Biomarker-assisted Decision Rule for Risk Classification under a positive predictive value - constraint: Application to disease screening
Principal Investigator
Name
Yingqi Zhao
Degrees
PhD
Institution
Fred Hutchinson Cancer Center
Position Title
Associate Professor
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-1556
Initial CDAS Request Approval
May 14, 2024
Title
Biomarker-assisted Decision Rule for Risk Classification under a positive predictive value - constraint: Application to disease screening
Summary
Novel biomarkers, when combined with existing clinical data, are being actively pursued to improve clinical decision-making across various medical fields, including screening, surveillance, and prognosis. In many applications, investigators are often interested in combining these biomarkers to construct a rule that is useful under certain practical and clinical constraints. In this work, we formulate the problem of developing an optimal biomarker combination rule that maximizes benefit under a pre-specified positive predictive value constraint. We propose several estimators of the optimal decision rule and also study asymptotic properties of the proposed estimators. We compare our method to the standard approach used in practice through extensive simulation studies and show that our approach has good finite sample performance. Overall, our method provides a valuable tool for medical decision making that takes into account patient heterogeneity and practical constraints. We want to apply and validate our proposed method to identify patients at higher risk of ovarian cancer using ovarian cancer data from the PLCO study.
Aims
1. Develop biomarker combination rule for disease risk classification.
2. Specialize the method to situations where the disease prevalence is rare (e.g., screening ovarian cancer, pancreatic cancer, etc)
3. Prove asymptotic properties of our proposed estimator.
4. Show finite sample performance of our proposed estimator through simulation studies and compared with a performance of a standard method of biomarker combination rule (logistic regression).
5. Apply and validate our proposed method using ovarian cancer date from the PLCO study.
Collaborators
Albert Osom (aosom@uw.edu) - University of Washington, Seattle
Ziding Feng (zfeng@fredhutch.org) - Fred Hutchinson Cancer Center