Using machine learning techniques to predict antimicrobial resistance in stone disease patients.

Area of Science:

• Medical Informatics
• Computational Biology
• Urology

Background:

• Artificial intelligence (AI) and machine learning (ML) are increasingly integrated into healthcare.
• Predicting antimicrobial resistance is crucial for effective patient management in urology.

Purpose of the Study:

• To evaluate the performance of ML techniques in predicting bacterial resistance within a urology department.
• To assess the impact of data features (Gram stain vs. bacterial species) on prediction accuracy.

Main Methods:

• Utilized laboratory information system data from 239 urolithiasis patients (2019).
• Compared ML classifiers using tenfold cross-validation on datasets with Gram stain information only and with bacterial species information.
• Evaluated models including multinomial logistic regression and bagging classifiers.

Main Results:

• ML models achieved a weighted average ROC area of 0.768 and F-measure of 0.708 using Gram stain data.
• Models incorporating bacterial species achieved a higher weighted average ROC area of 0.874 and F-measure of 0.783.
• The bagging classifier demonstrated superior performance when bacterial species data was included.

Conclusions:

• AI and ML can effectively predict antibiotic resistance patterns in urology.
• Prediction accuracy reaches ~77% with Gram staining and ~87% with specific microorganism identification.
• These predictions can guide urologists in selecting appropriate antibiotics 24-48 hours prior to definitive laboratory results.