Predicting bloodstream infection outcome using machine learning.

Area of Science:

• Infectious Diseases
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Bloodstream infections (BSI) are a significant global cause of illness and death.
• Early identification of high-risk BSI patients is crucial for timely interventions and effective management.
• Current methods for predicting BSI outcomes require improvement for better patient stratification.

Purpose of the Study:

• To develop and validate machine learning models for predicting patient outcomes in bloodstream infections.
• To leverage electronic medical record (EMR) data for early risk assessment of BSI patients.
• To provide a tool for earlier clinical decision-making and patient stratification.

Main Methods:

• Development of machine learning models utilizing comprehensive EMR data, including demographics, lab results, and medical history.
• Training and validation of models on a large cohort of 7889 hospitalized patients diagnosed with BSI.
• Evaluation of model performance using the area under the receiver-operating characteristics curve (AUC).

Main Results:

• A fully featured model achieved an AUC of 0.82, indicating strong predictive capability.
• A more compact model, using only 25 features, demonstrated comparable performance with an AUC of 0.81.
• The models effectively predicted patient outcomes based on integrated EMR data.

Conclusions:

• EMR-based machine learning models show significant potential for predicting BSI patient outcomes.
• These models can facilitate selective rapid microbiological identification and earlier appropriate antibiotic therapy.
• Implementation of these models may lead to reduced BSI development, adverse outcomes, and complications.