|
Activity Number:
|
20
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #305177 |
|
Title:
|
Using Super Learner for Robust Model Selection in Causal Effect Estimation
|
|
Author(s):
|
Eric Polley*+ and Mark J. van der Laan
|
|
Companies:
|
University of California, Berkeley and University of California, Berkeley
|
|
Address:
|
1468 5th ave, San Francisco, CA, 94122,
|
|
Keywords:
|
causal effects ; model selection ; machine learning ; targeted maximum likelihood
|
|
Abstract:
|
Most model selection algorithms for estimating causal effects are not reproducible and are based on subjective evaluation of the model fit. Additionally, many confounders have complex relationships with the treatment and outcome which cannot be accurately modeled with main terms regression. The Super Learner is an data-adaptive and robust machine learning algorithm for prediction. The prediction models are used as the estimates for the targeted maximum likelihood causal effect estimate. The method is presented with an example of estimating the effect of once-daily therapy on adherence with HIV patients. We demonstrate the improved precision and reduction in bias by using the super learner compared to standard causal effect methods.
|
