Large Language Models, Peritoneal Dialysis (PD)
Conditions
Brief summary
This study conducted a randomized controlled trial using clinical vignettes to evaluate differences in the quality of PD management among two large language model-assisted workflows and a physician-only decision-making process, and to identify potential risks (e.g., generating clearly erroneous or even harmful suggestions).
Interventions
The peritoneal dialysis-specialized large language model (PD-LLM) used in this study was jointly developed by the Department of Nephrology at the First Affiliated Hospital of Sun Yat-sen University and Digital Health China (DHC).
OTHERGeneralized LLM tools such as deepseek
Participants were permitted to ask the LLM any questions related to the clinical scenarios. However, they were explicitly instructed to critically evaluate the model's suggestions and to take full responsibility for the final clinical plans.
Sponsors
First Affiliated Hospital, Sun Yat-Sen University
Vantive Health LLC
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
TREATMENT
Masking
SINGLE (Outcomes Assessor)
Eligibility
Sex/Gender
ALL
Healthy volunteers
No
Inclusion criteria
1. Internal medicine or nephrology standardized training residents, licensed residents, or attending physicians. 2. Independent experience in PD management ≤ 3 years. 3. Provided signed informed consent and agreed to comply with the trial procedures.
Exclusion criteria
1. Direct involvement in the development or training of the specialized PD large language model, or in the construction of the clinical scenarios/ scoring criteria used in this trial. 2. Participation in the pilot testing of all clinical scenarios used in this trial. 3. Inability or unwillingness to access the study platform or use online resources during the study period. 4. Experienced PD experts.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Management Reasoning | Within 90-min study | Percent correct (range: 0 to 100) for each case. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Domain-Specific Scores | within 90-min study. | Percent correct (range: 0 to 100) for each case. |
| Severity of Potential Harm | Within 90-min study. | The severity of potential harm will be classified as none, mild-to-moderate, or severe. |
| Time Spent on Management | Within 90-min study. | The time participants spend per case. |
| Self-Reported Confidence per Case | Within 90-min study. | Scale 1-10. 10 represents being very confident. |
Countries
China
Outcome results
None listed