ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: FR-PO1094

Use of an Artificial Neural Network for the Prediction of Urine Culture Positivity from Urine Dipstick in Children

Session Information

Category: Pediatric Nephrology

  • 1700 Pediatric Nephrology

Authors

  • Feber, Janusz, Children's Hospital of East Ontario, Ottawa, Ontario, Canada
  • Blasutig, Ivan, Eastern Ontario Regional Laboratory Association, Ottawa, Ontario, Canada
Background

Urine dipstick results are important for clinical decision making regarding the presence or absence of urinary tract infections. The aim of the study was to analyse the performance of artificial neural network (ANN) in the prediction of positive urine culture from an automated urine dipstick test.

Methods

We retrospectively analysed all available automated urine dipstick (UD) and urine cultures (UCULT) tests performed at our institution over 2-year period (2015-2017). The final dataset (after merging and cleaning) consisted of 5912 complete UD and UCULT performed on the same date and time. Predictors of UCULT included: age, gender, and all UD results: glucose, ketones, specific gravity, blood, pH, protein, nitrates and leukocytes.
ANN model (sequential, feedforward with backpropagation) consisted of 30 neurons in 2 hidden layers (Tensorflow Keras). Data samples (n=5912) were randomly divided into training (70%) and validation set (30%). ANN prediction probabilities thresholds for positive UCULT results were set to 0.5 (ANN05) and 0.3 (ANN03). The performance of both ANN models was assessed by accuracy scores, sensitivity, specificity, positive and negative predictive value (PPV, NPV).

Results

Out of 1773 children (aged 4.3 ±5.3 years) in the validation set, 449 had a positive UCULT. ANN05 correctly predicted 1272 negative + 254 positive UCULT; with ANN03 the prediction was: 1229 negative + 289 positive UCULT.
The mean accuracy of both ANN05 and ANN03 models was 86% (95% CI = 84-88%). Sensitivity and specificity of ANN05 was 0.57 and 0.96 with PPV of 0.83 and NPV of 0.87. The corresponding numbers for the ANN03 model were: 0.64, 0.93, 0.75, and 0.89. The four most important UD variables for prediction of UCULT were: leukocytes, blood, nitrates and protein.

Conclusion

ANN predicted positive urine cultures from a simple urine dipstick (without urine microscopy) with 84-88% accuracy, PPV of 75-83% and NPV of 87-89%. ANN-powered automated urine dipstick lab prediction of positive urine cultures can be considered to facilitate clinical decision making.