In this research, we will study two related statistical topics aiming at evaluating longitudinal biomarkers that have impacts on clinical outcome and individualized medical decision. The results of this study will be applicable to classifying patients? response rates based on marker evaluation, predicting disease onset, and selecting treatment protocol that would benefit individual patients. In treating colon cancer, the patients? survival rates might be associated with the interaction of surgery and biomarkers, e.g. c-myc gene expression levels. In HIV studies, HIV-1 RNA and CD4 levels may result in different outcomes for various treatment groups. Such biomarkers could be indicators/predictors of which patient groups may benefit more from specific treatment options. Marker evaluation methods and tools are critical in finding the optimal treatment protocol for patients. We will develop a comprehensive and user-friendly environment for analyzing marker data, and the first goal focuses on marker evaluation in treatment selection. Typically, a selection criterion is based on a threshold of a marker. An innovative approach is to evaluate a selection policy by the Selection Impact (SI) score based on the treatment assignment proportion of a given threshold. By plotting such proportions and the SI score of a given threshold, one can easily find the optimal threshold of biomarker as well as the best marker in marker comparison. Our second goal focuses on the classification capability of markers in selecting diagnostic tests based on marker values. Receiver Operating Characteristics (ROC) curve is the traditional method in evaluating images in radiological studies. The challenges are that the level of biomarker may vary over time and the clinical outcome might be a time-to- event with censored values. Typically, the conventional image diagnostic tests focus on binary outcome with fixed marker values at the baseline, e.g. positive or negative results in tumor evaluation. We will emphasize more on ROC curve for longitudinal markers and survival outcome. At the end, we will develop various methods for SI and ROC curves and integrate them into a user-friendly statistical software. With such tools, we will promote the applications of time-varying markers and predictors in personalized treatment decision.
This project will provide statistical software with advanced methods useful for marker selection, evaluation, and personalized treatment. Statistical methods, graphics, computational environment and case studies will be developed.
