Colorectal cancer is a leading cause of cancer death, yet it is among the most preventable cancers when detected early via screening. The guidelines for initiation of colorectal cancer screening are currently only based on two risk factors: age and family history of colorectal cancer. This strategy, which leads both to substantial under- and over-utilization of colorectal cancer screening, does not account for substantial variation of risks in the population besides age and family history of disease.
To develop improved, individually tailored screening guidelines, we build comprehensive risk prediction models based on our large genome-wide genetic data, and harmonized lifestyle and environmental risk factor data using machine learning algorithms. This work is led by Dr. Li Hsu. We use our models, which are the most comprehensive and best performing models, which we used to estimate personalized starting age of screening (see Figure). These models are more cost-effective than standardized screening, as demonstrated by collaborators at the University of Rotterdam. We conduct the calibration and validation in an independent, external and community-based cohort of over 100,000 participants from Kaiser Permanente Northern California.
We demonstrate that our polygenic risk prediction model is substantially more predictive of early onset (<50 year of age) versus late onset colorectal cancer, which is relevant given that the rapid increase in early onset colorectal cancer incidence rates stirred an intense debate about personalized screening recommendations for those below the age of 50.
The fact that most genetic research across complex diseases, including colorectal cancer, has been conducted in individuals of European ancestry led to the unavoidable consequence that the polygenic risk score is substantially more predictive in this population, as shown by us and others. To overcome this key ethical and scientific challenge, we develop polygenic risk models that are predictive across all major U.S. populations by expanding genetic research to include African Americans, Asians, and Latinxs.
To accelerate the translation of several decades of genetic and epidemiologic research on colorectal cancer risks, we are collaborating with eMERGE (Electronic Medical Records and Genomics) network investigators to evaluate the genetic risk prediction models in the clinic setting. Furthermore, we are developing a pragmatic trial to test the impact of personalized screening regimens on screening uptake. These steps toward precision medicine may increase adherence, maximize the appropriate use of invasive technologies and reduce colorectal cancer mortality.