A Bayesian Look At Clinical Risk Prediction

Abstract: Building reliable predictive and prognostic models that leverage the growing scale of medical data and clinical records can make tremendous impact to the healthcare industry. Traditional survival analysis originated from clinical research focuses on identifying variates and factors that affect the hazard function; time series modelling approaches emphasize predicting future values based on previous observations; machine learning models are often formulated as mapping between large feature spaces to binary outcomes. However, all of these methods have their unique limitations and there are still much more to explore in the context of treating clinical survival analysis with machine learning models. In this talk, we will present our recent work on a new Bayesian framework that uniquely connects machine learning tasks (classification/regression) with event time analysis to provide risk prediction capabilities. We validate and demonstrate the utility of this approach with simulation data where the ground truths are known. We will then show a specific use case of this approach to perform risk prediction with real medical datasets. We will also discuss how this model can be implemented into clinical solutions.

Bio: Dr. Luo works at Wolters Kluwer Health as an applied data scientist since 2017. With a strong belief in the power of AI and big data, he brings his expertise in machine learning and natural language processing to solve the most challenging problems in healthcare, for example, early prediction of diseases from EHR data. Before Wolters Kluwer, he worked at T2 Biosystems, developing novel medical devices for sepsis detect and blood test. Dr. Luo got his Ph.D. in physics from UNC Chapel Hill and had a dozen publications and three patent applications.