|
Activity Number:
|
361
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 2, 2016 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #319006
|
|
|
Title:
|
Classification and Regression Tree Modeling of Correlated Binary Outcomes
|
|
Author(s):
|
Jaime Speiser* and Valerie Durkalski-Mauldin and Dongjun Chung and Bethany Wolf
|
|
Companies:
|
Medical University of South Carolina and Medical University of South Carolina and Medical University of South Carolina and Medical University of South Carolina
|
|
Keywords:
|
classification and regression tree ;
generalized linear mixed model
|
|
Abstract:
|
Correlated binary outcomes are frequently encountered in clinical research. Often the goal is to develop a prediction model of the outcomes using clustered or repeated measurements. Generalized linear mixed models (GLMMs) typically employed for correlated outcomes require assumptions which are inappropriate for some datasets. We develop an alternative method with fewer assumptions which combines classification and regression trees (CARTs) and GLMMs. Simulation studies show that the new method offers similar or superior predictive (test dataset) accuracy compared to CARTs and GLMMs. The method is applied to a real dataset from the Acute Liver Failure Study Group (1064 patients with up to seven days of repeated measures variables). The resulting tree model predicting outcome of improved or worsened condition using clinical variables provides an accurate, easily-implemented model for use at the bedside. Novel methodology offers an alternative framework for modeling correlated binary outcomes which may be applied in myriad research settings for datasets with clustered or repeated measurements.
|
Authors who are presenting talks have a * after their name.
