No Disease State or Condition
Conditions
Brief summary
Colorectal cancer survivors often face unique nutritional challenges and require support in their recovery and long0term health. While human experts have traditionally provided that support, there has been an increase in the use of Large Language Models (LLM) in medicine and in nutrition. The LLM offers a potential supplementary resource for generating personalized nutritional advice, specifically in personalized messaging. However, the efficacy and reliability of these AI-generated messages in comparison to human expert advice remain underexplored specific to this population. This study aims to compare the nutrition-related content generated by popular LLMs-ChatGPT, Claude, Gemini, and Co-Pilot-against messages crafted by human experts. By evaluating the generated content in terms of readability, thematic relevance, medical relevance, perceived effectiveness, and implementation of participants' clinical practice, this research will provide insights into the strengths and limitations of using AI for nutritional guidance in colorectal cancer care.
Interventions
OTHERNutritional Messaging
Dieticians will evaluate nutritional messages created by LLM and Human Experts.
Sponsors
University of Minnesota
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
HEALTH_SERVICES_RESEARCH
Masking
NONE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
* 18+ years of age * Currently practicing Registered Dietitian Nutritionist with at least five years of experience working with oncology patients and survivors in their practice. * Must have access to computer and internet access.
Exclusion criteria
* Non-English speakers, as the study materials and assessments are in English. * Experts with conflicts of interest related to any of the LLMs that are being evaluated.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Outcome Measure Title: Readability of Nutrition Messages | 8 to 12 months | Description: The readability of AI-generated and human expert-generated nutrition messages will be measured using the Flesch-Kincaid Grade Level tool. Unit of Measure: Grade level score (numerical score indicating reading difficulty level). Measurement Tool: Flesch-Kincaid Grade Level formula. Scale values: The values vary from 0 to 18, where 18 represents the most difficult text. |
| Outcome Measure Title: Thematic Relevance of Nutrition Messages | 8 to 12 months | Description: Thematic relevance of nutrition messages will be assessed by experts in nutrition using a thematic coding framework specifically designed for this study. Unit of Measure: Percentage (%) of messages that align with pre-determined thematic codes relevant to colorectal cancer survivorship. Measurement Tool: Thematic coding framework created by the research team. Scale values: The themes are capability (C), opportunity (O), and motivation (M) as three key factors capable of changing behavior (B). |
| Outcome Measure Title: Medical Relevance to Colorectal Cancer Survivors | 8-12 months | Description: Medical relevance will be rated by specialists using a 0-5 relevance rating scale. Unit of Measure: Mean relevance score (0-5). Measurement Tool: Dietitians/Participants review using a relevance rating scale. Scale value: 1-5 (1- least, 5- most) |
| Outcome Measure Title: Perceived Effectiveness of Nutrition Messages | 8-12 months | Description: Perceived effectiveness will be measured using a mean relevance score (1-5) administered to dietitians and participants. Unit of Measure: Mean relevance score (1-5). Measurement Tool: Dietitians/Participants survey. Scale value: 1-5 (1- least, 5- most) |
| Outcome Measure Title: Potential for Implementation in Clinical Practice | 8-12 months | Description: Feasibility for clinical implementation will be rated by dietitians using a 1-5 feasibility scale. Unit of Measure: Mean feasibility score. Measurement Tool: Dietitians/Participants survey. Scale value: 1-5 (1- least, 5- most) |
Countries
United States
Outcome results
None listed