Discovery, Evaluation, and Clinical Utility of Biomarkers for the Early Detection of Disease

This project will develop new statistical methods to help discover biomarkers that are missed by current methods. This work will focus on high-throughput technologies— these are technologies that allow us to assay thousands of blood-based biomarkers—for the early detection of cancer. However, current statistical techniques may ignore biomarkers of potentially great promise. In addition, most statistical techniques can only be based on a binary disease status (which implies the presence or the absence of the disease or cancer) without accounting for potentially different stages of it. This is often inappropriate because different stages of the disease may need different care; by considering only two stages at a time, we do not use all available information.

Common statistical tools involve the so-called Receiver Operating Curve or ROC. The ROC curve is a statistical tool that describes the accuracy of a diagnostic test (biomarker). The ROC curve needs to be extended to a surface when we are dealing with three stages of the disease status (see Figure 1 that illustrates an example of two ROC surfaces, one dominating the other) , or to a hypersurface when we are dealing with more than three stages. This allows us to use all available information provided by the biospecimens or blood of the patients at the same time.

Figure 1. Two ROC surfaces that correspond to 2 different biomarkers referring to a trichotomouse setting. This involves simultaneously: 1. Healthy individuals, 2. Individuals with pancreatitis, 3. Individuals with early-stage pancreatic cancer. We observe that biomarker 1 (yellow) dominates/outperforms biomarker 2 (blue).

Our project will focus on the following:

1. We are developing new mathematical tools that will reveal more promising biomarkers that are currently being ignored.
2. We will look for ways of combining the top-performing biomarkers to get to biomarker panels that will include more than two disease stages.
3. Finally, we will develop the mathematical framework as well as the corresponding software to obtain optimized decision cutoffs, tailor-made to the aggressiveness of the disease and the specific profile of an individual. We are constructing these mathematical tools to be useful for any form of cancer or even other diseases. So far, we have studied new biomarkers and biomarker panels for the early detection of prostate cancer (Bantis and Feng 2016, 2018), lung cancer (Guida et al. 2018), pancreatic cancer (Capello et al. 2017), Alzheimer's disease (Bantis et al. 2017) and late‐onset sepsis in neonates (Bantis et al. 2019).

Leonidas Bantis, Ph.D.
Assistant Professor, Department of Biostatistics and Data Science, University of Kansas School of Medicine
Junior Faculty, Kansas Institute for Precision Medicine COBRE