Biostatistics Seminar Series: Yun Li, PhD

Abstract: Breast cancer is the most commonly diagnosed cancer among American women. Newly diagnosed patients often undergo many medical tests to facilitate treatment selections. The impact of tests on treatment selections is of interest and often not well studied. First, selection bias occurs frequently since studies on the impact of testing are often observational. Second, causal inference methods used to address selection bias are often limited to estimating a net effect of testing. However, the net effect ignores the heterogeneity of treatment effects. Third, missing data are common, especially when multiple data sources are combined. We propose a statistical method to quantify the causal influence of medical testing on treatment selection, while addressing selection bias, missing data and individual effect heterogeneity. To illustrate, we will examine the impact of the 21-gene recurrence score assay, the most commonly used genomic assay in the US, in breast cancer. The analyses will use combined data sets from the NCI-funded Surveillance, Epidemiology, and End Results databases (SEER) registry, genomic testing laboratories, and patient/physician surveys.