The goal of this clinical trial is to learn how artificial intelligence (AI) may help doctors make diagnoses in kidney medicine. The researchers want to know whether an AI tool called a large language model (LLM) can help doctors choose the correct diagnosis more often and feel more confident in their answers.

Before starting the study, the research team tested several AI models and chose one of the best performers, a GPT-5-class model set to use high reasoning effort.

The main questions this study aims to answer are:

1. Do doctors make more correct diagnoses when they can see AI suggestions?
2. Does seeing AI suggestions change how confident doctors feel about their diagnosis?

Researchers will compare doctors who receive AI suggestions with doctors who do not receive AI suggestions to see how the AI affects accuracy, confidence, and decision-making.

Participants will complete up to 10 online clinical cases. For each case, they will:

1. Read a short medical scenario
2. Suggest up to three possible diagnoses

(If in the AI group) Review the AI's suggestions and decide whether to change their answer

The study will also look at how long participants take to answer each case and how the AI's performance compares to the human answers.

Reasoning Enrichment With Feedback From IA in NEphrology Trial

Diagnosis

Clinical Decision-making

Artificial Intelligence (AI) in Diagnosis

Decision Support Systems, Clinical

This study evaluates whether providing clinicians with real-time diagnostic suggestions from a high-reasoning large language model (GPT-5) improves diagnostic accuracy, confidence, and efficiency when solving nephrology clinical vignettes. Prior to selecting the model for the trial, the research team benchmarked several state-of-the-art models across a pilot set of nephrology cases, including: GPT-5, GPT-5-mini, O3, GPT-4o, Llama-4 Maverick-17B, Gemini-2.5-Pro, Qwen-3 VL-235B Thinking, DeepSeek-V3.2-Exp, MedGEMMA-27B, Claude Sonnet-4.5, and Magistral-Medium-2509. GPT-5 (high-reasoning) demonstrated the highest diagnostic performance, stability, and interpretability, and was selected as the AI system used in the intervention arm.

Participants include medical students, residents, fellows, and practicing physicians. After creating an account, participants complete a demographic questionnaire (specialty, years of experience, practice type, age category, AI familiarity) and must explicitly agree to the use of these data for research purposes before accessing the vignettes. No directly identifying information is collected.

Participants are randomized (with stratification by professional status) to either the AI-supported arm or the control arm. Each participant is assigned 10 nephrology vignettes in French or English and may complete them over multiple sessions. Once a vignette is submitted, it cannot be revisited ("no backtracking"). Completion time per vignette is automatically recorded.

Control Arm

Participants view each vignette and provide up to three diagnoses ("Top-3"), followed by a confidence rating (0-10).

AI-Supported Arm

Participants first enter an initial Top-3 diagnosis and confidence rating without AI assistance. The system then displays GPT-5's diagnostic suggestions, after which participants may revise their diagnoses once. The vignette is locked after submission.

The study collects:

* initial and final diagnoses,
* confidence ratings before and (if applicable) after AI suggestions,
* completion times,
* participant demographic variables,
* and the AI model's own diagnostic outputs.

Partial completion is permitted; all completed vignettes contribute to the analysis.

Primary and secondary outcomes include diagnostic accuracy (Top-3 and Top-1), accuracy improvement before vs. after AI, changes in diagnostic confidence, AI-induced diagnostic errors, human-versus-AI benchmarking, completion-time efficiency metrics, and the proportion of assigned vignettes completed.

The primary analysis will compare diagnostic accuracy between the control arm (physicians alone) and the experimental arm (physicians assisted by the AI model). Accuracy is analyzed as a binary outcome (correct vs incorrect diagnosis). Because each participant evaluates multiple clinical vignettes, accuracy will be modeled using a mixed-effects logistic regression with a fixed effect for study arm and random intercepts for both participant and vignette. This approach accounts for clustering and varying difficulty across cases. The primary hypothesis test uses a two-sided α = 0.05. Effect sizes will be reported as odds ratios with 95% confidence intervals. Secondary analyses will explore whether accuracy varies by demographic factors (e.g., experience level, specialty) using interaction terms.

Because each participant evaluates multiple vignettes, the team also performed simulation-based power analyses using mixed-effects logistic regression models with random intercepts for both participant and vignette, assuming an intra-participant ICC of 0.10. Under these assumptions, a total sample of 100 participants (50 per arm) with 10 vignettes per participant provides >99% power to detect a clinically meaningful improvement in diagnostic accuracy. The investigators therefore plan to enroll approximately 100 participants overall.

This study aims to quantify whether AI-augmented reasoning meaningfully improves diagnostic performance and decision-making when clinicians evaluate complex nephrology cases.

AI suggestion

Reasoning Enhancement With Feedback From a Generative AI in Nephrology (REFINe): A Randomized Evaluation of Generative AI Support in Nephrology Diagnosis

For each participant, proportion of vignettes where the correct main diagnosis is included in the participant's final top-3 diagnoses. Compare final top-3 accuracy between the AI arm (after AI suggestions) and the control arm (no AI).

Percentage of correctly diagnosed cases (top-3).

For each participant, proportion of vignettes where the correct main diagnosis is included in the participant's final top-1 diagnoses. Compare final top-1 accuracy between the AI arm (after AI suggestions) and the control arm (no AI). Percentage of correctly diagnosed cases (top-1).

In the AI-supported arm, participants first provide an initial answer (up to three diagnoses) without AI suggestions, then see AI-generated suggestions and may revise their answer once; they cannot return to that vignette later. For each participant, the investigators compute the difference in top-3 accuracy between initial and final answers across all completed vignettes.

Percentage-point change in Top-3 diagnostic accuracy

In the AI-supported arm, participants first provide an initial answer (up to three diagnoses) without AI suggestions, then see AI-generated suggestions and may revise their answer once; they cannot return to that vignette later. For each participant, the investigators compute the difference in top-1 accuracy between initial and final answers across all completed vignettes.

Percentage-point change in Top-1 diagnostic accuracy

Participants in both arms rate their confidence (0-10 scale) in their Top-3 diagnostic proposal before any AI suggestions.

In the AI arm, this is the "pre-AI" rating. In the Control arm, this is the single confidence rating (since no AI is shown).

The investigators compare the pre-AI confidence between arms, aggregated across all completed vignettes per participant.

Final diagnostic confidence (0-10 scale) in the Top-3 diagnostic proposal across all completed vignettes, compared between arms.

In the AI arm, this is the post-AI confidence rating. In the Control arm, this is the same confidence rating (participants do not receive AI suggestions).

In the AI arm, participants provide confidence ratings (0-10 scale) for their Top-3 diagnoses both before and after seeing AI suggestions.

For each participant, the investigators compute the within-participant change (post-AI minus pre-AI) across all completed vignettes.

Change in confidence score (0-10 scale)

Among completed vignettes where the participant's initial Top-1 diagnosis is correct, proportion for which the final Top-1 diagnosis becomes incorrect after AI suggestions.

Among completed vignettes in the AI arm, the proportion where the Top-3 diagnosis differs between pre-AI and post-AI answers.

For each vignette, the Top-3 diagnostic accuracy of human participants before any AI suggestions (combining participants from both study arms at their pre-AI stage) is compared with the Top-3 diagnostic accuracy of the AI model for the same vignette. The reported Outcome is the accuracy difference, defined as AI Top-3 accuracy minus human pre-AI Top-3 accuracy, expressed in percentage points and computed at the vignette level across all completed vignettes.

Percentage-point difference in Top-3 diagnostic accuracy

For each vignette completed in the AI-supported arm, the Top-3 diagnostic accuracy of human participants after viewing AI suggestions is compared with the Top-3 diagnostic accuracy of the AI model.

(Top-3 accuracy is a single measure) The reported Outcome is the accuracy difference, defined as AI Top-3 accuracy minus human post-AI Top-3 accuracy, expressed in percentage points and computed at the vignette level across all completed vignettes in the AI arm.

Percentage-point difference in Top-3 diagnostic accuracy between AI and human

For each vignette, the platform records the time from vignette opening to answer submission. In the control arm, a single completion time is recorded for each vignette. In the AI-supported arm, completion time is recorded before viewing AI suggestions and again after viewing AI suggestions. The Outcome reports the difference in completion time between study arms, expressed in seconds and calculated across all completed vignettes.

Seconds (difference in completion time)

For each participant, the proportion of the 10 vignettes completed within the study period, compared between arms.

Participants in this arm will complete the same clinical case vignettes as the control group. For each case, they will receive a suggested diagnosis generated by a large language model (GPT-5, high-reasoning configuration), which was selected after internal benchmarking. Participants can review the AI suggestion before entering their own final diagnostic answer. No additional information, prompts, or coaching is provided. The intervention consists solely of displaying the AI-generated diagnostic suggestion during the case-solving task.

Participants in this arm will complete the clinical case vignettes independently, without any AI-generated diagnostic suggestions. They will read each vignette and provide their own diagnostic answer based solely on the information presented. No external decision support or additional materials are provided.

Reasoning Enrichment With Feedback From IA in NEphrology Trial

About this trial

Study ID

Status

Type

Phase

Placebo

Accepting

Trial Timing

What is this trial about?

What are the participation requirements?