Using a two-source capture-recapture approach, we generated a preliminary estimate of approximately 1,613 PWID served through CNHS, with a 95% confidence interval ranging from 404 to 2,821 PWID. This study was the first attempt at estimating the number of PWID in Cherokee Nation, and one of the few existing studies to estimate the number of AI/AN people who inject drugs in the United States.
The proportion of reported PWID was high. Among the 906 individuals who were asked if they had injected drugs in the past six months, as reported in the CNHS EMRs, 14.2% of individuals responded affirmatively. Among the 947 patients who were asked the same question through the HCV Elimination Program, 7.9% responded affirmatively. Each of these proportions were much higher than existing PWID estimates among AI/AN people in the literature. However, past studies have also focused on particular groups within AI/AN communities, and not the population of PWID as a whole (11–14).
There are multiple limitations to this study design, related to patient population, health systems, data sources, and methods. Research has demonstrated that social desirability biases lead individuals to underreport drug use, and, in particular, injection drug use (25). It is possible that underreporting of injection drug use led to a falsely low estimated number and proportion of PWID within CNHS. Alternatively, although injection drug use in the past six months is intended to be asked across all CNHS patients, most CNHS patients were not asked this question during their visits. Due to the small percentage of total CNHS patients who were asked the question, it is possible that clinicians were selectively asking this question. This may have introduced selection bias, leading to a falsely high estimated number and proportion of PWID within CNHS and violating the capture-recapture assumptions that all individuals included in the databases were correctly classified, truly had an equal, non-zero probability of being captured, and were representative of the true PWID population within CNHS.
Although the two sources used for the capture-recapture estimate were from different databases, the sources are inherently connected because both sources are abstractions from CNHS patient databases and, thus, are only representative of individuals who receive health services through CNHS. In addition, those who test positive for HCV from within a CNHS clinic are later included in the HCV Elimination Program Database, and PWID are more likely than the general population to become infected with HCV. Although PWID-related data are collected separately in these databases, the inherent connection violates the capture-recapture assumption that samples are independent from one another, which may have led to a falsely low PWID estimate. Although this is a key study limitation, it is also a frequently identified limitation in capture-recapture studies, as reliance on related institutions or services to provide data sources is common (23).
In addition, this study violates the capture-recapture assumption of homogeneity of the data sources. Individuals are only captured in the HCV Elimination Program Database if they are enrolled in HCV care, whereas all individuals served through CNHS are captured through CNHS EMRs. However, these were the two available data sources that captured injection drug use in Cherokee Nation and violation of homogeneity is common when using healthcare data (26).
Finally, CNHS is not a closed population and the number of PWID served through CNHS is not stagnant. Individuals can register for health services, leave the healthcare system, start injecting drugs, and stop injecting drugs over time. We attempted to reduce the potential bias of this violation of capture-recapture methods by limiting the data abstraction time period for both data sources to one year. However, when we attempted this analysis, the sample size of PWID was too small to conduct the necessary calculations. Even when including all available data on PWID from CNHS EMRs and the HCV Elimination Program Database, the sample sizes of PWID from the two sources were very small.
Although there were limitations, this study adds to the increasing body of research surrounding PWID estimates among indigenous communities in the United States and provides the first attempt at generating a PWID estimate in Cherokee Nation. There is a need for continued assessment of PWID in Cherokee Nation, including utilizing data sources outside of the healthcare system and reducing the length of time for data abstraction as the sample size of PWID recorded across CNHS databases increases, to inform accuracy of estimates. There is also a need to assess the impact of social determinants of equity on injection drug use and infectious disease among AI/AN people.