AI-empowered Nudge to Improve Colonoscopy Uptake (AINC)

We will conduct a two-arm, parallel-group, cluster-randomized controlled trial to evaluate the effectiveness of an AI-empowered nudge model in improving colonoscopy uptake (AINC) among high-risk individuals aged 45 to 74. The AI-empowered nudge model combines default screening nudging with an AI chatbot on colorectal cancer screening. We will also evaluate the feasibility of this AINC model, and identify the facilitators and barriers to its real-world adoption.

The colonoscopy uptake rate is approximately 15% in China, and the proposed intervention is expected to increase this rate by 10%. Sample size calculation, based on detecting an increase in colonoscopy uptake from 15% to 25% with 90% power (α=0.05, two-sided), an ICC of 0.05, and 30 clusters per arm, indicates a need for 24 participants per cluster. There are 720 per arm, and 1440 in total. Allowing for 15% attrition, the final sample size is determined to be 1680 from 70 clusters. As a pragmatic trial in real world, the number of participants each cluster depends on the population size of the respective villages or communities. All eligible participants in the participating villages or communities will be included in the study.

Participant recruitment will be conducted across 70 villages/communities in three representative counties/cities in China, covering urban, suburban, and rural areas. Cluster randomisation will be performed at the level of villages or communities using a stratified block design to ensure balanced allocation across the two trial arms. Stratification factors include geographic access to colonoscopy hospital and the size of individuals aged 45 to 74 for each cluster. Clusters with comparable levels of these factors will be grouped into blocks within each city and then randomly assigned within each block to the AINC or control group. The random allocation sequence will be generated by an independent statistician using a computer-based random number generator in R software and implemented via a secure centralised system.

The study procedure involves first identifying high-risk individuals for CRC through an initial risk assessment questionnaire and a fecal immunochemical test (FIT). Those who meet the criteria will then receive the intervention corresponding to their village's assigned study arm. Participants in the intervention group will receive an AI-powered nudge for colonoscopy (AINC), featuring an AI chatbot providing real-time personalized responses and a nudge environment with default screening option, followed by message reminders once per two weeks. The control group will receive usual care. Colonoscopy uptake will be collected via the hospital information system at the 3-month follow-up.

The primary analysis will follow the intention-to-treat (ITT) principle, while the per-protocol (PP) analysis will serve as the secondary analysis. In the ITT analysis, all subjects randomized to each group will be included. Between-group comparisons for continuous and categorical variables will utilize t-tests and chi-square tests. The primary outcome (colonoscopy uptake) will be analyzed using Generalized Estimating Equations (GEE), and the secondary outcome (colonoscopy duration) will be analyzed using Generalized Linear Mixed Models (GLMM); both approaches adjust for cluster effects and relevant covariates to obtain robust estimates. Covariates include region, age, sex, smoking history, Body Mass Index, history of bowel-related symptoms or diseases, and family history. The timing of colonoscopy uptake will be analyzed using Kaplan-Meier survival curves and log-rank tests, and the intervention effects on the time-to-event will be quantified with a Cox proportional hazards model. Subgroup analyses will be conducted to elucidate the effect heterogeneity across populations stratified by pre-specified characteristics, including region, age, sex, smoking history, Body Mass Index, history of bowel-related symptoms or diseases, and family history.