CanadaHealthcare and life sciences2024 – ongoingThe Hospital for Sick Children
Hydronephrosis ultrasound AI silent trial
SickKids reported a clinician's invasive-scan referral rate fell from 80% to 58% after observing an ultrasound AI model
Problem
Children with hydronephrosis could require invasive nuclear scans to investigate kidney obstruction. Ultrasound interpretation helped clinicians decide which children needed further investigation.
Approach
SickKids developed an AI model to predict hydronephrosis severity from ultrasound images. A nurse practitioner repeatedly observed its inputs and outputs during a silent trial, in which model predictions did not directly inform care decisions. She subsequently anticipated its classifications.
Results
Takeaway
Observing a model can change clinical judgment even during a silent trial. The reported referral change concerned a single clinician and did not establish a hospital-wide treatment effect.
Categories
- Size
- Large enterprise
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Computer visionPredictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
SickKids described clinician learning during a hydronephrosis AI silent trial. It reported referrals falling from 80% to 58% and warned that model exposure could influence clinical behavior.
Full details are in the original sources.