|
Activity Number:
|
209
- Statistical methods for genomic and epigenetic data analysis
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Section on Medical Devices and Diagnostics
|
|
Abstract #318470
|
|
|
Title:
|
Combining Multiple Biomarkers Linearly to Minimize the Euclidean Distance of the Closest Point on the ROC Surface to the Perfection Corner in Trichotomous Settings
|
|
Author(s):
|
Brian R Mosier* and Leonidas E Bantis
|
|
Companies:
|
University of Kansas Medical Center and University of Kansas Medical Center
|
|
Keywords:
|
ROC;
Biomarker;
3-class;
Euclidean Distance;
Perfection Corner;
Youden index
|
|
Abstract:
|
Individual biomarkers may exhibit limited discriminatory ability and thus, there is interest to develop strategies for combining biomarkers to increase diagnostic performance. There is extensive literature regarding combining biomarkers in the two-class setting. The literature for combining biomarker scores for diseases with k progressive stages is limited. It is important that combinations are built so that: 1) estimation of combination coefficients and cutoff values account for all disease stages simultaneously and 2) all promising markers can simultaneously contribute to the estimation of cutoffs. In our study, we provide parametric and nonparametric frameworks that allow investigators to optimally combine biomarkers that seek to discriminate between 3 or more classes by minimizing the Euclidean distance of the perfection corner to the closest point on the ROC surface/hypersurface. We illustrate that the derived cutoffs exhibit narrower confidence intervals than a Youden-based approach since all data contributes to estimation of both the optimized combination coefficients and the cutoffs. We apply our approaches to real data sets that refer to liver and pancreatic cancer.
|
Authors who are presenting talks have a * after their name.
