For each survey we analyzed publicly available data using SUDAAN software version 9.1 (RTI International, Research Triangle Park, NC) [13] to account for the complex sample designs. These surveys allowed for 2 estimates of the population proportion of PWID based on recall period: an estimate for “lifetime” injection drug use and an estimate for “past year.” Due to the small number of PWID in each survey year, we aggregated data across years for more robust estimates, resulting in a single estimate for each data source. Prior to aggregating, we examined data for each year and determined no pattern of increasing or decreasing trend during the 2000–2008 time period. Small sample sizes and limited data points (surveys collected data every 2 years or aggregated data into 2-year periods) may have limited power to detect minor changes over time. However, the lack of trend in the 4 surveys is consistent with stable estimates of PWID reported by Tempalski and colleagues for the period 1992–2007 [10]. For the lifetime estimate, NSDUH, NHANES, and GSS contributed data; for the past year estimate, NSDUH, NHANES, and NSFG contributed data.
