Evaluation of the detection ability of uropathogen morphology and vaginal contamination by the Atellica UAS800 automated urine microscopy analyzer and its effectiveness

Abstract Background To help combat the worldwide spread of multidrug‐resistant Enterobacterales, which are responsible for many causes of urinary tract infection (UTI), we evaluated the ability of the Atellica UAS800 automated microscopy system, the only one offering the capability of bacterial morphological differentiation, to determine its effectiveness. Methods We examined 118 outpatient spot urine samples in which pyuria and bacteriuria were observed using flow cytometry (training set: 81; cross‐validation set: 37). The ability of the Atellica UAS800 to differentiate between bacilli and cocci was verified. To improve its ability, multiple logistic regression analysis was used to construct a prediction formula. Results This instrument's detection sensitivity was 106 CFU/ml, and reproducibility in that range was good, but data reliability for the number of cocci was low. Multiple logistic regression analysis with each explanatory variable (14 items from the Atellica UAS800, age and sex) showed the best prediction formula for discrimination of uropathogen morphology was a model with 5 explanatory variables: number of bacilli (p < 0.001), squamous epithelial cells (p = 0.004), age (p = 0.039), number of cocci (p = 0.107), and erythrocytes (p = 0.111). For a predicted cutoff value of 0.449, sensitivity was 0.879 and specificity was 0.854. In the cross‐validation set, sensitivity was 0.813 and specificity was 0.857. Conclusions The Atellica UAS800 could detect squamous epithelial cells, an indicator of vaginal contamination, with high sensitivity, which further improved performance. Simultaneous use of this probability prediction formula with urinalysis results may facilitate real‐time prediction of uropathogens and vaginal contamination, thus providing helpful information for empiric therapy.


| INTRODUC TI ON
Urinary tract infection (UTI) is an infectious disease frequently encountered in daily life and is a representative infection that can cause serious sepsis. 1 The multidrug-resistant Enterobacteriaceae, which produce extended-spectrum β-lactamase (ESBL) and carbapenemase, have recently spread worldwide. [2][3][4] The risk factor of intestinal colonization of multidrug-resistant Enterobacteriaceae is the use of broad-spectrum antibacterial drugs such as fluoroquinolone and carbapenem. [5][6][7] In addition, UTIs are very difficult to diagnose, and unnecessary treatment with antibacterial drugs of conditions with positive urine culture due to asymptomatic bacteria and vulvar-contaminated urine is often wasteful. 8 Currently, the gold standard method of identifying the causative organism of UTI is by bacterial culture, but such tests require long testing time, and obtaining test results concurrently with outpatient treatment is generally impossible. Therefore, in most UTIs, antimicrobial therapy is presently performed without knowing the causative organism. In addition, Gram staining, which is a rapid test, is complicated, and its specimen processing capacity is poor. A system in which the causative organism of UTI can be predicted from automated urinary analysis, which is the initial tool used in the diagnosis of UTI, is needed.
In recent years, automated microscopy has become the main tool currently used worldwide for automated urinary analysis. 9 The Atellica UAS800 (Siemens K.K., Tokyo, Japan), which was evaluated in this study, is an automated urine microscopy analyzer whose method is based on the principle of capturing and analyzing microscopic images with a digital the only morphological instrument that distinguishes urinary tract pathogens into bacilli and cocci, but its performance has not been evaluated.
Therefore, the purpose of this study was to evaluate the bacterial detection and bacterial morphological discrimination ability of the Atellica UAS800 automated microscopy analysis system and to calculate a prediction equation using multiple logistic regression analysis to improve its ability.

| Materials
Among fresh outpatient urine samples submitted to the general urinalysis laboratory of Tenri Hospital (a 1001-bed primary care hospital in Nara, Japan) between March and April in 2018, 118 samples (from 47 men and 71 women) were chosen in which pyuria (>5-10 WBCs/ high-power field) was confirmed and bacteriuria (>1+) was observed based on microscopic testing. Among them, 37 samples (from 14 men and 23 women) were defined as the cross-validation set. This study was approved by the ethical committees of Tenri Hospital and Tenri Health Care University (approval nos. 899 and 115, respectively).

| Measurement of automated microscopy
We used the Atellica UAS800 system to qualitatively measure the following 14 items in the target samples: number of BAC, BACr, BACc, YEA, RBC, WBC, WBCc, NEC, EPI, PAT, HYA, MUC, SPRM, and CRY.

| Verification of detection sensitivity and reproducibility using ATCC strains
To test the detection sensitivity and reproducibility of the Atellica UAS800, we used Escherichia coli ATCC25922 and Staphylococcus aureus ATCC25923. For the detection sensitivity test, bacterial dilutions of 10 0 colony-forming units (CFU)/ml to 108 CFU/ml were prepared, and the linearity of the mean of three measurements was confirmed. The reproducibility test was performed five times using bacterial solutions at concentrations above the detection sensitivity, and the coefficient of variation was calculated and evaluated.

| Microbiologic testing
We performed Gram-stain microscopy analysis and urinary culture.
For urinary culture, we inoculated 5% sheep blood agar/Drigalski medium with 10 μl of fresh urine using a loop and aerobically cultured each sample at 37°C for 18 to 24 h. We conducted strain identification of the grown colonies by matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS). We used a MALDI Biotyper (Bruker Daltonik, Bremen, Germany) and conducted ethanol-formic acid protein extraction as the pretreatment method. The quantification of bacteria by Gram staining and urine culture was as previously reported. 10

| Statistical analysis
To distinguish between the bacilli group, cocci group, and polymicrobial group, we performed bivariate analysis on the basis of the 14 qualitative urinalysis items and age and sex as explanatory variables and each bacterial morphology by Gram staining as the response variable using training set data. We further investigated the discriminant characteristics of three items, BAC, BACr, and BACc, with receiver operating characteristic curve (ROC) analysis. In addition, we performed multiple logistic regression analysis using the 14 qualitative urinalysis items and age and sex. Multiple logistic regression was also performed to construct predictive equations to improve the ability to differentiate bacterial morphology. Moreover, the calculated prediction formula using multiple logistic regression analysis was verified in the cross-validation set. In addition, we used the stepwise method to select explanatory variables in the multiple logistic regression analysis.
To distinguish between the vaginal contamination group and non-contamination group, as described above, we performed bivariate analysis, multiple logistic regression analysis, and ROC analysis on the basis of the results of vaginal contamination as judged using Gram staining and culture analysis.
We used StatFlex Ver. 6.0 (Artech Co., Ltd., Osaka, Japan) software for the statistical analysis, and the level of significance was set at p = 0.05.

| Verification of detection sensitivity and reproducibility using ATCC strains
Results of verification of the detection sensitivity of the Atellica UAS800 using the ATCC strains are shown in Figure 1. The detection limit of the Atellica UAS800 was 106 CFU/ml for both bacilli and cocci. Results of reproducibility for bacteria counts above 106 CFU/ ml are shown in Table 1. Reproducibility in that range was good, with an average of coefficient of variation percent (CV%) = 10.3, but the reliability of the BACc value was low because this value, which indicates the number of cocci, was high even when E. coli ATCC25923 was measured. In the single-species group, bacilli were detected in 17 specimens (bacilli group), which included the following bacterial strains:

| Results of vaginal contamination following Gram staining and urinary culture
Among the 81 target specimens in the training set, 60 specimens were included in the non-contamination group, and 21 were included in the contamination group. Among the 37 target specimens in the cross-validation set, 28 specimens were included in the noncontamination group, and 9 were included the contamination group.

| Bivariate and ROC analysis using the bacilli group, cocci group, and polymicrobial group in training set data
The results of bivariate analysis using the bacilli group, cocci group, and polymicrobial group in training set data are shown in Table 2, and the box-and-whisker plots of Gram staining and the BAC, BACr, and BACc values from the Atellica UAS800 are shown in

| Multiple logistic regression analysis for discrimination of the bacilli group and cocci or polymicrobial group
The results of final model selection using multiple logistic regression analysis for discrimination of the bacilli group and cocci or polymicrobial group are shown in When the cutoff value for the predicted prediction value Y was 0.449, the sensitivity was 0.879 and the specificity was 0.854 (Table 4). In addition, in the cross-validation set, the sensitivity was 0.813 and the specificity was 0.857.

| Statistical analysis of the distinguishability of vaginal contamination
Results of the bivariate analysis of the distinguishability of vaginal contamination using the Atellica UAS800 are shown in When the cutoff value for the predicted prediction value Y was 0.159, the sensitivity was 0.905 and the specificity was 0.900 (Table 7), and in the cross-validation set, the sensitivity was 0.889 and the specificity was 0.857.

| Workflow for discrimination of bacterial morphology and vaginal contamination
The workflow and its performance for discrimination of bacterial Gram staining reported at present for automated microscopy systems. 11,15,16 In this study, to contribute to the appropriate selection of antimicrobial therapy in the treatment of UTI, we evaluated the ability of the Atellica UAS800 automated microscopy system to detect bacteria and discriminate bacterial morphology and devised a predictive formula that can predict causative uropathogens of UTI using multiple logistic regression to improve its ability.
In the discrimination of bacterial morphology and vaginal con- containing ≥105 CFU/ml were used. However, the sensitivity, specificity, PPV, and NPV of the Gram-positive bacteria group were 81.3%, 80.0%, 64.4%, and 90.6%, respectively, and their performance was inferior to that of the Gram-negative bacteria. Besides, morphological instruments such as the iQ200 urine analyzer and cobas u 701 cannot differentiate bacilli and coccobacilli because they do not have items or flags to differentiate them.
The uropathogen prediction equation developed in this study using the Atellica UAS800 could not differentiate between the Cocci group and polymicrobial group, but it could differentiate between the Cocci or polymicrobial group and Bacilli group with 87.9% sensitivity and 85.4% specificity. This performance was also confirmed with the validation set. The PPV of bacilli and that of cocci or polymicrobials were 82% and 80%, respectively, which were comparable to those of flow cytometric analysis. In addition, because the Atellica UAS800 is a microscopy analyzer, it can detect squamous epithelial cells and thus is superior in differentiating vaginal contamination.
These are very useful advantages of the Atellica UAS800.
This study has two limitations. First, we used fresh urine of outpatients suspected of having a UTI as the targeted material for this study, but we did not consider patient backgrounds. Therefore, it is possible that patients such as catheterized patients and pregnant women may have asymptomatic bacteriuria. However, as we usually do not consider the patient's background in routine urinalysis, the probability prediction equation of this study, which does not consider patient background, is optimal when used in daily workflow. Second, we analyzed only the data obtained from an automated microscopy system, but as the dip-stick test is also performed in the actual inspection

ACK N OWLED G M ENTS
We thank Siemens Healthcare Diagnostics K. K (Tokyo, Japan) for use of the Atellica UAS800 system.

CO N FLI C T O F I NTE R E S T
None.

AUTH O R CO NTR I B UTI O N
All authors meet the ICMJE authorship criteria. AN, MK, and HY developed the trial design and contributed to the writing of the final manuscript. AN, TS, and NN were involved in data analysis and data interpretation.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.