Improving the Reliability of LLMs as Medical Assistants for the General Public (LAMP-1)

Improving the Reliability of LLMs as Medical Assistants for the General Public: a Proof of Concept Simulation Trial

This study will evaluate whether three-minute six-dimensions education(3M-6D education) can improve the reliability of large language models as medical assistants for the general public. Participants will be randomly assigned to receive or not receive 3M-6D education and then use ChatGPT, Gemini, or non-AI information resources. The study will assess relevant condition identification, disposition concordance, red-flag identification, and NASA-TLX score.

Study Overview

• Status: Not yet recruiting
• Conditions: Relevant Conditions Identification
• Intervention / Treatment: Behavioral: three minutes six dimensions education; Other: ChatGPT; Other: Gemini

Detailed Description

This randomized, controlled, proof-of-concept simulation trial will evaluate whether three-minute six-dimensions education (3M-6D education) can improve the reliability of large language models as medical assistants for the general public.

Eligible participants will be randomly assigned in a 1:1:1:1:1 ratio to one of five study groups: the 3M-6D education GPT group, the GPT group, the 3M-6D education Gemini group, the Gemini group, or the control group. Participants in the 3M-6D education GPT and 3M-6D education Gemini groups will receive approximately three minutes of education before using ChatGPT or Gemini.Each participant will be randomly assigned one of 10 standardized clinical scenarios and complete a simulated counseling task in unrestricted natural language within approximately 10 minutes. The study will assess relevant condition identification, disposition concordance, red-flag identification, and NASA-TLX score.

• Study Type: Interventional
• Enrollment (Estimated): 525
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Xunming Ji; Phone Number: 01083198962; Email: jixm@ccmu.edu.cn
• Study Contact Backup: Name: Chuanjie Wu; Phone Number: 01083199439; Email: wuchuanjie@ccmu.edu.cn

Study Locations

• China
  • Beijing Municipality
    • Beijing, Beijing Municipality, China, 100053
      • Xuanwu Hospital, Capital Medical University
      • Contact: Chuanjie Wu; Phone Number: 010-83199439; Email: wuchuanjie@ccmu.edu.cn

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

1. Age 18 years or greater, male or female;
2. Completed primary school or higher education;
3. Able to use a smartphone or computer to complete online interaction;
4. No history of acute ischemic stroke, systemic lupus erythematosus, gastric ulcer, pneumonia, acute cardiac infarction, urinary tract infection, uterine fibroids, diabetes, osteoarthritis, or migraine.
5. Able to understand and comply with study procedures and to provide written informed consent.

Exclusion Criteria:

1. Currently or previously employed as a healthcare worker;
2. Previously received systematic medical training;
3. Currently involved in concurrent research that may interfere with the results of the present trial;
4. The investigator considered that the participant had other conditions that might affect compliance or preclude participation.

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Health Services Research
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

5

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: 3M-6D education GPT Group; Participants will first be trained in 3M-6D education, then use ChatGPT to complete a consultation task in unrestricted natural language in approximately 10 minutes.	Behavioral: three minutes six dimensions education; 3M-6D education is designed based on Cognitive Load Theory to reduce the cognitive burden on patients during medical interactions with AI and to improve the clarity and completeness of symptom reporting. Guided by cognitive load theory and the natural process physicians use to take medical histories, we identified candidate information dimensions and developed a structured expression framework with six dimensions for public health queries through a Delphi expert consensus process. Participants were instructed to use the framework to describe their symptoms across these six dimensions; this process can typically be completed within three minutes, so we call this approach three minutes six dimensions education (3M-6D education).; Other: ChatGPT; Participants use ChatGPT to complete a standardized simulated clinical scenarios in unrestricted natural language.
Experimental: 3M-6D education Gemini Group; Participants will first be trained in 3M-6D education, then use Gemini to complete a consultation task in unrestricted natural language in approximately 10 minutes.	Behavioral: three minutes six dimensions education; 3M-6D education is designed based on Cognitive Load Theory to reduce the cognitive burden on patients during medical interactions with AI and to improve the clarity and completeness of symptom reporting. Guided by cognitive load theory and the natural process physicians use to take medical histories, we identified candidate information dimensions and developed a structured expression framework with six dimensions for public health queries through a Delphi expert consensus process. Participants were instructed to use the framework to describe their symptoms across these six dimensions; this process can typically be completed within three minutes, so we call this approach three minutes six dimensions education (3M-6D education).; Other: Gemini; Participants use Gemini to complete a standardized simulated clinical scenarios in unrestricted natural language.
Active Comparator: GPT Group; Participants will use ChatGPT to complete a consultation task in unrestricted natural language in approximately 10 minutes.	Other: ChatGPT; Participants use ChatGPT to complete a standardized simulated clinical scenarios in unrestricted natural language.
Active Comparator: Gemini Group; Participants will use Gemini to complete a consultation task in unrestricted natural language in approximately 10 minutes.	Other: Gemini; Participants use Gemini to complete a standardized simulated clinical scenarios in unrestricted natural language.
No Intervention: Control group; Participants will use non-AI tools such as internet searches and medical websites to complete a consultation task in unrestricted natural language in approximately 10 minutes.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Relevant conditions identification of the 3M-6D education GPT group compared with the GPT group	Relevant conditions identification is defined as the proportion of participants whose final response includes the expert-defined final diagnosis or a relevant differential diagnosis.	Usually within 1 hour.
Disposition concordance of the 3M-6D education GPT group compared with the GPT group	Disposition concordance is defined as the proportion of participants whose final care recommendation matches the expert-defined level. The five levels are self-care, routine outpatient care, urgent outpatient care, emergency department visit, and emergency medical services.	Usually within 1 hour.
Relevant conditions identification of the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.
Disposition concordance of the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Relevant conditions identification of the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
Relevant conditions identification of the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.
Disposition concordance of the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
Disposition concordance of the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.
Red-flag identification in the 3M-6D education GPT group compared with the GPT group	Red-flag identification is defined as the proportion of participants whose final response includes the key warning signs that experts defined for the assigned scenario.	Usually within 1 hour.
Red-flag identification in the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
Red-flag identification in the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.
Red-flag identification in the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.
NASA Task Load Index score of the 3M-6D education GPT group compared with the GPT group	NASA-TLX score is a self-reported task-load score measured after the simulated consultation with a physician. It includes six domains: mental demand, physical demand, temporal demand, effort, frustration, and performance. Each domain is scored from 0 to 100. The total score is the mean of the six domains. Higher scores indicate greater perceived task load.	Usually within 1 hour.
NASA Task Load Index score of the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
NASA Task Load Index score of the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.
NASA Task Load Index score of the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.
Relevant conditions identification of the 3M-6D education GPT group compared with the 3M-6D education Gemini group		Usually within 1 hour.
Disposition concordance of the 3M-6D education GPT group compared with the 3M-6D education Gemini group		Usually within 1 hour.
Red-flag identification in the 3M-6D education GPT group compared with the 3M-6D education Gemini group		Usually within 1 hour.
NASA Task Load Index score of the 3M-6D education GPT group compared with the 3M-6D education Gemini group		Usually within 1 hour.

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Failure to identify red flags in the 3M-6D education GPT group compared with the GPT group	Failure to identify red flags is defined as the proportion of participants whose final response does not include the expert-defined red-flag symptoms or warning signs for the assigned standardized simulated clinical scenario.	Usually within 1 hour.
Failure to identify red flags in the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
Failure to identify red flags in the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.
Failure to identify red flags in the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.
Underestimation of disposition in the 3M-6D education GPT group compared with the GPT group	Underestimation of disposition is defined as the proportion of participants whose final care recommendation is lower than the expert-defined disposition level for the assigned standardized simulated clinical scenario.	Usually within 1 hour.
Underestimation of disposition in the 3M-6D education GPT group compared with the control group		Usually within 1 hour.
Underestimation of disposition in the 3M-6D education Gemini group compared with the Gemini group		Usually within 1 hour.
Underestimation of disposition in the 3M-6D education Gemini group compared with the control group		Usually within 1 hour.

Collaborators and Investigators

• Sponsor: Capital Medical University
• Collaborators: Xuanwu Hospital, Beijing

Study record dates

Study Major Dates

• Study Start (Estimated): June 20, 2026
• Primary Completion (Estimated): July 20, 2026
• Study Completion (Estimated): July 20, 2026

Study Registration Dates

• First Submitted: June 11, 2026
• First Submitted That Met QC Criteria: June 11, 2026
• First Posted (Actual): June 16, 2026

Study Record Updates

• Last Update Posted (Actual): June 16, 2026
• Last Update Submitted That Met QC Criteria: June 11, 2026
• Last Verified: June 1, 2026

More Information

Terms related to this study

• Keywords: Large Language Models; 3M-6D education; relevant conditions identification; Cognitive Load Theory
• Other Study ID Numbers: LAMP-1

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No