Abstract
Non-alcoholic fatty liver disease (NAFLD) is the most global frequent liver disease, with a prevalence of almost 20% in the overall population. NAFLD may progress to fibrosis and later into cirrhosis in addition to other diseases. Our objective is to stratify patients' risks for NAFLD and advanced fibrosis over time and suggest preventive medical decisions. We used a cohort of individuals from the Tel-Aviv medical center. Time-series clustering machine learning model (Hidden Markov Models (HMM)) was used to profile fibrosis risk by modeling patients’ latent medical status and trajectories over time. The best-fitting model had three latent HMM states. Initial results show that tracking individuals over time and their relative risk for fibrosis at each point of time provides significant clinical insights regarding each state (and its group of individuals). Thus, longitudinal risk stratification can enable the early identification of specific individual groups following distinct medical trajectories based on their routine visits.
Recommended Citation
Ben-Assuli, Ofir; Shenhar-Tsarfaty, Shani; Goldman, Orit; Jacobi, Arie; Rogowski, Ori; Zeltser, David; Shapira, Itzhak; Berliner, Shlomo; and Zelber-Sagi, Shira, "Analyzing Non-Alcoholic Fatty Liver Disease Risk Using Time-Series Model" (2020). Proceedings of the 2020 Pre-ICIS SIGDSA Symposium. 3.
https://aisel.aisnet.org/sigdsa2020/3
Abstract Only