Measuring Network Dynamics of Opioid Overdose Deaths in the United States

TL;DR

The statistical robustness of the primary variable of interest, deaths in social proximity, supports the hypothesis of a social network effect on OODs and provides a pathway for public health interventions informed by social network structures.

Abstract

The US opioid overdose epidemic has been a major public health concern in recent decades. There has been increasing recognition that its etiology is rooted in part in the social contexts that mediate substance use and access; however, reliable statistical measures of social influence are lacking in the literature. We use Facebook's social connectedness index (SCI) as a proxy for real-life social networks across diverse spatial regions that help quantify social connectivity across different spatial units. This is a measure of the relative probability of connections between localities that offers a unique lens to understand the effects of social networks on health outcomes. We use SCI to develop a variable, called "deaths in social proximity", to measure the influence of social networks on opioid overdose deaths (OODs) in US counties. Our results show a statistically significant effect size for deaths in social proximity on OODs in counties in the United States, controlling for spatial proximity, as well as demographic and clinical covariates. The effect size of standardized deaths in social proximity in our cluster-robust linear regression model indicates that a one-standard-deviation increase, equal to 11.70 more deaths per 100,000 population in the social proximity of ego counties in the contiguous United States, is associated with thirteen more deaths per 100,000 population in ego counties. To further validate our findings, we performed a series of robustness checks using a network autocorrelation model to account for social network effects, a spatial autocorrelation model to capture spatial dependencies, and a two-way fixed-effect model to control for unobserved spatial and time-invariant characteristics. These checks consistently provide statistically robust evidence of positive social influence on OODs in US counties.

Figures (9)

• Figure 1: A) The spatial distribution of overdose death rates per 100,000 population in contiguous US from 2018 to 2019. Superimposed on this map is a social network diagram with edge widths representing the state-level social proximity weights . B) The two middle maps show the social proximity weights of alter states to California (on the left) and Pennsylvania (on the right). C) The bottom two maps show the spatial proximity weights of alter states to California (on the left) and Pennsylvania (on the right).
• Figure 2: A) The top left map shows the spatial spread of state-level opioid overdose death rates in the contiguous US. B) The top right map shows the spatial dispersion of "deaths in social proximity" for states in the contiguous US. C) The bottom left map shows the geographical spread of "deaths in spatial proximity". D) The bottom right map shows the difference between deaths in social and spatial proximity from top right and bottom left maps.
• Figure 3: The figure shows the distribution and relationships between the primary variables of interest (death rates $y_{i}$, deaths in social proximity $s_{-i}$, and deaths in spatial proximity $d_{-i}$). The histograms on the main diagonal depict the distributions of $y_i$, $s_{-i}$, and $d_{-i}$. Moving to the upper triangle, we observe the degree of linear dependence between these variables, while the lower triangle displays scatter plots.
• Figure 4: A) The plot shows the coefficient confidence interval plots for western and central US counties. The coefficient for $s_{-i}$ for cluster-robust linear regression (Supplementary Table \ref{['tab:simple_linear_model_with_robust_std_clustered_error-western-united-states']}), network and spatial autocorrelation (Supplementary Table \ref{['tab:network_spatial_autocorrelation_model_for_western_united_States']}), and two-way fixed effects model (Supplementary Table \ref{['tab:two-way-fixed-effect-western-united-states']}), all indicate a positive, significant (p$<$0.001) coefficient for $s_{-i}$. B). Shows the coefficient plot for social and spatial proximity for counties in the contiguous US. The coefficient for $s_{-i}$ for cluster-robust linear regression (Supplementary Table \ref{['tab:simple_linear_model_with_robust_std_clustered_error-contiguous-united-states']}), network and spatial autocorrelation (Supplementary Table \ref{['tab:network_spatial_autocorrelation_model_for_entire_united_States']}), and two-way fixed effects models (Supplementary Table \ref{['tab:two-way-fixed-effect-entire-united-states']}) are all positive and significant (p$<$ 0.001). C) Shows the coefficient plot for $s_{-i}$ and $d_{-i}$ for counties in the eastern US. The coefficient for $s_{-i}$ for cluster-robust linear regression (Supplementary Table \ref{['tab:simple_linear_model_with_robust_std_clustered_error-eastern-united-state']}), network and spatial autocorrelation models (Supplementary Table \ref{['tab:network_spatial_autocorrelation_model_for_eastern_united_States']}), and two-way fixed effects models (Supplementary Table \ref{['tab:two-way-fixed-effect-eastern-united-states']}) are all positive and significant (p $<$ 0.001). The effect sizes for standardized $s_{-i}$ in the cluster-robust linear regression models indicate that a one-standard-deviation increase, equal to $11.69523$, $12.2417$, and $5.7145$ more deaths per $100,000$ population in the social proximity of the ego counties in contiguous, eastern and western-central United States, respectively, is associated with thirteen more deaths per $100,000$ population in contiguous and nine more deaths eastern and six more deaths in western-central US counties.
• Figure 5: The diagram depicts the data pipeline for our analysis, including the data streams for the primary variable of interests, $s_{-i}$ and $d_{-i}$, as well as the relevant socioeconomic and clinical covariates. It also outlines our regression models for estimating the effect of peer influence as measured by SCI on county-level OOD outcomes.