# A prospective single-masked, non-inferiority, parallel-group randomized controlled trial of the efficacy of a ChatGPT-based AI chatbot to improve Boston bowel preparation scores for colonoscopy preparation: A trial protocol

## TL;DR

This study will test if a ChatGPT-based AI chatbot is as effective as standard counseling in helping patients prepare for colonoscopies.

## Contribution

The novel use of ChatGPT 4.0 as a counseling tool for colonoscopy preparation is evaluated in a randomized controlled trial.

## Key findings

- The trial will compare ChatGPT-based counseling to standard counseling for bowel preparation adherence.
- Primary outcome is the Boston Bowel Preparation Score (BBPS) to assess non-inferiority.
- Secondary outcomes include patient anxiety and satisfaction measured using standardized questionnaires.

## Abstract

Artificial intelligence (AI) is transforming healthcare through tools like large language model chatbots. AI chatbots can simulate human conversation, provide personalized information, and interact with patients in real time. Their ease of use and conversational interface make them attractive for healthcare education, especially in resource-limited settings. We propose a prospective, single-masked, randomized controlled trial to evaluate whether an AI-based chatbot (ChatGPT) is non-inferior to standard counseling in terms of patients’ adherence to pre-colonoscopy bowel preparation instructions and thus enhance the Boston Bowel Preparation Score (BBPS). Patients undergoing colonoscopy (ntotal = 96) will be randomized to ChatGPT 4.0 Large Language Model (LLM)-aided Colonoscopy Counseling (NChatGPT = 48) or standard counseling (nsc = 48) arms at a 1:1 ratio using a central block randomization scheme of varying block sizes. In the first group, participants will interact with ChatGPT 4.0 for bowel preparation counseling before colonoscopy, whilst the second group will receive standard counseling from trained clinicians. Only the outcome assessors will be masked to the intervention allotment. The primary endpoint is the BBPS, assessed for non-inferiority. Secondary endpoints are patient anxiety (DASS-21) and patient satisfaction assessed using DASS-21 and PSQ-18 questionnaires, respectively and the findings will be reported descriptively with two-sided 95% confidence interval and any p-values will be considered exploratory without multiplicity adjustment. The primary endpoint data will be analyzed using the intention-to-treat (ITT) analysis and non-inferiority framework based on the analysis of covariance (ANCOVA) to control the confounders (age, gender (male as the risk factor), prior colonoscopy experience, colonoscopy indication, and baseline constipation score). The results will be compared with the findings based on the per-protocol (PP) analysis as part of the sensitivity analysis. The protocol adheres to SPIRIT 2025 and the SPIRIT-AI extension guidelines to ensure comprehensive reporting of this AI-based intervention. This trial has received ethics approval and the trial protocol has been registered with the clinicaltrials.gov registry (NCT06905782).