Dynamic prediction incorporating time-varying covariates for the onset of breast cancer

PROJECT SUMMARY Triple negative breast cancer (TNBC) is an aggressive subtype of breast cancer with limited treatment options and poor survival. Approximately 12% to 17% of women with breast cancer are diagnosed with TNBC. Women with TNBC have relatively poor outcomes and cannot be treated with targeted therapies. However, the risk of TNBC is not uniform across all race-ethnic groups of

the US population. Review of epidemiologic risk factors and TNBC incidence shows limited insight to variation in risk or risk reduction with the exception of history of breast feeding and higher vegetable and grain intake. We aim to bring personalized dynamic prediction to improve the current TNBC risk classification paradigm to make full use of the longitudinal information, in

addition to the baseline information, where risk prediction/stratification can be updated as new observations are gathered to reflect the woman’s latest health- and behavioral-related status. Specifically, we aim to investigate 5- and 10-year TNBC risk prediction performance by proposing novel statistical methods that fully utilize the personalized mammogram-based risk factors from

repeated mammogram images. The proposed study capitalizes on the WashU TNBC cohort with rich digital mammograms with well-studied BC risk factors, 10-years of follow-up and pathology confirmed incident TNBC. All proposed statistical methods will be supplemented by R code that we will make publicly available.