Electronic Theses and Dissertations Archive

Date

2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Public Health

Committee Chair

Abu Mohammad Naser Titu

Committee Chair

Fanta Gutema

Committee Member

Dale Sanders

Committee Member

Debra Bartelli

Committee Member

Meredith Ray

Abstract

Environmental exposure and public health surveillance data are often collected or generated across repeated observations, shared populations, and broader geographic contexts that introduce hierarchy across multiple levels. When these data are analyzed without accounting for that structure, between-unit variation may be mischaracterized or misattributed, which may affect the conclusions ultimately drawn. Although the utility of random effects models, linear mixed-effects models, and their extensions has long been recognized for addressing these correlations, their application in epidemiologic studies may remain underutilized in descriptive settings where the analytical goal is not causal effect estimation. Thus, this dissertation examined how their application can improve descriptive inference across three retrospective observational studies involving large, clustered datasets. The first study evaluated the contribution of drinking water salinity to 24-hour urinary electrolyte excretion using paired samples from 9,748 person-visits across three cohort studies in southwest coastal Bangladesh. Three-level linear and linear quantile mixed-effects models showed that, as salinity changed, daily urinary electrolyte excretion varied in its proportion attributable to drinking water. While contributions were modest for sodium, chloride, and potassium, they were more substantial for calcium and magnesium. The second study characterized spatiotemporal patterns in 9,834 single-state foodborne outbreaks reported to the National Outbreak Reporting System in the United States between 2011 and 2023. Negative binomial, linear, and log-linear Poisson mixed models showed that reported outbreaks remained largely stable over time, reported illnesses declined nonlinearly, and reported hospitalizations declined modestly. Reported deaths showed no meaningful temporal change, reporting completeness improved slightly, and reporting ambiguity changed little. For all outcomes, substantial heterogeneity persisted across states without a discernible regional pattern. The third study explored state-level characteristics in 3,393 single-state restaurant-associated foodborne disease outbreaks reported between 2014 and 2023. Negative binomial mixed models guided by a hierarchical conceptual framework showed that adding state-level predictors to models with only outbreak-level predictors reduced residual between-state variation by nearly half for illnesses and a quarter for hospitalizations. These findings demonstrate that hierarchical mixed-effects models can improve descriptive epidemiologic inference in environmental exposure assessment and surveillance analyses involving clustered data.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest/Clarivate.

Notes

Embargoed until 06-05-2028

Download

Available for download on Monday, June 05, 2028

Share

COinS