Machine Learning techniques for work development: Testing urine analysis for urine culture

Abstract

Background: Urine analysis (UA) is a simple and cost-effective test. The presence of WBC indicates bacterial infection and inflammation. Applying machine learning techniques to pre-screen urine samples before sending them for culture can reduce workload and costs.

Objective: To develop and compare the performance of the most suitable machine learning model for predicting urinary tract infections (UTIs) using urine analysis data.

Methods: A retrospective study was conducted using urine culture data from 2022. Data from 345 urine samples, including UA results and culture results, were analyzed. Various machine learning models were developed and evaluated using Orange software. Model performance was assessed using metrics such as sensitivity, specificity, positive predictive value, negative predictive value, and accuracy.

Results: The decision tree model demonstrated the highest performance in predicting UTIs. A WBC count of ≥ 3-5 cells/HPF was significantly associated with UTIs. When evaluated using the decision tree model, a WBC count cutoff of ≥ 3-5 cells/HPF yielded a sensitivity of 96.0%, specificity of 80.0%, positive predictive value of 81.5%, negative predictive value of 96.0%, and an overall accuracy of 88.0%.

Conclusion: The decision tree model is a highly effective tool for predicting UTIs. Using a WBC count of ≥ 3-5 cells/HPF as a cutoff for urine culture can serve as a guideline for screening urine samples.