Personalized Colorectal Cancer Prevention: Integrating Individual Screening and Follow Up Information with Genetic Data

This is the first resubmission of a BLR&D CDA-2 proposal to provide five years of salary support to Brian Sullivan, M.D. M.H.S., toward his goal of developing an independent VA research career. Dr. Sullivan is an Investigator at the Durham Cooperative Studies Epidemiology Center (CSPEC), Directory of Quality at the Durham VA

Medical Center, and Assistant Professor at Duke University in the Department of Medicine. The long-term goal of this CDA is to provide Dr. Sullivan with the training and resources needed to lead a multi- disciplinary VA research program focused on colorectal cancer (CRC) prevention. There is considerable

variability in individual risk of CRC that could impact age at initiation of CRC screening, screening modality choice, and frequency of follow-up. Yet, guidelines do not recognize this variability, performing too much colonoscopy screening and surveillance in low risk individuals and not providing enough or timely screening and

surveillance in high-risk individuals. More precise risk stratification based on clinical and genetic factors offers a promising strategy to improve CRC prevention by targeting colonoscopy resources to individuals at increased risk for CRC, while reducing the costs and harms of invasive procedures in those at low risk who could undergo

delayed or non-invasive screening. This CDA will leverage powerful VA resources, including linked clinical repositories and genomic biobanks, to enhance current CRC risk-assessment algorithms with individual-level information to develop cost-effective and “smarter” CRC risk prediction tools that can guide cancer prevention.

Building on his prior work, Dr. Sullivan's short-term CDA goal is to create an accurate CRC risk prediction tool based on the clinical and genetic profiles of Veterans to improve CRC risk prediction and better target CRC screening resources. Specifically, he will establish a CRC risk cohort derived from natural language processing

(NLP)-based phenotype algorithms to more accurately identify cases of CRC and advanced precancerous lesions in large VA administrative databases, then apply emerging statistical models for longitudinal cohorts that incorporates clinical information from prior screenings (if available) or other testing to allow informative estimates

of CRC risk over time while taking competing risks of mortality into account (Aim 1). He will then link this database to the racially diverse Million Veteran Program (MVP) biorepository to test if additional CRC genomic discovery can bridge disparities and identify those at risk for CRC by creating a trans-ancestry polygenic risk score (Aim

2). Finally, he will augment genetic tools with known longitudinal CRC risk factors to create an accurate CRC risk prediction tool (Aim 3). This work will support future Merit applications for prospective validation and implementation studies of this tool. Dr. Sullivan has a team of expert mentors (including Drs. Elizabeth Hauser,

Jason Dominitz, David Lieberman, Samir Gupta, and Andrew Gawron) and consultants (Drs. Hillary Mull, Jason Vassy, Li Hsu, and Leah Zullig) who are qualified and strongly committed to helping him complete the CDA goals and become an independent VA investigator through future collaborative research. Together, his team has

designed a career development plan which includes comprehensive training in bioinformatics and large data acquisition, genomic analysis, and risk prediction. This mentorship, along with the rich Durham-CSPEC resources, will ensure that Dr. Sullivan has all the support necessary to meet his research and training aims.

In summary, this award will provide Dr. Sullivan critical support for his larger research goal of developing a novel CRC risk prediction tool based on clinical and genomic biomarkers to better inform individualized CRC prevention recommendations. Improving CRC risk prediction will be increasingly important as screening and surveillance

access becomes increasingly strained. This tool could both reduce CRC risk and unnecessary health care utilization by more precisely tailoring these CRC prevention resources. Our results will also provide much needed information about personalized risk prediction, and thus can also be broadly applicable to other cancer

prevention paradigms, which is something we will explore in future research.