TY  - EJOU
AU  - Ostrowski, David A. 
AU  - Logan, Joseph R. 
AU  - Thompson, Austin 
AU  - Broms, Reiley 
AU  - Syphan, Thomas 
AU  - Antony, Maria 
AU  - Rickard, Mandy 
AU  - Erdman, Lauren 
AU  - Head, Dennis 
AU  - Hannick, Jessica H. 
AU  - Woo, Lynn L. 
AU  - Grant, Frederick D. 
AU  - D’Souza, Neeta 
AU  - Viswanath, Satish E. 
AU  - Flask, Chris A. 
AU  - Lorenzo, Armando J. 
AU  - Fan, Yong 
AU  - Tasian, Gregory E. 
AU  - Weaver, John K. 

TI  - Random survival forest models predict renal complications in antenatal hydronephrosis
T2  - Canadian Journal of Urology

PY  - 
VL  - 
IS  - 
SN  - 1488-5581

AB  -  <b>Objectives:</b> This study aims to build machine learning models to analyze Technetium-99m mercaptoacetyltriglycine (MAG3) renal scans and clinical variables from patients with antenatal hydronephrosis (ANH) with concern for ureteropelvic junction obstruction (UPJO) and equivocal initial MAG3 renal scans to predict the risk of renal complication development. <b>Methods:</b> A retrospective cohort of patients under 1-year-old with ANH concerning for UPJO with an equivocal MAG3 renal scan evaluated at our institution from January 2009 to June 2021 was identified. Equivocal renal scans were defined as studies where the affected kidney demonstrated split function greater than 39%, and subsequent clinical recommendation was continued surveillance instead of surgery. Random survival forest models were c	onstructed to predict the risk of renal complications (decreased function, increased parenchymal thinning, worsening hydronephrosis) using clinical variables (Clinical Model), or numerical features extracted from MAG3 renal scans (Imaging Model), or clinical and numeric renal scan features (Ensemble Model). <b>Results:</b> One-hundred fifty-two patients were included; 62 patients developed a renal complication, and 90 did not develop a renal complication with a minimum of 3-year follow-up. The Imaging Model (C-index 0.74; 95% CI: 0.68–0.79) and the Clinical Model (C-index 0.70; 95% CI: 0.66–0.75) performed similarly, while the Ensemble Model (C-index 0.76; 95% CI: 0.75–0.86) demonstrated improved performance compared to either individually. Differential function per unit volume of the affected kidney was the most important driver of the Clinical Model’s predictions. All models identified high-risk patients at significantly higher risk of renal complications on Kaplan-Meier log-rank analysis (<i>p</i> &lt; 0.0001). <b>Conclusion:</b> Random survival forest models using clinical and/or renal scan features may improve the prediction of renal complication risk in patients with ANH, with concern for UPJO and an equivocal initial MAG3 renal scan.
KW  - diuretic renal scan; antenatal hydronephrosis; machine learning; random survival forest

DO  - 10.32604/cju.2026.077246