Speaker "Joseph Lucas" Details Back



Predictive medicine


Appropriate targeting of high risk patients is critical to the success of preventive medicine interventions. One challenge to building risk models for morbidity/mortality in patients with chronic disease is the complexity of the landscape of potential adverse outcomes. For example, diabetic patients are at risk for multiple comorbidities (examples are heart disease, kidney disease, and depression); the choice of outcome can have strong effects on which patients are identified as high risk. We will describe a model for concurrently predicting multiple outcomes of unknown relatedness. The model identifies which outcomes are related, quantifies the level of relatedness and automatically borrows strength across related outcomes to improve the overall accuracy of prediction for all outcomes. We demonstrate the effectiveness of the model in a cohort of diabetic patients. Predictive models will be built on independent variables built from the electronic medical record with the full model predicting future risk of many of the sequelae of diabetes including macular degeneration, heart disease, kidney disease, depression, chronic skin ulcers and stroke.


Dr. Lucas is a Research Assistant Professor and Associate Director of Health System Operations in the Information Initiative at Duke. He has 8 years of experience directing analytics projects at the intersection of biology/medicine and “big data”. His statistical expertise includes variable selection, factor modeling, experimental design, topic models and deep learning; His domain expertise includes proteomics, metabolomics, genomics, personalized medicine and electronic health records.