Home
Resource
Some effects of exposure misclassification on epidemiological studies
PDF
PDF format is widely accepted and good for printing.
Plug-in required
PDF-1(1.87 MB)
Citation & Export
View Usage Statistics
Staff View

Citation & Export
Hide

Simple citation

Zou, Jun. Some effects of exposure misclassification on epidemiological studies. Retrieved from https://doi.org/doi:10.7282/T3JM2DKM

Export

Click here for information about Citation Management Tools at Rutgers.

Statistics
Hide

Description

TitleSome effects of exposure misclassification on epidemiological studies
NameZou, Jun (author); RHOADS, GEORGE (chair); LIN, YONG (internal member); Demissie, Kitaw Demissie (internal member); Masaki, Kamal (outside member); Rutgers University; Graduate School - New Brunswick
Date Created2017
Other Date2017-05 (degree)
SubjectPublic Health, Epidemiology
Extent1 online resource (viii, 75 p. : ill.)
DescriptionIn many epidemiological studies the risk factor or exposure of interest is measured with significant error. In well-designed studies this error is non-differential with respect to the outcome, but it nevertheless makes it more difficult to detect associations and it biases estimates of effect toward the null. Less well recognized is that it increases the probability that a significant result, when found, will be a false positive. This is obvious if one considers the extreme example where the observed measure bears little association with the true value and is essentially random, in which case any significant result would have to be an alpha error. The traditional error model is not realistic in the presence of substantial error because with a fixed observed variance and a large error variance, the parameter variance is constrained. We propose a bivariate normal model, which makes fewer assumptions than the traditional model and does not constrain the underlying “true” variance. The model implies the need for larger sample sizes to assure that an effect associated with a misclassified variable is sufficiently unlikely to have occurred by chance that it implies the underlying true variable also shows the effect. A minimal estimate of misclassification can be obtained from the correlation between repeated measurements. When this correlation is low it implies a low correlation of the measurement with the true value and the need for large sample size increases that may make the use of such variables impractical. We use data from the Honolulu Heart Program, a large prospective study of cardiovascular disease to show that risk factors for heart attacks that have stood the test of time mostly are repeatable across a two- years time span with correlations exceeding 0.7. Other risk factors such as diet and physical activity that are believed to cause heart attacks but have been difficult to demonstrate within homogeneous populations have substantially lower repeatability correlations. These considerations emphasize the importance of good measurement of exposure in epidemiological studies.  
NotePh.D.
NoteIncludes bibliographical references
Noteby Jun Zou
Genretheses, ETD doctoral
Persistent URLhttps://doi.org/doi:10.7282/T3JM2DKM
Languageeng
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Version 8.5.5
Rutgers University Libraries - Copyright ©2024