Health Research and Policy

Workshop in Biostatistics - Abstract

DATE: April 17, 2014
TIME: 1:15 - 3:00 pm
LOCATION: Medical School Office Building, Rm x303
TITLE: Data, Predictions, and Decisions: Leveraging Electronic Health Records for Clinical Care
SPEAKER: Eric Horvitz
Microsoft Research

I will describe recent efforts on harnessing clinical data to build predictive models, focusing on two difficult challenges in healthcare. First, I will review work on predicting the onset of hospital-associated infections. 5% of inpatients acquire a healthcare-associated infection and such nosocomial infections are considered to be in the top ten of contributors to death in the US. I will focus on efforts with predicting active infection with C.difficile using snapshot and temporal models, including work on leveraging data from single and multiple hospitals. Next, I will present work on models that predict the risk of readmission of patients a short time after their discharge. I will review methods and results on coupling predictions with decision models to guide actions, considering the cost and efficacy of interventions. I will include discussion of our experiences with the real-world fielding of predictive models for both readmission and hospital-associated infection, and reflect about opportunities ahead that frame challenges with data availability and modeling.

Joint work with Mohsen Bayati, Mark Braverman, Wayne Campbell, John Guttag, Paul Koch, Hank Rappaport, and Jenna Wiens.

Bio:

Eric Horvitz is distinguished scientist and managing director at Microsoft Research. His interests include machine learning and statistics, decision analysis, and machine intelligence. His work has led to fielded applications and services in healthcare, search and retrieval, transportation, aerospace, operating systems, and ecommerce. He has been elected fellow of the NAE, AAAI, and AAAS. He received MD and PhD degrees at Stanford University.

Suggested readings:

J. Wiens, J. Guttag, and E. Horvitz. A Study in Transfer Learning: Leveraging Data from Multiple Hospitals to Enhance Hospital-Specific Predictions, Journal of the American Medical Informatics Association: 0:1–8, January 2014.

J. Wiens, J. Guttag, E. Horvitz. Patient Risk Stratification for Hospital-Associated C. Diff as a Time-Series Classification Task, NIPS 2012, Lake Tahole, CA, December 2012.

Jenna Wiens S.M., Wayne N. Campbell M.D., Ella S. Franklin R.N., John V. Guttag Ph.D., Eric Horvitz Ph.D. M.D., Learning Risk Stratification Models for Clostridium difficile, [Under Review]

Stanford Medicine Resources:

Footer Links: