Introduction

An estimated 326,000 people in the UK are thought to have chronic hepatitis B (CHB) 1, which can lead to cirrhosis, hepatocellular carcinoma (HCC), and death 2. Current treatment options in the UK and Europe include nucleosides (entecavir [ETV], lamivudine [LAM], and telbivudine [LdT]), nucleotides (adefovir dipivoxil [ADV] and tenofovir disoproxil fumarate [TDF]), and interferons (peginterferon-alpha-2a and interferon-alpha-2a/b).

Although interferons are effective for carefully selected patients 2, many people do not tolerate treatment 3-5. A year's treatment with peginterferon-alpha-2a leads to hepatitis B e antigen (HBeAg) seroconversion in approximately 32% of HBeAg-positive patients, and suppresses viral load to <20,000 copies/mL in around 43% of HBeAg-negative patients 6; the remaining patients are likely to require nucleos(t)ides to achieve sustained viral suppression.

Although nucleos(t)ides are well tolerated, resistance to LAM arises rapidly, with up to 70% of patients becoming resistant after 4 years of continuous therapy 7. Nevertheless, newer nucleos(t)ides (particularly TDF and ETV) are associated with substantially lower resistance rates 8-10, and are significantly more effective than LAM or ADV 11-14.

TDF is a nucleotide reverse transcriptase inhibitor with potency against hepatitis B virus (HBV) 14, including LAM-resistant HBV 15-17. In the UK, TDF was licensed for use in CHB in 2008, although it has been used to treat HIV since 2002. Recent registration randomized, controlled trials (RCTs) have shown that TDF is superior to ADV in treatment-naive patients 14, and is also effective in patients with persistent viremia during ADV therapy 18. TDF displays a favorable resistance profile: no cases of virological HBV resistance have been identified to date during intensive surveillance, which includes 8 years of clinical experience in HIV/HBV coinfected patients (who received TDF alongside other antiretroviral drugs) 16, 17, 19, 20 and up to 96 weeks of continuous use in controlled clinical trials on CHB 14, 21. Nevertheless, the cost-effectiveness of TDF in the treatment of CHB remains to be determined in a UK setting.

We set out to identify the most cost-effective first-line nucleos(t)ide treatment for CHB from the perspective of the UK National Health Service (NHS), and assess which drug(s) should be given to patients who develop resistance to first- or second-line treatment. This analysis focuses on nucleos(t)ides as they represent the most commonly used treatments for CHB in the UK.

Methods Outline of the Economic Model

A Markov model was constructed to model the progression of CHB, and the costs and benefits of nucleos(t)ide treatment, taking account of drug resistance and HBeAg-negative, as well as HBeAg-positive, CHB. Health benefits were measured in quality-adjusted life-years (QALYs), which take account of changes in both length and quality of life.

In addition to calculating the cost per QALY gained, we used the net benefit approach 22, 23 to compare the total monetary benefits of each treatment with those for all other possible treatment strategies whenever this simplified the interpretation of results. Total net benefits are calculated by multiplying the number of QALYs accrued over a lifetime by the ceiling ratio (the maximum amount that society is willing or able to pay to gain one QALY) and subtracting the total NHS costs accrued over a lifetime. The treatment with the highest total net benefit at the ceiling ratio of interest is the optimal choice for managing that population and will always correspond to the treatment that is optimal based on the cost-effectiveness ratio approach.

The National Institute for Health and Clinical Excellence (NICE) generally considers treatments costing less than £20,000 to £30,000 per QALY gained to be cost-effective 24. Net benefits were therefore calculated at ceiling ratios of £20,000 and £30,000 per QALY; results at a £10,000/QALY threshold are also presented to test the impact of using a lower ceiling ratio.

The analysis was based on a heterogeneous cohort of nucleos(t)ide-naive adults (aged ≥18 years) with compensated CHB, detectable HBV DNA, and evidence of active liver disease for whom nucleos(t)ide therapy is considered appropriate. CHB was defined as persistence of hepatitis B surface antigen (HBsAg) for at least 6 months. Patients coinfected with HIV or hepatitis C were excluded. Based on an audit conducted by the authors of this article in which anonymized hospital records of 85 HBsAg-positive adults attending a London hepatology outpatient clinic were reviewed 25, it was assumed that 5.3% of the patients were cirrhotic at baseline, and that 50% of cirrhotic patients and 55.5% of non-cirrhotic patients had HBeAg-negative CHB. Additional details on this audit are available on request.

Disease progression was modeled as movements between 18 disease states (Fig. 1). Treatments that slow disease progression were assumed to reduce disease management costs and extend patients' healthy life expectancy, and may therefore be cost-effective compared with less potent drugs. Furthermore, drug resistance may increase the risk of disease progression; subsequently, treatments with higher resistance rates may be less cost-effective. Each year, patients were assumed to move between disease states based on the transition probabilities shown in Appendix 1 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp. A lifetime time horizon was used in the model; because the mean age of patients in the London clinic audit was 38, of whom 63% were men, the model was run over a 42-year time horizon (the life expectancy of this age group 26).

Patient flow diagram for the Markov model. The model uses a 1-year cycle length, such that patients may move between disease states along the arrows shown once each year. Patients enter the model in one of the four states with a bold black border; as stated in the text, 5.3% of the patients were assumed to be cirrhotic at baseline, with 50% of cirrhotic patients and 55.5% of noncirrhotic patients having HBeAg-negative CHB. Patients in the disease states shown in white ovals will be indicated for any treatments available in the relevant treatment strategy. Although patients in the states indicated by * would not be eligible to start therapy with one or more of the agents considered in the analysis, it was assumed that treatment would not be discontinued if they entered these states after the start of treatment. Viral suppression was defined as HBV DNA <300 copies/mL occurring without seroconversion, generally as a result of antiviral medication. The liver transplant state encompassed the year of the transplant operation (from 3 months before the operation to 9 months afterward), after which time surviving patients progress to the post–liver transplant state. CC, compensated cirrhosis; CHB, chronic hepatitis B; c/ml, copies per milliliter; HBeAg, hepatitis B e antigen; HBsAg, hepatitis B surface antigen; HBV, hepatitis B virus.

As resistance to drugs such as LAM develops rapidly 7, 27, it is important to consider second-line (or third-line) treatment options that are used after resistance develops. The model allowed for switches between nucleos(t)ide therapies following development of drug resistance. Patients were assumed to receive sequences of up to three nucleos(t)ides (or nucleos(t)ide combinations) followed by best supportive care (BSC), which comprised monitoring with no antiviral therapy. The nucleos(t)ide treatments included in the model are shown in Table 1. For simplicity, only three combination therapies were included (Table 1), which represented the most plausible combinations for ADV, ETV, and TDF. It should be noted that the UK license indications for nucleos(t)ides neither specifically mention combination therapy nor advise against use in combination therapy 28-31. Nonetheless, in practice, the drug combinations listed in Table 1 may be considered clinically appropriate.

Table 1. Nucleos(t)ide treatments included in the model and their respective daily dosages and costs Drug Dose Drug cost BSC (no antiviral therapy) — £0.00/patient-day LAM (Zeffix, GSK) 100 mg/day £2.79/patient-day 5 TDF* (Viread, Gilead Sciences) 300 mg/day £8.50/patient-day 5, 66 ADV (Hepsera, Gilead Sciences) 10 mg/day £10.50/patient-day 5 TDF* plus LAM 300 mg/day TDF100 mg/day LAM £11.29/patient-day 5, 66 ETV (Baraclude, BMS) 0.5 mg/day for treatment-naive patients and 1 mg/day for patients resistant to ≥1 nucleos(t)ide £12.60/patient-day for both doses 5 ADV plus LAM 10 mg/day ADV100 mg/day LAM £13.29/patient-day 5 ETV plus ADV 10 mg/day ADV0.5 mg/day ETV for treatment-naive patients and 1 mg/day for patients resistant to ≥1 nucleos(t)ide £23.10/patient-day 5 * 300 mg tenofovir disoproxil fumarate (TDF) is equivalent to 245 mg tenofovir disoproxil (as fumarate). ADV, adefovir; BSC, best supportive care; ETV, entecavir; LAM, lamivudine; TDF, tenofovir disoproxil fumarate.

All sequences of the treatments shown in Table 1 other than those in which patients will be resistant to their third-line agent before starting that treatment were considered in the analysis, giving a total of 211 different treatment strategies for which costs and benefits were calculated. Nevertheless, for clarity, results are only presented for 20 treatment pathways that represent the most cost-effective of each of the main options; results for other strategies are available on request.

Historically, LAM followed by BSC with no antiviral treatment was the only treatment option available, and this strategy has been shown to be cost-effective 32, although it is no longer considered clinically appropriate for patients with high levels of HBV replication because potent second-line agents are now available 2, 33-35. BSC and LAM followed by BSC were therefore included in the analysis to enable assessment of whether TDF is the most cost-effective strategy out of all plausible nucleos(t)ide strategies (including those no longer used), and to ensure that TDF is compared against low-cost strategies with proven cost-effectiveness.

Interferon-alpha and peginterferon-alpha were not considered as comparators because these are given as short-term initial options in selected patients rather than maintenance treatments 2, and are used by less than 10% of patients 12. LdT was excluded because it is rarely used in the UK and is not recommended by NICE 36.

The development of drug resistance and switches between treatments were modeled by duplicating the 18 disease states shown in Figure 1, such that separate replica sets of these 18 disease states were used for first, second, and third-line treatments, fourth-line treatment with BSC, and for the year(s) in which resistance developed. Separate sets of states were also used to allow for variations in transition probabilities and resistance rates over time.

For simplicity, the risk of resistance was assumed to be the same for all disease states; the resistance rate data used in the model are described in Appendix 2. Resistance was defined as ≥ 1 log10 copies/mL rise in HBV DNA from nadir. Virological resistance was assumed to be identified an average of 1.5 months after the rise in HBV DNA, because interviews with five consultant gastroenterologists suggested that levels are monitored quarterly in UK clinical practice. During the year in which resistance developed, patients were therefore assumed to spend 10.5 months with the same risk of progression/improvement as drug-sensitive treated patients. When viral load rose at 10.5 months, patients in the “HBeAg-positive/negative viral suppression” states and the “HBeAg-positive/negative compensated cirrhosis HBV DNA <300 copies/mL” states were assumed to switch to the corresponding states for patients with detectable HBV DNA. For the remaining 1.5 months of the year, patients who developed resistance were assumed to have the transition probabilities of untreated patients, before treatment was switched to the next treatment in the pathway at the start of the next annual cycle. As there is some evidence that patients are at increased risk of hepatic flares and/or decompensation after developing LAM resistance 27, this assumption may bias the analysis slightly in favor of the treatments with high resistance rates (e.g., LAM), although this assumption is unlikely to have any significant effect on the results as patients spend only 1.5 months in this state before starting their new therapy.

The key assumptions used in the analysis are outlined below:

• 

The analysis was conducted from the perspective of the NHS 37.

• 

Costs and benefits were discounted at a rate of 3.5% per year 37.

• 

The reference year for costs was 2006/2007.

• 

A half-cycle correction was applied, whereby patients were assumed to move between states halfway through each year.

• 

It was assumed that patients who were initially infected with HBeAg-positive HBV could only develop HBeAg-negative CHB from the HBeAg-seroconverted disease state 38.

• 

It was assumed that once patients develop HBeAg-negative CHB (excluding the HBeAg-seroconverted carrier state), they could not move back to any HBeAg-positive disease state.

• 

Patients were assumed to continue treatment for an average of 5.6 (range: 0–9) months after HBsAg seroconversion, and 10.2 (range: 6–12) months after HBeAg seroconversion, based on estimates by four UK clinicians.

• 

Transition probabilities were assumed to be constant over time, except for the probability of HBeAg seroconversion, viral suppression, and reversion from decompensated to compensated cirrhosis, which were assumed to be higher in the first year of treatment than subsequent years (see Appendix 1 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp).

• 

Resistance rates were assumed to vary over the first five years of treatment with any given therapy, and remain constant at the year 5 values for all subsequent years (see Appendix 2 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp).

• 

HBeAg seroconversion was assumed to have the same outcomes regardless of whether the patient had previously been cirrhotic. Nevertheless, movement directly from the HBeAg-seroconverted state to compensated cirrhosis was permitted because this has been observed in natural history studies 32, 39-41.

• 

Because literature reviews identified no papers quantifying the costs, utilities, and transition probabilities for patients with both HCC and decompensated cirrhosis, it was assumed that patients with HCC could not undergo hepatic decompensation and that preexisting hepatic decompensation did not affect outcomes, costs, or quality of life in HCC; this simplification is unlikely to affect the results as only 1.2% to 1.6% of life-years experienced by the cohort are spent in the HCC state.

• 

In cases where two nucleos(t)ides were used in combination, it was assumed (given a shortage of published data) that the probability of viral suppression or HBeAg seroconversion with any combination therapy was equal to that of the most effective component of that combination, because RCTs conducted to date do not suggest that combination therapy increases viral suppression 18, 42, 43.

• 

It was conservatively assumed that nucleos(t)ide treatment does not affect the probability of HBsAg seroconversion because few trials identified in a systematic review 11 reported data on HBsAg seroconversion. Nevertheless, this assumption may bias the analysis against TDF, which has a higher incidence of HBsAg seroconversion than ADV in HBeAg-positive patients 44.

• 

For the decompensated cirrhosis, liver transplant, and post–liver transplant disease states, data on total mortality (including deaths from causes other than hepatitis B) were used in the model, and no all-cause mortality was applied; for other disease states, the annual risk of all cause mortality 26 was added to the excess mortality associated with the disease state in question to give the total risk of death each year (see Appendix 1 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp).

• 

Nucleos(t)ides were conservatively assumed to have no impact on mortality in patients with HCC or compensated cirrhosis.

• 

The (generally mild 28-31) adverse events associated with nucleos(t)ides were assumed to have no effect on costs, mortality, or quality of life, although the cost of renal monitoring was included in the analysis (Table 2). The cost of osteopenia monitoring was excluded as this is not routinely conducted in UK clinical practice 28-31.

Table 2. Frequency and cost of consultations for patients in precirrhotic disease states Disease state Mean (range) no. consultations per year Cost per consultation* (range) Active CHB or viral suppression (treated) 3.3 (2, 4) £114.69 (£46.23, £236.95) Active CHB or viral suppression (untreated) 2.8 (1, 4) £121.21 (£16.12, £251.26) HBeAg seroconverted 2.0 (1, 4) £121.21 (£16.12, £251.26) HBsAg seroconverted 0.03 (0, 1) £121.21 (£16.12, £251.26) Number of additional consultations required in the year when resistance develops† 1.17 (0.5, 2) £114.69 (£46.23, £236.95) Number of additional consultations required in year 1 (excluding the consultation in which treatment is initiated) 1 (1, 1) £240.60 (£6.25, £742.04) * The cost per consultation includes the cost of staff, clinic overheads, and laboratory tests such as full blood count, liver function profile, and HBV DNA quantification by polymerase chain reaction (PCR). All treated patients were assumed to receive quarterly renal monitoring (urea and electrolytes) and, where appropriate, testing for the presence of HBeAg, HBe antibody, and/or HBsAg. The total cost of tests/investigations was calculated by multiplying the unit cost per test by the proportion of consultations in which particular tests are conducted (which was estimated by clinicians) and summing across all tests. In line with its product license 29, TDF-treated patients were assumed to receive renal monitoring (assumed to comprise urea and electrolyte tests on blood samples taken in practice nurse consultations) every 4 weeks in their first year of treatment, rather than the quarterly monitoring assumed for all other treated patients. It was assumed that patients would not receive bone scans, because no clinicians interviewed felt that these would be conducted routinely. † In addition to the cost of a consultation, 33% (range: 0–100%) of patients developing drug resistance were assumed to receive HBV sequencing in the year resistance developed. The numbers of consultations/patient-year are based on the mean (and range) of estimates from five gastroenterologists working in the UK. Full details of the unit costs and quantities used in the analysis are available on request. CHB, chronic hepatitis B; HBeAg, hepatitis B e antigen; HBsAg, hepatitis B surface antigen; HBV, hepatitis B virus; TDF, tenofovir disoproxil fumarate.

Further technical details about the model are available from the authors on request.

Data Sources Used in the Analysis

The majority of model inputs were based on studies identified in a systematic literature review of all nucleos(t)ide therapies 11. Additional literature searches were conducted to identify studies on the natural history of CHB.

All studies identified in the systematic review that reported the incidence of drug resistance over at least 1 year's follow-up were pooled to generate overall estimates of the annual risk of developing resistance to each medication (see Appendix 2 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp). Trials on LAM-resistant patients 15, 45-52 were analyzed separately from those on nucleos(t)ide-naive patients (see Appendix 2). Nevertheless, resistance rates calculated from studies on LAM-resistant patients were also applied to patients who were resistant to nucleos(t)ides other than LAM.

Although no cases of virological resistance to TDF have been observed to date, experience with older nucleos(t)ides suggests that drug resistance may eventually be observed. To enable calculation of resistance rates for TDF and for other drugs where no resistance was observed in any particular year, it was assumed that the next patient to be treated and monitored would develop virological resistance. For example, because 0% (0/130) of LAM-resistant patients receiving TDF at year 1 developed resistance (see Appendix 2), the highest that the incidence of resistance with TDF can be is 0.76% (1/131). The resistance rates calculated in this way therefore represent the maximum rates that we can expect to see given available evidence and are subsequently likely to overestimate the actual risk of resistance.

Transition probabilities for the key transitions that differ between active treatments (the probability of achieving undetectable HBV DNA or HBeAg seroconversion with each nucleos(t)ide therapy or nucleos(t)ide combination) were based on a mixed treatment comparison meta-analysis conducted by the authors, which is described in the accompanying paper 11. Transition probabilities for untreated patients were based on data from natural history studies, economic evaluations, or the placebo arms of meta-analyses or RCTs (see Appendix 1 at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Dakin.asp).

For some of the transitions that may be influenced by treatment (predominantly those affecting patients with severe liver disease), data were only available for ADV or LAM. In these cases, all treated patients were assumed to have the same chance of improvement/progression regardless of which nucleos(t)ide was used.

Because the cost of managing severe liver disease is likely to differ little between hepatitis B and C, costs for the compensated cirrhosis, decompensated cirrhosis, HCC, liver transplant, and post–liver transplant disease states were based on large, retrospective UK microcosting studies on patients with hepatitis C 53-55 (Table 3). Costs were inflated to 2006/2007 values 56, and the cost of hepatitis B immunoglobulin was included in the cost of liver transplantation and posttransplant follow-up. The cost of nucleos(t)ide therapy was also added where applicable.

Table 3. Disease management costs for severe liver disease Disease state No. pts included in mean cost Cost per patient or patient-year (inflated to 2006/2007 values using HCHS 56) Mean cost Lower 95% CI Upper 95% CI Compensated cirrhosis: cost/patient-year 55 115 £1,341 £807 £1,876 Decompensated cirrhosis: cost/patient-year 55 40 £10,750 £7,240 £14,261 HCC: cost/patient-year 55 20 £9,580 £5,167 £13,992  Liver transplant: cost/patient for waiting list phase lasting 3 months 55 67 £4,393 £2,604 £6,182  Liver transplant: cost of transplant operation per patient (excluding HBIg) 55 67 £32,215 £25,550 £38,880  Liver transplant: cost/patient for first 8 months' posttransplant follow-up (excluding HBIg) 55 67 £7,432 £3,508 £11,357  Liver transplant: cost/patient for HBIg during year of transplant* — £16,250 £13,750 £18,750 Total cost/patient for liver transplant (over 12 months) — £60,291 — —  Post–liver transplant: cost per patient-year (excluding HBIg) 55 67 £1,633 £812 £2,453  Post–liver transplant: HBIg in posttransplant: cost per patient-year* — £5,000 — — Total cost of posttransplant state per patient-year — £6,333 — — * Cost of HBIg was based on personal communications with a UK transplant center. CI, confidence interval; HCHS, Hospital and Community Health Services; HCC, hepatocellular carcinoma; HBIg, hepatitis B immunoglobulin.

The cost of managing patients in other disease states was based on clinicians' estimates of the frequency of outpatient consultations for each patient group and the tests that would be performed at each consultation (Table 2), which were derived from telephone interviews with five consultant gastroenterologists. The resources used in each consultation were valued, based on published sources 56, 57 and provider tariffs, to produce the total consultation costs shown in Table 2. Further details on unit costs and resource use quantities are available on request.

Health state preference values or utilities for most disease states were based on standard gamble valuations of each health state from a study involving 93 UK patients with CHB 58, 59 (Table 4). It was conservatively assumed that achieving undetectable HBV DNA (in the absence of seroconversion) would not improve quality of life. The quality of life of HBsAg-seroconverted patients was based on UK population norms 60; however, the utility of patients in the HBeAg-seroconverted state was assumed to be 1% lower than populations norms, based on a previous economic evaluation 61.

Table 4. Utilities used in the economic evaluation State Mean Lower 95% CI Upper 95% CI Reference HBsAg seroconverted 0.86 0.85 0.87 Age-dependent population norm for all ages, based on a representative sample of 3395 members of the UK general population 60 HBeAg seroconverted 0.85 — — Age-dependent population norm multiplied by 0.99 (an adjustment for presence of HBsAg based on Wong et al. 61). The quality of life of the general population was varied over its 95% CI, while the disutility associated with detectable HBsAg was varied between 0% and 15% in sensitivity analyses, but was not varied in PSA. Active CHB 0.77 0.71 0.81 Ossa et al. 58* Viral suppression 0.77 0.71 0.81 Assumed to be the same as for active CHB Compensated cirrhosis, HBV DNA-negative 0.73 0.65 0.77 Ossa et al. 58* Compensated cirrhosis, HBV DNA-positive 0.73 0.65 0.77 Assumed to be the same as for compensated cirrhosis with detectable HBV DNA Decompensated cirrhosis 0.34 0.25 0.39 Ossa et al. 58* HCC 0.36 0.28 0.41 Ossa et al. 58* Liver transplant 0.56 0.49 0.62 Ossa et al. 58* Posttransplant 0.67 0.59 0.73 Ossa et al. 58* * Utilities from the study by Ossa et al. comprise standard gamble valuations by a sample of 93 UK patients with CHB 58, 59. CHB, chronic hepatitis B; CI, confidence interval; HBeAg, hepatitis B e antigen; HBsAg, hepatitis B surface antigen; HBV, hepatitis B virus; HCC, hepatocellular carcinoma.

All model parameters other than unit costs were varied independently over the range of values that they could plausibly take in one-way sensitivity analyses. We also conducted probabilistic sensitivity analysis 62, in which all model parameters except unit costs were varied over distributions defined by the mean and 95% confidence intervals shown in Tables 2–4 and Appendices 1–2. Costs and relative risks were assumed to follow gamma distributions, while utilities and probabilities were assigned beta distributions, in line with best practice 62, 63