Study data
The study population was identified from five NHANES cohorts from 2007 to 2016. Conducted by the National Center for Health Statistics (NCHS), NHANES is a continuous cross-sectional survey with data released biannually and is effective in determining the prevalence of major diseases and associated risk factors among adults and children in the U.S. [25, 26]. The NHANES data is rich and unique in two ways. First, it combines information collected from both interviews and physical examinations that are necessary to answer the research questions. The interviews include demographics, socio-economic status, drug use information, and health-related questions; and the physical examinations include medical measurements and results of laboratory tests. Second, each survey cycle examines a nationally representative sample, and findings from the study are generalizable to the U.S. Further details are described elsewhere [26].
Study population and sampling procedure
The study population comprised people who reported methamphetamine use in their lifetime. We chose to use the lifetime methamphetamine use (DUQ330 - ever used methamphetamine) over the most recent methamphetamine use (DUQ350 - last time used methamphetamine) because the latter variable has had up to 95% of data missing. The flow chart in Figure 1 illustrates the process of case selection. The study included anyone who completed testing for any of the three sets of tests including HBV, HCV and HIV, and also answered “yes” to both questions “ever used cocaine/heroin/methamphetamine” and “ever used methamphetamine”. The study excluded anyone whose age was not between 20 and 59 years as they were not eligible to answer drug use questions.
Data sources
The primary outcome measure was positive/negative detection of any of the three BBVs (HBV, HCV, and HIV) which were determined according to the results of a set of serological tests. Three HBV serological markers were tested in the NHANES study: antibody to hepatitis B core antigen (anti-HBc), hepatitis B surface antigen (HBsAg), and antibody to hepatitis B surface antigen (HBsAb) [27]. The presence of HBsAg for at least six months indicate a chronic HBV infection [28]. Positive HBV detection was defined as a positive result of anti-HBc; while negative HBV detection was defined as negative for all HBV serological markers including anti-HBc, HBsAg and HBsAb. Indeterminate serological test results were coded as negative since we used a conservative definition to determine positive detection. The HBsAg is tested only when the anti-HBc test is positive. Participants who were HBsAb positive but anti-HBc negative and HBsAg negative were not considered as a population at risk of HBV infection, but susceptible to HIV and HCV.
Two HCV markers were tested: hepatitis C antibody and hepatitis C RNA [29]. The hepatitis C RNA test is only conducted when the hepatitis C antibody test is positive. Current HCV infection was indicated by both hepatitis C antibody and RNA being positive. Chronic HCV infection was defined as hepatitis C RNA positive six months after an acute infection. Positive HCV detection was defined as a positive result for both hepatitis C antibody and hepatitis C RNA, while negative HCV detection was defined as negative for the hepatitis C antibody. Those with a positive antibody HCV test, but a missing RNA test, were also considered negative. Similarly, indeterminate serological test results were coded as negative.
Two HIV serological markers were tested: HIV-1 and HIV-2 antibody [30]. Specimens were initially tested by a combo set of HIV-1/2 Enzyme Immunoassay (EIA), and then repeated reactive specimens are tested with HIV-1/2 supplemental assay. Positive HIV detection was defined as positive results from the two rounds of tests. If EIA is positive but following supplemental tests are not positive (e.g., negative, indeterminate), a confirmatory test is performed for a final decision: HIV detection is positive with a positive confirmatory test result, and HIV detection is negative with a negative confirmatory test result.
According to previous literature [4, 5, 12, 31, 32], demographic characteristics (age, gender, race/ethnicity), socio-economic status (poverty index, health insurance, healthcare access, education), sexual activity (number of sexual partners in the past year, sexual identity), and drug use behaviors (number of drugs used, IDU, number of times methamphetamine used in lifetime, age started using methamphetamine) were known factors associated with infection of BBV. Therefore, these variables were included as potential confounders in the analyses.
Demographics (age, gender, and race), health insurance, hospital utilization, and access to care information were collected through Sample Person Questionnaire. Socio-economic status (poverty index, education) was obtained through Family Questionnaire. Drug use information (e.g., number of drugs used, IDU, number of times methamphetamine used in lifetime, and age started using methamphetamine, etc.) was obtained through Audio Computer Assisted Personal Self Interview (ACASI) Questionnaire. Sexual behaviors (number of sexual partners, sexual identity) were collected through both ACASI and computer assisted personal interview (CAPI) questionnaires during participants’ visit to the examination center. All three BBV-related measures were obtained from corresponding laboratory tests. The specific laboratory methods can be found elsewhere [25]. Responses to questions including education, drug use, and sexual activity were limited to participants aged 20 to 59 years.
Statistical analysis
Descriptive analyses include both crude and weighted frequency and percentages of all covariates mentioned above. The appropriate sample weights for combined NHANES 2007-2016 data were constructed using National Center for Health Statistics guidelines [33]. Weighted frequencies and percentages were calculated by multiplying the sample weight WTMEC2YR by 0.2. WTMEC2YR is the full sample two year MEC exam weight, which indicates the weighted variable for laboratory measurements. We chose WTMEC2YR as the appropriate weight in our analysis as WTMEC2YR has the least common denominator. The Rao Scott Chi-squared statistic was calculated to assess the association between each covariate and outcome measure. Bivariate logistic regression and three multivariable logistic regression models were developed to examine the risk factors associated with BBV positive results among people who used methamphetamine. The outcome was tested positive for BBV or negative to BBV as identified from laboratory tests. The main risk factors of interest were drug use behaviors (number of drugs used, IDU, number of times methamphetamine used in lifetime, and age started using methamphetamine).
Model I, which only includes demographics, evaluated the effect of demographic characteristics on the BBV positive results. Model II further added a set of socio-economic characteristics and sexual behavior information into the modelling to evaluate their effect on the BBV positive results, controlling for demographics. Although health insurance, healthcare access, and number of sexual partners were not statistically significant in our model, they are, in general, confirmed risk factors for BBV infection according to previous literatures, and they were included in the model to adjust for their effects. Model III further explored how drug use affects the BBV positive results while taking into consideration all previous variables, which is our key research interest. The rationale to include them are two-fold: i), they are statistically significantly associated (p<0.05) with the BBV positive results in the unadjusted analyses; ii), they are suggested to have influence on the likelihood of a positive BBV test.
Unadjusted odds ratios (uORs) and their 95% confidence intervals (CIs) were reported from bivariate logistic regression models, and adjusted odds ratios (aORs) and their 95% CIs were reported from the three multivariable logistic regression models. Missing data were not included in the modelling. For all ORs reported, statistical significance was considered as CI not crossing 1 and corresponding p-value being less than 0.05. Chi-square goodness-of-fit test was used to assess the deviance between the statistical models.
R programming (RStudio, version 3.6) was used for all analyses. Library “tidyverse” was used to clean data and generate appropriate subsets for statistical analyses. Library “survey” and “srvyr” were used to analyze weighted NHANES data. Survey functions “svytotal”, “svymean”, “svychisq” and “svyCreateTableOne” were used to perform descriptive analyses; “svyglm” was used to perform logistic regression modeling, and “jtools” was used to draw figure 2.