Structured nonparametric methods for mixtures of exposures
- Funded by National Institutes of Health (NIH)
- Total publications:0 publications
Grant number: unknown
Grant search
Key facts
Disease
COVID-19Start & end year
20182022Known Financial Commitments (USD)
$438,776Funder
National Institutes of Health (NIH)Principal Investigator
PendingResearch Location
United States of AmericaLead Research Institution
DUKE UNIVERSITYResearch Priority Alignment
N/A
Research Category
Epidemiological studies
Research Subcategory
Disease transmission dynamics
Special Interest Tags
N/A
Study Type
Non-Clinical
Clinical Trial Details
N/A
Broad Policy Alignment
Pending
Age Group
Not Applicable
Vulnerable Population
Not applicable
Occupations of Interest
Not applicable
Abstract
The parent R01 focuses on developing reliable and interpretable statistical methods for theassessment of simultaneous health effects of multiple chemicals. This is challenging due to thestatistical curse of dimensionality, to moderate to high correlation in levels of exposure todifferent chemicals, and to missing data and limit of detection issues. Current statisticalmethods for nonparametric regression fail to adequately address these challenges, and canproduce uninterpretable dose response surfaces and high error rates in detecting interactions.The parent R01 is developing transformative methods that incorporate mechanistic constraintson response surfaces, allow for the complications inherent in epidemiology studies of mixtures,produce interpretable results including for interactions, and borrow information across differentdata sources. This R01 has already produced new statistical tools that clearly improve upon thestate-of-the-art, and that can be implemented routinely by epidemiologists using publicly-available software packages (e.g., Ferrari and Dunson, 2020a,b).This proposal describes a competitive revision of the parent R01 to provide a transformativestatistical toolbox for epidemiologists studying risk factors for COVID-19 infection,morbidity and mortality. This toolbox builds on the Bayesian modeling frameworks developedby the parent R01, while crucially accounting for the types of large spatially and temporallystructured datasets that are now being collected as part of the COVID-19 monitoring effort. Anew class of computational algorithms is proposed for rapid analysis of massive andcomplex spatial-temporal data, these algorithms are used to develop statistical tools forepidemiologists studying COVID-19 including an R package, and the approach is applied tostudy interactions between environmental exposures, age, and other comorbidities withCOVID-19 mortality.