The latest cancer statistics report high mortality and morbidity rates for colorectal cancer (CRC) worldwide [1]. Colonoscopy aids in the detection and resection of precancerous lesions, thereby reducing CRC-related morbidity and mortality [2-8]. Colonoscopy quality indicators include the adenoma detection rate (ADR), cecal intubation rate, withdrawal time, and bowel preparation (BP) [9-11]. Notably, the quality of BP is crucial for the other indicators to be maintained appropriately [12-15]. For instance, BP is associated with an improvement in ADR and with a decreased risk of CRC mortality [16].

Multiple factors affect BP quality, including patient adherence to laxatives and dietary instructions [17-19]. Accordingly, enhancing patient education using booklets, visual aids, short messages, or smartphone apps has proven effective [20-25]. Additionally, evaluating the stool state during BP is important to achieve appropriate BP quality. In Japan, BP is commonly performed at home or in the hospital. At home, patients use visual scales to assess their stool; in hospitals, the assessment may be conducted by the medical staff, contingent on the facility’s protocols. In both methods, colonoscopy is initiated when the stool is clear and watery. However, evaluating the stool state during BP can be burdensome for patients and medical staff, and an accurate evaluation may sometimes be difficult. Some patients may also feel embarrassed about having their stool evaluated by the medical staff; likewise, the medical staff may experience psychological discomfort when assessing the patient’s stool. Therefore, we previously developed a smartphone app incorporating AI to evaluate stool state during BP to address these issues, and the effectiveness of this app was demonstrated in a single-center study [26]. While the app was effective in assessing BP at the hospital, its feasibility at home or under diverse laxatives and toilet environments remains unknown. Furthermore, we developed a viewer system that connects to the app and enables the medical staff to easily monitor the patient’s stool state during routine tasks without directly checking the app screen. This system is intended to improve workflow efficiency in clinical settings.

Accordingly, in this study, we aimed to assess the feasibility of performing colonoscopy of appropriate quality using our app and viewer system.

This multicenter prospective study successfully evaluated the performance of the app across various types of laxatives, toilet bowls, and diverse patient populations. Among 326 patients who successfully used the app, 99.1% (n=323) patients achieved a BBPS score ≥6, which exceeded the set expected value. Additionally, several key quality indicators, including an ADR of 46.9% (n=153) and a cecal intubation rate of 99.7% (n=325), exceeded the target values recommended by American and European guidelines [9,10]. These results demonstrated the feasibility of app-guided BP management while maintaining high-quality colonoscopy indicators across multiple institutions and environments.

Notably, this study demonstrated the effective use of our app across a wide range of clinical settings. Specifically, we validated our app under diverse BP protocols and toilet environments and demonstrated that it could be applied to BP management in various clinical contexts. The following 2 main findings of this study support the versatility of the app, which is critical for its implementation in real-world clinical practice.

First, we evaluated the app’s assessment of BP using a variety of laxatives. In Japan, 6 major types of laxatives are commonly used for BP [29], of which 5 were used in this study. Additionally, the timing of administration, use of a low-fiber diet, and administration of laxatives on the day before colonoscopy were determined according to the protocols of each institution. Regardless of the type of laxative used and a wide variety of BP methods, a stool state evaluation rated as 3 stars using the app indicated that colonoscopy could be performed with a BBPS score ≥6 in 100% (260/260) of cases. Moreover, the mean BBPS score of patients rated as 3 stars was ≥8 (SD 0.8; excellent BBPS score). Even in real-world clinical settings, which exhibit variability in the types of laxatives used and BP methods, the 3-star rating by the app was shown to be highly correlated with an excellent BBPS score. These findings suggest that the app can assist BP across a broad range of institutional practices (eg, clinics and general hospitals) and diverse patient needs (eg, preference and tolerability for specific laxatives and low-fiber diets) while maintaining an adequate quality of colonoscopy. Although the app evaluations were associated with BP quality, the app was not designed to directly predict BBPS score. Instead, the app was intended to evaluate stool state to verify whether BP was sufficient to proceed with colonoscopy. In the present study, 95.1% (39/41) of the patients with a 2-star evaluation and 83.3% (5/6) of those with a 1-star evaluation still achieved a BBPS score ≥6. Consistent with these findings, the app showed a high positive predictive value (100%) but a low negative predictive value (4.5%) for predicting a BBPS score of ≥6 above. These results suggest that while a 3-star evaluation reliably identified patients with adequate BP (BBPS score ≥6), lower star evaluations did not necessarily indicate inadequate BP (BBPS score <6). Although the number of patients with a 1-star evaluation was small, these findings suggest that a substantial proportion of patients with lower star evaluations ultimately achieved adequate BP. This may reflect limitations of the star-based evaluation, as well as the imperfect relationship between stool state immediately before colonoscopy and overall bowel cleanliness assessed by the BBPS. Nevertheless, further refinement of the stool-state assessment algorithm may improve its ability to distinguish between different stool states during BP. In the present study, some patients with 1-star or 2-star evaluations still achieved a BBPS score of ≥6, suggesting that additional preparation may have been unnecessary in a subset of cases. Therefore, there remains room for improvement in the diagnostic discrimination of the app, particularly in reducing potential additional preparation (eg, additional laxative intake or enema). To further clarify the diagnostic characteristics of the app, sensitivity, specificity, positive predictive value, and negative predictive value for predicting a BBPS score of ≥6 are summarized in Table S2 in Multimedia Appendix 1.

Second, the study included both hospital-based and home-based BP settings, reflecting diverse real-world environments with potential variability in toilet conditions. Unlike our previous study [26], this study also included patients who underwent BP at home. Evaluations for BP at hospitals were conducted using toilet bowls available at each of the 10 participating institutions. Meanwhile, evaluations for BP at home were performed using toilets installed in each patient’s home. Thus, the app assessed a wide variety of toilet bowls. In both settings, when the evaluation of the app was 3 stars, colonoscopy was performed with a BBPS score of ≥6 in 100% (260/260) of the cases. Furthermore, the mean BBPS score for patients rated as 3 stars was ≥8 (SD 0.8) in both settings, indicating that an excellent BBPS score was achieved. Therefore, even across a wide variety of toilet environments with differences in shape, color tone, and lighting conditions, the app was able to evaluate stool state consistency; when the stool was rated as 3 stars, colonoscopy could be performed with an adequate BBPS score. These findings suggest that the app may be applicable across diverse clinical and home-based environments and support its potential feasibility in real-world clinical practice.

Using a similar approach to this study, 2 recent reports have demonstrated the effectiveness of AI-based smartphone apps in improving BP quality [30,31]. In previous studies, the developed app had a binary classification function directly determining whether the bowel cleansing quality was adequate based on the stool state. Consequently, the mean BBPS score achieved with those apps was 7.2 (SD 1.4) and 6.7 (SD 1.3), respectively. In contrast, our app evaluated each defecation individually during the BP process, considering the number of defecations. The app was used to assess stool state throughout the BP process, followed by colonoscopy. In our study, a mean BBPS score of 8.5 (SD 1.0) was observed, which was higher than the values reported in previous studies. These findings suggest that it is difficult to accurately predict the quality of BP solely based on a single stool state and that evaluating all defecations may support more appropriate BP management throughout the BP process.

The observed rate of the BBPS score of ≥6 in this study was notably high (99.1%, 323/326), exceeding both the minimum (≥90%) and target (≥95%) standards recommended by the European guideline [9,10]. Several factors may have contributed to this finding. First, patients aged 70 years or older were excluded from this study. Because advanced age is a known risk factor for inadequate BP, this exclusion may have contributed to the favorable BBPS outcomes. However, this exclusion criterion was not intended to select patients with better BP quality. Instead, this decision was based on the low smartphone ownership rates among older individuals in Japan at the time of study protocol development (38.3% among individuals in their 70s and 11.0% among those in their 80s), as well as our previous single-center experience showing that all patients who were unable to appropriately operate the app were aged 70 years or older [26,32]. Therefore, older patients were excluded because they were considered more likely to have difficulty appropriately using the app in this study. In addition, the use of the app itself and the awareness that stool images and app evaluations could be reviewed by medical staff through the viewer system may have increased patient awareness and adherence to BP instructions (ie, observation effect).

Although the mechanisms underlying the potential benefits of the app were not directly evaluated in this study, several possible explanations and future perspectives may be considered. Several mechanisms may explain how the app could support BP management. First, the viewer system enabled real-time monitoring of stool state during BP, allowing medical staff to review temporal changes and provide timely instructions when needed. Second, the use of the app itself may have promoted behavioral reinforcement and improved adherence to BP instructions. In addition, future expansion of app functions, such as automated reminders for examination schedules and dietary instructions, may further support behavioral modification during BP. Third, the app may help reduce uncertainty in BP assessment. In Japan, stool state is commonly evaluated during BP; however, assessment methods may vary across institutions and between patients and medical staff. By providing standardized stool state assessment and visualization of temporal changes in stool state, the app may help reduce variability and uncertainty in BP management.

Although our app evaluates the stool state over time during BP, the responsibility for deciding whether to proceed with colonoscopy based on the app’s evaluations or to implement additional procedures if the BP is insufficient lies with the medical staff. Therefore, a viewer system was introduced in this study to enable an efficient review of temporal changes in stool state and the corresponding evaluations during BP. In the questionnaire survey, most medical staff reported no issues with the usability of the viewer and responded that the viewer was necessary during BP. Particularly, the reasons cited for the necessity of the viewer were that its use enabled real-time monitoring of BP, allowing timely instructions for patients during the BP process, and that it had the potential to reduce patient waiting times before colonoscopy. Thus, by enabling a standardized assessment of the stool state and real-time monitoring, the combined use of the app and viewer system introduces a new framework for quality assurance in BP.

Finally, the usability of our app was evaluated using questionnaires administered to patients and medical staff. Most patients were able to use the app without any issues, and discomfort with having their stool state assessed by medical staff, as well as anxiety about evaluating their own stool state, was alleviated using the app. The observed reduction in patient anxiety during BP suggests that the app may also help lower the psychological barriers to undergoing colonoscopy. Furthermore, answers from medical staff suggested that using the app might help reduce their physical and psychological burdens. This reduction implies that the app can support a more efficient clinical workflow. In health care systems experiencing workforce shortages, especially in aging populations, tools that facilitate task shifting and reduce manual workloads are essential. In this regard, our app may provide a practical solution for supporting efficient BP management without compromising the quality of care.

Nevertheless, this study has some limitations. First, the study was not designed as a randomized controlled trial and therefore lacks direct comparison with BP without the use of the app. Future randomized controlled trials comparing outcomes with and without the app are essential to validate the clinical usage of the app more robustly. Second, we did not specifically assess the usability of the app in patients aged 70 years or older. This subset of patients was excluded from the study because smartphone ownership rates among older individuals in Japan were relatively low at the time of study initiation, and many older patients were unfamiliar with smartphone app operation. In addition, in our previous single-center study, all patients who were unable to appropriately operate the app were aged 70 years or older. However, smartphone ownership among older individuals in Japan has increased substantially in recent years. According to national data from 2024, smartphone ownership rates have risen to 67.5% among individuals in their 70s and 30.7% among those in their 80s [33]. Therefore, with further improvements in user interface design and usability support, the applicability of the app to older patients and populations with lower digital literacy may expand in the future. Third, this study was conducted exclusively in Japan. Therefore, the generalizability of the results to other countries remains unclear. Furthermore, because BP protocols and toilet environments differ across countries and regions, further validation studies are needed to confirm whether the app can achieve similar performance outside Japan. Fourth, endoscopists who assigned BBPS were not blinded to study participation status or app evaluations. Therefore, the possibility of observer bias in BBPS assessment cannot be excluded because the primary end point relied on endoscopist-assigned BBPS. Although all participating endoscopists received standardized instruction on BBPS assessment, including washing and suctioning of residues before scoring, such training cannot completely eliminate the possibility of assessment bias in a nonblinded study. Fifth, the primary end point analysis included only patients who successfully used the app. Therefore, the observed results may overestimate the feasibility of app-guided BP management in real-world settings.

In conclusion, the proportion of patients with a BBPS score of ≥6 for colonoscopy using the app exceeded the expected value in addition to the threshold value. In this multicenter prospective study, AI-based stool state assessment using the app and viewer during BP was feasible across diverse BP methods and clinical environments. Favorable BP outcomes and high usability among patients and medical staff support the potential usage of this approach in real-world colonoscopy practice.