Pimentel M. Abstract 276. Presented at: Digestive Disease Week; May 21-24, 2022; San Diego (hybrid meeting).
Pimentel reports no conflicts of interest.
SAN DIEGO — A smartphone app using artificial intelligence was more effective than patient reporting in assessing stool samples in a cohort of individuals with irritable bowel syndrome, according to data presented at Digestive Diseases Week.
“In IBS, the outcome measure that the FDA has determined for using their guidance relies on the patients reporting an improvement in stool consistency, regardless of whether you are constipated or have diarrhea,” Mark Pimentel, MD, of the GI Motility Program at Cedars-Sinai Medical Center, in Los Angeles, told Healio. “The problem is that whenever you have a patient reporting an outcome measure, it becomes subjective rather than objective. This can impact the placebo effect.”
The limits of that outcome measure, the Bristol Stool Scale (BSS), are seen in clinical trials assessing the form and frequency of stools.
In the current study, they aimed to evaluate and validate a smartphone application that employs artificial intelligence (AI) to characterize digital images of stool.
IBS. They were asked to capture an image of every stool during 2-week screening phase.
The researchers then divided the study population into two groups. Thirty-five percent were used to validate the AI app, while the other 65% were used as an independent assessment group for the app.
Five stool characteristics underwent analysis, including BSS, consistency, edge fuzziness, fragmentation and volume.
For the validation set, two expert gastroenterologists were used as the gold standard to blindly assess stools by these parameters. The aim was to determine whether the AI met this gold standard. Once the AI was validated in this way, it was used to characterize images for the remainder of the study population.
“The app used AI to train the software to detect the consistency of the stool in the toilet based on the five parameters of stool form,” Pimentel said. “We then compared that with doctors who know what they are looking at.”
The researchers next compared AI-generated and patient self-reported daily average BSS scores.
The final analysis included 39 participants who completed study protocols. The first 14 of these patients were used in the validation set.
Results showed that validation images covered BSS scores from 1 to 7.
A moderate to high degree of agreement was seen between the validation images and the AI assessments (ICC = 0.78-0.85).
Agreement between the validation images and AI was also reported in terms of stool consistency (ICC = 0.873-0.890), edge fuzziness (ICC = 0.836-0.839), fragmentation (ICC = 0.837-0.863) and volume (ICC = 0.725-0.851).
Moreover, BSS scores reported by the AI app correlated better with physician gold standard assessment than patient-reported scores. The specificity of the app was 27% higher than the patient-reported outcome, while the sensitivity of the app was 23% higher. “Patients did not do a good job of assessing their stools compared with the app,” Pimentel said.
Following the validation protocol, the researchers then compared BSS scores of the AI ap with self-reported scores from the remaining 25 patients. Results showed that this agreement was only moderate (ICC = 0.61).
Other findings showed minimal agreement between the AI app and patient-reported assessment of categorical diarrhea, constipation or normal stool days ( = 0.21). However, AI assessment of stool BSS showed better correlation with abdominal pain, bloating and diarrhea severity as compared with patient self-reporting for these outcomes.
“I would love to see this tool used in clinical trials to determine stool consistency based on FDA guidance,” Pimentel said. “It will better show the effectiveness of a drug in a trial.”
The app can also be used to assess GI-related adverse events of GI drugs, according to Pimentel.
“The next step is to get the FDA to agree that this is a sufficiently valid tool to quantify the outcome measure they assess in IBS,” he said.