Novel markers have the potential to dramatically change the decision making process in many branches of medicine, such as in early diagnosis of disease and in selecting patient specific treatments. Most biomarker tests are not perfect, and the costs of incorrect test results can be enormous, both in financial and human terms. To translate putative markers into standard medical care, rigorous evaluation is required. Prospec- tive cohort studies are crucial for such evaluation as time to event carries more information about a marker's value on early detection and prognosis than a simple measure of disease status. However prospective biomarker evaluation is challenging. Until now there has been little guidance provided for statistical de- sign and analysis of these studies.
We aim to develop novel statistical tools for designing and analyz- ing biomarker studies with time to event outcome. The proposal will emphasize three unique aspects in prospective biomarker evaluation. First, to fully characterize the diagnostic and prognostic potential of a novel marker, accuracy summaries must incorporate additional dimension of time.
In Aim 1 we propose time-dependent accuracy summaries and generalize the estimating methods which enable researchers to identify factors that influence marker performance, optimally combine several predictive factors and account for competing risk events. We also consider flexible procedures to quantify the updated predictive accuracy of longitudinal markers, and derive decision rules on the basis of risk, incorporating both cross-sectional and longitudinal marker information. Second, the ascertainment of marker information in a large cohort requires enormous resources. Cost-effective cohort sampling is therefore highly desirable.
In Aim 2 we will develop estimating and inference procedures for calculating accuracy summaries under cohort sampling designs and conduct simulation studies to shed light on the limitations and strengths of different sampling strategies under a variety of practical situations. Applications in cancer biomarker development provide a context for our research. Data from the Early Detection Research Network and from several large cohort studies will be analyzed. Programs and algorithms developed in this proposal will be made available to public.
The research proposal addresses the pressing need for strong statistical input in cancer biomarker field, providing comprehensive statistical tools which will enable investigators to conduct valid and more powerful biomarker validation studies and to evaluate the diagnostic and prognostic potential of novel biomarkers. The research will demonstrate lasting usefulness as more biomarkers are developed and studied for poten- tial impact on public health.
|Zhou, Qian M; Zheng, Yingye; Cai, Tianxi (2013) Assessment of biomarkers for risk prediction with nested case-control studies. Clin Trials 10:677-9|
|Zheng, Yingye; Cai, Tianxi; Pepe, Margaret S (2013) Adopting nested case-control quota sampling designs for the evaluation of risk markers. Lifetime Data Anal 19:568-88|
|Sun, Jianping; Zheng, Yingye; Hsu, Li (2013) A unified mixed-effects model for rare-variant association in sequencing studies. Genet Epidemiol 37:334-44|
|Schenk, Jeannette M; Hunter-Merrill, Rachel; Zheng, Yingye et al. (2013) Should modest elevations in prostate-specific antigen, International Prostate Symptom Score, or their rates of increase over time be used as surrogate measures of incident benign prostatic hyperplasia? Am J Epidemiol 178:741-51|
|Zhou, Qian M; Zheng, Yingye; Cai, Tianxi (2013) Subgroup specific incremental value of new markers for risk prediction. Lifetime Data Anal 19:142-69|
|Zheng, Yingye; Heagerty, Patrick J; Hsu, Li et al. (2010) On combining family-based and population-based case-control data in association studies. Biometrics 66:1024-33|
|Zheng, Yingye; Cai, Tianxi; Stanford, Janet L et al. (2010) Semiparametric models of time-dependent predictive values of prognostic biomarkers. Biometrics 66:50-60|