P3800 - AI Classification of Stool Characteristics Using the Dieta App Could Guide the Management of Hepatic Encephalopathy: A Cross-Sectional and Longitudinal Analyses
Virginia Commonwealth University and Richmond VAMC Richmond, VA
Andrew Fagan, BS1, Asaf Kraus, MBA2, Jacob Anstey, BS2, David Tiede, BS2, Travis Mousel, RN1, Mary Leslie Gallagher, FNP1, Edith Gavis, RN1, Jasmohan Bajaj, MD, MS, FACG3 1Richmond VAMC, Richmond, VA; 2Dieta Health, Los Angeles, CA; 3Virginia Commonwealth University and Richmond VAMC, Richmond, VA
Introduction: Pts with cirrhosis & hepatic encephalopathy (HE) are often prescribed lactulose titrated by bowel movement(BMs). BMs quality through the Bristol stool scale (BSS) could improve HE Rx but accurate assessment may be difficult. Dieta App records BMs & analyzes consistency, fragmentation, edge fuzziness & volume to generate an AI-BSS(FigA). Aim: Define acceptance, learning, & lactulose self-titration using Dieta App in cirrhosis.
Methods: Cross-sectional: healthy controls(Ctrls) & pts with cirrhosis on/not on lactulose were trained on the App to take their stool pictures after each BM & self-titrate lactulose according to BM no. & BSS for 2 wks. We calculated BMs/AI-BSS ratio & compared groups. Comfort with App was tested with a semi-quantitative questionnaire (max 84=good) before/after App. Pts could also enter self-BSS daily. Longitudinal: In a subset post-2 wks, we removed the App & re-downloaded it after >30 days but without pts having access to AI-BSS to gauge their self-BSS scoring.
Results: Cross-sectional: 36 subjects (11 ctrl, 25 cirrhosis; 13 on lactulose) were included. Ctrls were younger & rifaximin use was highest in lactulose users; rest was similar(Table). Lactulose users had ↑ daily BMs. On the App, these pts trended towards lower consistency, ↑ fragmentation, & edge fuzziness. While AI-BSS were similar, a higher BM/AI-BSS ratio was seen in lactulose pts (Table). Self BSS was only completed initially by 60%. This ↑ over time & correlated with pre-App comfort on the questionnaire (r=0.57, p=0.04). Patients self-adjusted lactulose doses based on App results, through a negative correlation with BM/AI-BSS(FigB). Personal comfort level with the App ↑significantly within controls (pre 68 vs post 73, p=0.01) & lactulose users (pre 62 vs post 67, p=0.05) but not others.
Longitudinal: 12 subjects (5 controls, 7 cirrhosis; 2 on lactulose), were given the App twice. Self-BSS initially did not correlate with AI-BSS (FigC). This correlation significantly improved during the 2nd App administration, even without access to AI-BSS(FigD), indicating learning of accurate BSS scoring.
Discussion: AI-generated stool quality adjusted for daily BMs using the Dieta App associates with self-titration of the lactulose dose in pts with HE. Exposure to the App resulted in lasting education regarding accurate stool quality scoring. Pts with HE were comfortable using this App, which grew over time. Using this AI-based App could improve dosing, follow-up & patient insight into HE therapy.
Figure: Overview of the App and the Experience with Cross-sectional Adjustment of Lactulose Dose and Longitudinal Improvement in Learning by Subjects to Accurately Code Bristol Stool Scale compared to AI-generated results.
Disclosures:
Andrew Fagan indicated no relevant financial relationships.
Andrew Fagan, BS1, Asaf Kraus, MBA2, Jacob Anstey, BS2, David Tiede, BS2, Travis Mousel, RN1, Mary Leslie Gallagher, FNP1, Edith Gavis, RN1, Jasmohan Bajaj, MD, MS, FACG3. P3800 - AI Classification of Stool Characteristics Using the Dieta App Could Guide the Management of Hepatic Encephalopathy: A Cross-Sectional and Longitudinal Analyses, ACG 2023 Annual Scientific Meeting Abstracts. Vancouver, BC, Canada: American College of Gastroenterology.
.jpg "Jasmohan Bajaj, MD, MS, FACG photo")
