An Artificial Intelligence-Guided Strategy to Reduce Poor Bowel Preparation: A Multicenter Randomized Controlled Study

Christopher Vélez, MD

Program Director, Advanced Fellowship in Neurogastroenterology/Motility, Massachusetts General Hospital Center for Neurointestinal Health, Division of Gastroenterology, Hepatology, and Endoscopy, Mass General Brigham Department of Medicine, Harvard Medical School, Boston, MA

This summary reviews Gimeno-García AZ, Benítez-Zafra F, Redondo-Zaera I, et al. An artificial intelligence-guided strategy to reduce poor bowel preparation: A multicenter randomized controlled study. Am J Gastroenterol. 2026;121(4):890-898.

Access this article through The American Journal of Gastroenterology.

Keywords: artificial intelligence, bowel preparation, colonoscopy

STRUCTURED ABSTRACTQuestion: Can artificial intelligence (AI) be used to reduce poor bowel preparation?

Design: Multicenter, randomized, controlled, parallel-group trial.

Setting: Four hospitals in Spain.

Patients: Inclusion criteria included those older than 18 years of age, those who had a scheduled outpatient colonoscopy, and those who agreed to participate in the intervention. Exclusion criteria included subjects who had reasons for which colonoscopy could not be performed, diagnosis of certain metabolic conditions which could make preparation challenging, those who had had a history of total or subtotal colectomy, those who would have had difficulty ingesting bowel preparations such as those as dementia, those lacking a smartphone in order to be able to be given the intervention those who were not technologically literate, and those who refused participation.

Interventions/Exposure: Eligible subjects were randomized 1 to 1 using a computer-generated scheme at a research university in Spain. Group 1 included those individuals who would be deemed control who received standard oral and written bowel preparation instructions. Group 2 deemed the intervention received the same instructions but in addition were instructed to use a mobile phone application. Each subject was provided an individual code to access the application, or a code was provided if any participant had any difficulty. These subjects then followed the software instructions and were then told to take an image of their last rectal effluent in the toilet bowl under specific conditions. This would be deemed as adequate or inadequate for the purposes of endoscopic evaluation. If the preparation was deemed to be adequate, participants were advised to proceed with colonoscopy as previously scheduled. If the preparation was deemed to be inadequate, additional bowel preparation was recommended. Colonoscopies were performed by experienced endoscopists who were blinded to the assigned group. The Boston Bowel Preparation Scale was used to assess the quality of bowel cleansing.

Outcome: The primary outcome of this study was to assess the proportion of subjects in each arm with adequate bowel preparation as measured by the Boston Bowel Preparation Scale. According to the scheme each bowl segment is scored from zero to three points with the maximum possible score of nine and a score of two or more points per segment is considered adequate. Secondary outcomes included proportion of subjects with excellent bowel preparation, the use of mobile application among patients assigned to the intervention group, as well as satisfaction with the intervention using a questionnaire. Finally, both groups were compared via the adenoma detection rate.

Data Analysis: Associations were reported as odds ratios with typical characteristics surrounding confidence intervals. There was also a per protocol analysis of the primary outcome as well as an intention to treat analysis.

Funding: Funding for this study was provided in part from the Spanish Society of Gastrointestinal Endoscopy. The foundation was not involved in any phase of the research.

Results: In this study, 774 participants were eligible and randomized. The intention to treat analysis demonstrated statistically significant differences in bowel cleansing quality favoring the intervention group receiving the application examining rectal effluent through AI. Namely 91% of the individuals in the intervention arm had favorable bowel cleansing quality parameters versus 84.2% in the control arm. The differences between right and left exhibited better cleaning in the intervention group as well (90.4% versus 84.8%).

Table 1: Comparison of adequate bowel cleansing in control vs intervention/AI group.

COMMENTARY

Why Is This Important?
This study¹ demonstrates the use of AI as a way in which to bolster clinical productivity as well as the success of a relatively invasive procedure like a colonoscopy. Bowel preparation remains one of the most challenging elements of colorectal examination through endoscopy. Many patients find the process to be unpleasant. Individuals also find out often at the last minute that they are unable to go through colonoscopy evaluation due to inadequate prep. From the perspective of clinical throughput as well as revenue generation, inability to perform a scheduled examination results in a wasted spot as well as a potential for lost income. This is due to the relatively static nature in which bowel preparation instructions are typically administered, in the form of written or verbal instructions. This study demonstrates a more dynamic process as being potentially successful and easily adoptable throughout the world. Namely, by having patients able to interact with an application at home prior to endoscopy they can alert offices to difficulty with preparation or only present to endoscopy suites when their preparation has completed. That the Boston Bowel Preparation Scale was used makes the study easy to interpret as well as applicable in other centers².

Key Study Findings

The randomization scheme appears to be successful in this study with no major differences between the intervention and control groups. In both the intention to treat and per protocol analysis group, there were improvements in bowel preparation in different segments as well as overall in the entire colon in the intervention AI group. As expected, given that an expert group of endoscopists were selected for the study there were no major differences in this study as it related to other colonoscopy outputs such as adenoma detection rate period.

Caution
While overall I find that the study was very well executed and has the potential to provide additional tools for bowel preparation prior to colonoscopy, I do have some reservations regarding its use in clinical practice. I am concerned about the types of groups that were excluded because these are the types of groups that tend to have difficulty with bowel preparation and thus a higher chance of having incomplete colonoscopy. Namely those with cognitive difficulties or those who have had prior surgical interventions are individuals that perhaps would be better served by identifying additional tools in which to reduce the risk of failed bowel preparation. Given the need to be technologically savvy to participate this also can worsen inequity, namely among older individuals who are not as able to navigate newer technologies, or people with reduced resources who are unable to use or access a smartphone. Additionally, while this is outside of the scope of the study, it would seem to be difficult to implement this in a high-volume practice. For example, if an early morning case needs more bowel preparation and the rest of the morning and afternoon is fully scheduled, one could be left with the troubling situation where one takes additional purgative only to be told that there is no availability later in the day to perform the procedure. I would recommend that practices considering use of an AI or software-based approaches to determine bowel prep operation quality make sure that there is some flexibility to allow for changing of schedules the day of procedures to allow for people to be given additional bowel preparation. I would also state as a potential limitation use of this application in people who have reduced vision as well or difficulties with literacy. Additionally, a recent United States task force³ would have the control arm fall below accepted quality standards (>90%); this may exaggerate the reported benefit in the intervention arm.

My Practice
In my practice I could see the immediate possible benefits of such a strategy as employed in the intervention arm. Like many colleagues at my institution, I have a heavy portion of my endoscopy practice dedicated to colorectal cancer screening. I have taken care historically of two special populations who are at risk of poor bowel preparation outcomes. One population includes those individuals I care for in a community health center with limited English proficiency. An application could, when appropriately translated, allow for some of the inequity that exists related to language proficiency to be reduced by being able to give additional administrations of bowel preparation in a language concordant fashion without needing English. The second population of individuals that care for at risk of poor bowel preparation are people with cystic fibrosis given the intraluminal nature of the disease. While a priori such individuals are recommended to take more aggressive purgative at times it may not be successful, particularly in those with the history of prior surgery such as surgical correction of meconium ileus. Use of this application could potentially spare those patients with CF who may not need additional bowel preparation but are blanket recommended to take additional bowel preparation. It could allow for a more dynamic administration to reduce the burden in people with CF as well.

For Future Research
For future research, it would be valuable to see this application validated in other languages, as well as further exploration of different software use with different cell phone types. A lower technology effort should alsob be developed to improve access to such advances for those who historically have been marginalized both in research and in clinical care.

Conflict of Interest
Dr. Vélez has no relevant conflicts.

REFERENCES

Gimeno-García AZ, Benítez-Zafra F, Redondo-Zaera I, et al. An artificial intelligence-guided strategy to reduce poor bowel preparation: A multicenter randomized controlled study. Am J Gastroenterol. 2026;121(4):890-898.
Lai EJ, Calderwood AH, Doros G, Fix OK, Jacobson BC. The Boston Bowel Preparation Scale: A valid and reliable instrument for colonoscopy-oriented research. Gastrointest Endosc. 2009;69(3 Pt 2):620-5.
Jacobson BC, Anderson JC, Burke CA, et al. Optimizing bowel preparation quality for colonoscopy: consensus recommendations by the US Multi-Society Task Force on colorectal cancer. Am J Gastroenterol. 2025;120(4):738-764.