Statistical methods for prospective evaluation of biomarkers:

Investigators:

Yingye Zheng, PhD, Anna S. Lok, MD

Hypotheses/Aims

The broad, long term goal of this research is to develop useful regression methodologies, graphical summaries and software tools for analyzing modern biomarker data and to provide recommendations for efficient design of marker validation studies.

The specific aims of this proposal are:

Aim 1:  To develop statistical methods for evaluating the clinical utility of biomarkers evaluated in a prospective cohort study.

The proposed analysis has three major aspects: (1) quantify the time-dependent accuracy of biomarkers; (2) generalize the methods to a combination of several predictive factors in regression framework; and (3) evaluate the accuracy of the biomarker in the presence of other competing outcomes.

Aim 2:  To develop statistical methods for analyzing marker data using cohort sampling and to provide recommendation on optimal sampling strategies in different clinical settings.

The proposed analysis has two major aspects: (1) develop estimating and inference procedures for calculating the accuracy summaries (time-dependent TPF, FPF, PPV and NPV) using cohort sampling designs such as case-cohort or nested case-control studies; and (2) evaluate the performance of different sampling strategies in biomarker studies to establish optimal study designs in a variety of clinical settings.

Aim 3: To develop statistical methods for evaluating the predictive accuracy of prognostic markers using longitudinal data.

The proposed analysis has two major aspects: (1) develop robust and flexible procedures to quantify and update the predictive accuracy of longitudinal markers; and (2) develop and evaluate a decision rule on the basis of risk, incorporating both cross-sectional and longitudinal marker information.