Adaptation and External Validation of Pathogenic Urine Culture Prediction in Primary Care Using Machine Learning

Area of Science:

• Clinical informatics
• Machine learning in healthcare
• Diagnostic accuracy

Background:

• Urinary tract infection (UTI) symptoms are common, but antibiotic use requires confirmed infection.
• Urine culture is definitive but delayed; existing machine learning models (NeedMicro) require microscopy unavailable in primary care.
• Need for a rapid, accessible diagnostic tool for UTI in primary care settings.

Purpose of the Study:

• To develop and validate a machine learning classifier (NoMicro) for predicting pathogenic urine cultures without urine microscopy.
• To assess the performance of NoMicro in both emergency department and primary care populations.
• To evaluate the potential of NoMicro to reduce antibiotic overuse in primary care.

Main Methods:

• Redesigned a machine learning classifier (NoMicro) excluding urine microscopy features.
• Retrospectively validated NoMicro using extreme gradient boosting (XGBoost), artificial neural networks, and random forests (RFs) on large datasets (80,387 ED, 472 PC adults).
• Predictor variables included patient demographics, dipstick urinalysis results, and UTI symptoms; primary outcome was pathogenic urine culture (≥100,000 CFU/mL).

Main Results:

• Internal validation showed NoMicro/XGBoost achieved a high ROC-AUC of 0.86, only slightly lower than NeedMicro (0.88).
• External validation in primary care demonstrated excellent performance with NoMicro/RF achieving a ROC-AUC of 0.85.
• Simulations indicated NoMicro/RF could safely guide antibiotic withholding for low-risk patients, reducing unnecessary prescriptions.

Conclusions:

• The NoMicro classifier demonstrates appropriate performance for primary care UTI diagnosis.
• NoMicro offers a promising solution for accurate UTI assessment without microscopy, aiding antibiotic stewardship.
• Further prospective trials are warranted to confirm the clinical utility and safety of the NoMicro classifier.