Machine learning-based clinical decision support for infection risk prediction.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Decision Support

Background:

• Healthcare-associated infections (HAIs) pose significant risks to hospitalized patients and healthcare systems.
• Early detection and intervention are crucial for managing HAIs.
• Current diagnostic methods may not identify infections before they become clinically apparent.

Purpose of the Study:

• To develop and validate a machine learning model for predicting HAIs before overt symptoms emerge.
• To create a clinical decision support tool for proactive patient assessment.
• To enable earlier diagnosis and treatment of infections in hospitalized patients.

Main Methods:

• Utilized ensemble-based boosted decision trees on a large retrospective hospital dataset.
• Extracted a dataset of 36,782 healthcare-associated infection patients.
• Leveraged vital signs, laboratory measurements, and demographics to predict HAI risk.

Main Results:

• The best performing model achieved a cross-validated AUC of 0.88 at 1 hour before clinical suspicion.
• The model maintained an AUC >0.85 for up to 48 hours prior to suspicion.
• A reduced model using fewer features still achieved an AUC of 0.86 at 1 hour before suspicion, outperforming existing methods.

Conclusions:

• The predictive model effectively aggregates physiological data to provide a continuous infection risk score.
• This tool can be deployed in hospitals to offer advance warning of patient deterioration due to infection.
• Early prediction of HAIs can lead to timely interventions and improved patient care.