Using Negative Exposures to Partially Control for Unmeasured Confounders in Time-Series Analysis of Air Pollution and Health Restricted; Files Only

Abstract

Background: Observational studies face the challenge of confounding. Residual confounding may persist if known confounders are not properly measured or if unknown confounders are present. A regression-based method was proposed to directly reduce residual confounding in observational studies based on negative control exposures.

Objectives: The objective of this study is to investigate additional forms of negative exposure controls, including those that are lagged and the use of multiple negative exposures, as methods for reducing residual confounding by unmeasured or mis-measured confounders in time-series studies. These methods were evaluated in a time-series analysis of daily air pollution and asthma emergency department visits, as well as in simulation studies.

Methods: The study employed a log-linear model to estimate short-term effects of air pollution and observed counts of asthma ED visits using Poisson regression with overdispersion based on observed ozone level data during the study period. Model misspecifications were intentionally created by omitting one of the measured confounders and reducing the degree of freedom of nature cubic spline term for calendar time. We also fitted the model by adding the negative exposure, by adding each of the four negative exposures one-at-a-time, or combining a lagged negative exposure with the future negative exposure[ES1] [CX2] . Bias, root mean square error (RMSE), and coverage of the 95% confidence interval were calculated.

Results: Simulation studies showed that adding the negative control exposure can reduce bias in some scenarios. The standard error associated with estimated log relative risks for ozone is negligibly different when comparing the inclusion of negative control exposure versus excluding it. In some scenarios, the negative control exposure slightly increased coverage.

Conclusions: In simulation studies, we have shown that in estimating short-term health effects of air pollution with Poisson log-linear time-series model, the bias resulting with unmeasured confounders can be smaller when negative control exposure is included the model. In some cases, we also find that the use of multiple negative controls can further reduce bias. But this reduction of bias is not guaranteed.

Table of Contents

1.                  Introduction

2.                  Methods

2.1              Data

2.1.1 Asthma emergency department data

2.1.2 Ozone and meteorological data

2.2              Statistical Analysis

2.2.1 Time-Series Analysis of Ozone and Emergency Department Visit

2.2.2 Proposed Negative Exposure Indicators for Unmeasured Confounder Adjustment

3.                  Simulation Study

4.                  Results

4.1              Descriptive analyses

4.2              Results of Real data analysis

4.3              Results of Simulation studies

5.                  Discussion

Reference

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics
Degree	M.P.H.
Submission	Master's Thesis
Language	English
Research Field	Health Sciences, Public Health
Palavra-chave	Environmental Epidemiology Simulation Study Pollutant Asthma
Committee Chair / Thesis Advisor	Howard Chang, Emory University
Committee Members	Stefanie Ebelt, Emory University

Última modificação

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 18 May 2025	2023-04-10 10:16:25 -0400	File download under embargo until 18 May 2025

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions