The Data Analytics and AI for Advanced Liver Disease Laboratory is pursuing several avenues of investigation. Highlights of the lab's current research projects include development in the following areas:

  • ECG-based detection of cirrhosis using the AI-Cirrhosis-ECG (ACE) score. This deep-learning model can accurately detect the presence of cirrhosis and grade the severity of liver disease on digitized 12-lead ECGs. The ACE score has demonstrated an outstanding classification performance for distinguishing cirrhosis from no cirrhosis (area under the curve of 0.908). In addition, the magnitude of the ACE score is closely associated with clinical disease severity and liver-related outcomes in cirrhosis. The ACE score was presented as oral presentations at national meetings (Digestive Disease Week 2021 and American Association for the Study of Liver Disease Liver Meeting 2021). The initial manuscript was published in the American Journal of Gastroenterology.
  • Digital phenotyping in alcohol-associated liver disease and alcohol use disorder. Improving our understanding in this area can help us predict clinical outcomes, including cravings and relapse events. This pilot study uses readily available technologies to identify signals of association between digital biomarkers and predictors of relapse. The study's overall aim is to develop a patient-centric, personalized approach to management of alcohol use in liver disease.
  • Remote patient monitoring for improved health outcomes in cirrhosis. Designed to reduce hospital readmission in patients with decompensated cirrhosis, this program is offered to all patients hospitalized with decompensated cirrhosis. It involves at-home monitoring, including daily vital signs and automated symptom checks, and is monitored by trained nursing staff who can perform targeted interventions such as diuretic titration and lactulose dose adjustments. We are determining the impact of this program in reducing hospital readmissions. Initial results show a significant decrease in the 90-day readmission rate.
  • Predicting cirrhosis-related readmissions using machine learning. Creating algorithms to predict 90-day readmission in patients hospitalized with cirrhosis is desirable. We have retrospectively collected variables to use in our model. They include laboratory studies, vital signs, radiology reports, reason for hospitalization, cirrhosis severity and related complications, comorbidities, psychosocial factors, and functional status. Currently, the lab is working with Mayo's Enterprise Agile Transformation Office to develop this novel machine-learning model.
  • Classification of histopathologic features of liver disease using deep learning. To improve our classification of steatohepatitis on liver biopsy, the lab is training a convolution neural network to identify alcoholic steatohepatitis (ASH) versus nonalcoholic steatohepatitis (NASH). We are building a collaboration with an industry partner in computational pathology to upscale our development and validation of the deep-learning architecture. Also, the lab is collaborating with data scientists to achieve the vision of a multimodal deep-learning architecture, incorporating clinical, laboratory and radiographic data into our AI classifier.