|
Activity Number:
|
397
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #307753 |
|
Title:
|
Dealing with Control Group Drop-Out: Using Historical Patient Information to Supplement a Randomized Trial
|
|
Author(s):
|
Elizabeth A. Stuart*+ and Donald B. Rubin and Samantha Cook
|
|
Companies:
|
Johns Hopkins Bloomberg School of Public Health and Harvard University and Google
|
|
Address:
|
624 N Broadway, 8th Floor, Baltimore, MD, 21205,
|
|
Keywords:
|
propensity score ; causal inference ; matching methods ; missing data ; Bayesian methods ; patient registries
|
|
Abstract:
|
The goal of a clinical trial is often to compare a new treatment with currently available treatments. However, in some trials, some control group members may receive the new treatment and thus can no longer be considered controls. We describe a method of utilizing information from a large collection of "traditionally treated" patients (e.g., available through patient registers) to impute the control patients' outcomes as if they had stayed on the traditional treatment. This is done by modeling trends in the disease using the individuals in the patient register who look most similar to the study patients, selected through propensity score matching. This talk focuses on two particular challenges to the matching: missing covariate data and the need to define baseline for patients in the register. The methods are illustrated using data from a clinical trial for a rare disease.
|
