- This event has passed.
Biostatistics Student Seminars: Daniel McGuire, Renan Sauteraud and Lin Qiu
December 5 @ 12:00 pm - 1:00 pm
“Effective representation for categorical variables in modeling datasets” will be presented by Daniel McGuire, fourth-year Biostatistics PhD student at Penn State College of Medicine.
Categorical predictor variables may represent valuable information in predictive models. In real-world datasets, some high-dimensional categorical variables may be extremely sparse and have hundreds of levels. Various strategies might be employed in the way a categorical variable is encoded numerically in a modeling dataset, which might affect a model’s predictive performance, efficiency and interpretation. McGuire’s summer internship project at Travelers Insurance focused on comparing a few different methods of encoding categorical predictors. During his presentation, he will discuss several common strategies and a systematic approach for comparing them using a real high-dimensional dataset with insurance claims information. Some background on Travelers and data science in the insurance industry will be discussed.
“Standardized patient profiles visualization tool for clinical trials” will be presented by Renan Sauteraud, fourth-year Biostatistics PhD student at the College of Medicine.
The patient profile is a highly flexible, general-purpose shiny module aimed at providing high-quality and value-added patient-level data visualizations. The application comprises several submodules that facilitate the exploration of clinical trial data, including adverse events, concomitant medications, drug exposure, laboratory measurements, ECG and PK information. Importantly, the application was designed with the ability to work seamlessly with the popular AVA safety explorer and the liver function plots known as eDISH, as well as with popular interactive visualizations such as those using plot.ly. This functionality allows users to identify and select patients with clinically interesting measurements and easily drill down to additional data to provide further patient characterization. As part of the mission of Novartis’ Scientific Computing and Consulting group to push for the use of open-source software in the pharmaceutical industry, this module follows FDA data standards and is intended to be used industry-wide.
“An Exhaustive Search Based Approach for Subgroup Identification” will be presented by Lin Qiu, fourth-year Biostatistics PhD student at the College of Medicine.
Heterogeneity caused by genetic and environmental factors can make it very difficult to develop treatments that benefit all patients. A central goal of precision medicine is to identify patient subgroups whose average response to a treatment is much higher or lower than the population average. Because of the availability of vast troves of data of various types, the subgroups should be defined in terms of biomarkers (genetic profiles, laboratory test results and history and severity of illness) as well as demographic variables. Qiu’s research proposed a novel exhaustive search-based subgroup identification approach at AbbVie during a summer internship. The extensive simulation studies and real data applications will be discussed.