series data.
• The model facilitates early identification of high-risk patients, enabling timely intervention.
• Potential for optimized ICU resource allocation through improved prognostic accuracy.