Events login

Main Events Calendar

Warning Icon This event is in the past.
November 28, 2018 | 1:45 p.m. - 2:45 p.m.
Category: Seminar
Location: Integrative Biosciences Center 1D
Cost: Free
Audience: Academic Staff, Alumni, Community, Current Graduate Students, Current Undergraduate Students, Faculty, Parents, Prospective Students, Staff

The Center for Urban Responses to Environmental Stressors (CURES) presents their Wednesday afternoon seminar series on November 28, 2018 from 1:45 to 2:45 p.m. at the IBio Building in Seminar Room 1D, located at 6135 Woodward Avenue.  The seminar is free and open to the entire university community.

The guest speaker will be Roman Jandarov, PhD, Assistant Professor of Biostatistics and Bioinformatics in the Department of Environmental Health, College of Medicine at the University of Cincinnati.  Dr. Jandarov will speak on "A Unified Exposure Prediction Approach for Multivariate Spatial Data".

Prior to joining the University of Cincinnati, Dr. Jandarov was a postdoctoral senior research fellow in the Department of Biostatistics at the University of Washington.  At the University of Washington, his postdoctoral research focused on statistical problems arising in air pollution epidemilogy.  Dr. Jandarov obtained his PhD from the Department of Statistics at Pennsylvania State University.  In his doctoral dissertation, Dr. Jandarov developed inferential methods and statistical tools for infectious disease and population dynamics models and collaborated with leading experts in the field of infectious diseases.  Dr. Jandarov also holds a Specialist degree in Mathematics (combined undergraduate and graduate degree equivalent of M.S. in U.S.) from Lomonosov Moscow State University.

ABSTRACT

In epidemiological cohort studies of health effects of air pollution, a well-known challenge is the lack of availability of exposure data at subject locations. While the most accurate way to characterize individual-level exposures is to conduct field studies using personal monitors, measuring each person's exposure using the personal monitors is cost prohibitive and often burdensome on the study subjects.  The most common approach used in air pollution epidemiology studies is the use of statistical modeling to predict the individual-level exposures by utilizing ambient measurements of pollution concentrations at monitoring locations and health analysis of these predictions as surrogates for personal exposure.  When presented with a multi-pollutant exposure prediction problem, it is possible to adapt univariate methods by applying them independently to each pollutant.  While these advanced statistical models for predicting pollution exposures can incorporate all important meteorological, geographical and land-use information and allow for spatiotemporal dependence between the pollution concentrations at close distances in space and time to obtain accurate exposure preditions for a single pollutant, applying the univariate models to each pollutant of the multi-pollutant scenario will ignore a potentially important source of information that lies in the correlations between the pollutants.  We develop novel unified exposure prediction approaches for multi-pollutant data based on the idea of linking models in a chain.  In the proposed approaches, we apply univariate models sequentially and the predicted exposures from each model are used in the subsequent models as an input.  We also incorporate dimension reduction and variable selection techniques before applying each model.  The methods are applied to simulated data with different covariance structures and to monitoring data from the U.S. Environmental Protection Agency.  The results demonstrate that chain-based approaches can lead to increased prediction accuracy compared to traditional univariate models.

For more information about this event, please contact Christina Cowen at 313-577-6590 or mzchris@wayne.edu.