Dissertation/Thesis Abstract

Is age too automatically controlled for as a confounder in epidemiologic studies?
by Ulfers, Margaret Inskeep, Ph.D., The George Washington University, 2008, 196; 3311324
Abstract (Summary)

Differing age distributions can confound comparisons between groups due to the close relationship between age and mortality (or morbidity) making age an important variable in epidemiologic studies and leading to its common treatment as a potential confounder. Theoretical work has shown that inappropriate control of confounding can seriously bias results, but this hasn't been specifically aimed at control of age. An extensive literature search including major epidemiologic textbooks, found few guidelines for the best ways to include age in analyses (even alongside advice to always include it). To assess if the criteria for confounder selection are being applied to age, and if not, if there are any adverse consequences, a literature survey was conducted and comparisons made between age-adjusted and unadjusted analyses of simulated data. The literature survey also looked for historical changes in age control and described practices related to age inclusion. Original research epidemiologic articles were randomly selected from the American Journal of Epidemiology (AJE) for the years 1973, 1985, 1995 (48 articles each year) and 2005 (72 articles) and in New England Medical Journal for 2005 (72 articles). 2.8 percent included age explicitly in some part of the study design or analyses. Of 223 analytical studies, age was treated as a potential confounder in 89.7% (95% CI: 84.9, 93.4) and adjusted for in the final results in 64.6% (95% CI: 57.9, 70.8). Of the 200 studies including age as a possible confounder, 68.5% (95% CI: 61.6, 74.9%) gave some evidence of possible confounding. Only 7.6% gave evidence of both age-exposure and age-outcome associations. Effect modification by age was addressed in 29.6% of all analytical studies surveyed. Almost no studies justified choice of age form (i.e. nominal, continuous, or different strata size) used. Age adjustment appeared to be automatic in many studies since there were no or insufficient reasons given for doing so. Simple models of populations with yearly disease accumulation and fixed exposure or yearly accumulation of both exposure and disease were analyzed as cross-sectional studies of disease prevalence comparing the adjusted and unadjusted prevalence ratios (PR) and prevalence odds ratios (POR). The emphasis was on a qualitative assessment since this was not designed to model the exact age associations. Results demonstrated that in the first model with no age-exposure association, adjustment moved the POR further from the yearly relative risk (RR) but did not change the PR. If confounding were judged solely by a change in estimate and the PR not examined, this would have been misleading. In the second model, when age is associated with both exposure and outcome, adjustment always removed a confounded association when no true association existed between exposure and disease, but did not always improve estimation of the model RR compared to the unadjusted measure. Restricting the age range in these models increased the bias of the prevalence measures in some cases. Results suggest that a more careful consideration of age control is needed and meeting the criteria of being associated with exposure and disease is insufficient if effect modification or bias is present.

Indexing (document details)
Advisor: Hirsch, Robert P.
Commitee: Cleary, Sean D., Hoffman, Daniel, Rice, Madeline M., Verme, Dante, Wirtz, Philip
School: The George Washington University
Department: Epidemiology
School Location: United States -- District of Columbia
Source: DAI-B 69/06, Dissertation Abstracts International
Subjects: Epidemiology
Keywords: Age, Confounders, Effect modification, Epidemiologic methods, Prevalence ratios
Publication Number: 3311324
ISBN: 9780549619673
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy