Using artificial intelligence to reduce diagnostic workload without compromising detection of urinary tract...

Area of Science:

• Clinical Microbiology
• Artificial Intelligence in Healthcare
• Diagnostic Efficiency

Background:

• A large proportion of microbiological screening involves suspected urinary tract infections (UTIs).
• Approximately two-thirds of urine samples yield negative culture results, indicating inefficiency.
• Improving laboratory efficiency is crucial given high demand and limited resources.

Purpose of the Study:

• To evaluate machine learning models for optimizing urine sample screening prior to culture.
• To compare the performance of machine learning against a heuristic model for workload reduction.
• To identify strategies for improving diagnostic service efficiency in microbiology.

Main Methods:

• Retrospective analysis of 212,554 urine microscopy, culture, and sensitivity reports.
• Comparison of a heuristic model (WBC and bacterial counts) with machine learning algorithms (Random Forest, Neural Network, Extreme Gradient Boosting).
• Inclusion of demographic, historical, and clinical data as independent variables.

Main Results:

• Machine learning algorithms outperformed the heuristic model in workload reduction at >95% sensitivity.
• Independent evaluation of pregnant patients and children improved classification sensitivity.
• An optimal solution using three Extreme Gradient Boosting algorithms achieved 41% relative workload reduction and 95% sensitivity for stratified groups.

Conclusions:

• Supervised machine learning models offer significant potential for improving diagnostic laboratory efficiency.
• The implemented heuristic model demonstrated considerable time and cost savings without compromising diagnostic performance.
• This work highlights the applicability of machine learning in resource-constrained public healthcare settings.