Activity Number:
|
494
- Statistical Methodologies for Identifying, Modeling, and Managing Subpopulations at Risk
|
Type:
|
Topic Contributed
|
Date/Time:
|
Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics in Epidemiology
|
Abstract #324280
|
View Presentation
|
Title:
|
Machine Learning Methods in the Statistical Prediction of Health Outcomes
|
Author(s):
|
William Padula*
|
Companies:
|
Johns Hopkins Bloomberg SPH
|
Keywords:
|
machine learning ;
health outcomes ;
health economics ;
classification ;
data mining ;
dimension reduction
|
Abstract:
|
The introduction of Big Data in health care through electronic health record (EHR) systems and aggregation of centralized clinical data from many sources provides a potential wealth of information for health services research methods. However the management of these data to develop models that can predict complex, and potentially rare patient outcomes becomes increasingly more challenging. We will present an array of data dimension reduction and classification methods that can be applied to instances of health analytics research questions, including: decision trees; lasso and ridge regression; random forests; boosting and bagging; and neural networks. In addition, we will specifically show how data can be obtained from an EHR of an academic medical center to develop ad-hoc Markov models for analyzing the cost-effectiveness of preventing hospital-acquired conditions. What we have found thus far is that up-front investment in careful data management and learning these methods saves additional time in the long-term to develop complex economic and predictive models of health outcomes that more accurately indicate real-world effectiveness.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2017 program
|