Colloquium Details
Predicting Clinical Events from Electronic Medical Records (EMR) Data
| Author: | David Page University of Wisconsin |
|---|---|
| Date: | May 30, 2013 |
| Time: | 15:30 |
| Location: | 220 Deschutes |
Abstract
Machine learning from clinical data has been discussed as a tool to enable better health care decisions, reduce health care costs, and prevent dangerous adverse drug events. Achieving these goals requires using machine learning to build accurate predictive models for diseases and other health care events. This talk discusses specific applications of machine learning to prediction of myocardial infarction (heart attack), atrial fibrillation, and several kinds of adverse drug events. From these case studies the talk raises issues in learning accurate models from clinical data, statistical relational learning, incorporation of genetic data, protection of patient privacy, andcausal inference that must be addressed if we are to achieve the promise of machine learning to improve health care.
Biography
David Page received his Ph.D. in computer science from the University of Illinois at Urbana-Champaign in 1993 and did a post-doc in the Oxford University Computing Laboratory. He is a professor of Biostatistics and Medical Informatics at the University of Wisconsin-Madison, where he also holds an appointment in the Computer Sciences Department. He served on the NIH study section on BioData Management and Analysis during its first 3 years as a standing study section and was on the steering committee of the International Warfarin Pharmacogenetics Consortium. He directs the Informatics Shared Service for Wisconsin's Carbone Comprehensive Cancer Center and is on the scientific advisory board for OMOP, a joint PhARMA, FDA and FNIH initiative on identifying adverse drug events. David is a member of the Genome Center of Wisconsin, a co-director of the CIBM training program in biomedical informatics, and is UW-Madison's scientific lead in the Wisconsin Genomics Initiative. David's algorithm developments with colleagues include the SAYU and LUCID algorithms for change of view in statistical relational learning, skewing for learning correlation immune functions, and structure learning in continuous-time Bayesian networks via functional gradient boosting. David’s algorithm development is motivated by machine learning applications to various biomedical data types including electronic health records, SNP genotypes, gene expression from next generation sequencing and micorrays, mass spectrometry proteomics, and high-throughput assays for ligand-protein binding.