Modelling the impact of improving screening and treatment of chronic hepatitis C virus infection on future hepatocellular carcinoma rates and liver-related mortality

English population estimates were calculated as a percent of the total UK population, obtained as 5-year age and sex cohorts based on the United Nations (UN) population database [25]. The UN database combines reliable local data sources and contained estimates for 1950 to 2100 in a format consistent with the model (i.e. annual 5-year age and gender cohorts). A limitation of this approach was the need to estimate the English population from the total UK population figures. The percentage of the UK population residing in England in 1991, 2001 and 2011, was obtained from the ONS and was used to estimate the English population [26].

PHE, formerly the Health Protection Agency (HPA), uses a Bayesian model to produce annual estimates of HCV prevalence in England and Wales [27]. A published evidence synthesis estimated the excess risk among ethnic minorities to generate a more robust calculation of HCV prevalence, based on the presence of antibodies, in England: 0.54% (95% confidence interval [CI]: 0.4%-0.75%) among the English population over 15 years of age in 2005 [27]. This is equivalent to 0.4% (95% CI: 0.3%-0.6%) of the total population, including people less than 15 years of age, in England. Figure 1 shows the HCV prevalence in England in 2005 stratified by age and gender.

The present model includes only viraemic cases (defined as patients who test positive for HCV RNA). Antibodies indicate previous HCV exposure, not necessarily current infection. Therefore, limiting the model to individuals positive for HCV RNA ensures that the model includes only current infections. A rate of 68.7% was used [28] to calculate the proportion of people with antibodies to HCV that are also RNA positive. This corresponded to 151,600 (95% CI: 105,100–196,500) viraemic cases in 2005 equivalent to a prevalence of 0.3% (95% CI: 0.2%-0.4%) of the English population. HCV genotypes 1 and 3 account for 45% and 46% of infections in England respectively, according to data collected by 18 sentinel centres between 2007 and 2010 [28]. Table 1 shows the full genotype distribution.

The annual number of new cases in England peaked in 1990 due to infections associated with blood products, then decreased after the introduction of blood screening [29]. The model also included a second peak in 2000 to reflect self-reported needle sharing between 1991 and 2011 [30]. IDU accounted for 90% of laboratory confirmed HCV cases between 1996 and 2012 [4]. The remaining risk factors reported by diagnostic laboratories in England during this period were: blood transfusion or receiving a blood product (2% of laboratory confirmed HCV cases); sexual exposure (2%); renal failure (0.4%); vertical transmission from mother to child; household (0.2%); occupational (0.2%); and other (4.9%) [4].

The model accounts for mortality in the general HCV population and a high-risk subpopulation. Mortality in the high-risk population may be higher than in the background population because of causes unrelated to HCV (e.g. overdoses and accidents). Active IDUs are a high-risk population and, therefore, an increased mortality rate was applied to ensure that the model does not overestimate the burden of disease. Background mortality was estimated using a database hosted by the Max Planck Institute for Demographic Research [31]. We assumed that the peak IDU occurs between the ages of 15 and 44 years. Therefore, increased mortality among active IDUs was estimated for individuals between 15 and 44 years of age [32‐37].

HCV notification reports from 1992–2010 (n = 8,147) [38] were adjusted to account for mortality and for patients who were cured (i.e. no detectable HCV RNA after antiviral treatment), so that only individuals with current chronic infection remained. The number of cured patients was calculated by multiplying the number of treated patients each year by the average SVR. Based on these figures, an estimated 67,200 individuals had been identified living with an antibody-based HCV diagnosis in 2010. Adjusting this value using a viraemic rate of 68.7%, an estimated 46,200 individuals had been identified with RNA-confirmed HCV in 2010. Each year, an estimated 5600 individuals are newly diagnosed with chronic HCV. SVR inputs for the base-case were set at 70% for genotypes 1–3 and 48% for genotype 4 in 2013 (Table 2). The average SVR for genotype 1 took into consideration treatment with protease inhibitors (ie boceprevir and telaprevir).

The model estimates that 5,430 HCV patients were treated in England during 2010. This was based on the number of standard units of peg-IFN sold between 2006 and 2011, with an estimated 26,670 patients treated in this period (IMS Health Incorporated; Danbury, Connecticut). This estimate is in line with those of PHE, which estimates that a total of 27,500 patients were treated between 2006 and 2011 [4]. The English genotype distribution was used to estimate the average number of weeks of treatment and, therefore, the number of units of peg-IFN used per patient, assuming 80% adherence. The base-case scenario assumed that the diagnosis rate, treatment rate and SVR remained constant from 2013 to 2030.

Once the base case was developed, the model was used to analyse alternative inputs for the number of patients treated, eligibility, age-related treatment restrictions, SVR by genotype (G1, G2, G3, G4), and the total number of newly diagnosed HCV cases at different times. This modelled the impact of increasing rates of diagnosis and treatment of HCV as new DAAs become available.

The model examined the impact of future therapies becoming available in four waves:

The model assumed that uptake of new therapies was immediate. To reflect the increased efficacy of new therapies, the average SVR was modified for each wave of new therapies (Table 2).

The model estimated the impact of increasing the number of patients that are treated with new therapies on HCV-related morbidity and mortality. To estimate the ‘best case’ of producing an optimal impact with new therapies on the burden of disease, increases in the number of treated patients were required. For the purposes of this analysis, the number of patients treated was modelled to increase from 5,430 in 2013 to 8,150 in 2014. The number of patients treated then increased to 10,190 in 2016, and 11,710 in 2018 (Table 2). This increase in treatment was selected to accomplish a substantial impact of new therapies on the burden of HCV infection. The increase in treated patients was initially modelled in those with F2 or higher fibrosis, before the treated cohort included F1 or higher from 2018 to provide sufficient numbers of patients to be treated. In the model, eligibility refers to the number of patients without contraindications who are willing to accept treatment. New therapies were assumed to increase eligibility for treatment, increasing from 60% in 2013 to 95% in 2018. In addition, the upper age-limit for treatment increased from 69 years old to 74 years old in 2016 (Table 2). To allow for increases in treatment, the model required that the number of patients newly diagnosed increased from 5,600 in 2013 to 8,400 in 2014 and to 13,430 in 2016 and beyond (Table 2).

Sensitivity analyses and Monte Carlo simulations were performed using Crystal Ball, an Excel add-in by Oracle, to quantify the impact of uncertainties on modelled outcomes. Under Monte Carlo simulation, uncertain variables are represented as probability distributions and model outputs are recalculated 1,000 times to estimate a range of possible outcomes. Each recalculation uses a new randomly selected set of values from the input probability distributions and the likelihood of each outcome is recorded to generate 95% uncertainty intervals (95% UI). Beta-PERT distributions were used for all uncertain variables. Additionally, the impact of variations in the assumptions made in the ‘best case’ scenario (including the number of treated and newly diagnosed patients, and the segment of the population treated) on HCC cases and the number of liver-related deaths by 2020 was measured.

The model estimated that there were 144,000 (95% UI: 103,000-174,000) viraemic cases of HCV in England in 2013 (Figure 2A). The age distribution in 2013 is shown in Figure 2B and estimates for treatment and diagnosis in 2013 are shown in Table 3.

Peak viraemic prevalence of chronic HCV infection was reached in 2007 with 153,000 infected individuals. Since then, the model suggested a decline in overall prevalence reaching an estimated 76,300 in 2030 (Table 4). Due to the lag between infection and the onset of HCV-related hepatic complications, the number of cases of compensated cirrhosis is projected to peak in 2029 at 14,800, increasing from 9,500 in 2013 (Table 4). The population with decompensated cirrhosis is estimated to increase from 860 in 2013 and peak in 2029 at 1400 cases. The number of individuals with HCV-related HCC is projected to increase from 410 in 2013 to 880 in 2030 (Table 4). Unless current practices of care in England change, the model forecasts that liver-related mortality will increase by 90% to 740 by 2030 (Table 4).

Table 5 compares the impact of increasing diagnosis and treatment of HCV as new therapies become available compared to continuing with the current strategy. In the latter case, the model predicts a total of 122,000 viraemic cases of HCV in 2020. Improving diagnostic and treatment rates by a total of 140% and 115% in 2018 respectively, would reduce the number of cases by 30% to 89,400 (Figure 3A).

Similarly, by increasing diagnosis and treatment, the number of HCV-related fibrosis cases in 2020 would decrease by 20% compared to the base case from 107,000 to 83,500 (Figure 3B). The number of patients with HCV-related cirrhosis and decompensated cirrhosis in 2020 would also decrease compared to base case, declining from 12,600 to 4,850 cases (60% reduction) and from 1,140 to 410 cases (65% reduction) respectively (Figure 3C and D). Meanwhile, the number of HCV-related HCC cases would decline from 640 to 310, a 50% reduction (Figure 3E). Finally, liver-related deaths caused by HCV would decrease by 50% from 570 to 280 cases in 2020 (Figure 3F).

Monte Carlo analysis identified the range around the prevalence estimate as the largest driver of uncertainty in the model, accounting for 95% of explained variability. The results of the sensitivity analyses are shown in Table 6. Changes in the number of treated patients (from 12,000 annually to 3,000 annually) had the largest impact on HCC cases and HCV liver-related deaths in 2020. The disease state of treated patients (treating all ≥ F1 patients) and the age of the treated patients considered eligible for treatment (limiting to patients ≤69 years) showed the next largest impact on HCC cases and HCV liver-related deaths in 2020.