|
Activity Number:
|
75
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #305037 |
|
Title:
|
A Comparison of Tree-Based Methods with Logistic Regression to Classify Cardiovascular Disease Risk
|
|
Author(s):
|
Layla Parast*+ and Latha Palaniappan and Helena Kraemer and Michael Pencina and Ralph D'Agostino
|
|
Companies:
|
Harvard University and Palo Alto Medical Foundation and Stanford University and Boston University and Boston University
|
|
Address:
|
677 Huntington Avenue, Boston, MA, 02115,
|
|
Keywords:
|
risk classification ; CART ; Framingham ; cardiovascular disease
|
|
Abstract:
|
Logistic regression is a commonly used method to estimate risk of cardiovascular disease. We compare the predictive ability of logistic regression, a classification and regression tree (CART) and signal detection regression (SDR) tree analysis in the Framingham Offspring Study cohort. The cohort consists of 3104 individuals with 270 cardiovascular disease events in a 10 year period. Cross-validation was used to determine optimal depth for the tree-based methods using a test subset of the data. Area under the receiver-operating curve and discrimination slope were used to evaluate predictive ability using a validation subset.
|
