Integrating an AI-Driven Hydronephrosis Decision-Making Tool

N/ANot Yet RecruitingNCT07581223

The Hospital for Sick Children322 enrolled

Overview

Hydronephrosis is a common congenital kidney anomaly. While most cases resolve on their own, some require surgery. Clinicians rely on repeated ultrasounds and sometimes invasive tests to decide if surgery is needed, but predicting outcomes is difficult. Researchers at SickKids developed an AI model that analyzes ultrasound images to assist in diagnosing and managing hydronephrosis. This study tests how well the AI integrates into real-world care. Clinicians will first make care decisions without AI and then review the AI's prediction before deciding whether to change their plan. A separate expert, unaware of whether AI influenced the first clinician's plan, will make the final decision to ensure care remains unchanged. The study will assess whether AI improves decision-making, reduces unnecessary tests, and fits into clinical workflows. If successful, the AI model could serve as a complementary tool to make diagnoses more efficient and precise while minimizing invasive procedures.

Study Type

INTERVENTIONAL

Allocation

Purpose

OTHER

Masking

NONE

Enrollment

322

Conditions

Hydronephrosis Hydronephrosis Congenital

Interventions

Machine learning modelOTHER

The AI intervention is a deep learning algorithm used to predict obstructive hydronephrosis. It was developed at SickKids and has recently completed the silent trial phase. This clinical trial aims to validate the model's clinical integration by assessing its impact on clinician decision-making and care plan recommendations. To uphold standard care, a blinded clinician will make final decisions.

Eligibility

Sex: ALLMin age: 0 MonthsMax age: 24 Months

Medical Language ↔ Plain English

Inclusion Criteria: * Seen for HN in-person in the Pediatric Urology clinic with ultrasound scans taken at SickKids * New and follow-up patients 0-24 months. Exclusion Criteria: * Older than 24m * Concurrent urinary tract anomalies (duplex configurations; PUV etc.) * History of renal surgical intervention (post-op patients)

Outcomes

Primary Outcomes

Change in Clinician Management Decisions Following Exposure to the AI Model

The proportion of clinician management decisions revised immediately after exposure to the AI model output. Management decisions include: (1) discharge, (2) monitor with ultrasound, (3) additional invasive testing, or (4) referral for surgery.

Time frame: Immediately after AI model exposure during each case review session, through study completion (average of 6 months)

Secondary Outcomes

Agreement Between Clinician Decisions and Expert Reference Decisions Using Cohen's Kappa

Agreement between clinician management decisions and the expert reference decision will be assessed before and after AI exposure using Cohen's kappa statistic. Higher kappa values indicate greater agreement.

Time frame: Immediately after clinician review and AI model exposure during each case review session, through study completion (average of 6 months)

Proportion of Management Decision Changes Stratified by Clinician Experience Level

The proportion of clinician management decisions revised after AI model exposure will be compared across clinician subgroups, including training level and years of experience.

Time frame: Immediately after AI model exposure during each case review session, through study completion (average of 6 months)