BACKGROUND: Rising rates of injection drug use (IDU) have led to a co-epidemic of opioid overdose and hepatitis C virus (HCV) infection in the United States, with increases in incidence of both conditions over the last decade. In 2014 IDU was associated with an outbreak of HIV and HCV in rural Indiana, resulting in 212 new HIV infections as of December 2016. In response, the CDC developed a nationwide vulnerability index to determine which counties are at risk for similar IDU-associated disease outbreaks. To obtain a state-specific perspective we recreated the analysis using national census data enhanced with Illinois data sources which provided additional indicators, more recent data, and a different proxy for unsterile IDU.
METHODS: We used multilevel Poisson regression models to select indicators associated with emergency department (ED) visits for heroin use in Illinois from July 15 to December 14, 2016 as a proxy outcome for IDU. This outcome measure was captured by the state syndromic surveillance system, which receives and processes free-text chief complaint data from EDs. Potential indicators were selected from several topic areas including sociodemographic characteristics, access to care, drug-related criminal activity, and IDU-associated disease. Descriptive statistics were calculated for fifteen indicators chosen based on the availability of recent county-level data. Heroin use ED visits were modeled against each indicator and in a multilevel Poisson regression model with counties as random effects and total ED visits as the offset. Backwards stepwise regression was used to select a parsimonious set of indicators significantly associated with the proxy outcome for IDU.
RESULTS: The final multivariable model included four indicators positively associated with heroin use ED visits: controlled substances arrests, per capita income, insurance coverage, and heroin seizures. The selected indicators and magnitude of effects are different from the national model developed by the CDC.
CONCLUSIONS: We created a model to identify indicators associated with ED visits related to heroin use in Illinois. Two of the four final indicators were from criminal justice datasets, which are only available at the state level. This, in addition to other differences from the national model, indicates that locally generated models may reveal additional information about IDU and outbreak vulnerability. Future steps include assessing linearity and collinearity of the indicators, creating standardized regression coefficients, and developing a vulnerability index. This information will allow public health authorities to more effectively allocate resources to prevent disease outbreaks and other adverse outcomes associated with IDU.