NAFLD Diagnosis and Outcomes

PROJECT SUMMARY Nonalcoholic fatty liver disease (NAFLD) affects 25-30% of US adults and has a major impact on healthcare burden and population health due to higher morbidity and mortality than the general population. Even though NAFLD can progress to cirrhosis, decompensation, and liver cancer, these outcomes affect a small proportion of patients, but there are no accurate methods that are accessible in primary care to identify these patients early. The heterogeneity of clinical phenotypes, lack of universal screening, and risk-stratification approaches lead to delayed diagnosis, phenotype-specific prophylactic or therapeutic intervention, and poor patient outcomes. The long-term goal is to develop easily accessible methods for NAFLD screening and prediction of disease trajectory. The objective of this application is to leverage large electronic health record (EHR) datasets and analytics to enhance the early identification of NAFLD in general healthcare settings. The central hypothesis is that targeted screening with machine-learning (ML)/artificial intelligence (AI) models applied to longitudinal healthcare data (diagnoses, laboratory values, medications, anthropometrics, demographics) can identify predictors of NAFLD risk and, subsequently, a progressive phenotype toward liver events. The rationale that underlies the proposed research is that EHR-based clinical algorithms which identify NAFLD and phenotype the disease trajectory will guide clinicians in selecting patients who need liver-related diagnostic evaluation and enable timely intervention to prevent hard outcomes. Guided by strong preliminary data, the hypothesis will be tested by pursuing two specific aims. In Aim 1, a predictive model for NAFLD will be developed using retrospective data from a population-based EHR-linkage system with validated diagnosed NAFLD and non-NAFLD controls by chart review. Among those with NAFLD, an EHR-based model to predict a clinical phenotype at risk for future liver-related events or death will be developed, using death from other causes as competing risk. In Aim 2, the two models will be validated and calibrated to screen for NAFLD in the population and to predict a liver phenotype at early stage (steatohepatitis or stage≥2 fibrosis). In this aim, the cohort will be prospectively enrolled from the community to undergo NAFLD screening by magnetic resonance imaging and assessment of liver disease severity by biopsy. Lastly, an AI-based NAFLD care model using the predictive models introduced in the EHR system will be implemented in the primary care practice to measure impact on disease identification and management. The approach is innovative because it expands the analytical toolbox beyond conventional methods and challenges the current clinical paradigm which targets only late-stage liver disease identification. These multimodal EHR-based algorithms of NAFLD screening and clinical phenotyping will serve as scalable and unbiased clinical decision-making tools that enable personalized intervention, thereby fulfilling a critical unmet need in the healthcare burden of NAFLD.