Improving bowel preparation for colonoscopy with a smartphone application driven by artificial intelligence
Yan Zhu,Yan Zhu,Dan-Feng Zhang,Dan-Feng Zhang,Hui-Li Wu,Pei-Yao Fu,Pei-Yao Fu,Li Feng,Kun Zhuang,Zi-Han Geng,Zi-Han Geng,Kun-Kun Li,Xiao-Hong Zhang,Bo-Qun Zhu,Bo-Qun Zhu,Wen-Zheng Qin,Wen-Zheng Qin,Sheng-Li Lin,Sheng-Li Lin,Zhen Zhang,Zhen Zhang,Tian-Yin Chen,Tian-Yin Chen,Yuan Huang,Yuan Huang,Xiao-Yue Xu,Xiao-Yue Xu,Jing-Zheng Liu,Jing-Zheng Liu,Shuo Wang,Shuo Wang,Wei Zhang,Quan-Lin Li,Quan-Lin Li,Ping-Hong Zhou,Ping-Hong Zhou
DOI: https://doi.org/10.1038/s41746-023-00786-y
IF: 15.2
2023-03-14
npj Digital Medicine
Abstract:Optimal bowel preparation is a prerequisite for a successful colonoscopy; however, the rate of inadequate bowel preparation remains relatively high. In this study, we establish a smartphone app that assesses patient bowel preparation using an artificial intelligence (AI)-based prediction system trained on labeled photographs of feces in the toilet and evaluate its impact on bowel preparation quality in colonoscopy outpatients. We conduct a prospective, single-masked, multicenter randomized clinical trial, enrolling outpatients who own a smartphone and are scheduled for a colonoscopy. We screen 578 eligible patients and randomize 524 in a 1:1 ratio to the control or AI-driven app group for bowel preparation. The study endpoints are the percentage of patients with adequate bowel preparation and the total BBPS score, compliance with dietary restrictions and purgative instructions, polyp detection rate, and adenoma detection rate (secondary). The prediction system has an accuracy of 95.15%, a specificity of 97.25%, and an area under the curve of 0.98 in the test dataset. In the full analysis set (n = 500), adequate preparation is significantly higher in the AI-driven app group (88.54 vs. 65.59%; P < 0.001). The mean BBPS score is 6.74 ± 1.25 in the AI-driven app group and 5.97 ± 1.81 in the control group (P < 0.001). The rates of compliance with dietary restrictions (93.68 vs. 83.81%, P = 0.001) and purgative instructions (96.05 vs. 84.62%, P < 0.001) are significantly higher in the AI-driven app group, as is the rate of additional purgative intake (26.88 vs. 17.41%, P = 0.011). Thus, our AI-driven smartphone app significantly improves the quality of bowel preparation and patient compliance.
health care sciences & services,medical informatics