|
Activity Number:
|
480
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #310131 |
|
Title:
|
The Exclusion of Prevalent Cases from Incident Dementia Risk Factor Studies Sacrifices Lots of Information
|
|
Author(s):
|
Daniel Tancredi*+
|
|
Companies:
|
University of California, Davis
|
|
Address:
|
2600 Kline CT, Davis, CA, 95618,
|
|
Keywords:
|
dementia epidemiology ; case-2 interval censoring ; parametric survival analysis ; Fisher information ; bias--variance tradeoff ; risk factor estimation
|
|
Abstract:
|
Population-based studies of senile dementia risk factors typically collect and analyze data at two timepoints. Baseline data are used to study prevalent disease. Follow-up data are used to study incident disease among those free of disease at baseline. An alternative approach, using all of the data together in a single analysis, is tempting but rarely done because of concerns that the prevalent data are biased by selection effects (e.g., differential mortality). Could the risk factor analysis of incident disease benefit from including the prevalent cases? Parametric failure time regression models are specified for a simplified setting to address this question. Theory and computer simulations examine how truncation, interval censoring and selective mortality affect the bias and variance of risk factor estimators. Results show that including the prevalent cases offers big potential gains.
|
