Using AI feedback to improve diagnostic reasoning in kidney medicine
Official title: Reasoning Enhancement With Feedback From a Generative AI in Nephrology (REFINe): A Randomized Evaluation of Generative AI Support in Nephrology Diagnosis
Run by University Hospital, Lille · NCT07352475 · Data updated 2026-11-12
This test sees if an advanced AI language model can help medical students, residents, fellows, and physicians make more accurate and confident kidney-related diagnoses using online clinical cases.
Quick facts
| Study type | Interventional |
|---|---|
| Ages | 18 Years and up |
| Sex | All |
| Enrollment | 100 (estimated) |
| Sponsor | University Hospital, Lille Academic / other |
| Locations | 1 site (Lille) |
| Status | Recruiting |
| Started | November 20, 2025 |
| Main results expected | October 31, 2026 |
| Study ends | December 31, 2026 |
| Registry record updated | January 20, 2026 |
| Trial ID | NCT07352475 on ClinicalTrials.gov |
Is this trial a fit for you?
✓Likely a good fit
Ideal participants are adults with at least basic medical training (medical students, residents, fellows, or practicing clinicians) who can read English or French and have internet access.
✕Probably not a fit
People without medical training, those under 18, or anyone unable to complete online procedures or who lacks English/French ability or internet access are unlikely to benefit from participating.
A plain-English reading of the study's rules. Only the study team can confirm whether you qualify.
What this trial studies
Clinicians are randomized to either receive real-time diagnostic suggestions from a high-reasoning GPT-5-class model or to complete cases without AI support, and each participant completes up to 10 nephrology clinical vignettes presented in English or French. The research team previously benchmarked multiple state-of-the-art models and selected GPT-5 (high-reasoning) based on diagnostic performance, stability, and interpretability. Participants create an account, provide demographic information, consent to data use, and complete vignettes online without backtracking while completion time and responses are recorded. Analysis compares diagnostic accuracy, confidence, and decision-making between the AI-supported and control arms with stratified randomization by professional status.
Why it matters
Potential benefit: If successful, clinicians could arrive at correct nephrology diagnoses more often and with greater confidence, which may translate into faster or more appropriate patient care.
How similar studies have gone: Previous vignette and reader studies using earlier large language models and decision-support tools have sometimes improved diagnostic accuracy but have produced mixed results, so this application is promising but not yet proven.
Who can join: the full rules
Show full inclusion / exclusion criteria
Inclusion Criteria: Adults aged 18 years or older. Able to read and answer clinical vignettes in English or French. Access to a computer or smartphone with an internet connection. Provides informed consent online. Participants are expected to have at least basic medical training (e.g., medical students, residents, fellows, or practicing clinicians), although no formal verification is required. Exclusion Criteria: Individuals under 18 years of age. Inability to complete online study procedures. Prior involvement in the design, development, or evaluation of the AI system used in this study.
Where it's running
-
Lille University Hospital (online study)
Lille, France Recruiting
Study contacts
- Study coordinator: Raphaël BENTEGEAC, MD, MPH
- Email: raphael.bentegeac@univ-lille.fr
- Phone: +33651204000
Similar studies
Other studies for the same condition or treatment. Recruiting studies are listed first.
-
AI analysis of ECGs to detect structural heart disease
This project will test whether AI can read traditional and wearable ECGs to find undiagnosed valve disease, pulmonary hypertension, and heart failure in adults.
-
Benchmarking AI language models against tumor boards for oncology treatment decisions
This project tests whether frontier AI language models can match the treatment recommendations of multidisciplinary tumor boards for common cancers using standardized synthetic cases.
-
Does AI-based cephalometric tracing match digital manual tracing for orthodontic measurements?
This study tests whether AI-powered cephalometric analysis can produce measurements similar to traditional digital tracing in 100 patient radiographs.
-
Virtual Reality Simulation for Psychiatric Nursing Decision-Making
This study tests whether a schizophrenia-focused virtual reality simulation helps psychiatric nursing students make better clinical decisions and feel more confident, while easing worry before and during clinical practice.
-
Heart score calculator test using GPT-4o and Claude in chest pain patients
This study tests whether two large language models can calculate HEART scores from patient notes and predict 30-day adverse heart events in adults with chest pain in the emergency department.
-
How cervical ultrasound settings affect AI prediction of preterm birth
This study tests whether an AI model for predicting preterm birth performs differently when cervical ultrasound images are taken with different imaging settings.
How to take part
- Review the eligibility criteria above with your treating physician.
- Check the official listing on ClinicalTrials.gov for the latest contact details and recruitment status.
- Contact the study coordinator or principal investigator to ask for pre-screening. Pre-screening is free and never obligates you to enroll.