BACKGROUND: The public health burden associated with injection drug use (IDU) in Clark County (Nevada) demands increasing awareness and targeted interventions. Rates of newly diagnosed HIV increased over the past five years in the county, with IDU exposure implicated in about 9% of these cases. There was a concurrent increase in mortality due to hepatitis C among residents over the same period, indicating its transmission through IDU. To enable a harm reduction strategy aimed at predicting and mitigating IDU-related health consequences, we obtained small-area estimates of injection practices associated with HIV acquisition using a spatial process model. The spatial model provides an approach to implement evidence-based syringe service programs (SSP) and other harm reduction efforts underway by the Southern Nevada Health District (SNHD).
METHODS: IDU practice data were derived from in-person needs assessment conducted by SNHD and partners, and aggregated to the zip code level. Chronic hepatitis and liver cirrhosis (excluding alcohol causes) attributable hospitalizations were extracted from hospital discharge databases, and cross-sectoral data on heroin seizures obtained from local police. These small-area covariate data were incorporated into a cokriging model to help explain intra-urban variations in past/present IDU among persons living with HIV/AIDS (IDU+PLWHA) (reported to eHARS as of 2015). Cross-covariance between IDU+PLWHA and covariates was specified as a function of separation distance, and local covariance structure selected on the basis of cross-validated error statistics. Model-estimated risk of IDU in HIV acquisition was mapped using ArcGIS v10.3.
RESULTS: The distribution of IDU+PLWHA in Southern Nevada displayed positive spatial association (Moran’s I=0.128, p-value=0.000). The multivariate kriging model using neighborhood characteristics including needle users, chronic hepatitis and liver cirrhosis morbidity, and heroin seizures captured nearly 70% of the variance in IDU+PLWHA. The key feature of the prediction surface is concentration of IDU+PLWHA in central parts of the Las Vegas valley, particularly in neighborhoods of lower socioeconomic status.
CONCLUSIONS: We employed a spatial stochastic model to evaluate patterns in injection drug activity associated with HIV acquisition in Southern Nevada. By synthesizing data from multiple sources (i.e., eHARS, inpatient discharge, IDU needs assessment, law enforcement), this spatial modeling structure facilitated a better understanding of local differences in IDU-associated HIV transmission and behavioral covariates, which enabled SSP resource planning and prioritization at the neighborhood scale.