Prevalence of Hepatitis C Virus Infection in US States and the District of Columbia, 2013 to 2016

Key Points Question During 2013 to 2016, what proportion of adults were living with hepatitis C virus (HCV) infection in each US state? Findings In this survey study, US national HCV prevalence during 2013 to 2016 was 0.93% and varied by jurisdiction between 0.45% and 2.34%. Three of the 10 states with the highest prevalence and 5 of the 9 states with the highest number of HCV infections were in the Appalachian region. Meaning Regions with long-standing HCV epidemics, and those with newly emergent ones partly driven by the opioid crisis, face substantial HCV prevalence.

We explored the potential for some drug deaths coded as suicides to be accidental overdoses. The proportion of narcotic and unknown drug deaths coded as suicides varies by state (eTable 2). Since these vary only modestly, we did not include suicides in the primary analysis. Additionally, drug intoxication does not result in a majority of suicide deaths, relative to other (more violent) methods. 10 It is actually possible that our inclusion of deaths of undetermined intent includes misclassified suicides that did not have enough evidence to be reported as suicides. 10,11 Level 2: Overdose deaths by drug class by state Level 2 depicts a bit more detail that is available in NVSS mortality data with regards to drug overdose deaths. Within each category of intentionality, ICD-10 codes are separated by drug class. These classes include: poisoning by and exposure to non-opioid analgesics, antipyretics, and antirheumatics; poisoning by and exposure to antiepileptic, sedative-hypnotic, antiparkinsonism and psychotropic drugs, not elsewhere classified; poisoning by and exposure to narcotics and psychodysleptics (hallucinogens), not elsewhere classified; poisoning by and exposure to other drugs acting on the autonomic nervous system; and poisoning by and exposure to other and unspecified drugs, medicaments, and biological substances. For this analysis, our definition was restricted to deaths due to poisoning by narcotics and psychodysleptics (X42, Y12) and exposure to other and unspecified drugs (X44, Y14).
Drug overdoses due to narcotics were included because this drug class includes cannabis, cocaine, codeine, heroin, methadone, morphine, and opium. While not all of these drugs are typically administered via injection, this class provides a more specific definition than merely using all drugs.
Death investigation and drug toxicology processes vary by state. [12][13][14] In order to account for this variation, and to provide a more sensitive definition, we included overdoses due to other and unspecified drugs. 8,15 Level 3: Overdose deaths by specific drugs by state The third level describes specific drugs that are more likely to be used via injection than other drugs. NVSS mortality data includes specific drug toxicology codes (T codes) for heroin, natural and semisynthetic opioids (morphine, codeine, hydrocodone, and oxycodone), methadone, synthetic opioids excluding methadone (fentanyl, fentanyl analogs, and tramadol), cocaine, and psychostimulants with abuse potential (methamphetamine, amphetamine, Ritalin, caffeine, and ecstasy). While inclusion of T codes for injection-related drugs such as heroin and synthetic opioids excluding methadone would be an improved signal of injection-related overdose, the toxicology completion regarding specific drug codes on death certificates varies greatly by state and by year. [12][13][14]16,17 A second issue with using specific drug codes for this analysis is that these are not mutually exclusive. Many drugs, particularly heroin and fentanyl, are found together in toxicology and subsequent death certificates. 18 This becomes an issue since some drugs (i.e., fentanyl) have higher fatality rates from injection than others, and some drugs are more frequently used via injection than other routes of administration. 19,20 This is particularly important for assessing the geographic distribution of overdose deaths since the distribution of fentanyl varies, in part due to the relative ease of incorporating fentanyl into the white powder heroin supply east of the Mississippi River, compared to black tar heroin. 21 These have not been incorporated into the present analysis due to the variation of completion by state, 22 but this is a critical issue that should be explored in future research, in order to reduce biases introduced by non-specificity for injected drugs and by spatial heterogeneity in highly-lethal substances such as fentanyl.

Level 4: Overdose deaths by specific drugs and injection status by state
Finally, for the fourth level, the ideal measurement of injection-related overdose is overdose by drug by injection. The most relevant literature-based estimate of opioid overdose deaths attributable to injection is not generalizable to all states. 23 This is the ideal measure to use for those born after 1970 for the HCV work. However, due to the lack of literature as well as the above-mentioned limitations in the specific drugs, we cannot achieve this level of detail.

eAppendix 4: Equation for Estimator of the Total Persons With HCV Infection in Each US State
The above equation details our estimator for the total persons with HCV in each state ( ) in the NHANES population. Within 12 strata representing above-defined levels of sex, race/ethnicity, and birth year, we computed the standardization estimate ̂ , where is the 2016 ACS population in the state's stratum ("state-stratum") and ̂ the national 2013-2016 HCV RNA prevalence in stratum by direct estimation from a weighted logistic regression model of NHANES, which included terms for these strata, era (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016), and poverty. To yield standardized estimates for the 12 demographic strata that accounted for poverty, we weighted logistic model estimates according to the ACS poverty distribution for the 12 strata in each state.
Next, we estimated the state-stratum-specific likelihood of HCV-related mortality ( ), using a logistic model of NVSS-derived mortality counts, per person-years ( * ), that approximated full-stratification with main effects for state, sex, race/ethnicity, birth cohort, era; two-way interactions for state by each sex/ethnicity, race, birth cohort, and era; two-way and three-way interactions for each combination of sex, race/ethnicity, birth cohort, and era; and four-way interaction of sex, race/ethnicity, birth cohort, and era. These were divided by the national stratum-specific average, yielding a mortality ratio for the state-stratum. This process was repeated for the narcotic overdose mortality ( ). The two mortality ratios per stratum were averaged according to weights (values described in the main manuscript text and eAppendix 5) and then multiplied by the standardization-based value to yield adjusted totals . Summing these across all 12 state-strata yielded , which when divided by the ACS state population yielded the estimated prevalence rate.

eAppendix 5: Description of analytic weight derivation
To represent the spatial distribution of both older prevalent HCV infections (those existing during 1999-2012) and newer HCV infections (during 2013-2016) resulting from injection drug use, we separately modeled mortality rates from HCV infection and narcotic overdose. Mortality rates were used to calculate state-level mortality ratios for both HCV infection and narcotic overdose. Within each age group (defined by birth year), we calculated a single weighted state-level mortality ratio.
For the first weighting scenario, we only used state-level mortality ratios from the HCV death model in order to compare to our previous method (eTable 3, results depicted in Figure 2).
For the second scenario, we used available data and expert knowledge of HCV epidemiology to derive the weights. First, we assumed all HCV infections among persons born <1945 are a result of older exposure and used a weight of 1.0 for that age group (w 1 ). For the other two age groups, we used trends of HCV antibody prevalence in NHANES data to make inferences and assumptions about HCV incidence from 2013-2016. For persons born ≥1970, we estimated there were 411,449 persons with a history of infection (HCV antibody) prior to 2013 and 1,253,938 after 2013 (eTable 4). We assumed ~0% mortality for this age group during this time frame, suggesting 37.8% of persons with HCV exposure in this age group acquired HCV prior to 2013 and 62.2% thereafter. This is similar to other observations around 70%. 24 Therefore, we used a weight of 0.378 for the HCV death state effect ratio for persons born ≥1970 (w 3 ). Persons with HCV antibody, but who are not currently infected, and those who are currently infected but are not diagnosed (and presumably relatively asymptomatic), may experience death rates higher than the background mortality rates, although such data are unavailable. Recognizing that this age group (persons born 1945-1969) has the highest HCV burden and may disproportionally impact prevalence results, we conducted a sensitivity analysis examining a third scenario that considered an overall mortality rate of 20% for person with HCV antibody (w 3 = 0.80) (eTable 5).

eAppendix 6: Further descriptions of analyses for additional populations not in NHANES sampling frame
This analysis used the results of a literature review from Hofmeister et al., 26 which used articles that reported HCV prevalence from 1/1/2013-12/31/2017. Search terms used for the incarcerated population were ("hepatitis C" or "HCV") and ("prison" or "jail" or "correctional") and for the homeless population were ("hepatitis C" or "HCV") and ("homeless" or "homeless persons" or "housing unstable" or "housing insecure"). Details on motivation for additional populations and prevalence and population size sources used in Hofmeister et al. 26 are described in eTable 6.

Alternative Approach to Additional Population Estimates
The alternative approach to estimating state-level HCV RNA prevalence among additional populations involved two steps. First, we generated a national prevalence ratio for each population component (incarcerated, unsheltered homeless, and nursing home residents) by taking the national HCV prevalence in the population component divided by the national HCV prevalence in NHANES. Then, we multiplied this national prevalence ratio by the each state's HCV prevalence in the NHANES population and each state's population size of each population component. This provided an estimate of HCV infections among additional populations that reflects each state's underlying HCV prevalence rather than the national HCV estimate. This assumes that the state epidemics are echoed in these additional populations. Full results including both the primary approach for the additional population estimate and the alternative are shown in eTable 7. There was a median difference in prevalence between methods 1 and 2 of 0.004% (relative multiplicative change of -0.5%).

eFigure 1: Conceptual overview of method for estimating Hepatitis C virus (HCV) RNA prevalence in US states
We used a multistep, statistical approach that first generated estimates for each state using National Health and Nutrition Examination Survey (NHANES) national prevalence in sex, race, birth cohort, and poverty strata (1A). To represent the spatial distribution of older and recent infections respectively, we separately modeled mortality rates from HCV infection and narcotic overdose in the National Vital Statistics System (NVSS), yielding stratified state-level mortality ratios (1B). We weighted these ratios according to birth cohort-specific trends in HCV exposure history (1C) and used them to adjust initial NHANES-based estimates (1D). Finally, we estimated additional infections among populations not included in NHANES' sampling frame, by applying literaturebased estimates of prevalence in these groups to state-specific population estimates (1E).  31 Cocoros et al., 32 de la Flor et al., 33 Kuncio et al., 34 Mahowald et al., 35 Schoenbachler et al., 36