Hepatitis C transmission in young people who inject drugs: Insights using a dynamic model informed by state public health surveillance.

Increasing injection of heroin and prescription opioids have led to increases in the incidence of hepatitis C virus (HCV) infections in US young adults since the early 2000s. How best to interrupt transmission and decrease HCV prevalence in young people who inject drugs (PWID) is uncertain. We developed an age-stratified ordinary differential equation HCV transmission model of PWID aged 15-64, which we fit to Michigan HCV surveillance data among young PWID aged 15-29. We used Latin hypercube sampling to fit to data under 10,000 plausible model parameterizations. We used the best-fitting 10% of simulations to predict the potential impact of primary (reducing injection initiation), secondary (increasing cessation, reducing injection partners, or reducing injection drug use relapse), and tertiary (HCV treatment) interventions (over the period 2017-2030) on acute and chronic HCV cases by the year 2030. Treating 3 per 100 current and former PWID per year could reduce chronic HCV by 27.3% (range: 18.7-30.3%) and acute HCV by 23.6% (range: 6.7-29.5%) by 2030 among PWID aged 15-29 if 90% are cured (i.e. achieved sustained virologic response [SVR] to treatment). Reducing the number of syringe sharing partners per year by 10% was predicted to reduce chronic HCV by 15.7% (range: 9.4-23.8%) and acute cases by 21.4% (range: 14.2-32.3%) among PWID aged 15-29 by 2030. In simulations of combinations of interventions, reducing injection initiation, syringe sharing, and relapse rates each by 10% while increasing cessation rates by 10% predicted a 27.7% (range: 18.0-39.7%) reduction in chronic HCV and a 38.4% (range: 28.3-53.3%) reduction in acute HCV. Our results highlight the need for HCV treatment among both current and former PWID and the scale up of both primary and secondary interventions to concurrently reduce HCV prevalence and incidence in Michigan.

We developed an age-stratified ordinary differential equation HCV transmission model of PWID aged 15-64, which we fit to Michigan HCV surveillance data among young PWID aged 15-29. We used Latin hypercube sampling to fit to data under 10,000 plausible model parameterizations. We used the best-fitting 10% of simulations to predict the potential impact of primary (reducing injection initiation), secondary (increasing cessation, reducing injection partners, or reducing injection drug use relapse), and tertiary (HCV treatment) interventions (over the period 2017-2030) on acute and chronic HCV cases by the year 2030.
Our results highlight the need for HCV treatment among both current and former PWID and the scale up of both primary and secondary interventions to concurrently reduce HCV prevalence and incidence in Michigan.

Introduction
The epidemiology of hepatitis C virus (HCV) infections in the United States (US) has changed dramatically over the last decade, with notable increases in HCV incidence among young people aged approximately 15-29 years. [1][2][3][4][5] These changes in incidence have been associated with increases in opioid and injection drug use (IDU). [1][2][3][4][5] In the US, up to 2.6% of adults have injected drugs in their lifetime and more than half of US people who inject drugs (PWID) have HCV infection. [6][7][8][9][10][11] IDU is the primary risk factor for new HCV infections in the US. 5 After decades of asymptomatic chronic infection, HCV leads to liver-associated morbidity and mortality (e.g. cirrhosis and hepatocellular carcinoma). 12,13 In the US, transmission modeling studies have shaped HCV screening, treatment, and prevention policies by increasing our understanding of HCV transmission dynamics, forecasting prevalence of HCV-related liver diseases, and simulating the impact, costs, and benefits of highly effective direct-acting antivirals (DAAs) among PWID and other groups disparately burdened by HCV. 12,[14][15][16][17][18][19] Multiple studies support the cost-effectiveness of treating PWID with DAAs to interrupt HCV transmission, a strategy known as 'treatment as prevention.' 15,16,18, HCV treatment may be particularly impactful in reducing HCV prevalence among young PWID. Echevarria et al. suggested that treating just 5 per 1,000 young PWID (<30 years) in Chicago could halve HCV prevalence in this age group over 10 years (from 10% to 5%). 15 This large predicted reduction from a modest intervention stemmed in part from the low baseline HCV prevalence among young PWID in Chicago compared with their older counterparts. 15,44,45 HCV seroprevalence studies in several US cities (Baltimore, Chicago, Los Angeles, New York City, San Diego, and Seattle) suggest that HCV prevalence among young US PWID varies widely across the US, with estimates ranging 10-53%; however, estimates are unavailable for a majority of states and cities. 15,[44][45][46][47][48][49] This absence of local HCV seroprevalence estimates limits our ability to evaluate prevalence trends and the potential impact of interventions in modeling studies. We demonstrate here that HCV public health surveillance data collected as part of nationally notifiable and state-reportable condition surveillance might be used to evaluate the potential impact of interventions among young PWID in locations without a systematic characterization of HCV prevalence. 50 HCV incidence increases in young adults were first identified using HCV public health surveillance data. 2,51,52 As part of HCV surveillance, laboratories and physicians report positive HCV lab results to state health departments, who apply standard case definitions to stage HCV as acute or chronic. [53][54][55][56] Underreporting of HCV cases limits use of surveillance data for purposes other than description and outbreak monitoring. 2 57 In addition to under-reporting, there is high variability in capacity to collect risk factor or demographic information, trace contacts, and connect people with HCV infection to further testing and treatment. 2,5 These limitations have discouraged use of public health surveillance data for HCV transmission modeling.
We developed an HCV transmission model fit to HCV surveillance data among [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29] year olds in Michigan during 2000-2016 that adjusts for case under-reporting. We reviewed the literature to identify ranges for model parameters and simulated the model across 10,000 plausible scenarios using Latin hypercube sampling, a form of stratified random sampling. 50,58 We then evaluated the potential impact of several interventions implemented during 2017-2030, including primary prevention (reduced injection initiation), secondary prevention (behavioral initiatives), and tertiary initiatives (HCV treatment) in a counterfactual framework by summarizing the predicted reduction in HCV prevalence (chronic cases) and incidence (acute cases) in the year 2030. This modeling framework could be applied to HCV surveillance data from other states and/or adapted for use in other nationally or state-notifiable conditions.

Methods
A detailed discussion of the model parameters, initial conditions, surveillance data, and parameter estimation process is available in the supplemental material. Matlab (The MathWorks Inc, Natick, MA) code for model simulation and R (R Foundation for Statistical Computing, Vienna, Austria) code for figures is freely available at https://github.com/ epimath/Hepatitis-C-in-Young-PWID.

Model Structure
An HCV ordinary differential equation (ODE) transmission model of PWID with preferential age mixing was developed and implemented in Matlab R2017b. The model consists of 11 states per age group that reflect the natural history and transmission of HCV ( Figure 1). The model is simulated among four age groups: 15-19 years, 20-25 years, 26-29 years, and 30-64 years. The 11 model compartments per age class include uninfected, non-PWID with substance abuse or dependence (Z i ), uninfected current or former PWID (S i or S Ni , respectively), acutely infected current or former PWID (A i or A Ni , respectively), chronically infected current or former PWID (C i or C Ni , respectively), immune current or former PWID (I i or I Ni , respectively), and treated current or former PWID (T i or T Ni , respectively). Current PWID are those who currently inject drugs whereas former PWID are those who injected drugs in the past but do not currently inject. We multiplied US national prevalence estimates by the Michigan population size for each age group (P i ) as initial conditions for the number of people with substance use disorders, current PWID, and former PWID (Supplemental Table 1). 1,6,59 Movement through the 11 compartments is governed by a set of 44 differential equations shown in (1), where subscript i denotes the age class (1: 15-19 years, 2: 20-25 years, 3: 26-29 years, 4: 30-64 years) of individuals moving through compartments and subscript j denotes the age class of effective contacts. In the youngest age class, non-PWID without HCV infection are added each year to the Z 1 compartment. The number of non-PWID introduced per year is based on the prevalence of substance abuse or dependence among [15][16][17][18][19] year olds from the National Survey on Drug Use and Health (NSDUH, ψ 0 ) and the average population size of 15 year-olds during 2000-2016 in Michigan (ν 0 P 0 , see Supplemental Table 1 for a complete description of parameters). [59][60][61] Individuals exit each compartment through aging (individuals age to the subsequent age class at a rate ν i ) or death. To account for the elevated mortality among people who use substances relative to the general population, we multiplied age-specific mortality rates in Michigan (μ i ) by standardized mortality ratios for people who use substances, current PWID, and former PWID (η Z , η P , and η N , respectively). 60,[62][63][64][65][66][67] Non-PWID (Z i ) begin injecting drugs (transition to S i ) at an estimated injection initiation rate (θ i ) calibrated to fit acute case data (described further below and in the Supplemental Methods). Susceptible PWID (S i ) acquire new infections through effective contact with an acutely (A j ) or chronically (C j ) infected individual in any age class. A proportion (τ) of treated current PWID (T j ) also transmit HCV during intervention simulations (described further below). We model the number of new infections per year as the product of the probability of infection per contact (β), the number of susceptible current PWID (S i ), the total number of syringe sharing partners per susceptible per year (σ i ), and the proportion of contacts that are infected. We incorporate age-assortative syringe sharing by adjusting the proportion of infected individuals per age class by a scaling factor that reflects ageassortative syringe sharing (π j,i ) using a study of PWID in Baltimore (Supplemental Methods). 68 Other HCV transmission modes (e.g. perinatal acquisition, unregulated tattoos, sexual transmission) are not considered as IDU is the primary risk factor for new HCV infections in the US. 5 Newly infected PWID (A i ) have acute HCV infection for a duration of 6 months (ε −1 ), at which point 50-85% (1-δ) develop chronic infection (C i ). [69][70][71][72] We assume that the acute phase lasts exactly 6 months to be consistent with national case definitions for acute HCV used to fit our model, though some heterogeneity in HCV clearance time exists. 71-73 A fraction (ξ, 0-45%) of acutely infected individuals who spontaneously clear their HCV infection have sterilizing immunity and move to the immune class (I i ) while the remaining individuals (1-ξ) move back to the susceptible class (S i ) where they can be re-infected at the same rate as infection-naïve individuals. 74 During treatment intervention simulations described below, chronically infected individuals can be treated (T i and T Ni ) at rates φ P and φ N for a duration of ω −1 . Treated current PWID become susceptible to HCV re-infection after treatment (i.e. return to the susceptible class, S i ). We assumed that there was no HCV treatment during the period 2000-2016.
Individuals from any of the current PWID classes stop injecting drugs and move to their adjacent former PWID class at a cessation rate γ i . Former PWID can begin injecting again after a period of injection abstinence and enter the current PWID class at a relapse rate κ i .
To maintain a realistic ratio of current to former PWID during model fitting to data, we calculated age-specific relapse rates for each simulation using the sampled cessation rates and prevalence of current and former PWID in the US based on national survey data (Supplemental Methods). 1,6,59

Surveillance Data, Parameter Estimation, and Parameter Sampling
The Michigan Department of Health and Human Services (MDHHS) receives reports of HCV diagnoses from healthcare providers and laboratories and stages cases as acute or chronic using standardized national case definitions. 53,54 We obtained the number of newly identified acute and chronic HCV cases per year during 2000-2016 and made two adjustments to the acute case series to facilitate model fitting to data (Supplemental Table 2 To incorporate parameter uncertainty, we drew a stratified random sample of 10,000 parameter sets across plausible ranges using Latin hypercube sampling (Supplemental Table  1). All parameters were sampled from uniform distributions created from the minimum and maximum bounds we identified through literature review. We used an initial ( year olds was sampled (rather than estimated) as we had no data to fit to for this age group and because the focus of our analysis was on young PWID. Demographic parameters (i.e. population sizes and mortality rates, and age group sizes) were not sampled. The duration of acute HCV infection was also not sampled to be consistent with the HCV data we used in model fitting. 53,54 Residual sum of squares (RSS) values provided a summary of model fit to acute HCV data after parameter estimation. To determine if a certain range appeared more consistent with data, we plotted histograms by quartile of RSS (Supplemental Figure 1). Parameter estimation and simulations were run using fminsearchbnd and the ODE15S solver in Matlab (due to stiffness in some simulation runs). 76 Details for the selection of initial conditions and bounds for estimated parameters used in fminsearchbnd and a discussion of our reasons for expanding injection initiation rate bounds from values found in the literature are discussed in Supplemental Methods. 77-87

Intervention Simulations
We selected the best-fitting 10% of parameter sets (i.e. the 1,000 simulations with the lowest RSS values) to simulate the potential impact of interventions on acute and chronic HCV in Michigan during the period 2017-2030. We simulated interventions by scaling one or more parameters in each parameter set during the period 2017-2030. We summarized the expected percent reduction in acute or chronic HCV cases in the presence of interventions compared to predicted cases with no intervention at the end of the simulation period (year 2030).
Supplemental Table 3 outlines the three types of interventions we conducted, which include single interventions wherein just one parameter was altered, combinations of interventions that altered multiple parameters, and sensitivity analyses that further examined HCV treatment interventions. We simulated several interventions, including reduced injection initiation (θ i ), decreased syringe sharing (σ i ), decreased IDU relapse (κ i ), increased IDU cessation (γ i ), and treating former (φ N ) or current PWID (φ P ) for HCV infection. Injection initiation, syringe sharing, and IDU relapse reduction interventions involved reducing each parameter by 10%, 20%, and 40% of their 2000-2016 value whereas IDU cessation interventions involved increasing 2000-2016 cessation rates by 10%, 20%, or 40%. To facilitate comparison with prior work, we simulated treatment rates of 3, 6, and 24 per 100 PWID per year. 88 To facilitate comparison, we also classify interventions by their most immediate roles as primary (reduced injection initiation), secondary (syringe sharing, IDU cessation, and relapse), and tertiary prevention interventions (HCV treatment). 89 We conducted four intervention sensitivity analyses further characterize HCV treatment interventions and the duration of IDU. First, we examined the impact of sustained virologic response (SVR), a measure of the percentage cured by HCV treatment. 90 We assumed that approximately 90% of individuals achieved SVR in main intervention simulations (parametrized by α) in alignment with a recent meta-analysis of DAAs, and examined 60-100% cure in sensitivity analyses to summarize the potential impact of treatment incompletion among PWID. [90][91][92][93][94][95][96] Second, we examined expected case counts when a proportion of current PWID receiving HCV treatment also transmitted HCV (parametrized by τ). We assumed that 50% of treated current PWID transmitted HCV in the main results, and compared values of 0-100% in sensitivity analyses. Third, we simulated a treatment duration of 12 weeks for the main analysis, and examined 8 and 16 week durations in sensitivity analyses. All are typical treatment durations for currently approved DAAs. 13 Finally, we compared the predicted reductions in HCV cases for parameter sets with a long versus short duration of injecting drugs during the pre-intervention period (i.e. 2000-2016).
Long injecting durations were defined as parameter sets with >median cessation rate and ≤median relapse rate whereas short durations had ≤median cessation and >median relapse rates. Long injecting durations included scenarios with permanent cessation wherein relapse rates were so low that there was no possibility of relapse given the duration of our simulation period.

Model Fit to Acute HCV Surveillance Data
The model fit acute HCV surveillance data well, especially for the 1,000 best-fitting parameter sets used to simulate interventions ( Figure 2). Acute cases were expected to increase through 2030 for the best-fitting scenarios. Similar to prior work, chronic HCV prevalence generally declined early in the simulation period due to increases in PWID prevalence and high mortality among current PWID, and later increased under some conditions (Supplemental Figure 2). 97 There was a wide range in predicted prevalence due to the variability of conditions assessed. A variety of scenarios across sampled parameter ranges fit the data well (Supplemental Figure 1). Better-fitting parameter sets had a slight tendency towards lower total contacts among 15-19 year olds and more total contacts among 20-25 and 30-64 year olds (Supplemental Figure 1).

ntervention Simulations
We simulated the potential impact of primary (reducing injection initiation), secondary (increasing cessation, reducing injection partners, or reducing injection drug use relapse), and tertiary (HCV treatment) interventions on acute and chronic HCV cases among young (aged 15-29 years) PWID during 2017-2030. On average, decreasing the number of syringe sharing partners was associated with the highest predicted reduction to acute and chronic HCV among the primary and secondary interventions we simulated. Reducing the number of syringe sharing partners per year by 10% predicted a 15.7% (range: 9.4-23.8%) reduction in chronic cases and a 21.4% (range: 14.2-32.3%) reduction in acute cases among PWID aged 15-29 relative to their expected values in 2030 without intervention ( Figure 3).
In general, the predicted case reductions from treatment were highly dependent on parameter values, and therefore the predicted case reductions associated with treatment spanned a large range, especially among former PWID and for the expected reduction in acute cases. Treating current and former PWID resulted in similar median predicted reductions in chronic HCV cases among 15-29 year olds by 2030, with more variability among former than current PWID treatment ( Figure 3). These predicted reductions were more variable among PWID aged 30-64 (Supplemental Figure 3).
When simulated in combination, adding secondary interventions enhanced the predicted reductions from primary prevention initiatives (reduced injection initiation) and HCV treatment ( Figure 4). Treating former and current PWID together reduced some of the uncertainty in treatment effect compared to treatment of former or current PWID alone. Treating 3 per 100 current PWID per year and 3 per 100 former PWID per year predicted a reduction in chronic HCV cases by 27.3% (range: 18.7-30.3%) and acute cases by 23.6% (range: 6.7-29.5%) among PWID aged 15-29 when we assumed that 90% were cured (i.e. achieved SVR). Reducing injection initiation, syringe sharing, and relapse rates each by 10% while increasing cessation rates by 10% predicted a 27.7% (range: 18.0-39.7%) reduction in chronic HCV and a 38.4% (range: 28.3-53.3%) reduction in acute HCV by 2030.

Sensitivity Analyses
We assessed the impact of three aspects of treatment on reduction to acute and chronic HCV among 15-29 year olds. Reducing the proportion of treated current PWID who achieved cure (i.e. sustained virologic response) modestly attenuated the percent reduction in HCV prevalence and incidence (Supplemental Figure 4). The magnitude of these effects was similar for former PWID (data not shown). Treatment duration and the proportion of treated current PWID who shared syringes (and contributed to transmission during treatment) did not impact predicted treatment results (Supplemental Figure 4).
We also examined how the duration of injecting may have impacted the predicted reductions in PWID aged 30-64 (Supplemental Figures 5&6). As expected, treating former PWID for HCV led to greater predicted reductions in chronic HCV than treating current PWID when injection duration was short (i.e. high cessation rates and low injection relapse). Conversely, the predicted reductions from treating current PWID approached (and sometimes exceeded) the predicted reductions from treating former PWID when the duration of injecting was long.

Conclusions
We developed and implemented an HCV transmission model among PWID informed by Michigan HCV surveillance data, and leveraged our model to evaluate the potential benefits of primary, secondary, and tertiary interventions for reducing HCV prevalence and incidence. The incorporation of state-level surveillance data allowed us to evaluate interventions in a framework consistent with the HCV incidence trends in Michigan. Simulation results suggested that HCV treatment could be a highly effective strategy to reduce HCV prevalence among young PWID, especially when both former and current PWID receive treatment. Especially in combination with primary and secondary prevention measures, such as decreasing injection initiation, decreasing relapse, increasing cessation, and decreasing syringe sharing, treatment could substantially reduce chronic HCV infections by 2030. In line with the concept of 'treatment as prevention,' treatment also yielded reductions to acute HCV, though the impact was highly uncertain. We found no differences in the predicted impact of treatment when PWID continued to share syringes and contribute to transmission during treatment, supporting current recommendations that all PWID be provided treatment, regardless of their IDU behaviors. 13 Our findings suggest that treatment rates >0.24/year may be required to eliminate HCV among young PWID in Michigan by 2030. These treatment rates were higher than prior research by Zelenev et al., who found that HCV could be eliminated in Hartford, CT by treating >12 per 100 PWID per year for 10 years when initial HCV prevalence was <60%, and with Fraser et al., who found that treating 15.9 per 100 PWID would be required to eliminate rural Indiana by 2030 in the absence of other interventions. 88,97 There are several notable caveats to our findings about HCV treatment. First, the predicted impact of treatment was highly uncertain across scenarios tested, all of which represent reasonable approximations to the current HCV epidemic and PWID prevalence, given their parametrization using findings from empirical studies. In addition, compartmental HCV models that do not account for the dynamic network structure of syringe sharing are known to overestimate treatment effects, including their treatment as prevention potential. 98 Incorporating network structure was beyond the scope of this model; however, we did incorporate some heterogeneity in IDU contact patterns based on age to examine HCV treatment's potential effectiveness under several potential scenarios. Nonetheless, the predicted impacts of treatment should be taken as an upper bound of the potential treatment effect and the integration of a network model within our current framework is an area for future model development. Finally, treatment results were somewhat sensitive to cure rates. Taken together, these results emphasize the importance of engaging PWID in treatment until cure, and highlight the need to incorporate behavioral interventions that complement treatment, given the uncertainty about its population-level effects.
Secondary prevention interventions that reduce syringe sharing, promote injection cessation, and prevent relapse, and primary prevention interventions that reduce injection initiation consistently predicted reductions in acute and chronic HCV. These results were robust across tested scenarios (i.e. parameter sets) and were associated with less uncertainty than treatment or relapse prevention interventions. Thus, these interventions should be implemented alongside HCV treatment and relapse prevention programs (e.g. addiction treatment).
Though we cannot implement them directly, our simulated primary and secondary interventions are meant to represent real-world initiatives known to decrease HCV transmission. For example, expanded access to addiction treatment services and medicationassisted treatments promote cessation of drug use and prevent relapse. 99,100 Several realworld harm reduction interventions, including the "Staying Safe," "Break the Cycle," or "Change the Cycle" interventions, help prevent people who use drugs from initiating IDU while concurrently promoting safe drug injection. 101,102 Simulated reductions in the number of syringe sharing partners emulate syringe services programs that provide PWID with sterile injecting equipment to reduce transmission of HCV and other bloodborne viruses. 103 With respect to treatment, several studies have found that providing HCV treatment within existing addiction treatment or syringe services programs led to similar cure rates to those seen in randomized controlled trials of DAAs among non-PWID, even when patients continue to use drugs. 90,93,95,96 A necessary step before HCV treatment, regardless of the location, is the identification of HCV infection through screening. All PWID are recommended to be screened for HCV infection. 12 Similar to HCV treatment, addiction treatment and syringe services programs are settings increasingly recognized as integral to improving HCV diagnosis rates. 103,104 In addition to reducing the public health impacts of HCV directly, interventions that include behavioral risk reduction focused on IDU could concurrently reduce the economic costs of HCV, which increase with earlier age of infection, and the economic and societal costs of heroin and other opioid use disorders. 105,106 Our results therefore reinforce the need for an integrated care model. For example, harm reduction programs could provide co-located services that include provision of sterile injecting equipment, bystander overdose response training, bloodborne virus testing and referral to (or coordination of) treatment for HCV and Human Immunodeficiency Virus, and other services, such as mental health counseling, substance use disorder treatment, or opioid maintenance therapy. 107

Limitations
Like all modeling exercises, we made several simplifying assumptions, and we were limited by existing data. Our model only considers HCV acquisition through IDU and we focus results on PWID aged 15-29 given available data. Our results may not be generalizable to other risk groups and simulated intervention outcomes among 30-64 year olds should be interpreted cautiously. Cases with missing risk factor data were assumed to be PWID, consistent with PWID being the most common risk factor for HCV. 3,5,108 Because surveillance data suffers from under-reporting and missing data, we applied a reporting rate to adjust for under-detection of cases by surveillance during the time period under study, although the true number of undetected cases is unknown. 57 We were unable to stratify our analysis by HCV genotype given the available data, though, like the rest of the US, we expected that nearly three-quarters of infections are with genotype 1a. 109 Similarly, we were unable to incorporate geographic heterogeneity in HCV prevalence, transmission, or syringe sharing in Michigan. To address uncertainty in model parameter values, we sampled nearly all parameters and used values most consistent with surveillance data to simulate interventions.
We also made several simplifying assumptions about the natural history of HCV infection. We used Latin hypercube sampling to include several possibilities for spontaneous HCV clearance and sterilizing immunity; however, we did not incorporate partial or waning immunity. Susceptibility to reinfection was the same as susceptibility to primary infection and we assumed that HCV was similarly infectious across the acute, chronic, and treatment stages, and for first or reinfections. These parameters can impact transmission dynamics in the context of HCV elimination and should be examined in future work. 110 Our model assumed homogeneous mixing beyond age and used a syringe sharing contact matrix from Smith et al. among PWID in Baltimore, which may not reflect the age-related patterns in syringe sharing in Michigan. 68 We therefore expanded the assortative age-based syringe sharing patterns from Smith et al. to capture many plausible syringe sharing scenarios, including proportional mixing by age. However, treatment effects may still be overestimated and should be interpreted as the maximum possible impact of treatment. 98 Treatment interventions also assumed a treatment-naïve population during 2000-2016, which is likely realistic given the historically low treatment rates among PWID during that period. [111][112][113][114] Further, our model examined the potential impact of interventions on chronic and acute HCV. Other outcomes, such as hepatocellular carcinoma and liver-related mortality, could be examined in future work. While simulated intervention are an approximation to real-world initiatives and provide an estimate of the potential efficacy of interventions, they do not include the full scope of intervention implementation and adherence challenges that impact intervention effectiveness in the real-world. Finally, in some simulations, our model's estimated injection initiation rates exceeded values found in the literature. [77][78][79][80][81][82][83][84][85][86][87] This discrepancy in injection initiation rates could be for several reasons, including that injection initiation rates were not available for Michigan, that available studies were conducted among specific populations (e.g. Canadian street youth), or because high estimated values of injection initiation rates were artefacts of poorly fitting simulations or ill-suited initial conditions to data. Intervention simulations were conducted using the best fitting 10% of simulations, and these include injection initiation rates consistent with empirical studies and higher values.

Conclusions-HCV
surveillance data is a valuable source of information for understanding HCV transmission and identifying local intervention opportunities among young PWID. In Michigan, HCV treatment could reduce prevalence and incidence by 2030. The impact of treatment is more certain when both former and current PWID are treated and when treatment is combined with behavioral interventions that reduce injection drug use initiation or syringe sharing, increase injection cessation, and reduce relapse. PWID at all stages of use or recovery should be connected to HCV treatment alongside primary and secondary prevention interventions.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Hepatitis C Transmission Model among People who Inject Drugs: Model States and Parameters
This model diagram outlines states and parameters controlling flows between states of an HCV transmission model among PWID in Michigan. Non-PWID with substance abuse or dependence (Z i ) begin injecting drugs (S i ) at a rate θ i and acquire acute infection (A i ) through effective contact with an HCV-infected PWID (A j or C j or T j ) of the same or discordant age (informed by a contact matrix Π j,i ) at a transmission rate (b). Chronic infection (C i ) develops at a rate ε among a proportion of acute cases, (1-d), and resolves in the remaining d acute cases, of whom ξ develop sterilizing immunity and the remaining (1ξ) become susceptible to reinfection. Chronic infection can be treated (T i ) at a rate φ P , for a duration of ω −1 years, after which point susceptibility to reinfection ensues. PWID stop injecting drugs at a rate γ i and transition to former PWID states (denoted by State Ni ). Former PWID can begin injecting drugs after a period of abstinence at a rate k i . Death occurs at a rate μ i , which is elevated among current PWID by a factor h P and among non-PWID by a factor h Z . Former PWID mortality rates are closer to mortality rates from the general Michigan population (μ i ) by multiplying the current PWID mortality increase factor by a protective factor h N . Subscript i denotes parameter or state age class (1: 15-19 years, 2: 20-25 years, 3: 26-29, 4: 30-64 years) and subscript j denotes the age group of contacts from whom susceptible current PWID (S i ) can acquire infection. Individuals move through age groups based on the duration predicted by the age range captured in each group (n i , not depicted for simplicity) and new 15 year-olds are added to the non-PWID compartment each year at a rate (n 0 ψ 0 P 0 , not depicted for simplicity).  Predicted Reductions in Chronic and Acute HCV Cases during 2030 among Young PWID Aged 15-29 from Simulated Primary, Secondary, and Tertiary Interventions The distribution of predicted percent reduction in chronic HCV (top) and acute HCV (bottom) among young PWID aged 15-29 years for 6 interventions for the best-fitting 10% of parameter sets to data are depicted as violin plots. Diamonds denote the median percent reduction and percent reductions are relative to the predicted case count in 2030 from no intervention. On average, decreasing PWID contacts reduced cases more than other primary and secondary interventions. Treatment of current and former PWID predicted the largest reductions to chronic HCV case counts on average, though there was high variability in predicted results across tested scenarios. Histograms of the predicted percent reduction in acute (top) and prevalence (bottom) among 15-29 year olds during the year 2030 are plotted using the best fitting 1,000 parameter sets. All percent reductions are relative to the predicted case count in 2030 from no intervention and interventions are sequentially added from a base of primary (left) versus tertiary (right) interventions. Diamonds denote the median percent reduction. Treating both former and current PWID treatment predicted both prevalence and incidence reductions and the precision of predicted reductions grew with higher intensity interventions or when secondary interventions were combined with treatment.