Prediction of health outcomes lies at the heart of medical practice. The emergence of massive patient-level medical data opens up the possibility of dramatically improved predictive models that can profoundly alter medical practice. This talk presents two such models and demonstrates their applicability using increasingly available large-scale medical data. The first approach provides patients with interpretable predictions for which symptoms they may be likely to experience next, given a list of previous symptoms. Specifically we propose the Hierarchical Association Rule Model (HARM) that generates a set of association rules such as "dyspepsia and epigastric pain" imply "heartburn," indicating that dyspepsia and epigastric pain are commonly followed by heartburn. Next, we present a dynamic modeling framework which makes nearly real-time predictions using streaming, rapidly changing data. This model accounts for uncertainty in the underlying data-generating mechanism through Dynamic Model Averaging, an extension of Bayesian Model Averaging that allows posterior model probabilities to change with time. We apply our method to data from children with appendicitis who receive either a traditional (open) appendectomy or a laparoscopic procedure. Our results indicate that the factors associated with which children receive a particular type of procedure changed significantly over the seven years of data collection, a feature that is not captured using standard regression modeling.
Big data meets medical care: Patient-level predictive models for large-scale medical data
Room
409