Prediction of Uropathogens by Flow Cytometry and Dip-stick Test Results of Urine Through Multivariable Logistic Regression Analysis

PURPOSE: Multidrug-resistant Enterobacteriaceae in urinary tract infection (UTI) has spread worldwide; one cause is overuse of broad-spectrum antimicrobial agents such as fluoroquinolone antibacterials. To improve antimicrobial agent administration, this study aimed to calculate a probability prediction formula to predict the organism strain causing UTI in real time from dip-stick testing and flow cytometry. METHODOLOGY: We examined 372 outpatient spot urine samples with observed pyuria and bacteriuria using dip-stick testing and flow cytometry. We performed multiple logistic-regression analysis on the basis of 11 measurement items and BACT scattergram analysis with age and sex as explanatory variables and each strain as the response variable and calculated a probability prediction formula. RESULTS: The best prediction formula for discrimination of the bacilli group and cocci or polymicrobial group was a model with 5 explanatory variables that included percentage of scattergram dots in an angular area of 0–25° (P<0.001), sex (P<0.001), nitrite (P = 0.002), and ketones (P = 0.133). For a predicted cut-off value of Y = 0.395, sensitivity was 0.867 and specificity was 0.775 (cross-validation group: sensitivity = 0.840, specificity = 0.760). The best prediction formula for P. mirabilis and other bacilli was a model with percentage of scattergram dots in an angular area of 0–20° (P<0.001) and nitrite (P = 0.090). For a predicted cut-off value of Y = 0.064, sensitivity was 0.889 and specificity was 0.788 (cross-validation group: sensitivity = 1.000, specificity = 0.766). CONCLUSION: Simultaneous use of the calculated probability prediction formula with urinalysis results facilitates real-time prediction of organisms causing UTI, thus providing helpful information for empiric therapy.