Future complications of chronic hepatitis C in a low-risk area: projections from the hepatitis c study in Northern Norway

In 1992, a screening and medical follow-up programme of patients with HCV infection was established in the Health Region of Northern Norway (460.000 inhabitants). In brief, general practitioners were encouraged to screen patients with former or present risk behavior for HCV infection. If chronic HCV infection was detected, the general practitioners were encouraged to refer the patient for follow-up at one of the 11 medical centers in the region. An estimate of the year of transmission was made for all referred patients based on either the year of acute HCV infection or the first year of high-risk behaviour [23]. Liver biopsies were performed, and fibrosis was graded (0–6) according to Ishak et al. [25]. Presence of concomitant alcoholic liver disease was assessed by clinical judgment.

Individuals with a positive anti-HCV test registered at the two microbiological departments in Northern Norway between 1998 and 2012 were included. The year of diagnosis was defined as the first year of a positive anti-HCV test (ARCHITECT Anti-HCV Reagens kit. Abbott System, Wiesbaden, Germany). Until 2004, a positive anti-HCV test was directly confirmed with a recombinant immunoblot assay (RIBA HCV 3.0 SIA test, Chiron Cooperation, Emeryville, CA, USA). Individuals with a positive RIBA (two or more positive bands) or indeterminate RIBA (one band) were included, while individuals with a negative RIBA (no bands) were excluded. The result of the HCV RNA test was recorded if available: an in-house reverse transcriptase polymerase chain reaction (RT-PCR) until 2004, where after the ROCHE RT-PCR (Cobas Amplicor Hepatitis C Viral Polymerase Chain Reaction, Roche Molecular System Inc., Branchburg NJ, USA) was used. The ROCHE PCR test replaced the RIBA test for confirmation of HCV infection from 2005. HCV genotyping was performed as a hybridization assay on products from the HCV RNA PCR according to the manufacturers’ instructions (INNO-LIPA HCV II kit, INNOGENETICS, Ghent, Belgium). Individuals without a registered residence in Norway were excluded. The results of HBsAg (hepatitis B surface antigen) and anti-HIV (human immunodeficiency virus) were recorded if available.

In order to predict HCV-related morbidity and mortality an open Markov model was constructed (Fig. 1). Of infected patients, 20% were assumed to recover spontaneously during the first year of infection [26]. Yearly transitions between categories were done according to probability estimates generated from different sources. The effect of medical treatment was modelled in three scenarios where 0%, 15% and 50% of all patients were assumed to receive medical treatment. Before the introduction of direct acting antivirals (DAAs), treatment was offered to eligible patients irrespective of fibrosis stage, i.e. patients with Metavir fibrosis score F0/F1 were also treated. The treatment coverage at this time was approximately 15%. When the highly expensive DAAs became available, the Norwegian national guidelines restricted treatment to patients with fibrosis score ≥ F2. However, since DAAs offer a simple, tolerable, short-term and highly effective therapy, we consider an increased treatment coverage to 50% as possible. Recently (March 2017), updated national guidelines recommend treatment for all patients with genotype 1, irrespective of fibrosis grade. Treatment of genotype 2 and 3 is still restricted to patients with fibrosis score ≥ F2.

To estimate the rate of fibrosis progression in the cohort, 237 records with a positive HCV RNA test, known duration of infection, and available liver biopsy results were entered into an ordinal regression analysis, which can be described as a multilayer logistic regression. This method estimates points of transition on an ordinal scale (like the Ishak fibrosis scale). Sex, time after contraction of the disease and HCV genotype were included as predictors of Ishak fibrosis grade. Non-significant terms were removed from the model leaving time after contraction and genotype as significant predictors. Genotype 1 and 4 were analyzed as a single group due to few observations of genotype 4. The model predicted 70% of the observed fibrosis grades correctly within a margin of error of one Ishak grade, but overestimated fibrosis grade in 13% and underestimated in 17%. From this model, a matrix of estimated probabilities of transition to a higher fibrosis grade versus staying in the present grade for each year of infection could be constructed. It was assumed that fibrosis development during ongoing HCV infection could only either stay the same or change to a higher stage each year of infection. Additional file 1 shows the regression model in more detail. The Markov model was adjusted for genotypes in the estimation of fibrosis progression, according to the prevalence of the different genotypes in our population. For transitions from Ishak 6 (compensated cirrhosis) to more severe states, fixed probabilities were used. We did not have exact data from Northern Norway for the various transitions, and have thus used data from a Scottish HCV population of drug-injecting abusers (as in our study), as described by Hutchinson et al. [13], with transition probabilities similar to that reported by others [14, 27], as shown in Table 1. The model was corrected for standardized mortality rate according to Norwegian population characteristics. Non-cirrhotic subjects successfully treated for HCV with achieved sustained virological response (SVR) were removed from the model. Subjects with cirrhosis remain in risk of liver complications in spite of SVR, and of those, 50% were retained in the model [28]. All successfully treated subjects with decompensated cirrhosis and HCC were kept in the model.

The overall SVR rate until 2015 was 71.3%, based on the response of interferon-based treatment in our own clinical practice. For individuals treated from 2015 onwards, the overall SVR rate was set to 90% to reflect the improved treatment response provided by DAAs.

Figure 2 shows the actual incidence of newly diagnosed HCV infection per year in the period from 1992 to 2012 (dark bars). After 2004, the incidence decreased to a stable level of approximately 90 newly diagnosed cases each year. In view of this stable number, we have projected a stable occurrence of 90 newly diagnosed cases each year from 2013 and forwards (light bars).

HCV-positive individuals were entered into the model at time of contraction. Since only a subgroup had a known time of transmission, this had to be estimated for the remaining records. First, we investigated the 402 records with known times of transmission. In 2004, all general practitioners in the area were subject to a HCV campaign encouraging screening of their patients with known risk behaviour. We found an increased amount of early diagnoses (within the first year of transmission) in the years after this campaign compared to before (35.5% vs. 21.3%). From these data, probability distributions could be derived in order to estimate time of contraction for records diagnosed before and after 2004, respectively. Records without known time of transmission were then entered into the model using these distributions as weights. From 2013 forward, the estimated number of 90 newly diagnosed cases each year were allocated due to the late probability distribution in order to estimate their time of contraction.

The registration of HCV infection revealed incomplete records regarding confirmation testing. A positive anti-HCV test normally should be followed up with a new blood sample for the confirmation with RIBA and/or HCV RNA. Not all the persons with positive anti-HCV tests had a recorded confirmation test. The starting cohort in the model thus consists of patients with confirmed HCV infection (either a positive RIBA or a positive HCV RNA test), as well as individuals with unconfirmed HCV (either only a positive anti-HCV or an indeterminate RIBA). We therefore estimated the likelihood of a true positive test in incomplete records in the following way: In a sample of 326 records with a positive anti-HCV-test where RIBA had been measured, 207 subjects (63%) had a positive RIBA test, and the probability of a true positive anti-HCV test was estimated to 0.63. Similarly, in a sample of 14 records with an inconclusive RIBA and a HCV RNA test, we found three individuals with a positive HCV RNA test, and the probability of a true positive record in case of inconclusive RIBA was estimated to 0.21. Summarized, individuals with either a positive HCV RNA or RIBA were weighted 1.0, and individuals with only a positive anti-HCV test or an inconclusive RIBA were weighted 0.63 and 0.21, respectively.

In the subgroup that was the basis for estimation of fibrosis progression (n = 237), the sex distribution was the same as in the total cohort and the distribution between HCV genotypes 1 through 4 were approximately equal to the total cohort, 45%, 10%, 44%, and 1%, respectively. The median duration of infection was 13 years (range 0–42 years) and the mean age at liver biopsy was 40 years. The rate of fibrosis progression was relatively slow in the first 20–25 years of infection, followed by an accelerated fibrosis progression, especially in patients with genotype 3 (Fig. 3).

The Markov model projects the number of patients (per 100,000 inhabitants) in the different states of compensated cirrhosis, decompensated cirrhosis and HCC for the years 2013 to 2050, given various scenarios of HCV treatment coverage. It estimates an almost threefold increase in the incidence of cirrhosis (68 per 100,000), of decompensated cirrhosis (21 per 100,000) and of hepatocellular carcinoma (4 per 100,000) by 2050, as shown in Fig. 4a-c. Complications are expected to reach a peak around 2040. The model predicts a six-fold increase in the cumulated number of deaths from HCV-related liver disease (170 per 100,000 inhabitants), as shown in Fig. 4d. All estimates are made assuming an unchanged treatment coverage of approximately 15%. The estimated numbers can be reduced by approximately 50% for cirrhosis, and by approximately one third for the other endpoints if treatment coverage were scaled up to 50%.

Precision of dynamic modelling for prognosis of prevalence, morbidity and mortality depends on the quality of the data entered and the assumptions made. We have used locally acquired data whenever possible. The critical aspect in HCV pathophysiology is the progressive development of fibrosis towards cirrhosis, and in this aspect, our model also made use of local data. The further development from established cirrhosis is not likely to be very different from that of other cohorts, and the use of estimated transition rates from other studies [13] should not affect the prognosis measurably.

Our Markov model is based on an assumed low and stable incidence until 2050, whereas other reports estimate a declining or stable incidence [2]. The observed incidence peak in 2002–2004 in our data is probably a result of increased focus on HCV infection among general practitioners due to our encouragement of diagnosis in that period.

Our cohort is relatively young, and the liver biopsy data has a limited number of observations of severe fibrosis (Ishak grade 4–6). Hence, the estimates generated from these data are less precise regarding progression through high fibrosis stages.

Published data show variable rates of liver fibrosis progression, with 10%–40% developing cirrhosis after 20–35 years of infection [5, 8, 29]. A large systematic review supports nonlinear fibrosis progression [8] and other studies have shown that HCV genotype 3 is associated with a faster fibrosis progression [30‐33]. Male sex is a reported risk factor for disease progression [5, 31]. However, we found no effect of sex in the progression of fibrosis, which is also observed in other studies [34].