In the medical community, clinical trials are the gold standard for decision making: evidence that an intervention is effective and cost-effective in a cohort of patients provides a strong foundation for decision making. However, since hepatitis B usually takes decades to progress, a trial evaluating these policies would be very expensive, and any results – and valuable policy interventions – would be delayed for decades. We instead used operations research-based modeling to estimate the likely health and economic impacts of alternative hepatitis B prevention and control programs.
We evaluated the health and economic impacts using a cost-effectiveness framework. We present our results in the standard health economics format of an incremental cost-effectiveness ratio, where the strategies are compared, incrementally to each other, in terms of the incremental cost to gain an incremental unit of health, and health units are expressed in terms of quality-adjusted life years (QALYs) (Drummond et al. 2005
, Gold et al. 1996
, Muennig 2008
). This allows for ready comparison of alternative hepatitis B control strategies as well as comparison with other potential public health interventions. When thought of in a traditional operations research optimization framework, the incremental cost-effectiveness ratio is the value/cost metric used to solve a fractional knapsack problem (Hillier and Lieberman 2005
). Interventions that cost less per QALY gained than three times a country’s per capita GDP are considered to be cost-effective, and interventions that cost less than a country’s per capita GDP are considered to be very cost-effective (Murray and Lopez 2002
, World Health Organization 2009
To estimate the economic and health impacts of hepatitis B screening, treatment, and vaccination programs, we created a decision model of alternative policies () along with a Markov model of hepatitis B disease () (Hutton et al. 2010
, Hutton et al. 2007
). This allowed us to capture the long-term health and economic outcomes among chronically infected individuals with and without treatment. We modified the model (policy alternatives considered, epidemiological factors, costs, etc.) as appropriate for our analyses of policies in the United States and China.
Schematic of Markov model of hepatitis B infection and progression*
The decision model () captures various possibilities for screening, treatment, and vaccination. We consider a cohort of 10,000 individuals of a given age (e.g., age 40 for adult Asian and Pacific Islanders in the United States, or age 10 for school-age children in China). Individuals can be screened (via blood tests) to determine their infection status. If an individual is screened and found to be infected, the individual can be treated or not; if the individual is found to be susceptible, the individual can be vaccinated or not; and if the individual is found to be immune (either from previous vaccination or from previously resolving the infection), no further action is taken. If an individual is not screened, the individual can still be vaccinated or not vaccinated. For all individuals who were already infected, or who were susceptible and unvaccinated and subsequently acquired the infection, we model disease progression over the future lifetime of those individuals using a Markov model.
Our Markov model of hepatitis B disease () is a discrete time, time-inhomogeneous model. To construct the model, we examined previously published models of hepatitis B infection in the literature and used the expertise of Dr. So to create a model that reflects current knowledge of hepatitis B infection and treatment. Individuals who acquire acute hepatitis B infection can become immune by resolving the infection; if their immune system cannot resolve the infection, they develop chronic hepatitis B infection. Chronic infection can be asymptomatic for many years. For many individuals, chronic infection does not cause liver disease; in such cases an individual remains in the chronic infection state for his or her lifetime. Some individuals do develop liver disease, which is often asymptomatic until serious liver disease develops. Individuals identified as having liver disease (either through screening or via the development of symptoms) can receive treatment, which may successfully suppress the virus. Otherwise, the disease can progress to cirrhosis and/or liver cancer. We chose yearly transitions for the Markov model since chronic hepatitis B tends to progress slowly. Transitions follow the Markov property of being memoryless, with one exception: background mortality is based on the age of the individuals in the cohort, year by year.
The model calculates net present costs and health outcomes (measured in discounted quality-adjusted life years (QALYs)) of each policy, over the lifetime of the cohort of individuals. Quality-adjusted life years are a measure of both quality and length of life. Each year of life is multiplied by a utility value between 0 (representing death) and 1 (representing perfect health); this utility value is intended to reflect the relative “quality” of a given health state. In each time period, costs and utilities are assigned to individuals in each health state. The costs incurred and QALYs experienced are measured for each health state and time period over the time horizon of the problem (in this case, the remaining lifetime of all individuals) and then discounted back to the present.
Our model has several novel features. First, we combined a decision model of alternative screening and vaccination policies with probabilistic trees of infection and a Markov model of disease progression and treatment. No previous model has examined hepatitis B screening and treatment together. Some analyses have considered the cost-effectiveness of screening, without considering the effects of subsequent treatment for infected individuals (Jacobs et al. 2003
, Margolis et al. 1995
, Pisu et al. 2002
), while other analyses have considered only the cost-effectiveness of treatment for infected individuals who have already been identified (Kanwal et al. 2005
, Shepherd et al. 2006
). Second, our Markov model of hepatitis B infection captures the effects of treatment with recently developed antiviral drugs. These drugs, developed as an outgrowth of advances in treatment for HIV infection, are more effective than previous treatments for hepatitis B infection, but also more expensive. Although treatment with some of the newer drugs has been shown to be cost-effective (Kanwal et al. 2005
, Rajendra and Wong 2007
, Shepherd et al. 2006
), they have not been considered in conjunction with population-based screening to identify people needing treatment.
Because our intent was to make a model that could be shared with policy makers, we developed the model in Microsoft Excel (). We incorporated sufficient detail to capture important characteristics of hepatitis B disease progression and treatment so that it would be believable to a clinical audience – but we also tried to keep the model simple enough so that it could be easily understood by non-modelers. In fact, during the course of model development we shared the model directly with the Centers for Disease Control and Prevention (CDC) and reviewed and interpreted the results with them, as described below. In developing the model we used an iterative process: we began with a very simple representation of patient health states and then added details incrementally as suggested by Dr. So, our colleagues at the CDC, and other hepatitis B experts with whom we consulted. We continued to add detail until the clinicians and hepatitis B experts were satisfied that our model was an appropriate representation of the policy problem. In this way, we kept the model as simple as possible while still capturing the salient elements of the problem.
Screenshots from the Excel model