Dr. Dempsey’s research focuses on statistical methods for digital and mobile health. My current work involves three complementary research themes: (1) experimental design and data analytic methods to inform multi-stage decision making in health; (2) statistical modeling of complex longitudinal and survival data; and (3) statistical modeling of complex relational structures such as interaction networks. Current directions include (1) integration of sequential multiple assignment randomized trials (SMARTs) and micro-randomized trials (MRTs) and associated causal inference methods; (2) recurrent event analysis in the presence of high-frequency sensor data; and (3) temporal models for, community detection of, and link prediction using complex interaction data.
I have broad interests and expertise in developing statistical methodology and applying it in biomedical research. I have adapted methodologies, including Bayesian data analysis, categorical data analysis, generalized linear models, longitudinal data analysis, multivariate analysis, RNA-Seq data analysis, survival data analysis and machine learning methods, in response to the unique needs of individual studies and objectives without compromising the integrity of the research and results. Two main methods recently developed:
1) A risk prediction model for a survival outcome using predictors of a large dimension
I have develop a simple, fast yet sufficiently flexible statistical method to estimate the updated risk of renal disease over time using longitudinal biomarkers of a high dimension. The goal is to utilize all sources of data of a large dimension (e.g., routine clinical features, urine and serum markers measured at baseline and all follow-up time points) to efficiently and accurately estimate the updated ESRD risk.
2) A safety mining tool for vaccine safety study
I developed an algorithm for vaccine safety surveillance while incorporating adverse event ontology. Multiple adverse events may individually be rare enough to go undetected, but if they are related, they can borrow strength from each other to increase the chance of being flagged. Furthermore, borrowing strength induces shrinkage of related AEs, thereby also reducing headline-grabbing false positives.
Z. Tuba Suzer Gurtekin is an Assistant Research Scientist at the University of Michigan’s Institute for Social Research. Her research includes managing monthly surveys of consumer attitudes, expectations and behavior. Her published research focuses on methods to quantify nonresponse and measurement survey errors in probability and nonprobability sample surveys, and mixed-mode survey design and inference. Her research experience has included development of alternative sample, methodology and questionnaire designs, data collection and analysis methods for a general population in parallel survey modes. She also teaches Survey Sampling for Clinical Research at the University of Michigan’s Clinical Research Design and Statistical Analysis program (OJOC CRDSA).