Uncovering Patterns in the National Opioid Crisis: Investigating the Links Between Opioid-Related Mortality, Prescription Opioids, and Social Vulnerability

The opioid crisis remains one of the most complex and daunting public health challenges in the United States. Despite national efforts that reduced opioid prescribing by nearly 45\% from 2011 to 2021, opioid-related overdose deaths have more than tripled; such alarming trends raise important questions about what underlying social factors may be driving opioid misuse. This dissertation investigates the role of select social vulnerability index (SVI) variables in two phases. Using county-level data spanning 2014 through 2020, the first phase analyzes the relationships between opioid-related mortality rates, opioid prescription dispensing rates, and disability. To successfully estimate and predict trends in these opioid-related factors, we employ a spatially augmented Kalman filter. Using the filter's predictions, we create yearly heat map vulnerability profiles and identify hotspots. In this context, hotspots are defined on an annual basis as counties with rates in the top 5 percent nationally. The results from the first phase highlight consistent patterns of vulnerability in Appalachia and reveal a connection between disability and opioid-related mortality. Furthermore, the findings suggest a shift in the crisis from prescription opioids to illicit drugs.

The shift towards illegal opioids motivated the second phase of this dissertation to focus on a broader set of social vulnerability factors. Using county-level data from 2010 to 2022, this stage examines the role of thirteen more SVI variables in relation to opioid-related mortality. It begins with a preliminary analysis of how the rates of each SVI variable manifest in counties with both anomalously high and low mortality rates, identifying patterns that warrant further investigation. Building on these findings, two machine learning models—XGBoost and an autoencoder—are then employed to further assess the importance of the thirteen SVI variables. Both models take the SVI variables as input to predict mortality rates for each county, allowing us to leverage two distinct feature importance metrics: information gain for XGBoost and a Shapley gradient explainer for the autoencoder. These metrics offer two unique insights into the most important SVI factors in relation to opioid-related mortality. The findings from the second phase highlight key social factors that may exacerbate the opioid crisis.