Validation of an algorithm for sound-based voided volume estimation

Abstract

A voiding diary is commonly used in clinical practice to monitor urinary tract health. However, manual recording and use of a measuring cup can cause significant inaccuracy and inconvenience. Recently sound-based voided volume estimation algorithms such as proudP have shown potential to accurately measure the voided volumes of patients urination while overcoming these inconveniences. In order to validate the sound-based voided volume estimation algorithm, we chose bodyweight change after urination as a reference value. Total 508 subjects from the United States and Korea were enrolled. 584 data points that have matching bodyweights change data and urination sound data were collected, and fivefold cross validation was performed in order to evaluate the model on all data in the dataset. The mean voided volume estimated by the algorithm was 202.6 mL (SD: +/- 114.8) while the mean bodyweight change after urination was 208.0 g (SD: +/- 121.5), and there was a strong linear correlation with high statistical significance (Pearson's correlation coefficient = 0.92, p-value < 0.001). Two paired t-test showed the equivalence with bodyweight change data with 10 mL margin. Additionally, a Bland-Altman plot shows a mean difference of - 5.5 mL with LoA (- 98.0, 87.1). The results support high performance of the algorithm across the large population data from multi-site clinical trials.