|
Activity Number:
|
168
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #304136 |
|
Title:
|
Treatment Effects under Early Detection
|
|
Author(s):
|
Shih-Yuan Lee*+ and Alexander Tsodikov
|
|
Companies:
|
University of Michigan and University of Michigan
|
|
Address:
|
1420 Washington Heights, Ann Arbor, MI, 48109,
|
|
Keywords:
|
Early detection ; Lead time ; Misspecified model ; Proportional hazards ; Bias
|
|
Abstract:
|
Cancer-specific survival counted out from the point of diagnosis is the most common endpoint used in cancer trials. Estimated treatment effect in proportional hazards (PH) models may be biased if models are misspecified. We consider a misspecification associated with ignoring the random mechanism of the early detection of the disease. We study how the lead time resulting from the early detection of cancer affects the treatment effect as estimated by a PH model ignoring the advanced diagnosis. We study how the multiplicative treatment effect differs in screened vs. non-screened population and assess the magnitude and the direction of the bias under various conditions. We also investigate the behavior of the estimator in the Cox PH model using partial likelihood. A meta-analytic approach is proposed to correct the bias based on a joint cancer incidence and survival modeling approach.
|
