Abstract 5240: Validation of research cohort based penetrance estimates for multiple cancer types and multiple primary cancers on clinically ascertained families

Abstract

Abstract Age-at-onset penetrance is the probability of developing disease by a certain age given a patient’s characteristics. Accurate penetrance estimates are important for cancer screening and diagnosis, and for policy making in public health. We developed two Bayesian semi-parametric models: a model that estimates cancer-specific penetrances to the first primary (CS), and a model that estimates penetrances to the second primary without distinguishing between the cancer types (MPC). Model training requires a dataset that is enriched with enough MPC cases, as well as with competing risks from a variety of cancer types. Thus, we trained our models using data from families affected by the Li-Fraumeni Syndrome (LFS), a genetic disorder characterized by germline mutation in the gene TP53. This dataset consists of 189 families, all of which were ascertained via probands who were diagnosed with pediatric sarcoma at MD Anderson Cancer Center (MDACC) between 1944 and 1983. We have validated our penetrance estimates on independent LFS datasets with different ascertainment criteria, namely a patient cohort prospectively collected from high-risk clinics at MDACC, and another cohort from the National Cancer Institute (NCI). These datasets were meticulously collected for research purposes. However, clinical datasets, which resemble the data that genetic counselors encounter in counseling sessions, are severely impacted by missing data, most commonly missing ages at diagnosis and ages at last contact, and only represent a snapshot of family history without extended follow-up. To investigate the utility of our research-based risk prediction models in real-world clinics, we evaluated our penetrance estimates on clinically ascertained LFS families that were collected through the Clinical Cancer Genetics (CCG) program at MDACC in the past 10 years. The CCG dataset consists of 3,275 individuals across 124 families, of which 645 were diagnosed with at least one primary and 127 tested positive for TP53 mutation. Our CS model showed good performance when making cancer-specific predictions of the first primary, with Areas Under the Curve (AUCs) of 0.76, 0.81, and 0.68 for breast cancer, sarcoma, and all other cancer types combined. Our MPC model showed an AUC of 0.7 when predicting patients with MPC versus those with a single primary. Both the CS and MPC models perform better than the Chompret criteria when predicting TP53 mutation, with AUCs of 0.75 and 0.77 respectively. These results suggest that our models can be utilized for personalized risk predictions in clinical settings, despite being trained on a research-based dataset. Given that the Chompret criteria tends to predict TP53 mutation more conservatively, the ability of our models to cast a wider net is especially meaningful in the identification of potential carriers, which assists efficient genetic testing and screening. Citation Format: Nam H. Nguyen, Elissa B. Dodd-Eaton, Jessica L. Corredor, Jacynda Woodman-Ross, Nathaniel D. Hernandez, Angelica M. Gutierrez Barrera, Banu K. Arun, Wenyi Wang. Validation of research cohort based penetrance estimates for multiple cancer types and multiple primary cancers on clinically ascertained families. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5240.