Statistical Methods for Prospective Evaluation of Biomarkers

The overarching theme of this project is to develop novel statistical approaches for designing and analyzing biomarker discovery and validation studies. We consider here prospective studies in clinical settings where risk markers are used for disease surveillance and prognosis. Motivated by our collaborative research in the cancer biomarker field, we plan to address several new challenges in prospective marker evaluation. For many cancers, disease outcomes may be heterogeneous due to the multi-focal nature of the disease. The specific prediction of the risk of developing aggressive cancer as opposed to indolent cancer is of great clinical interest, yet it is analytically challenging. In Aim 1 we plan to develop statistical tools for risk stratification and individualized treatment rules in a population with a mixture of indolent and aggressive cancers. Aim 2 will address complications in disease outcome ascertainment. Efficient and unbiased estimation procedures will be developed to quantify the prognostic and predictive accuracy of biomarkers. These methods will be developed in the presence of interval censored data and surveillance-triggered outcome ascertainment and imperfect or surrogate outcome measurements. Finally, we will derive novel criteria for model selection with longitudinal data and develop novel approaches for deriving and validating dynamic surveillance regimens for disease monitoring. The proposed methods will be tested in a wide range of practice settings in cancer biomarker studies, including stratifying breast cancer survivors in risk of second primary breast cancer; developing and evaluating optimal biopsy interval regimen in the active surveillance of prostate cancer; accommodating surveillance-triggered outcome ascertainment schemes; and making treatment decisions among patients who are at high risk for liver cancer, or colorectal cancer recurrence.