Page 5 - CUA 2020_Endourology
P. 5
Podium 1: Endourology, Nephrolithiasis
POD-1.5. Fig. 2. Out-of-sample ROC curve for quintile stratification (AUC=0.70) and corresponding confusion matrix.
and Kidney Institute, Cleveland Clinic, Cleveland, OH, United States; POD-1.6
9 Urology, University of North Carolina School of Medicine, Chapel Hill, The effect of a bacterial urinary tract infection isolate and
10
NC, United States; Urology, Palmetto Health USC Medical Group, antibiotics on calcium urolithiasis: A novel role for zinc?
11
Columbia, SC, United States; Urology, University of California Irvine Jennifer Bjazevic , Kaitlin Al , Hassan Razvi , Jeremy Burton 2
2
1
1
12
School of Medicine, Orange, CA, United States; Urology, University of 1 Surgery, Western University, London, ON, Canada; Microbiology &
2
California San Francisco School of Medicine, San Francisco, CA, United Immunology, Western University, London, ON, Canada
States; Urology, Dartmouth Hitchcock Medical Center, Lebanon, NH, Support: Canadian Urological Association Scholarship Foundation
13
United States; Urology, University of Florida College of Medicine, Introduction: The formation of calcium-based stone disease may be
14
15
Gainesville, FL, United States; Urology, McGill University Health impacted by both urinary bacteria and antibiotics. Zinc (Zn), which is
Centre, Montreal, QC, Canada; Urology, University of California Davis involved as an early nidus for the mineralization process of urinary stones
16
17
School of Medicine, Sacramento, CA, United States; Urology, University and is known to be involved in both immune system function and bacte-
of Washington, Seattle, WA, United States; Urology, Université de rial pathogenesis, may play an integral role in this process. Our study
18
Montréal, Montreal, QC, Canada aims to further clarify the role of urinary bacteria and antibiotics on the
Introduction: The Wisconsin Stone Quality of Life (WISQOL) question- pathogenesis of urolithiasis, potentially through modulating Zn transport-
naire, a quality of life (QOL) measurement tool designed specifically for ers as part of the immune response in a Drosophila melanogaster (DM)
kidney stone patients, was recently validated. Using the WISQOL score as urolithiasis model.
the gold standard of the model, we built the WISQOL Machine Learning Methods: Wild-type DM were reared under standard conditions at 25°C,
Algorithm (WISQOL-MLA) to predict patients’ QOL based on demographic, 40% humidity and a 12–12-hour light/dark cycle with either a standard or
symptomatic, and clinical data collected for the validation of the WISQOL. lithogenic diet supplemented with 0.1% sodium oxalate. DM flies were
Methods: Using gradient boosting and deep learning models implemented in either treated with a combination of a non-urease-producing strain E.
Python, QOL scores for all 3206 patients were predicted. We also stratified coli (UTI89), ciprofloxacin (0.2 μg/mL), or TMP-SMX (30/10 μg/mL) for
QOL scores by quintiles. The dataset was split using a standard 70/10/20% seven or 14 days. Stone burden was assessed through measurement of
training/validation/testing ratio. Per usual machine learning practice, cat- pixel intensity of CaOx crystals in dissected Malpighian tubules. DM flies
egorical variables were numerically discretized. Variables were then stan- were anesthetized with CO and homogenized; total RNA was isolated
2
dardized to mean of 0 and standard deviation of 1 to aid model conver- and converted to cDNA; quantitative PCR was undertaken for each DM
gence. Regression performance was evaluated using Pearson’s correlation Zn transporter gene (CG3994, CG17723, CG11163), using ®–tubulin as
(r). Classification was evaluated with area under the ROC curve (AUC). an internal control.
Results: Gradient boosting obtained a test correlation of 0.622 (Fig. 1). Results: Treatment with UTI89 (p=0.005), ciprofloxacin (p<0.001), and
Deep learning obtained a correlation of 0.592. Multivariate regression only TMP-SMX (p=0.003) increased CaOx stone formation in DM flies at day
achieved a correlation of 0.4375. Quintile stratification on all WISQOL 7. Preliminary results demonstrate a trend towards increased expres-
patients obtained an average test AUC of 0.70 for the five classes. The sion of Zn transporter gene ZnT41F (CG1163) with UTI89 treatment.
model performed best in distinguishing between lowest (0.79) and highest Co-treatment of the DM flies with both E. coli UTI89 and either antibiotic
quintile (0.83) (Fig. 2). Feature importance analysis showed that the model reduced the expression of ZnT41F to baseline levels.
correctly weights in symptomatic status, body mass index, and age, as well Conclusions: These findings suggest that non-urease-producing E. coli,
as other medical and demographic features to estimate QOL. and the antibiotics ciprofloxacin and TMP-SMX impact CaOx stone for-
Conclusions: Harnessing the power of the WISQOL questionnaire, mation. Zn may be involved in this process through the modulation of
WISQOL-MLA can accurately predict a stone patient’s QOL from read- its transport proteins. Further investigation is required to confirm these
ily available clinical information and outperforms linear models. Future results and delineate the exact mechanism involved.
endeavors include scaling the tool as an aid to urologists that don’t have
the resources to collect precise QOL scores via questionnaire.
CUAJ • June 2020 • Volume 14, Issue 6(Suppl2) S27