Activity Number:
|
215
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, August 13, 2002 : 12:00 PM to 1:50 PM
|
Sponsor:
|
Biometrics Section*
|
Abstract - #301669 |
Title:
|
Combining Biomarkers for Early Detection of Disease
|
Author(s):
|
Yue Wang*+ and Jeremy Taylor
|
Affiliation(s):
|
University of Michigan and University of Michigan
|
Address:
|
1420 Washington Heights, Ann Arbor, Michigan, 48109-2029, U.S.A.
|
Keywords:
|
diagnostic marker ; penalized likelihood ; splines ; monotonicity constraints
|
Abstract:
|
It is reasonable to hypothesize that combinations of biomarkers may be significantly more effective for the early detection of cancer than a single biomarker. We assume that the risk of disease is a smooth and monotonic function with respect to each marker. A penalized likelihood method with linear inequality constraints is used for estimation of the probability of disease. Sequential quadratic programming method is implemented for the optimization given the value of smoothing parameter. An extended generalized approximate cross-validation (GACV) score is used to select the optimal amount of smoothing. Simulation studies are carried out where Kullback-Leibler distance is used to monitor the performance of the estimates. The efficacy of the extended GACV score is shown. The method is also compared to standard linear logistic regression using pancreatic cancer data.
|