Enhancing Antibiotic Stewardship: A Machine Learning Approach to Predicting Antibiotic Resistance in Inpatient Care.

Area of Science:

• Medical Informatics
• Computational Biology
• Infectious Diseases

Background:

• Antibiotic resistance poses a significant threat to global health, challenging current medical treatments.
• Innovative antibiotic stewardship strategies are essential to combat rising resistance rates.
• Electronic health records (EHRs) contain vast data that can be leveraged for clinical decision support.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting antibiotic resistance.
• To identify key clinical factors associated with antibiotic resistance.
• To assess the potential of these models in improving empirical antibiotic prescribing and de-escalation.

Main Methods:

• Development of machine learning models, termed 'personalized antibiograms', using Stanford's EHR data.
• Inclusion of 49,872 patient records for urine, blood, and respiratory infections.
• Utilized LightGBM algorithm incorporating demographics, prior resistance, prescriptions, and comorbidities as features.

Main Results:

• Models achieved notable discriminative ability with Area Under the Receiver Operating Characteristic Curves (AUROCs) ranging from 0.74 to 0.78.
• Prior antibiotic resistance and prescription history were identified as significant predictive factors.
• High specificity of the models suggests potential for safe antibiotic de-escalation.

Conclusions:

• Machine learning models can effectively predict antibiotic resistance, aiding clinical decision-making.
• Personalized antibiograms show promise in informing antibiotic de-escalation strategies.
• Leveraging EHR data with machine learning offers a feasible approach to enhance empirical antibiotic prescribing and combat resistance.