Back to 2026 Abstracts

---

Risk-Aligned Management Shifts After Exposure to an AI Hydronephrosis Severity Index
Samer Maher, MHSc¹, Adree Khondker, MD¹, Jin Kyu Kim, MD², Lauren Erdman, PhD³, Hadel Alsubaie, MD⁴, Laura Betcherman, MD⁴, Fabio Botelho, MD⁴, Valentina Bruno, MD⁴, Martina Bruneira, MD⁵, Joana Dos Santos, MD⁴, Usman Kahloon, MPH⁴, Jethro Kwong, MD¹, Mirriam Mikhail, MD⁴, Asmaa A. Milyani, MD⁴, Beverly Miranda, MN⁴, David-Dan Nguyen, MD¹, Innocent Nzeyimana, MD⁴, Mawuenyo Oyortey, MD⁴, Priya Saini, MD⁴, Nithiakishna Selvathesan, MD⁴, Chia W. Teoh, MD⁴, Marilyn Wong, MD⁴, Michael Chua, MD⁴, Armando J. Lorenzo, MD⁴, Mandy Rickard, MN⁴.
¹University of Toronto, Toronto, ON, Canada, ²Riley Hospital for Children, Indianapolis, IN, USA, ³Cincinnati Children's Hospital, Cincinnati, OH, USA, ⁴The Hospital for Sick Children, Toronto, ON, Canada, ⁵University of Padova, Padova, Italy.

Introduction
The Hydronephrosis Severity Index (HSI) is a convolutional neural network that predicts hydronephrosis severity with a continuous numerical score from renal ultrasound images and has undergone multi-centre external validation. We evaluated whether clinician exposure to the HSI altered simulated management recommendations in a risk-aligned manner.
Methods
Twenty-three paediatric urology and nephrology clinicians independently reviewed 293 retrospective, de-identified hydronephrosis cases (age 0-24 months), yielding 6,739 clinician-case observations. For each case, clinicians received standardized clinical information and two renal ultrasound images, then selected one of four plans: discharge, repeat ultrasound, diuretic renogram, or surgery. They then reviewed HSI output (predicted probability of surgery; traffic-light risk category: green, yellow, red) and recorded a revised or unchanged plan. Pre/post-model distributions were compared with the Stuart-Maxwell test. Directional changes were classified as escalation, de-escalation, or no change. Mixed-effects logistic regression with random effects for clinician and case evaluated whether model-predicted surgical risk was associated with escalation, de-escalation, and transition to surgery.
Results
Decisions shifted significantly across the four categories (Stuart-Maxwell p<0.001). Overall, discharge rose from 5.9% to 8.7% and surgery from 21.1% to 26.7%, while repeat ultrasound fell from 46.0% to 40.2% and diuretic renogram from 27.0% to 24.4%. Shifts were risk-aligned (Table 1): in green-risk cases, discharge increased from 18.2% to 32.5% and diuretic renogram fell from 7.8% to 3.2%; in red-risk cases, surgery increased from 47.7% to 67.6%. Yellow-risk cases showed the most balanced direction (7.2% escalation vs. 6.4% de-escalation), consistent with genuine clinical equipoise. Overall, 19.6% of decisions changed (12.2% escalating, 7.4% de-escalating). Each 1-SD increase in model-predicted surgical risk was associated with greater odds of escalation (OR 3.32, 95% CI 2.77-3.98) and transition to surgery (OR 11.27, 95% CI 8.10-15.68), and lower odds of de-escalation (OR 0.12, 95% CI 0.07-0.18).
Conclusion
HSI exposure significantly altered hydronephrosis management in a risk-aligned manner: low-risk cases shifted toward discharge with less diagnostic testing, and high-risk cases shifted toward surgery. AI risk stratification may reduce uncertainty in intermediate cases and support more definitive, risk-appropriate care pathways, though prospective workflow studies linked to patient outcomes are needed before deployment.

Table 1. Management shifts after HSI model exposure by AI risk category.

HSI risk group	Discharge Δ	Repeat US Δ	Renogram Δ	Surgery Δ	Escalation	De-escalation
Green	+14.3%	−9.6%	−4.6%	−0.1%	0.5%	19.7%
Yellow	−2.4%	+1.7%	+5.4%	−4.8%	7.2%	6.4%
Red	+0.0%	−10.6%	−9.3%	+19.9%	25.3%	0.1%

---

Back to 2026 Abstracts