近期活动

Joint Seminars

A Longitudinal Case-Control Study Design To Evaluate The Odds Ratio Of An Imaging Prognostic Marker

Ying Lu, Ph.D., Professor of Biostatistics, Stanford University, USA
Fri, 2012-07-13 10:30 - 11:30
601 Pao Yue-Kong Library

A retrospective case-control study can derive an unbiased estimate of an odds ratio quickly and inexpensively. However, it is not an option for new radiological tests because pre-diseased tests are rarely available. In this paper, we explore a longitudinal age-matched case-control design for a non-lethal disease with known time of disease onset for cases. For each pair at each time t after the disease-onset of the cases, we observe a pair [X(t), Z(t), Y] where Y is the binary disease status, X(t) is the test value, and Z(t) is the binary indicator of being sampled. Mixed random effects model is proposed between the missing X(0) and Y. A likelihood function and MLE can be derived. We performed simulation studies to show that such a design is reasonable to estimate and test the odds ratios for new radiological tests under some reasonable assumptions. The tradeoff between the losses of the statistical efficiency versus the cost of a prospective study suggest this design be an alternative to evaluate the low cost diagnostic tests, for which large scale longitudinal prospective studies are often not an option. This is a joint work with Dr. Johann Won at US VA Palo Alto Healthcare System.