Estimates of state-level chronic hepatitis C virus infection, stratified by race and sex, United States, 2010

We extended a previously-described small-area estimation approach that synthesizes NHANES national estimates of HCV infection with state-level data on HCV-related mortality from the National Vital Statistics System [2]. The data sources and analytic methods are briefly described below.

NHANES uses a complex, multistage sampling design to collect nationally representative questionnaire and laboratory data on the health of the non-institutionalized United States population [17]. Data and corresponding sampling weights are released every 2 years. Data from seven NHANES release cycles (1999–2012) were pooled in order to ensure sufficient data for all demographic groups. All analyses were restricted to respondents aged ≥18 years of age at time of survey. Race/ethnicity was categorized into non-Hispanic black, non-Hispanic white and Hispanic/other. Similar to previous analyses, birth year was classified into three cohorts: before 1945, 1945–1965 and after 1965 [18, 19].

NHANES tested lab samples for anti-HCV using an anti-HCV screening chemiluminescence immunoassay and a confirmatory recombinant immunoblot assay (RIBA) [20]. Samples with positive results on RIBA were tested for HCV RNA using an in vitro nucleic acid amplification test. Participants with a positive or indeterminate anti-HCV test and positive HCV RNA test were considered to represent chronic HCV infection. As done in previous analyses, participants who tested positive for anti-HCV but negative for HCV RNA (resolved infections; n = 112) and those who tested positive for anti-HCV but did not have a HCV RNA test (n = 160) were not included in this analysis [14].

The National Vital Statistics System (NVSS) collects demographic, geographic and cause of death data from all death certificates in the United States [21]. Mortality Multiple Cause Microdata files (from 1999 to 2012) were used and data were categorized into the same sex, race and birth cohort categories described above. Any death records that listed the ICD-10 code for acute viral hepatitis C (B17.1) or chronic viral hepatitis C (B18.2) as the underlying or a multiple cause of death were considered to indicate HCV-related mortality.

Annual intercensal population estimates (1999–2012) from the US Vintage 2000, 2009 and 2014 data sets were used as denominators for HCV-related mortality rates within each demographic strata [22]. Microdata from the 2010 5-year American Community Survey (ACS, years 2006–2010) were used to generate estimates of the non-institutionalized population [23, 24], which were combined with our estimated number of HCV cases to estimate HCV prevalence rates for each state. The incorporation of ACS population totals into this analysis is an update to our previously published approach and reflect current NHANES guidance [25]. Estimated population total for each sex and race group are reported in Additional file 1: Table S1.

The number of persons with chronic hepatitis C in each state were estimated using a standardization-based estimator described in detail earlier [2]. Briefly, we used NHANES data to calculate weighted estimates for national HCV RNA prevalence for 18 strata of sex (2), race/ethnicity (3) and birth cohort (3) [25]. We multiplied these weighted estimates by 2010 ACS 5-year population estimates for the corresponding demographic stratum for each state to generate crude state-by-stratum totals. We used NVSS mortality data and intercensal population totals to fit a high-order logistic regression model the average HCV-related death rates in the corresponding 12 strata over the 14-year period. We compared observed state-by-strata HCV-related mortality totals to model predictions to assess collinearity and model fit [26, 27]. A ratio of state-by-demographic stratum effects were calculated by comparing state specific HCV-related death rates to national HCV-related death rates (within the same strata). The crude state-by-stratum HCV RNA totals were adjusted by the state-by-stratum effects to calculate mortality-adjusted HCV RNA prevalence totals in each strata in each state. Totals were summed within single strata of race (white non-Hispanic, black non-Hispanic and Hispanic/other race) or sex and divided by corresponding population totals. This yielded HCV RNA prevalence rate estimates by sex and race for each state. Prevalence rate ratios were generated to compare HCV RNA prevalence for men vs women and non-Hispanic black vs non-Hispanic white. An analogous approach and model was used to estimate the prevalence rate and number of persons with HCV antibodies (anti-HCV), which is often used as an indicator of past or current infection [14]. The results for anti-HCV prevalence by race and sex in each state are presented in Additional file 1: Tables S2-S5.

We calculated 95% confidence intervals (CI) for state-level estimates to account for the joint statistical uncertainty in the NHANES prevalence estimates and HCV-related mortality rate estimates from the logistic regression model. Confidence intervals were calculated using a Monte Carlo simulation that sampled from logit-normal and normal distributions, respectively (k = 10,000). We calculated the coefficient of variation for all prevalence rate and ratio estimates. Similar to the guidelines used by NVSS, any estimates with a coefficient of variation of 23% or greater are potentially unreliable and marked accordingly [28].

This approach uses data from several large, public and population-based data sources. However, there are some limitations to consider when interpreting these results. First, these estimates only represent the non-institutionalized population. Data from NHANES does not represent homeless persons or persons in correctional facilities, nursing homes or other institutions in which chronic HCV prevalence may differ from the general population [47]. Geographic differences in incarceration rates by race and sex could influence statewide HCV prevalence. However, previous work on a national level demonstrates a promising approach for including underrepresented populations that can be extended to the state level in the future [47]. Second, we aggregated 14 years of NHANES data in order to have a sufficient sample size to produce reliable estimates. Slight changes in local HCV incidence over that time period might not be captured in this approach. In the NHANES data, the national overall prevalence of HCV RNA changes slightly from 1999 to 2006 (1.26%; 95% CI: 1.06–1.48%) to 2007–2012 (1.05%; 95%: 0.82–1.34%). However, there is not enough data to reliably determine if there are temporal changes in prevalence within subgroups. Interpreting these results in conjunction with further analysis of local risk behaviors will help state health departments better understand their epidemic. Additionally, we do not present separate results for American Indians/Alaska Natives (AI/AN), who face an elevated HCV burden and are a key population identified in the National Viral Hepatitis Action Plan [12]. Due to under sampling and sparse NHANES data within this group, we are not able to separate AI/AN into an additional race category. We combined AI/AN and other under sampled race groups with Hispanic in order to have sufficient data in each strata of NHANES data. However, 76% of the estimated HCV RNA infections in this combined group were among persons who identified as Hispanic. Similarly, although we control for birth cohort in our model, we do not present results stratified by age. It is challenging to present meaningful estimates for age because we pooled data across 14 years. Finally, despite pooling data, some individual point estimates might still be unreliable. However, those values have been indicated as potentially unreliable, and confidence intervals have been provided for all estimates. Improved direct surveillance data or additional model input data are needed to overcome the demographic limitations of using NHANES data.