PD35-02 APPLICATION OF ARTIFICIAL INTELLIGENCE TOOL TO IDENTIFY PATIENTS AT HIGH RISK FOR SYMPTOMATIC KIDNEY STONE RECURRENCE

Abstract

INTRODUCTION AND OBJECTIVE: Nephrolithiasis is a chronic disease and stone recurrence rates can be reduced by implementing the AUA kidney stone medical management guidelines. It is recommended that clinicians should perform additional metabolic testing in high-risk or interested first-time stone formers and recurrent stone formers. This investigation assessed the ability of artificial intelligence to identify patients who are at high risk for symptomatic kidney stone recurrence and predict when a patient will recur. METHODS: A 10 year period (January 2008 to December 2018) retrospective review of electronic medical records of patients with diagnosis of kidney stones. We used machine learning and developed two models to identify symptomatic stone recurrence: 1. Which patients are at risk of symptomatic recurrence 2. When will these patients present with a symptomatic recurrence. Symptomatic stone recurrence was defined as any kidney stone encounter occurring more than 90 days after initial diagnosis that involved an Emergency Department visit, admission, or surgery. The algorithm was developed on patients Vol. 203, No. 4S, Supplement, Saturday, May 16, 2020 THE JOURNAL OF UROLOGY
e717 Copyright © 2020 American Urological Association Education and Research, Inc. Unauthorized reproduction of this article is prohibited. with kidney stone diagnosis encounters. The algorithm was validated on members with kidney stones between January 2019 to June 2019. RESULTS: A total of 108,000 patients were identified with the diagnosis of kidney stones. The algorithm was applied to 516,000 kidney stone encounters attributed to this cohort. The algorithm was validated on 1,123 patients with kidney stones between January 2019 to June 2019. The models consider 655 attributes that were identified through data analysis and clinician input. The models can predict which patients are at risk for symptomatic kidney stone recurrence with AUCROC [ 0.83 (see Figure 1). The time to recurrence model prioritized patients within the high risk group. C-statistic value is 0.79. CONCLUSIONS: Artificial Intelligence can analyze large and complex data sets and improve clinicians ability to triage patients into low, moderate and high-risk groups. This will improve patient counseling and may improve quality of care by implementing preventive measures on high risk patients and reduce metabolic workup, invasive endoscopy and ionizing radiation exposure for low risk kidney stone patients.