Urine Flow Cytometry Parameter Cannot Safely Predict Contamination of Urine-A Cohort Study of a Swiss Emergency...

Area of Science:

• Clinical Chemistry
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Urine flow cytometry (UFC) analyzes urine samples to count various parameters.
• Predicting urine culture growth, including contamination (mixed growth), is clinically important.

Purpose of the Study:

• To predict different urine culture growth types (negative, positive, mixed) using UFC parameters.
• To evaluate the efficacy of machine learning in predicting urine culture outcomes.

Main Methods:

• Retrospective cohort study of adult emergency department patients (07/2017-09/2020).
• Machine learning model (extreme gradient boosting) trained on UFC parameters, sex, and age.
• Dataset split into training (75%) and validation (25%) sets.

Main Results:

• Urine flow cytometry parameters like squamous epithelial cells, bacteria, and leukocytes correlated with culture growth types.
• Achieved 80% overall prediction accuracy for negative, positive, and mixed cultures.
• Correctly predicted negative (96.3%) and positive (74.0%) cultures, but only 11.1% of mixed cultures.

Conclusions:

• Urine flow cytometry parameters can reliably rule out significant bacterial growth.
• UFC alone is insufficient for accurately predicting positive or mixed urine cultures.
• Squamous epithelial cells detected by UFC are associated with mixed culture growth, suggesting contamination.