Aging of Hepatitis C Virus (HCV)-Infected Persons in the United States: A Multiple Cohort Model of HCV Prevalence and Disease Progression
Gastroenterology Feb 2010
"our findings suggest that the CH-C that we have become familiar with during the last 30 years is much different than the hepatitis C we will come to know during the next decade or 2......Currently, only a small proportion of those with CH-C are aware of their infection and, of these, just 10% to 27% are offered treatment.......the proportion of cases with advanced fibrosis will continue to rise during the next 2 decades (Cirrhosis accounted for just 5% of all cases (diagnosed and undiagnosed) of CH-C in 1989, 10% in 1998, and 20% in 2006, the proportion with cirrhosis is projected to reach 24.8% in 2010, 37.2% in 2020, and 44.9% in 2030).....HCC in persons older than the age of 65 years with HCV infection has doubled during the last several years....(The model suggests that decompensation became more common after 1995 and is currently estimated to be present in 11.7% of persons with cirrhosis (Figure 4). The proportion of cirrhotics with decompensation is expected to continue to rise at least through 2030)......critical to identify infected persons and treat their disease before advanced fibrosis or liver failure ensues......It is in the immediate best interest of patients, providers, insurers, and governments to promote guidelines and encourage better screening for infection and early antiviral treatment.68 Without such a proactive policy, it is likely that we will spend a considerable amount of resources during the next 2 or 3 decades dealing with liver failure in our elderly population."
Gary L. Davis, Miriam J. Alter, Hashem El-Serag, Thierry Poynard, Linda W. Jennings
Division of Hepatology, Baylor University Medical Center and Baylor Regional Transplant Institute, Dallas, Texas
Infectious Disease Epidemiology Program, University of Texas Medical Branch, Galveston, Texas
Baylor College of Medicine and Michael E. DeBakey Veterans Affairs Medical Center, Houston, Texas
Assistance Publique Hopitaux de Paris, Universite Pierre et Marie Curie Liver Center, Paris, France
Received 15 July 2009; accepted 28 September 2009. published online 26 October 2009.
Background & Aims
The prevalence of chronic hepatitis C (CH-C) remains high and the complications of infection are common. Our goal was to project the future prevalence of CH-C and its complications.
We developed a multicohort natural history model to overcome limitations of previous models for predicting disease outcomes and benefits of therapy.
Prevalence of CH-C peaked in 2001 at 3.6 million.
Fibrosis progression was inversely related to age at infection, so cirrhosis and its complications were most common after the age of 60 years, regardless of when infection occurred.
The proportion of CH-C with cirrhosis is projected to reach 25% in 2010 and 45% in 2030, although the total number with cirrhosis will peak at 1.0 million (30.5% higher than the current level) in 2020 and then decline.
Hepatic decompensation and liver cancer will continue to increase for another 10 to 13 years.
Treatment of all infected patients in 2010 could reduce risk of cirrhosis, decompensation, cancer, and liver-related deaths by 16%, 42%, 31%, and 36% by 2020, given current response rates to antiviral therapy.
Prevalence of hepatitis C cirrhosis and its complications will continue to increase through the next decade and will mostly affect those older than 60 years of age. Current treatment patterns will have little effect on these complications, but wider application of antiviral treatment and better responses with new agents could significantly reduce the impact of this disease in coming years.
It is estimated that up to 4 million persons in the United States have chronic hepatitis C virus (HCV) infection (CH-C).1, 2 Despite the marked decrease in newly acquired infections in recent years, overall prevalence of CH-C has not fallen.2 Most individuals with CH-C acquired their infection 20-40 years ago, before identification of the virus and availability of screening tests.3 Because CH-C typically progresses slowly and does not result in morbidity for many years, most remain undiagnosed. We are only now beginning to recognize the magnitude of the consequences of infection that has persisted for decades.4, 5
Outpatient and hospital visits for CH-C have doubled in recent years and show no sign of leveling off.6, 7, 8 In the United States, complications of CH-C are the leading indication for liver transplantation and the disease is reported to contribute to 4600-12,000 deaths per year based on death certificate documentation,2, 6, 9, 10 despite the limitations of this method in estimating true death rates.11, 12 Although some have suggested that the health care burden resulting from complications of CH-C has reached a plateau,10 others have projected a further increase in cirrhosis and its complications for another 2 to 3 decades.13, 14
Several models have been developed during the last decade to predict the future course of CH-C.13, 14, 15, 16 However, simple transition-state (Markov) models have significant limitations in that the studied cohort is considered homogeneous and traverses through their disease at a fixed and predictable rate over time. In reality, however, the population is quite heterogeneous due to factors such as age at infection, gender, and disease duration; therefore, the course of disease is variable and nonlinear over time.17, 18, 19 Also, previous models used standard population mortality, which recent studies suggest might underestimate true mortality in a chronic disease population.11, 20, 21
Advances in computer software now permit construction of complex models that allow parallel cohorts with different disease states and probabilities to run over time in order to provide a more realistic estimation of end-point events. The purpose of this project was to utilize state-of-the-art statistical modeling techniques and the latest epidemiologic, demographic, and natural history data to more accurately model the evolution of CH-C during the last 60 years and project its course in the coming decades.
Our model estimated that the prevalence of CH-C in the United States peaked at 3.6 million in 2001 and will decline to about half this number by 2030. Although the decline in overall infections is encouraging, other trends in the data are of concern. First, similar to reports by others,13, 14, 16 the proportion of cases with advanced fibrosis will continue to rise during the next 2 decades, with the number of cases of cirrhosis and hepatic decompensation peaking after the year 2020. Second, the age of those with cirrhosis and its complications will continue to rise. Because about 40 years elapses from the peak incidence years of HCV infection until the peak prevalence of cirrhosis and other complications (Figure 1), it is not surprising that we found the group of persons aged 60 to 80 years to be those most affected. This phenomenon is already beginning to occur.49, 52, 53, 54 Indeed, Ferenci and colleagues found that 34% of infected paid plasma donors identified in the 1970s had bridging fibrosis, cirrhosis, or HCC 30 years later.51 In addition, Thabut and colleagues reported that cirrhosis was more prevalent in the elderly and 14% of them presented with decompensation compared to just 4% in persons younger than 65 years.52 Indeed, D'Souza and others have even suggested that patients who live long enough will almost invariably develop advanced hepatic fibrosis.54, 55 We also predicted a modest increase in HCV-related HCC in coming years, despite using a very conservative estimate of the annual risk in our model. HCV infection currently accounts for most of the HCC in the United States.49 HCC in persons older than the age of 65 years with HCV infection has doubled during the last several years,48, 56 consistent with our predictions. Taken together, our findings suggest that the CH-C that we have become familiar with during the last 30 years is much different than the hepatitis C we will come to know during the next decade or 2.
Although we purposefully chose conservative estimates of disease progression and complications, it is still possible that we and others might have overestimated the number of cases that will progress to liver failure from CH-C due to the influence of competing risks.11, 12, 20, 57 Indeed, when we increased the background (nonhepatic) mortality by just 50%, there was a striking reduction in the number of cases of cirrhosis, liver failure, and cancer. However, we believe that such high background mortality is unlikely in HCV-infected patients. If it occurs it is probably limited to a short period around the time of acute infection and would be unlikely to influence long-term outcomes. Furthermore, common comorbid conditions, such as diabetes, obesity, and alcohol, lead to more rapid progression of fibrosis, which can offset any potential impact of a change in background mortality.26, 58, 59, 60 Competing risks could certainly influence resource utilization and might explain why, for example, the number of liver transplantations related to hepatitis C has started to plateau, despite our prediction of more disease complications.
These projections emphasize how critical it is to identify infected persons and treat their disease before advanced fibrosis or liver failure ensues. Currently, only a small proportion of those with CH-C are aware of their infection and, of these, just 10% to 27% are offered treatment.61, 62, 63 Many physicians still do not ask their patients about risk factors for HCV infection and some remain confused about treatment options and efficacy.64 And yet, antiviral therapy is becoming increasingly effective65, 66 and eradication of virus clearly reduces risk of liver failure or cancer.27, 67 Certainly, as we have shown, a far higher proportion of cases will need to be identified and treated to impact the dire projections described here. It is in the immediate best interest of patients, providers, insurers, and governments to promote guidelines and encourage better screening for infection and early antiviral treatment.68 Without such a proactive policy, it is likely that we will spend a considerable amount of resources during the next 2 or 3 decades dealing with liver failure in our elderly population.
Population-Based Model Outcomes
There was a rapid increase in the prevalence of CH-C between 1970 and 1990, when the incidence of acute HCV infection was greatest (Figure 1). We estimated 3.49 million infected persons in 1994, which is similar to previous predictions based on NHANES III.13, 14 Estimated prevalence peaked at 3.57 million in 2001 and began to decline slowly thereafter, reaching about half its peak level by the year 2030 (Figure 1). Similarly, the model estimated that the number of persons who had ever been infected (resolved or chronic) peaked in 2001 at 5.04 million.
Figure 1. Estimates by year of prevalent cases ever infected (top line), with chronic hepatitis C (open circles), and cirrhosis (solid squares). Acute infections (solid gray line) peaked between 1970 and 1990. The peak of chronic hepatitis prevalence was 2001, while the highest prevalence of cirrhosis is projected to be between 2010 and 2030, about 40 years after the peak of acute infections.
Predicted distribution of histologic stages of fibrosis over time is shown in Figure 2. In 1970 and throughout the period during which HCV prevalence grew, the majority of cases were stage F0 or F1 (1970: 86.5%, 1980: 84.2%, 1990: 77.6%). Indeed, F0 and F1 fibrosis accounted for the majority of cases of CH-C until just recently. Currently, 41.8% of infected persons have minimal-to-mild fibrosis (F0 or F1), and 39.5% have F3 or F4 fibrosis.
Figure 2. Distribution of histologic stages of fibrosis by year in persons with chronic hepatitis C (F0 = closed black squares, F1 = closed gray diamonds, F2 = open triangles, F3 = solid gray, and cirrhosis (including decompensated and hepatocellular carcinoma) = fine line with closed circles).
Cirrhosis accounted for just 5% of all cases (diagnosed and undiagnosed) of CH-C in 1989, 10% in 1998, and 20% in 2006 (Figures 1 and 2). This proportion began to rise sharply after 1990, as the age and duration of infection of those infected began to increase. Indeed, the proportion with cirrhosis is projected to reach 24.8% in 2010, 37.2% in 2020, and 44.9% in 2030, although the total number of persons with cirrhosis is expected to peak at 1.04 million (30.5% higher than its current level) in 2020 and slowly decline thereafter. Men, particularly those infected before age 50, account for the majority of cases of cirrhosis today (73.6%) because of their more rapid rate of progression (Figure 3). Although men who acquired their infection after age 50 have the most rapid disease progression, they account for a small proportion of all cirrhosis (7.7%) because they often died of other causes before their fibrosis had a chance to progress. Although females who acquired infection before age 50 accounted for almost the same proportion of acute HCV infections as similarly aged men during the peak incidence years (43.0% vs 50.3%), they represent a much smaller proportion of those who have progressed to cirrhosis as of 2009 (16.1%) because chronicity was less likely and CH-C had a slower rate of progression than men.
Figure 3. Stacked prevalence curves showing number of cases by year with cirrhosis according to gender and age at time of initial hepatitis C virus infection.
Hepatic decompensation and HCC are late complications of CH-C occurring in persons with advanced fibrosis. The model suggests that decompensation became more common after 1995 and is currently estimated to be present in 11.7% of persons with cirrhosis (Figure 4). The proportion of cirrhotics with decompensation is expected to continue to rise at least through 2030, although the total number of persons with liver failure will start to decline after 2022. The number with HCC began to rise steeply after 1990 (Figure 4). The model estimated 37,697 cases between 1990 and 1999 compared to 86,765 (+130%), 130,366 (+50%), and 124,298 (-5%), respectively in each of the subsequent decades. The incidence of HCV-related HCC is projected to peak in 2019 at almost 14,000 cases per year if the risk in HCV-infected persons with fibrosis remains stable. Assuming that 55% of HCC cases are due to hepatitis C, the HCC projections in males and females in 2005 approximate estimates from a recent report based on the Surveillance, Epidemiology and End Results database (predicted vs estimated cases: male, 7700 vs 8053; female, 1608 vs 922).48, 49
Figure 4. Projected number of cases by year of decompensated cirrhosis (black) and hepatocellular carcinoma (gray). The model assumes a first year mortality of 80% to 85%, so in contrast to the decompensated cirrhosis projection, the number of cases of hepatocellular carcinoma the prevalence demonstrated here closely resembles annual incidence of liver cancer.
We projected that hepatic deaths due to HCV would continue to increase through 2022, although the rate of year-to-year change began to slowly decline after 1991. Consistent with the increasing average duration of infection in persons with CH-C and the severity of liver disease described here, we estimated 29,090 liver-related deaths from 1980 to 1989, 56,377 (93%) in 1990-1999, 145,667 (+158%) in 2000-2009, 254,550 (+74%) in 2010-2019, and 283,378 from 2020 to 2029.
The proportions with cirrhosis, decompensation, HCC, liver-related death, and non-liver-related death after 10, 20, and 30 years in each of the 6 cohorts are shown in Table 2. These proportions apply to the entire cohort that was ever infected, including those who resolved infection early. For example, although 1.33% of women in the youngest cohort were expected to have cirrhosis after 20 years, these cases occurred only among the 55% who had not resolved acute infection spontaneously; this represented 2.59% of persons with CH-C. Fibrosis and cirrhosis increased over time in all cohorts, but never exceeded 40% within the 30-year observation period. However, if only those who developed CH-C are considered, cirrhosis was predicted in more than half of the 2 older male cohorts after 30 years because progression rates were more rapid in these persons. The majority of these remained compensated.
Duration of infection before the peak prevalence of cirrhosis and its complications always varied inversely with the age at acute infection. Therefore, the estimated average ages at the peaks for disease complications were surprisingly consistent in the 6 cohorts (Supplementary Table 3). In the female cohorts, the peak for cirrhosis appeared at ages 71.5-87.0 years, decompensation at 74.5-82.5 years, and HCC at 72.5-87.0 years. In males, cirrhosis appeared at ages 64.5-79.0 years, decompensation at 68.0-82, and HCC at 65.5-80.0 years.
Effect of Treatment
Our original model was not designed to examine treatment effects so a second model was designed to test the effects of treating various proportions of persons with CH-C and no preexisting complications of liver disease. All treatment was administered in the year 2010. Assuming current estimates that 30% of cases of HCV are diagnosed and up to 25% of those are treated, we would anticipate just a 1.0% reduction in cirrhosis by 2020 compared to 7.8% or 15.6% reductions if half or all of individuals were treated, respectively (Figure 5A). However, if viral clearance increased to 80%, as appears possible with evolving treatments, treatment of half or all of infected persons would reduce cirrhosis by 15.2% or 30.4%, respectively, after just 10 years. The effects are more pronounced when looking at complications of liver disease. Indeed, treatment of half or all of infected persons in 2010 would result decrease cases of liver failure of 39.4% or 78.9%, HCC by 30.2% or 60.4%, and liver-related deaths by 34.0% or 68.0% over the next decade (Figure 5B).
One-way sensitivity analysis in the youngest female and male cohorts found the rate of chronicity after acute infection was dominant over all other variables in determining the risk of cirrhosis. No other transition rates significantly impacted the risk of cirrhosis after 20 years. The group's predicted estimates of cirrhosis after 17 and 24 years were 0.8% and 2.3%, respectively. These proportions are similar to the 2.0% and 3.1% reported in young Irish women by Kenny-Walsh and Levine39, 40 (Table 3). Similarly, we predicted that a 1.8% hepatic death rate in men infected before age 30, which is similar to the 5.9% observed in the small study of military recruits reported by Seeff and colleagues41 (Table 3). These projections appear to confirm the accuracy of the assumptions in the younger cohorts. In the oldest males, the initial transition from F0 to F1 had the greatest impact on the risk of cirrhosis. However, when the sensitivity analysis in this group was confined to stage-to-stage fibrosis transition rates, no single transition dominated, and risk of cirrhosis after 20 years varied from just 3.75-6.08% less to 2.39-5.69% more than our model projection. In addition, the projected prevalence of cirrhosis after 20 years in the older male group was 24%, consistent with the high rate of fibrosis progression reported by others in this group.19, 50, 51 (Table 3).
Another potential source of uncertainty stems from our use of tunnel states to slow progression during early fibrosis stages and acceleration of progression rates when younger cohorts reached age 50. We felt that these age- and duration-adjusted fibrosis progression rates best reflected the observations from previous reports.17, 18, 19, 33, 34, 39, 40, 41 However, we also tested a fixed stage-to-stage progression rate as reported in the meta-analysis by Thein and colleagues.23 As expected, this led to front-loading of morbidity and hepatic mortality, particularly in the younger cohorts, during the years when standard age-related mortality was low. As a result, the fixed rate model overestimated cases of cirrhosis, decompensation or HCC, and liver-related death by 41%, 86%, and 41%, respectively, compared to the more conservative base case model, and these estimates were much higher than those reported in previous prospective studies.39, 40, 41 Therefore, we maintained our more conservative assumptions for the base case model.
Finally, we looked at the impact of utilizing higher background (nonhepatic) age- and gender-related mortality rates on our projections of chronic hepatitis and cirrhosis in coming decades. Recent studies have suggested that comorbid medical conditions might increase background mortality in HCV-infected persons by as much as 3 times the reported actuarial rates.11, 12, 20 Increasing background mortality rates in the model by 50% or 100% had no impact on the proportion of infected persons with cirrhosis, but it significantly reduced the total number with cirrhosis. A 50% increase in background mortality decreased the number of persons with chronic hepatitis or cirrhosis in 2020 by 35% and 31%, respectively, compared to the base case model projections. A 100% increase decreased the number with chronic hepatitis or cirrhosis in 2020 by 46% and 43%, respectively, compared to base case projections.
Our sensitivity analysis did not include incidence data because no other estimates exist and these data have previously been shown to be consistent with prevalence data as estimated by NHANES III.1, 14
Materials and Methods
Construction and computer simulations of our model utilized TreeAge Pro 2009 Suite (TreeAge Software, Williamstown, MA) linked with Microsoft Excel 2007 (Microsoft Corporation, Redmond, WA). TreeAge allows construction of complex Markov models with progression rates that can be varied with time according to calendar year, patient age, year of disease, or dwell time within a specific state, eg, years of cirrhosis. It also allows the dynamic import of information such as disease incidence, population demographic change, and other factors from Excel spreadsheets.
The model was developed in stages starting with a traditional bubble diagram of disease states that served as the basis for developing a more detailed mathematical model that followed infected persons from the time of acute infection until death. Both the bubble diagram and the more detailed TreeAge model are provided in Supplementary Figure 1, Supplementary Figure 2 available in the online version of this article. Acute infection could resolve, evolve to fulminant hepatitis, progress to chronic hepatitis, or end with death due to background (nonhepatic) mortality. CH-C was modeled through fibrosis stages (Metavir F0 to F4), disease complications, and death. The model cycled at yearly intervals allowing individuals to move to another state; however, all states except acute infection and fulminant hepatitis could resolve back into themselves indefinitely.
It has become apparent in recent years that age at infection and gender greatly influence the risk of developing chronic infection and progressing to fibrosis.18, 19, 20, 21, 22, 23, 24, 25, 26 To accommodate this heterogeneity, we divided acutely infected individuals into 6 cohorts, each with their own cohort-specific transition states for chronicity, fibrosis progression, and complications (Table 1). The output of the 6 models was then collated and exported as an Excel spreadsheet defining cohort-specific and overall projections by year. Cohort and population projections included numbers with resolved infection, chronic infection, stage of fibrosis (F0 to cirrhosis), liver failure, hepatocellular carcinoma (HCC), liver-related deaths, and age- and gender-specific (nonliver) deaths.
Antiviral treatment, ablation of tumors, and liver transplantation could not be modeled because the indications for these practices and procedures are not standardized, their use varies considerably by geography, none is widely applied, and the outcomes are not well-defined in subgroups. Of these, only antiviral treatment has the possibility of significantly altering disease end points from a population perspective. Therefore, we subsequently modified the model to estimate the potential impact of antiviral treatment on cirrhosis and its complications. To do so, we considered treatment penetrance ranging from 0% to 100% in the infected population and sustained viral response rates of 40% to 80%. All cases were treated in 2010 and outcomes were calculated for the year 2020. It was assumed that patients with sustained viral response had no chance of progressing to cirrhosis or, if already cirrhotic, had no chance of subsequent hepatic failure.27 The risk of progressing from cirrhosis to HCC after sustained viral response was assumed to be 0.66% per annum.27 Other assumptions of the model were unchanged.
Medical literature was reviewed by the authors to provide best estimates of the range of probabilities for moving between disease states. Because reported progression rates vary between studies, a consensus rate was chosen (Table 1). We used conservative estimates when a particular transition was in doubt. Transition rates varied for some states according to the age and gender of the cohorts.23, 24, 25, 26, 27, 28 For example, in the young female cohort, progression from acute to chronic infection was lower (55%) and the rate of fibrosis progression was slower than the oldest male cohort.
Transition rates for moving from one fibrosis stage to another largely utilized the pooled rates from a meta-analysis reported by Thein and colleagues.17, 23 However, our model also included tunnel states that kept a portion of the cohort with minimal fibrosis (F0 and F1) from progressing to the next state for variable periods of time in order to simulate slower rates of progression or no progression in some persons. To accommodate the tunnel states, fibrosis transitions were adjusted accordingly such that overall rates remained consistent with those of Thein and colleagues23 (Table 1). Alcohol intake was not modeled separately because its influence in the different cohorts is unknown; rather our transition rates are based on observational studies in which excessive alcohol use was present in 19% of the cohorts.23 Risks of developing HCC in persons with bridging fibrosis or cirrhosis were taken from longitudinal studies in North America and Europe.19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32 HCC risk in those with bridging fibrosis was estimated to be 10% of that in cirrhosis. HCC risk in females was estimated to be 40% of that in males.33 Finally, transition rates for fibrosis progression and HCC risk were allowed to increase after age 50 or 10 years of cirrhosis, respectively, consistent with previous observations that these risks are not linear.17, 18, 19, 33, 34 Confirmation of the accuracy of the estimated transition rates was done by comparing the projected chronic infection prevalence in 1994 to results from the Third National Health and Nutrition Examination survey (NHANES III) and the projected cirrhosis and HCC prevalence in the different cohorts to published observations in similar groups.
Annual numbers with newly acquired HCV infections between 1960 and 2006 were generated from a previously published model that estimated the past incidence of acute HCV infection given the actual prevalence measured at the time (1988-1994) of the NHANES III.1, 14, 35 Annual infections were stratified by age and gender based on actual distributions of these variables for cases of acute hepatitis reported to the Centers for Disease Control and Prevention's Sentinel Counties Study for each year from 1979 to 20062, 36 (personal communication, MJ Alter, November 25, 2008). For years when data from the Sentinel Counties Study were not available, number and distribution were assumed to be the same as the closest year with data. We sequentially entered these incident infections in annual cycles into the cohort models and projected disease outcomes through the year 2030. Annual incidence data and cohort distribution utilized in the model are shown in Supplementary Table 1 available online.
Age- and gender-specific all-cause mortality was derived from standard US mortality tables.37 The ages within each cohort were tracked through the annual cycles of the model such that background mortality would be appropriate as the cohorts aged.
Sensitivity Analysis and Validation
Sensitivity analysis was performed to assess the extent to which the model's calculations were affected by uncertainty in our assumptions. The ranges utilized in the sensitivity analysis were derived from the medical literature.31, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 and are reported in Supplementary Table 2 available online. Sensitivity analyses were confined to the young female and oldest male cohorts because these were the most disparate in transition probabilities and, therefore, most likely to identify potential sources of weakness within the model assumptions. First, one-way sensitivity analysis with tornado diagrams were utilized to identify dominant variables and to rank the impact of different variables on disease outcome (cirrhosis). If no dominant variable was identified, a second one-way sensitivity analysis was done. Finally, we examined the impact of increasing background mortality, consistent with some reports that suggest higher all-cause mortality in persons with CH-C.5, 15