Background: With the recent increase in stimulant and cocaine drug overdose mortality in Georgia and the continuing impact of the opioid epidemic, it is critical for public health researchers to have reliable methods to estimate and evaluate drug overdose mortality rates and risk factors. This study used statistical modeling to predict the drug overdose public health burden in Georgia counties. A multivariable Poisson mixed-effects model was developed to estimate county-level drug overdose mortality rates using 2017 opioid epidemic data.
Methods: An empirical review was performed on 73 county-level indicators to assess each indicator for potential association with drug overdose mortality. Principal component analysis was implemented for dimension reduction and was followed by a multicollinearity assessment and univariate analysis. Stepwise selection methods were performed, and potential effect modification was explored to finalize the predictive model. Final model fit was assessed by comparing estimated drug overdose mortality cases and spatially smoothed rates to the observed values.
Results: The predictors significantly associated with drug overdose mortality were Race (β=0.022; p<0.001), Opioid and Benzodiazepine Prescription Overlap (β=0.062; p<0.001), STD Rate (β=0.001; p<0.001), Opioid Prescription Length (β=-0.090; p=0.010), and Vehicle Inaccessibility (β=-0.036; p=0.032). A significant interaction was found between STD Rate and Race (β=-1.136e-05; p=0.018). 83% of predicted drug overdose mortality estimates were within 1 case of the observed 2017 death cases, 91% were within 2 cases, and 95% were within 3 cases. After performing spatial smoothing on the estimated cases for 2017, Bacon, Gilmer, Pickens, Fannin, and Haralson were identified as counties with the largest estimated smoothed mortality rates.
Conclusion: The final model provides researchers with a tool to identify which Georgia counties may demonstrate high drug overdose mortality rates and counts based on race, opioid and benzodiazepine prescription overlap, vehicle inaccessibility, opioid prescription length, and STD rate. The model’s accuracy in estimating the mortality cases for 2016 and 2017 indicates it is a helpful tool to understand the spatial spread of drug overdose mortality throughout Georgia. It is important for public health researchers to explore the identified risk factors further to understand how preventative measures can be implemented for high-risk counties in Georgia.
Table of Contents
1. Introduction 1
1.1. Drug Overdose Mortality in the United States 1
1.2. Drug Overdose Mortality and Surveillance in Georgia 2
1.3. Public Health Responses to the Opioid Epidemic 3
1.4. Literature Review 4
1.5. Purpose 6
2. Methods 8
2.1. Study Design 8
2.2. Outcome Variable 8
2.3. Potential Model Predictors 9
2.4. Regression Modeling 13
2.5. Model Fit Assessment 14
3. Results 16
3.1. Risk Factors Associated with Drug Overdose Mortality 16
3.2. County-Level Drug Overdose Mortality Estimates 18
4. Discussion 20
4.1. Understanding the Final Model 20
4.2. Comparison with Prior Research 21
4.3. Limitations and Challenges 21
4.4. Next Steps 22
5. References 24
6. Tables and Figures 29
About this Master's Thesis
|Subfield / Discipline|
|Committee Chair / Thesis Advisor|
|Estimating County-Level Drug Overdose Mortality in Georgia with Mixed-Effects Poisson Regression Modeling ()||2020-04-15 16:54:56 -0400||