Machine Learning for Antibiotic Resistance Prediction: A Prototype Using Off-the-Shelf Techniques and Entry-Level...

Area of Science:

• Clinical microbiology
• Artificial intelligence in healthcare
• Infectious disease management

Background:

• Antimicrobial resistance poses a significant threat, necessitating rapid identification and treatment of infections.
• Traditional susceptibility testing methods are time-consuming (≥24 hours), delaying appropriate antibiotic selection for critically ill patients.
• Automated machine learning (AutoML) offers potential as a clinical decision support tool for predicting antimicrobial resistance.

Purpose of the Study:

• To evaluate the efficacy of AutoML in predicting antimicrobial resistance using patient data.
• To assess the utility of AutoML for guiding empirical antibiotic therapy selection.
• To explore the potential of AutoML in addressing the challenges posed by antimicrobial resistance.

Main Methods:

• A dataset of 11,496 antimicrobial susceptibility instances from 499 patients was analyzed using Microsoft Azure AutoML.
• Predictions were based on demographic characteristics and previous antibiotic susceptibility testing results.
• Both original and balanced datasets were processed to evaluate model performance.

Main Results:

• The stack ensemble technique demonstrated the highest performance.
• An area under the curve-weighted metric of 0.822 was achieved on the original dataset.
• An improved metric of 0.850 was obtained on the balanced dataset.

Conclusions:

• AutoML implementation for antimicrobial susceptibility data provides valuable insights into potential antibiotic resistance.
• This technology can assist clinicians in selecting appropriate empirical antibiotic therapy.
• Considering the local antimicrobial resistance patterns, AutoML can enhance treatment strategies.