Analytical and Methodological Challenges in the Study of Complex Liver Diseases
Organizers: Ayako Suzuki (Duke)
Chair: Christine M Hunt (Duke)
Vice Chair: Ayako Suzuki (Duke)
Ayako Suzuki (Duke)
Yunusa Olufadi (Washington State Univ.) and Ebenezer Olusegun George (UMem)
Minjun Chen (FDA)
Jimmy T. Efird (Durham VA)
Title: Current challenges and opportunities in studying complex liver diseases
Speaker: Ayako Suzuki (Duke)
Nonalcoholic fatty liver disease (NAFLD) is a multifactorial liver disease and a rapidly growing health threat worldwide. In the US, about a third of the adult population is estimated to have NAFLD. Once complicated by hepatocellular damage and inflammation (so-called nonalcoholic steatohepatitis or NASH), it could progress to NASH with fibrosis, cirrhosis, and end-stage liver disease, and lead to the development of liver cancer. Owing to the obesity epidemic, NAFLD is rapidly approaching as the leading cause of cirrhosis and liver cancer in the US. Despite the growing body of evidence in the literature, there are no FDA-approved pharmacological treatments for NAFLD to date nor non-invasive biomarkers that accurately identify patients with NASH or advanced NASH. These challenges are, at least in part, explained by the significant heterogeneity existed in the NAFLD pathogenesis and treatment responses.
On the other hand, drug-induced liver injury (DILI) is one of the most common adverse reactions and can lead to early termination of drug development and withdrawal from the market. Like other drug adverse reactions, DILI is relatively uncommon. However, once clinically significant DILI occurs, about 10% of patients develop serious outcomes such as acute liver failure or death. DILI is one of the leading causes of acute liver failure in the US. Mechanisms of human DILI are not fully understood. Multiple factors, including drugs and host factors, are thought to interact and determine the risk and phenotypes of DILI.
Several challenges exist in studying such complex liver diseases; sexually dimorphic disease mechanisms, explanatory pluralism, and multifactorial risks. In this talk, current knowledge gaps and challenges in studying these two complex liver diseases are discussed.
Title: A Procedure for Joint Estimation and Selection in Mixed Endpoints Patients Data
Speaker: Yunusa Olufadi (Washington State Univ.) and Ebenezer Olusegun George (UMem)
Multiple mixed endpoints consisting of discrete, continuous, count, and ordinal outcomes are often recorded in biomedical, clinical, and behavioral studies. A joint analysis of the endpoints is often needed because (1) a single endpoint is often not adequate to describe the disease complexities, (2) researchers are interested in studying the associations among these multiple endpoints, and (3) investigators are interested in identifying interactions between explanatory factors, such as age and gender. Correlations among these outcomes usually involve similar measurements on the same individual and those among the different endpoints taken over time. An approach that analyzes each outcome separately, ignoring the above correlations, is often misleading. In this talk, we present a procedure that jointly models several longitudinal (or clustered) mixed endpoints in a dataset from patients to guide both the model estimation and efficient extraction of potential active predictors of incident nonalcoholic, non-viral hypertransaminasemia, a surrogate of incident nonalcoholic fatty liver disease. We use this dataset to study the development of elevated aminotransferases in a nonalcoholic population and demonstrate the performance and sensitivity of this novel procedure.
Title: The strategies for developing predictive models to assess risk of drug-induced liver injury during drug development and in clinical setting
Speaker: Minjun Chen (FDA)
Drug-induced liver injury (DILI) presents a significant challenge for clinicians, drug developers, and regulators. DILI is a frequent cause of compound attrition in drug development and a leading cause of acute liver failure in the United States. Despite the recent progress in drug safety, there are still demands for innovative approaches and methodologies to identify drugs with potential human hepatotoxicity during drug development as well as reliable biomarkers for early detection of DILI in patients. Therefore, we have developed multiple computational models for predicting human DILI risk not only based on drug properties but also considering drug-host interplay to identify susceptible patients. This presentation will describe our strategies for developing predictive models and discuss the opportunities and challenges in the application of the computational models in regulatory science and clinical practice. The views in this presentation have not been formally disseminated by the US Food and Drug Administration.
Discussion by Dr. Efird:
Discussion (20 min)
(1) Common biological assumptions in translational and clinical studies and potential solutions without the assumption.
(2) Computer learning vs. theory-driven analytic approaches. Improve model convergence in highly dimensional, multifactorial data.
(3) Statistical consideration in integrating heterogeneous covariates to improve the model prediction
Summary of the discussion by Dr. Efird (5 min)