The lecture by Ruth Pfeiffer, NIH, HHS, will take place at the Complexity Science Hub Vienna.
If you are interested in participating, please email to firstname.lastname@example.org
Electronic medical records and linked large databases are important resources for observational research. However, the design and analysis of scientific studies based on such data require special considerations, as the data are usually not collected to answer scientific questions.
I have designed studies to assess risk factors for cancer by linking various study populations to cancer registries. Examples include linked data from U.S. Medicare, that covers approximated 97% of Americans aged 65 and older (roughly 45 million individuals) to cancer registries, data from patients in transplant registries to cancer registries, and data from the AIDS cancer registries to cancer registries.
Biased estimates of exposure associations can result from selection biases in the study population, failure to control for confounding factors, and measurement error in routinely collected data. I discuss the bias of estimates of exposure effects when summary scores of high dimensional confounders, e.g. propensity scores, instead of the confounders themselves, are used to design observational studies and to analyze them.
Recently, we used a “negative control condition” to quantify bias in estimates of association from linked cancer registry-Medicare data. We have devised methods to use prevalent cases to estimate association parameters for incident disease. I show how to combine data from these large databases that have coarse information on many individuals with studies that measure detailed information in small samples to estimate associations and to improve prediction models.
Dr. Pfeiffer is a tenured senior investigator in the Biostatistics Branch of the Division of Cancer Epidemiology and Genetics (DCEG), National Cancer Institute (NCI). She received an M.S. degree in applied mathematics from the Technical University of Vienna, Austria, and an M.A. degree in applied statistics and a Ph.D. in mathematical statistics—both from the University of Maryland, College Park. At NCI she is an active collaborator on many research projects and mentors several fellows and junior investigators.
Ruth’s research focuses on statistical methods for absolute risk prediction, problems arising in molecular and genetic epidemiologic studies, and the analysis of data from electronic medical records. She is the recipient of a Fulbright Fellowship, an elected Member of the International Statistical Institute, and an elected Fellow of the American Statistical Association.