Ray Han

Ph.D.

Peking University

Professor

PLCO (Learn more about this study)

PLCO-461

Mar 21, 2019

Novel Patient-Specific Hybrid Modeling for Personalized Early Detection of Cancer using Blood-Based Biomarkers

The early detection of cancers is of paramount importance as a cure is still possible through a surgical resection of the tumor. To achieve this goal, one plausible method is to exploit blood-based biomarkers complimented with mathematical modeling such that the earliest detectable time of a tumor by current imaging modalities can be predicted. Although promising, the parameterization of the mathematical model to each individual patient requires at least 3 measurements of the biomarker at 3 different time points, or in current clinical practices, predictions can only be made after 3 years of screening. This limitation limits the clinical utility of the mathematical model. In this project, we wish to address this limitation by developing a novel hybrid model that combines machine learning and mathematical models so that prediction can be done with only one biomarker measurement and the risk factors of the patient. Utilizing the serial PSA and CA125 data in the PLCO dataset, associations between the cancer risk factors and parameters will be learned through machine learning and subsequently, the predicted parameter will be fed into the mathematical model to enable a personalized detection of cancer. The performance of this hybrid model will be compared to existing longitudinal algorithms such as the method of mean trends (MMT) and the parametric empirical Bayes (PEB), as well as with the single thresholding method.