Risk Prediction and Value of Polygenic Risk Scores in Colorectal Cancer Screening

Abstract

Risk prediction models that are based on common genetic variants, known as Polygenic risk Score (PRS), have shown promises to guide personalized screening for colorectal cancer (CRC). Continuous efforts to improve PRS risk prediction are needed for clinical use, and understanding its added value of guiding CRC screening is needed to inform screening guidelines, clinical adoption, and reimbursement decisions. In Chapter 1, we assessed whether the clinical validity of PRS risk prediction models could be improved by a new approach, called Multiple Polygenic Score (MPS) approach. This approach leverages PRSs developed for other diseases to enrich the risk prediction model. We first used machine learning models and large datasets to develop the MPS risk prediction models. We then used an independent dataset to validate the clinical validity of these models, measured by Area under the Receiver Operating Curve (AUC). Our results showed that MPS was a statistically significant predictor of CRC risk. Additionally, the increment in AUC associated with the MPS approach was small but was statistically significant. Our findings suggested that the MPS approach is able to improve PRS risk prediction models for CRC, and yet more efficient approaches to improve the AUC in a more noticeable way should be explored in the future. In Chapter 2, we developed a decision analytic model to simulate the long-term clinical and economic value of a population-level genomic screening to inform CRC screening. The genomic screening interventions included (1) population-level screening for PRS, (2) population-level screening for Lynch Syndrome, a rare but very high penetrance genetic syndrome associated with high risk of CRC (lifetime risk up to 70%), (3) population-level screening for both PRS and Lynch Syndrome. We compared these interventions with standard of care. We found that genomic screening for both Lynch Syndrome and PRS was marginally cost-effective and yet genomic screening for PRS only or Lynch Syndrome only was unlikely to be cost-effective. Our study also found that potential harms associated with false reassurance, i.e., reduced screening due to negative genomic results, could nullify the clinical benefits of genomic screening. The findings suggested that both Lynch Syndrome and PRS are important components of the value of population-level genomic screening. Additionally, our study emphasizes that proper risk communication with patients is critical to reduce the harm of false reassurance. Lastly, we found that the age of genomic screening generated the largest clinical benefits when it was offered at age of 0 years, and delayed genomic screening had a greater negative impact on individuals with LS than individuals with a high PRS. Our studies first help inform methodological development of PRS risk prediction in CRC. Future studies should continue to develop and examine new methods to improve the clinical validity of PRS efficiently. Our findings also help understand the economic value of population-level genomic screening for CRC. Future studies should continue to assess the value of genomic screening for other diseases to facilitate the understanding of the value of genomics in disease screening and prevention.