Algorithm may predict health-related quality of life in patients with urolithiasis

July 28, 2020

The machine learning algorithm incorporates readily available clinical information.

A machine learning algorithm (MLA) incorporating readily available clinical information demonstrated promising performance for predicting quality of life in kidney stone patients.

The research was presented at the 2020 European Association of Urology Virtual Congress by David-Dan Nguyen, on behalf of the Wisconsin Quality of Life research group.1

The MLA was built using data collected for the development and validation of the Wisconsin Stone Quality of Life Questionnaire (WISQOL). In its first version, the MLA adequately estimated health-related quality of life of kidney stone patients and outperformed linear models. Assessment of its ability to classify patients into quintiles by quality of life scores showed that the MLA did best at identifying the highest and lowest quintiles, reported Nguyen, medical student, McGill University, Montreal, Quebec, Canada.

“Urolithiasis is a disease with complex symptoms that poses a significant burden on the quality of life of those affected by it. As such, an overarching goal of the management of urolithiasis is to improve quality of life. Patient-reported outcomes play an important goal in guiding the management of a disease, and there has been increasing efforts, including the development of certain tools, to specifically measure the quality of life of kidney stone patients,” he said.

“In the future with more data, hopefully our model can be improved to be used in more refined clinical settings.”

The WISQOL questionnaire is a disease-specific, validated health-related quality of life instrument capturing the unique symptoms of patients with urolithiasis and functional impact of their stone disease. It was developed by a team led by Kristina L. Penniston, PhD, and Stephen Y. Nakada, MD, both of the University of Wisconsin School of Medicine and Public Health, Madison.

Using the WISQOL score as the gold standard, 3 MLA candidate models were built to predict quality of life score from demographic, symptomatic, and clinical data. Nguyen reported that using Pearson’s correlation testing to evaluate regression performance, the gradient boosting model outperformed the deep learning and multivariable lasso regression models (out-of-sample r values for correlation between quality of life estimates and the WISQOL total score = 0.622, 0.592, and 0.438, respectively).

Classification performance of the MLAs was analyzed with area under the receiver operating characteristics curve (AUC). For the gradient boosting model, the out-of-sample AUC for quintile stratification was 0.70, and it performed best distinguishing between lowest and highest quintiles (AUC = 0.79 and 0.83, respectively).

“Our model underperformed in classifying patients in the middle quintile of quality of life,” Nguyen said.

“Future endeavors include scaling the tool as an aid to urologists that do not have the resources to collect precise quality of life scores via questionnaire and retrospectively generating quality of life scores in existing cohorts,” the investigators wrote.

Reference

1. Nguyen D-D, Luo JW, Lim JRZ, et al. Wisconsin quality of life machine learning algorithm for predicting quality of life in kidney stone patients. 2020 European Association of Urology Virtual Congress. July 17-26, 2020. Abstract 785

<<<More from the 2020 EAU Annual Congress