Peginterferon alfa and ribavirin for chronic hepatitis C in patients eligible for shortened treatment, re-treatment or in HCV/HIV co-infection: a systematic review and economic evaluation

Prevalence

The estimated global prevalence of chronic HCV is around 2%–3%, corresponding to about 130–170 million people. 1,8 In England and Wales, the HPA3 estimates that, based on statistical model data for the year 2003, around 191,000 [95% credible interval (CrI) 124.00 to 311.00] people aged 15–59 years are HCV antibody positive, with 142,000 people chronically infected, a prevalence of 0.44% (95% CrI 0.29% to 0.72%) in this age group.

Prevalence estimates vary geographically in England and Wales, with highest numbers of laboratory reports (from public health and UK NHS laboratories in England and Wales under a voluntary surveillance scheme) returned in the north-west, followed by London and the south-east of England. 3

The prevalence of chronic HCV also varies according to different population groups. For example, HCV is more common in men and in the 25–44 years age group. Estimates of the number of current injecting drug users (IDUs) in England vary between 100,000 and 217,000, and it is estimated that around 40% of IDUs are infected with chronic HCV, based on the Unlinked Anonymous Prevalence Monitoring Programme’s Survey of Injecting Drug Users in 2006. 9 There are limited data on prevalence in minority ethnic populations. However, it is thought that the prevalence of HCV is higher in migrants who will have acquired the infection while overseas, notably Pakistan. 3

Evidence suggests varying rates of HCV in people with HIV infection. For example, Mohsen and colleagues10 reviewed the international literature on the epidemiology of HCV/HIV co-infected patients. They included 12 HCV seroprevalence studies carried out in people infected with HIV-1 in Europe and the USA. HCV prevalence ranged from 7% to 57%, largely influenced by risk factors in the study populations. Prevalence was highest in people with a history of injecting drug use (> 80%). It has been suggested that up to 10% of all HCV-infected people are co-infected with HIV. 11

Prevalence is difficult to estimate because symptoms of HCV are frequently absent or non-specific and thus people can remain undiagnosed for many years. Between 1992 and 2007 there were 62,000 laboratory-confirmed diagnoses of HCV in England, and 3688 in Wales (from 1996). 3 It is thought that a proportion of those who are undiagnosed are ex-IDUs who used drugs transiently in the past. Sentinel surveillance by the HPA suggests that the number of people diagnosed with HCV in all settings is increasing, which may, in part, reflect awareness-raising campaigns to encourage uptake of testing. 3

Incidence

The incidence of chronic HCV is likely to be driven by two main sources – newly acquired infections in current UK residents (largely IDUs) and inward migration of chronically infected individuals from other countries. Up-to-date estimates of overall incidence are not yet available but recent studies in IDUs suggest that 3%–42% of susceptible injectors become infected each year. 3 The HPA reports that the number of laboratory-confirmed diagnoses of HCV in England and Wales in 2007 was 7540, representing a 12% increase from 2006. 3 This does not, however, necessarily represent an increase in rates of incidence but may be attributed to testing rates.

Recent rises in HCV infection in HIV-positive MSM have generated increased interest in the role of sexual transmission of HCV. HCV RNA can be detected in the semen of HCV-infected men, with higher levels in HIV-positive men, suggesting the possibility of transmission during certain sexual practices. Increases in cases of acute co-infection in HIV-positive MSM in urban centres in Europe and the USA have been reported in recent years. 12 A study of genitourinary medicine (GUM) clinics in London and the south-east of England found a 20% average annual increase in the number of HIV-positive MSM diagnosed with HCV between January 2002 and June 2006. 7 The prevalence of HCV in HIV-positive MSM is estimated to be between 4% and 11.5%. 12

Study design

Randomised controlled trials were included for the clinical effectiveness review. Trials published as abstracts or conference presentations from 2007 onwards were included only if sufficient details were presented to allow an appraisal of the methodology and the assessment of results to be undertaken. Systematic reviews were used only as a source of references. For the systematic review of cost-effectiveness, studies were eligible for inclusion if they reported the results of full economic evaluations [cost-effectiveness analyses (reporting cost per life-year gained), cost–utility analyses or cost–benefit analyses]. For studies reporting QoL and epidemiology/natural history, a range of study designs were eligible (e.g. cohort studies, cross-sectional surveys).

Interventions

• Combination therapy comprising ribavirin and either peginterferon alfa-2a or peginterferon alfa-2b.

• Peginterferon alfa-2a or peginterferon alfa-2b monotherapy (for patients who are unable to tolerate or are contraindicated to ribavirin).

Comparators

For patients who have been previously treated with combination therapy, and for HCV/HIV co-infected patients:

• BSC (e.g. symptomatic treatment, monitoring, treatment without any form of interferon therapy).

For patients who meet the criteria for receiving shortened courses of combination therapy:

• standard-duration courses of peginterferon alfa and ribavirin combination therapy (up to 24 or 48 weeks, as appropriate).

Population

Adults with chronic HCV, restricted to people who:

• have been previously treated with peginterferon alfa and ribavirin in combination but who relapsed/did not respond

• have HCV/HIV co-infection

• meet the criteria within the marketing authorisation for receiving shortened courses of peginterferon alfa and ribavirin in combination, namely patients with:

  1. – genotype 2 or 3 with LVL* at the start of treatment and an RVR (defined as HCV RNA undetectable by week 4) – shortened course of 16 weeks†

  2. – genotype 1 with LVL* and an RVR (defined as HCV RNA undetectable by week 4 and at week 24) – shortened course of 24 weeks

  3. – genotype 4 with an RVR (defined as HCV RNA undetectable by week 4 and at week 24) – shortened course of 24 weeks.†

(*For peginterferon alfa-2a, LVL is defined as ≤ 800,000 IU/ml;42 for peginterferon alfa-2b, LVL is defined as ≤ 600,000 IU/ml. 43 †Applies only to peginterferon alfa-2a.)

Outcomes

Studies had to report SVR (defined as undetectable HCV RNA at least 6 months after treatment cessation). The following outcomes were also included:

• virological response (e.g. during treatment)

• biochemical response (e.g. ALT levels)

• histological improvement (fibrosis and inflammation)

• survival

• adverse effects of treatment

• HRQoL

• cost-effectiveness (incremental cost per life-year gained) or cost–utility [incremental cost per quality-adjusted life-year (QALY) gained].

Sustained virological response

Sustained virological response was defined as undetectable serum HCV RNA (< 50 IU/ml,52,56 25 IU/ml,53 < 5.3 IU/ml59) at the end of 24 weeks’ follow-up in four trials, and as HCV RNA negative (< 50 IU/ml) at the end of treatment and end of follow-up in two trials. 54,55

Sustained virological response was the primary outcome in all six included RCTs. Four of the trials52–54,56 separately reported SVR in the subgroup of patients who achieved an RVR and had LVL at baseline, which is the patient subgroup meeting the licensed criteria for receiving shortened courses of combination therapy (Table 3). Yu and colleagues55 reported SVR for patients who achieved an RVR, but did not further stratify this subset by baseline viral load. However, it can be assumed that rates would be similar to SVR by RVR rates, as the mean baseline viral load was low for both treatment arms, and approximately 83% of the study population had LVL at baseline (< 800,000 IU/ml). Although the trial by Berg and colleagues59 reported SVR in the subgroup of patients who achieved an RVR and had LVL at baseline, the threshold used was either ≤ 800,000 IU/ml or > 800,000 IU/ml, which differs from the threshold of < 600,000 IU/ml specified in the SPC for the study drug peginterferon alfa-2b. 43 For this reason we do not present the results for this subgroup, but instead present the SVRs for the subgroup that achieved an RVR irrespective of the baseline viral load. As the mean viral load for the study sample, as a whole, was log₁₀ 5.7 IU/ml (calculated to be around 500,000 IU/ml), these SVRs can be considered, overall, to reflect LVL in accordance with the SPC.

Results for SVR for treatment groups as a whole, SVR by RVR, and SVR by viral load can be seen in the data extraction forms in Appendix 6.

In patients with LVL (≤ 800,000 IU/ml) who attained an RVR, SVR rates were comparable between groups who received the standard duration of treatment and those who received shortened courses, for both genotype 1 and genotypes 2 and 3. Rates were similar in five trials, ranging from 83% to 100% for standard treatment duration compared with 84%–96% for shortened treatment duration, with no statistically significant differences between treatment arms. In addition, SVRs were broadly similar regardless of genotype with the exception of the trial by Berg and colleagues,59 in which SVRs were lower than in the other studies. This may be due to the fact that these rates are only for those who first became HCV RNA negative at week 4 and do not include those who became HCV RNA negative during weeks 1–3 (as a consequence of the study design), whereas in all of the other trials the rates reflect all patients who became negative up to week 4. It should also be noted that patient numbers in these subgroups were small, and none of the trials was powered for this subgroup analysis. In the trial by Mangia and colleagues,52 in particular, only 10% of patients had an RVR and LVL.

For those with high baseline viral load, lower SVR rates were observed in patients who were treated for a shorter duration, although this was reported to be statistically significant in only two trials (100% vs 92%, p = 0.03, at < 1,000,000 IU/ml;53 100% vs 76.5% , p = 0.045, at ≥ 400,000 IU/ml54 for standard vs shortened treatment, respectively).

Virological response during treatment

The included trials varied in their lower limits of detection, with RVR defined as undetectable serum HCV RNA (< 25 IU/ml),53 serum HCV RNA negative (< 50 IU/ml),52,54,55 serum HCV RNA < 600 IU/ml56 or < 615 IU/ml,59 all at week 4 of therapy.

Table 4 presents RVR rates for each of the six included RCTs. There were no statistically significant differences between treatment groups who received the standard duration of treatment compared with those who received shortened courses, for both genotype 1 and genotypes 2 and 3.

There was a large range in reported RVR between the studies, with rates in genotype 1 patients generally being lower than in genotype 2/3 patients. In the four genotype 1 trials,52–54,59 26%–68% of patients achieved an RVR, although in the subset of patients treated for 24 weeks in the Mangia and colleagues trial52 all of the patients achieved an RVR as per the study design (see Description of the included trials). The rates in this trial were lower than in the other five trials, and this may be due to the smaller proportion of patients (24%) having LVL at baseline. In the trial of genotype 2 patients by Yu and colleagues,55 rates were much higher at 86%. In the study of genotype 2/3 patients,56 two of the three treatment arms had RVR rates of 100% owing to the nature of the study design, whereby patients who achieved an RVR at week 4 were randomised (at week 8) to a total of 16 or 24 weeks’ treatment. In the trial by Mangia and colleagues,52 it is also reported that RVR rates were not significantly different between those treated with peginterferon alfa-2a compared with peginterferon alfa-2b (24% vs 29%, respectively, p = 0.14) (see Appendix 6), although results were not reported for the different treatment arms for the two peginterferons.

Early virological response rates and EOT response rates were similar for patients receiving shortened and standard duration treatment, with no statistically significant differences (where significance values were reported). As these results were presented for all patients rather than the subgroup of patients with RVR and LVL of interest to this systematic review, we have not presented these data here. However, for information they can be found in Appendix 6.

Relapse rate

Relapse was defined as the re-appearance of serum HCV RNA during the 24-week follow-up period in patients who achieved an EOT response. The RCT by Yu and colleagues54 was the only included trial to report the relapse rate in the subgroup of patients with an RVR and LVL (Table 5). In this subgroup, rates of relapse were low and were not statistically significantly different between treatment arms [3.6% vs 0% for 24 vs 48 weeks, respectively, difference 3.6%, 95% confidence interval (CI) –7.2 to 6.6, p = 1.000]. In those with an RVR and high viral load, shortening the duration of therapy resulted in higher rates of relapse, reaching statistical significance (23.5% vs 0 for 24 weeks vs 48 weeks, respectively, p = 0.045).

Relapse rates for the other included RCTs can be found in Appendix 6. These have not been presented here because they were reported for the study groups as a whole rather than the subgroup of patients of relevance to this systematic review (i.e. those with both LVL and an RVR).

Biochemical response

Two RCTs reported biochemical response rate (normalisation of ALT levels) (Table 6). 53,56 In one trial of genotype 1 patients (Liu and colleagues53), data were analysed for 248 patients with available paired ALT levels (baseline and end of follow-up). Treatment for 24 weeks resulted in a lower ALT normalisation rate compared with 48 weeks of treatment, with the difference being statistically significant (51% vs 72%, respectively, p < 0.001). However, the study did not report the response rate for the subgroup of patients with an RVR or RVR and LVL. In the trial of genotype 2/3 patients (von Wagner and colleagues56) there was no statistically significant difference in sustained biochemical response rates between groups who achieved an RVR.

Histological response

Histological response was reported by one trial in patients with genotype 1 HCV (Liu and colleagues53), and was analysed for 295 patients with available paired liver biopsy specimens (baseline and end of follow-up). However, the numbers in each treatment arm were not reported by the authors. Patients who received the shortened treatment regimen had a significantly lower histological response than those treated for the standard duration of 48 weeks (59% vs 78%, respectively, p = 0.001). Again, the study did not report the response rate specifically for the subgroup of patients with an RVR or RVR and LVL (Table 7).

Adverse events

Adverse events for the included studies are presented in Table 8. All of the trials presented adverse events for treatment groups as a whole, not for the subgroup of patients achieving an RVR and with LVL.

The incidence of dose discontinuations as a result of adverse events was reported by all six RCTs and was low across treatment groups, ranging from 0% to 9%. For three trials (all genotype 153,54,59) there appeared to be fewer discontinuations due to adverse events in those patients treated for a shortened duration, although this was not significantly different in one trial (p = 0.10)53 and not statistically tested in the other two. 54,59 However, Yu and colleagues54 found a statistically significant difference in the total incidence of treatment discontinuations (for adverse events and other reasons combined) in favour of the shortened treatment regimen (10% vs 3%, p = 0.045). In the trial by von Wagner and colleagues56 in genotype 2/3 patients, the total incidence of treatment discontinuations also appeared to favour the shortened-treatment-duration group (1% vs 8% for 16 weeks vs 24 weeks, respectively), although this was not statistically tested.

For four trials,54–56,59 the incidence of drug dose modifications for adverse events/laboratory abnormalities [classified by the studies as either peginterferon alfa-2a, RBV (ribavirin), peginterferon alfa-2a or RBV, or peginterferon alfa-2b or RBV] was observed to be lower in patients treated for a shortened duration, as might be expected, although the differences were not statistically significant54,55 or not tested. 56,59 The same trend was observed in the two trials that presented the incidence of drug dose reductions for any adverse event, but differences between treatment arms were not statistically tested. 52,53

The incidence of serious adverse events was low (range 0%–7%) as reported by five trials. 53–56,59 Frequencies were not different between treatment arms although statistical tests were generally not reported. In two trials54,56 it is not clear whether the events were related to treatment or not, although in one trial53 12 of the 15 events were considered to be treatment related (3 out of 4 vs 9 out of 11 in 48 weeks vs 24 weeks, respectively, p = 0.11). von Wagner and colleagues56 did not differentiate between the three treatment groups when reporting this outcome, so the proportion in each group is unknown. Only one death was reported,53 which was due to reactivation of pulmonary tuberculosis in a patient with a history of pulmonary tuberculosis and prolonged fever, dyspnoea and weight loss.

All of the trials reported the frequency of specific adverse events (see full data extractions in Appendix 6 for more details) with the exception of Berg and colleagues,59 who did not present the data. Most adverse events reported were typical of those commonly associated with peginterferon-based treatment. The most frequently occurring adverse events were similar across trials and included influenza-like symptoms, such as headache, fatigue and fever, insomnia, anorexia, dermatological symptoms such as skin rash/dry skin and alopecia. On the whole, the frequency of adverse events was not statistically different between treatment arms, although in three studies53,54,56 there was a trend for a lower incidence of events in patients treated for a shorter duration. Two trials54,55 reported statistical tests for comparison between groups for all the reported adverse events, two trials reported statistical comparisons for some adverse events,52,53 while two trials56,59 presented no statistical comparison between treatment groups. Liu and colleagues53 found that body weight loss (weight reduction of > 10% from baseline weight) was encountered less frequently in those receiving treatment for 24 weeks than in those receiving treatment for 48 weeks (19% vs 30%, respectively, p = 0.03). In the trial by Yu and colleagues,55 the incidence of alopecia was significantly lower in the 16-week group than in the 24-week group (20% vs 49%, respectively, p = 0.001).

Clinical effectiveness: summary

• All six included RCTs were in patients who were eligible for shortened treatment duration. No RCTs comparing peginterferon alfa with or without ribavirin with BSC were identified for the HCV/HIV co-infection or re-treatment patient groups.

• In the subgroup of patients who achieved an RVR and had LVL at baseline, SVR rates were comparable (i.e. no statistically significant differences) between groups who received the standard duration of treatment and those who received shortened courses, for both genotype 1 and genotypes 2 and 3. This implies that this patient group can receive shortened courses of peginterferon combination therapy without compromising SVR rates.

• For both genotype 1 and genotype 2 and 3 patients, there were no statistically significant differences in rates of RVR between treatment groups who received the standard duration of treatment and those who received shortened courses. Rates of RVR in genotype 2/3 patients were observed to be generally higher than in genotype 1 patients.

• Relapse rates in the subgroup of patients with LVL and RVR (one trial) were low and not significantly different between those treated for 24 versus 48 weeks.

• Treatment for 24 weeks resulted in a significantly lower biochemical response rate (reduction of ALT to normal levels) and histological response rate than 48 weeks of treatment in one trial of genotype 1 patients. Shortening the treatment duration had no effect on biochemical response in one trial of genotype 2/3 patients. Rates of biochemical and histological response should be treated with caution, as the results relate only to those patients with available data and rates were not reported in the subgroup of patients with LVL and RVR.

• Adverse events were presented for treatment groups as a whole and the reporting of statistical tests varied. However, the most frequently occurring adverse events were similar across all the trials and included flu-like symptoms, insomnia, anorexia, dermatological symptoms and alopecia.

• There was a trend for a lower incidence of adverse events in patients who were treated for a shorter duration (three trials), although statistically they were comparable between treatment arms. The incidence of dose discontinuations was significantly lower in those receiving a shortened treatment regimen in one trial.

• None of the studies was powered for subgroup analysis and therefore the results should be interpreted with caution.

Relevance of the studies to the UK

The patient group in the model is similar to one of those currently of interest in this appraisal – patients co-infected with HIV. However, the Kuehne and colleagues study64 focuses on patients with moderate HCV liver-related disease, whereas the current NICE guidance covers patients with moderate to severe38 and mild chronic HCV. 33 The US health system, in which both of the studies are based, is not comparable to the UK NHS, and this will therefore extend to the costs incurred within it.

Sensitivity analyses

The authors of both studies report that the results are sensitive to the discount rate. In Kuehne and colleagues,64 this is a variable to which the results appear most sensitive; however, the discount rate applied in the base-case analysis was not reported. Campos and colleagues65 describe their results as sensitive to the discount rate: a 0% rate resulted in an ICER 60% lower than the base case, whereas a 5% discount rate resulted in an ICER of 140% higher than the base case.

The results in both the included studies are sensitive to the fibrosis progression rates. In a two-way sensitivity analysis with the effectiveness of combination peginterferon alfa plus ribavirin and disease progression, Campos and colleagues65 reported that cost-effectiveness ratios were < US$50,000 per YLS, regardless of fibrosis, when treatment efficacy exceeded 50%. This is higher than the base-case treatment efficacy of 40%. Where treatment efficacy was < 25%, cost-effectiveness ratios were < US$100,000 across the range of RRs, although these are described as having had ‘slightly more influence’ (p. 277). 65 No further details of how, or the degree to which, the RR is influential are reported. Kuehne and colleagues64 reported that in mild HCV and peginterferon plus ribavirin the difference in their ICER from the base case was largest when the RR was between 1 and 2, with less sensitivity to RRs > 3. Changes in the RR of progression had a greater effect on the ICER in patients with mild HCV than in those with moderate HCV.

The order of the strategies described in the study by Campos and colleagues65 remained the same when it was assumed that treatment was discontinued in the absence of an EVR – US$59,300 per YLS for men in genotype 1 versus US$33,100 per YLS for men in genotype non-1; the results in women again reflected this. The results were reported by the authors as being most sensitive to variation in the annual excess death rate owing to HIV, fibrosis progression rates and treatment efficacies in non-cirrhotic patients, and as being ‘moderately’ sensitive to drug costs. None of these was reported in detail across the patient subgroups or intervention strategies. Where no discount rate was applied this resulted in an ICER that was 60% lower than the base-case analysis; a 5% discount rate saw the ICER increase to 140% higher. The cost of the peginterferon alfa and ribavirin strategy was varied from 50% to 150% of the base-case value, which resulted in ICERs of between US$56,300 and US$88,000 per YLS, respectively. The variation in death rate owing to HIV was illustrated by an example of the excess mortality being reduced by 97%, reducing the ICER to US$41,000 per YLS. No justification for this reduction is described. However, where this was increased 11-fold to reflect death rates in patients with a history of severe opportunistic infections, treatment is dominated by non-treatment. No results are reported for the fibrosis progression rates or treatment efficacy one-way analyses.

Campos and colleagues65 further describe a two-way sensitivity analysis whereby the effectiveness of the combination therapy of peginterferon alfa plus ribavirin and the RR of fibrosis progression due to co-infection were varied. Where efficacy was increased by 50%, the cost-effectiveness ratios decreased to < US$50,000 per YLS, and this was not sensitive to the variation in fibrosis progression. Where efficacy was decreased by 25% the cost-effectiveness ratios decreased to < US$100,000 per YLS. The authors state that this was ‘slightly more’ sensitive to fibrosis progression.

Kuehne and colleagues64 performed a number of one-way sensitivity analyses in patients who were receiving interferon alfa plus ribavirin and peginterferon plus ribavirin. The ICERs were found to be most sensitive to the RR of progression to cirrhosis compared with mono-infected patients. In the figure in the study publication, the ICER appears most sensitive to the discount rate, HAART efficacy, relapse after a sustained response and cost of ribavirin in patients receiving interferon alfa combination therapy for 48 weeks compared with 24 weeks, as well as discount rate, relapse rate and HAART efficacy in peginterferon alfa plus ribavirin compared with interferon alfa plus ribavirin for 48 weeks. Minor adverse events are reported as having little impact on the ICERs, while major toxicity in 20% of patients receiving 48 weeks of peginterferon combination therapy increased this ICER from US$40,000 to US$69,000. Decreasing utility estimates by 10% during 48 weeks of therapy led to this strategy dominating the non-peginterferon-based treatments.

The authors further reported that the ICER was minimally sensitive to minor toxic effects, with no further details. Major toxicity could affect the effectiveness of HAART in this group, and a sensitivity analysis was undertaken: if the effectiveness of HAART was reduced by 50% in 20% of the patients receiving peginterferon and ribavirin, the ICER increased from US$40,600 to US$69,000 per QALY.

Overview

The Roche submission to NICE in support of peginterferon alfa-2a consists of a 226-page written document (containing submitted evidence on the clinical effectiveness and a cost-effectiveness analysis) and a fully executable, electronic copy of the manufacturer’s economic model. The MS reports cost-effectiveness results for the three populations covered by the scope for this NICE appraisal:

• Patients who have previously been treated with peginterferon alfa, including both those who did not respond to previous treatment (by viral genotype) and those who relapsed on previous treatment. Costs and outcomes for these patients are compared with supportive care, in line with the scope for this appraisal.

• Patients with LVL and RVR who receive shortened courses of treatment with peginterferon alfa (by viral genotype). Costs and outcomes for these patients are compared with treatment for the same group of patients receiving the standard duration of treatment, in line with the scope for this appraisal.

• Patients co-infected with HCV/HIV. Costs and outcomes for these patients are compared with treatment for the same group of patients receiving non-peginterferon alfa therapy, which is not consistent with the scope for this appraisal.

The perspective of the analysis is not stated, but appears to be consistent with the NICE reference case68 of the NHS and Personal Social Services (PSS), capturing direct costs and benefits only. The submission reports lifetime costs and outcomes (reported as life expectancy and QALYs) for each treatment arm and the incremental costs and outcomes for peginterferon alfa-2a combined with ribavirin compared with usual care (which varies between patient populations, as stated above).

Below we outline the approach taken by the manufacturer and provide an outline review based on a checklist suggested for the critical appraisal of cost-effectiveness analysis by Drummond and colleagues,67 the requirements of NICE for submissions on cost-effectiveness (reference case),68 and a suggested guideline for good practice in decision modelling by Philips and colleagues. 69

Modelling approach

The cost-effectiveness analysis model adopted for the MS is a state-transition model that is structurally similar to published models previously used in the population of patients with chronic HCV,76–81 including our previous assessment report17 for NICE (TA106). The model has a lifetime horizon (in the base-case analysis the cohort simulation is truncated at patient age of 99 years), with a cycle length of 1 year, and is used to estimate the morbidity and cost resulting from progressive liver disease and treatment costs (up to a maximum duration of treatment with peginterferon alfa-2a of 72 weeks). The model has five health states indicating progressive liver disease (HCV, compensated cirrhosis, decompensated cirrhosis, HCC and liver transplantation), one state representing a treatment response (SVR) and one absorbing state (death), although this last state is broken down to differentiate deaths from progressive liver disease and deaths from all other causes. Unlike the model adopted in our previous assessment,17 the model developed for the MS does not distinguish the stage of liver disease in non-cirrhotic patients with chronic HCV (i.e. there is no distinction between mild and moderate HCV). The impact of this structural assumption is not discussed in the MS.

The main treatment effect applied in the model is the SVR for treated patients, with the proportion of patients in each of the modelled populations achieving an SVR based on data from clinical trials conducted in the relevant patient populations, reported in the MS (discussed in Data inputs; see also Appendix 3). Patients who achieve an SVR are assumed in the model to be ‘cured’ and do not face any risk of reactivation of disease or any excess risk of progressive liver disease (above that of a general population). Age-specific mortality risks for the general population, weighted for the proportion of men in the baseline cohort, are applied to patients achieving an SVR. Patients who do not achieve an SVR are at risk of progressive liver disease and are assumed to face the same risks of disease progression as untreated patients. Risks of disease progression and, where relevant, excess mortality risks associated with advanced liver disease states in the model have been drawn from natural history studies.

The base-case population in each analysis is the same, with all patients entering the model being non-cirrhotic, with chronic HCV. The simulated patient cohort has a mean age of 45 years, with 70% being male. These assumptions have no impact on response to treatment (i.e. SVRs in the model are not broken down by age or sex), but affect the all-cause mortality rates applied in the model. Patient weight is assumed to be greater than 75 kg – again this has no impact on the patient response to treatment, but has an impact on the cost of treatment, as ribavirin dosage is weight related. The MS discusses these assumptions in relation to the characteristics of patients recruited to the clinical trials used to estimate the SVRs applied in the model. However, there is no discussion of the relevance of these characteristics to the population of UK patients with chronic HCV or in the modelled populations.

Health-state utilities applied to the chronic HCV and progressive liver disease states in the model were taken from the UK Mild Hepatitis C Trial. 82 Age-specific utility values [reported for a general population survey using the European Quality of Life-5 Dimensions (EQ-5D83) and valued using a UK general population tariff84] were applied for only the SVR state. The MS does not discuss the possible implications of using age-specific utility values for one state and not for others (discussed in Data inputs). The model does not include treatment-related adverse events, other than to reduce utility in the year of treatment by 0.11 (from 0.66 to 0.55).

The costs applied in the submission were made up of two components. Treatment-related costs (which for peginterferon alfa-2a combination therapy consist of drug acquisition costs, monitoring of patients on treatment and surveillance of patients once treatment has stopped) were estimated separately from health-state costs. The latter relate to service use associated with management of progressive liver disease, associated with chronic HCV infection in patients who do not respond to treatment and for patients whose disease progresses despite demonstrating a response to treatment.

Drug usage for peginterferon alfa-2a was based on a dosage of 180 µg/week, supplied in a prefilled syringe and self-administered by patients, at a cost of £126.91. The dose of ribavirin used in combination with peginterferon alfa-2a varies by patient group and by weight, in the case of genotype 1 patients (Table 15). Expected duration of treatment with the combination of peginterferon alfa-2a plus ribavirin also varies by patient group (see Table 15 for a summary).

Resource use for patient monitoring associated with peginterferon alfa-2a and ribavirin combination therapy and surveillance of patients following treatment cessation was estimated using management protocols, which were developed using expert opinion for our previous report17 for NICE (TA106). The original costing protocols were slightly modified (to include quantitative, rather than qualitative, HCV viral load at key assessment stages) and were inflated to 2007–8 prices using the Hospital and Community Health Services (HCHS) Pay and Prices Index85 (Table 16).

Health-state costs in the model are based on values adopted in our previous assessment,17 inflated from 2003–4 to 2007–8 prices using the HCHS Pay and Prices Index85 (Table 17).

Model/cost-effectiveness results

The MS reports total costs (broken down as treatment-related costs and future costs of medical care for HCV) and outcomes (life expectancy and QALYs) for peginterferon alfa-2a combination therapy and each comparator modelled separately, as well as an incremental analysis (these are summarised in Table 18). Scatter plots showing the cost-effectiveness plane (incremental cost and incremental QALYs for peginterferon alfa-2a combination therapy) from PSA are also reported for each patient population, as well as CEACs for re-treatment of patients who failed to respond to previous treatment with peginterferon.

The MS states that peginterferon alfa-2a in combination with ribavirin is cost-effective in all modelled comparisons for all populations (below a threshold of £15,000), emphasising that treatment dominates the ‘current standard of care’ for relapsed patients and for those with HCV/HIV co-infection. These conclusions are reflected in the manufacturer’s PSA where:

• the probability of peginterferon alfa-2a combination being cost-effective (at a threshold of £20,000) was 100% for re-treating patients who failed to respond to previous peginterferon treatment (both for genotype 1 and genotype non-1 patient subgroups)

• treatment for patients who relapsed on previous peginterferon alfa treatment and for HCV/HIV co-infected patients was dominated in the majority (99%) of simulations. However, as stated earlier, the comparator included in the model for HCV/HIV co-infected patients was non-peginterferon alfa combination therapy – not supportive care as specified in the scope.

The interpretation of the results of the model for patients receiving shortened duration of treatment is complicated by the fact that, although shortened treatment duration is associated with significant savings in treatment costs, it incurs a penalty in terms of a reduced SVR compared with standard durations (from 97% to 91% for genotypes 1 and 4 and from 94% to 89% for genotypes 2/3). As a result, the incremental cost and incremental QALYs associated with shortened treatment duration are negative (for genotypes 1 and 4 the total cost is reduced by £4500 and total QALYs are 0.29 lower, whereas for genotypes 2 and 3 the total cost is reduced by £660 and total QALYs are 0.25 lower), yielding a positive ICER. However, this cannot be interpreted using the commonly assumed decision rule – is the ICER below a given (arbitrary) threshold – as the manufacturer’s have done in their conclusions (section 7.5, p. 182 of the MS), selecting a threshold of £15,000 per QALY gained. In this situation the logic is reversed whereby ICERs below the threshold are rejected. 86 This can perhaps be better understood by considering the analysis using the net benefits framework where we would accept an intervention with positive incremental net benefit, i.e. where the value of incremental benefits exceeds the incremental costs (see Appendix 8 for more details). This requires costs and benefits to be valued on the same scale – commonly achieved by multiplying the incremental effect (incremental QALYs) by a given threshold value (willingness to pay per QALY gained), as below:

where ΔE is incremental QALYs, ΔC is incremental cost and λ is the threshold.

Applying this framework to the analysis of patients receiving shortened duration of treatment presented by the manufacturer, for a range of threshold values (λ) from £0 to £30,000 per QALY gained (Table 19), the incremental net monetary benefit for shortened duration of treatment is positive for genotype 2/3 at only comparatively low threshold values (below the ICER value of £2719). For genotype 1/4 patients the incremental net monetary benefit is positive over a wider range of willingness-to-pay values (below the ICER value of £15,472).

Outline appraisal of the cost-effectiveness analysis undertaken

The NICE reference case requirements (Roche) are shown in Table 20.

Outline review of modelling approach

Assessment of uncertainty

Uncertainty is addressed using deterministic sensitivity analysis (DSA) and PSA. The DSAs were limited in scope and focused on characteristics not included in the PSA, and might more accurately be termed ‘scenario analyses’, as they deal with alternative assumptions rather than variability in input parameters. These analyses address issues of methodological uncertainty (varying discount rates), parameter uncertainty (using alternative assumptions for baseline characteristics including patients’ mean age and weight as well as the proportion of women in the baseline cohort) and structural uncertainty (duration of surveillance for patients following cessation of treatment). The MS reports the incremental cost and effect, as well as the ICER, for each of the sensitivity analyses to facilitate interpretation of changes in the ICER in relation to alternative assumptions. The ICERs were largely insensitive to changes assessed in the DSA and none of these analyses would lead to a change in conclusion from the base-case analysis. The greatest variation was associated with differences in the starting age for the cohort (where incremental cost tended to reduce and incremental effect tended to increase with younger starting ages) and discounting practice (where re-treatment of non-responding patients became dominant for discount rates of 0% for both costs and effects). Additional DSAs were conducted for the HCV/HIV co-infected cohort to consider alternative assumptions regarding the excess death rate for co-infected patients. In the analysis presented in the MS peginterferon alfa-2a, treatment was dominant for all scenarios; however, this was for the comparison with non-peginterferon alfa, rather than with BSC.

Parameter uncertainty is also addressed in a PSA. The majority of parameters in the model are included in the PSA, including transition probabilities in the natural history model, health-state utilities, health-state costs, on-treatment monitoring and post-treatment surveillance costs, as well as SVR and (where appropriate) EVR probabilities. The choice of distribution applied to model parameters appears appropriate, beta distributions for utilities and probabilities and log-normal distributions for costs. However, the parameterisation for many of the distributions does not make best use of the available data. The SVR and EVR probabilities have been parameterised using the point estimate from the base-case analysis as the mean of the distribution, as would be expected, with the standard error (SE) assumed to be 0.02 (the implications of this assumption are discussed below). The rationale for this assumption is not discussed in the MS. The MS presents (for each trial used to derive the base-case SVR and EVR for each modelled population) the total number of patients in each arm, and the number achieving SVR and, where relevant, EVR, but it is not clear why these observed values were not used to parameterise the distributions. Similarly, the SEs for health-state costs have been assumed at 20% of the mean value, without any justification for this assumption. Standard deviations and the number of observations for the health-state costs are reported by the UK Mild Hepatitis C Trial82 and could have been used to parameterise the distributions. Scatter plots of incremental cost and incremental QALYs are presented for all comparisons, while CEACs are presented only for re-treatment of patients who failed to respond to previous peginterferon treatment. There is no discussion of this in the MS and the presentation of the PSA is generally inadequate in the context of current NICE methodological guidance. 68

The key source of heterogeneity in the modelled populations, in terms of response to treatment, has been taken into account through the presentation of separate analyses for viral genotype – either characterised as genotype 1 and genotype non-1 in the case of re-treatment of patients who did not respond to prior peginterferon treatment, or as genotype 1/4 and genotype 2/3 for shortened treatment duration. The remaining analyses (re-treatment of patients who did not relapse following prior peginterferon treatment and HCV/HIV co-infected patients) were not stratified by genotype. The MS does not discuss how representative the overall SVR from included clinical trials (which will reflect the genotype distribution of patients in the trial population) is of the overall SVR expected in a UK population of patients with chronic HCV, which may have a different genotype distribution. The MS has not considered another important source of heterogeneity, in terms of response to treatment, which is the stage of disease at treatment. Where trials have analysed SVR by stage of disease they tend to indicate that response is lower in patients with cirrhosis.

Summary of general concerns

• The manufacturer’s model appears likely to overestimate the QALY gain from achieving SVR by:

  1. – applying age-specific utilities to the SVR state and not applying age-specific utilities to other health states

  2. – collapsing the HCV state into one, rather than differentiating mild and moderate HCV (which appear to have different health-state values).

• The model assumes that all patients start treatment in the moderate HCV state. It is likely that some patients will present at other stages of liver disease, including compensated cirrhosis. The base-case results, applying to patients with moderate liver disease, may not apply to this group.

• The manufacturer’s model does not include the cost of the health state patients are in when they start treatment.

• The cost applied for surveillance of patients who achieve an SVR is low compared with that estimated in the UK Mild Hepatitis C Trial. This cost is applied only for the year following transition to the SVR state.

• The manufacturer’s model appears to be applying an incorrect cost for ribavirin (for genotype 2/3 patients and for the HCV/HIV co-infected group).

• The parameterisation of some distributions in the PSA is based on assumed values and could be improved on. Additionally, some logically related parameters appear to be sampled independently in the PSA, which is likely to give misleading results.

Additional analyses undertaken by SHTAC

The assessment group undertook additional analyses using the manufacturer’s model to address some of the concerns raised in the previous section. Table 21 reports the results of the additional analyses undertaken for the population of patients eligible for shortened duration of treatment. All of the changes made to the manufacturer’s model have the effect of increasing the value of the ICER. However, it needs to be borne in mind when interpreting these results that the incremental costs and outcome when comparing shortened with standard treatment duration are negative. The majority of the changes in assumptions in the model reduce the incremental QALYs between standard treatment and shortened duration – the exception is the change in the distribution of patients across stages of disease (to assume 32% of the cohort have cirrhosis prior to starting treatment).

Table 22 reports the results of the additional analyses undertaken for the population of non-responding or relapsing patients undergoing re-treatment. For non-responding patients the ICER increases in value for each of the scenarios examined, with the results for both genotype groupings being most sensitive to changes in the distribution of patients across stages of disease at baseline. However, although these analyses suggest that the ICER for re-treating patients with peginterferon alfa-2a combination therapy may be higher than the manufacturer’s base case, they do not substantially alter the conclusions from the analysis. In all of the alternative scenarios, re-treatment of relapsing patients remains dominant.

Table 23 reports the results of the additional analyses undertaken for the population of HCV/HIV co-infected patients. The results are similar to those for other patient groups – the ICER increases in value for each of the scenarios examined, with the results for both genotype groupings being most sensitive to changes in the distribution of patients across stages of disease at baseline. As before, while these analyses suggest that the ICER for treating HCV/HIV co-infected patients with peginterferon alfa-2a combination therapy may be higher than the manufacturer’s base case, they do not substantially alter the conclusions from the analysis.

Overview

The Schering-Plough submission92 to NICE consists of a 69-page written document (containing submitted evidence on the clinical effectiveness and a cost-effectiveness analysis) and a fully executable, electronic copy of the manufacturer’s economic model. The MS reports cost-effectiveness results for two populations covered by the scope of the NICE appraisal:

• Patients who have been treated previously with peginterferon, and who did not respond to previous treatment or who relapsed on previous treatment. This analysis is reported for all patients (a cohort including patients of all viral genotypes) and broken down by broad genotype categories (genotypes 1 and 4 combined or genotypes 2 and 3 combined). Costs and outcomes for these patients is compared with BSC, in line with the scope for this appraisal.

• Patients co-infected with HCV/HIV. This analysis is reported for all patients (a cohort including patients of all viral genotypes) and broken down by broad genotype categories (genotypes 1 and 4 combined or genotypes 2 and 3 combined). Costs and outcomes for these patients are compared with BSC, in line with the scope for this appraisal.

No assessment is presented on the cost-effectiveness of shortened versus standard treatment duration. The reason for this omission is not discussed by the manufacturer, though it may be due to peginterferon alfa-2b being licensed only for shorter treatment durations in genotype 1 (as opposed to genotypes 2, 3 and 4).

The perspective of the analysis is stated as being that of the NHS and PSS, consistent with the NICE reference case. 68 The submission reports lifetime costs and outcomes (reported as QALYs) for each treatment arm and the incremental costs and outcomes for peginterferon alfa-2b combined with ribavirin compared with BSC.

The MS does not report whether a systematic search was undertaken for economic evaluations of peginterferon alfa-2b or other treatments for chronic HCV in the patient populations covered by the scope, nor does it report any detail on the development and validation (including any details of clinical validation) of the model adopted for the MS.

Below we describe the approach taken for the model and provide an outline review based on a checklist suggested for the critical appraisal of cost-effectiveness analysis by Drummond and colleagues,67 the requirements of NICE for submissions on cost-effectiveness (reference case)68 and a suggested guideline for good practice in decision modelling by Philips and colleagues. 69

Modelling approach

The model consists of an initial decision tree covering the first year in the model, where patients are eligible to receive treatment. The decision tree incorporates two chance nodes: the first of these applies a probability of patients achieving an EVR, the second applies a probability of patients who achieved an EVR (and therefore remained on treatment) achieving an SVR. A state-transition model is then used to model patients’ costs and outcomes, depending on the state in which they emerge from the decision – with an SVR, remaining with HCV/compensated cirrhosis, or dead from all causes. The state-transition model is structurally similar to published models previously used in the population of patients with HCV, including the previous assessment report for NICE. 17 The model has six health states (mild HCV, moderate HCV, compensated cirrhosis, decompensated cirrhosis, HCC and liver transplantation) indicating progressive liver disease, one state representing a treatment response (SVR) and one absorbing state (death), although this last state is broken down to differentiate deaths from progressive liver disease and deaths from all other causes.

The model does not differentiate the SVR state according to patients’ stage of disease prior to SVR. However, QoL data reported by the UK Mild Hepatitis C Trial82 would suggest that there are differences in health-state utility for patients who enter the SVR state from both mild and moderate chronic HCV, and it may be more appropriate to structure the model to identify prior stage of disease (given that patients with compensated cirrhosis are eligible to receive treatment, as well as those with mild or moderate chronic HCV).

The main treatment effect applied in the model is the SVR for treated patients, with the proportion of patients in each of the modelled populations achieving an SVR being based on data from clinical trials conducted in the relevant patient populations, reported in the MS. Patients who achieve an SVR are assumed in the model to be ‘cured’ and do not face any risk of reactivation of disease or any excess risk of progressive liver disease (above that of a general population). Age-specific mortality risks for the general population, weighted for the proportion of men in the baseline cohort, are applied to patients achieving an SVR. Patients who do not achieve an SVR are at risk of progressive liver disease and are assumed to face the same risks of disease progression as untreated patients. Risks of disease progression and, where relevant, excess mortality risks associated with advanced liver disease states in the model have been drawn from natural history studies.

The base-case population characteristics (in terms of age at entry to the model, weight and the proportion that are male) differ between the patient subgroups modelled and are based on the baseline populations in the relevant clinical trials. These assumptions have no impact on the patient response to treatment (i.e. SVRs in the model are not broken down by age or sex), but age and the proportion of men affect the all-cause mortality rates applied in the model, while patient weight has an impact on cost of treatment, as peginterferon alfa-2b and ribavirin dosage is weight related.

The health-state utilities have been derived from the UK Mild Hepatitis C Trial,82 and a study of the cost-effectiveness of liver transplantation. 93 There is no systematic search for these values reported in the submission. The EQ-5D was completed by 130 patients within the Mild Hepatitis C Trial,82 at 24 or 48 weeks post treatment or control. A linked observational study used cases recruited to the costing study in order to estimate the HRQoL for patients with moderate disease, compensated cirrhosis or decompensated cirrhosis. HRQoL for liver transplant patients and post-liver transplant health states was taken from a cost-effectiveness study of liver transplantation. This latter study also utilised the EQ-5D to estimate HRQoL in liver transplant patients; however, its applicability may be limited, as the included patients did not have HCV.

The model applies a disutility of 0.13 (owing to treatment-related adverse effects) to the utility score for treatment-eligible health states for the year in which patients undergo treatment. In their example, moderate HCV was assigned a baseline utility of 0.66, and this was reduced to 0.53 during treatment. This was based on the overall mean difference in EQ-5D utility score for treated and control patients at 12 or 24 weeks following randomisation in the UK Mild Hepatitis C Trial. 82 The disutility associated with treatment was adjusted for duration of treatment, so that a lower utility decrement would apply for patients (who fail to demonstrate an EVR) stopping treatment at 12 weeks.

The costs applied in the submission were made up of two components. Treatment-related costs (which consist of drug acquisition costs and monitoring of patients on treatment) were estimated separately from health-state costs. Health-state costs include resource use associated with the management of progressive liver disease.

Drug usage for peginterferon alfa-2b was based on a dosage of 1.5 µg per kg per week and was assumed to be supplied in prefilled pens. As dosage of both peginterferon alfa-2b and ribavirin is weight based (Table 24) the MS needed to assume an average weight for each of the modelled patient populations. A mean weight of 80 kg was applied in the base-case analysis for the re-treated group, and of 63 kg in the HCV/HIV co-infected group. The values for the weight of the HCV/HIV co-infected group were taken from the Laguno study;94 however, the default value in the MS is reported as 68.29 kg.

The drug acquisition costs for peginterferon alfa-2b and ribavirin adopted in the Schering-Plough submission92 (taken from the BNF, No. 57, March 2009) are presented in Tables 25 and 26 below. With an assumed body weight of 80 kg for re-treated patients, and assuming a preference for prefilled pens for peginterferon alfa-2b and tablets for ribavirin, the weekly treatment costs are £165.73 for peginterferon alfa-2b and £114.85 for ribavirin. The equivalent cost for the HCV/HIV co-infected patients is £118 for peginterferon alfa-2b and £91.88 for ribavirin.

The costs of initial investigations and monitoring included liver biopsy, an overnight stay in hospital for this procedure, and regular outpatient consultations and investigations. These costs were all taken from the Mild Hepatitis C Trial. 82 The initial investigations were calculated to cost £822.27 per patient assessed. As a result of interviews with clinicians, suggesting that between one and five patients would be assessed for each patient treated, this was then tripled to account for that (range 1–5). The monitoring costs of patients being treated was £489, which was inflated to £587.85 per patient treated (2007–8 values).

Costs for each disease state were again taken from the UK Mild Hepatitis C Trial,82 or in the case of moderate and more severe disease from the observational costing study conducted within that trial.

The baseline population differs between the two patient groups modelled, and, for the majority of these characteristics, is based on the baseline population of the relevant clinical trials. The simulated cohort of re-treated patients has a mean age of 49 years, with 71% being male. Patient weight is assumed to be greater than 81 kg and it is assumed that 85% of re-treated patients have genotype 1 or 4 (the remainder with viral genotypes 2 and 3). This last assumption has an impact on outcome for the overall cohort of patients, as patients with viral genotypes 1 and 4 have a lower probability of SVR than those with viral genotype 2 or 3. For patients with HCV/HIV co-infection, the base-case characteristics are based on the RCT by Laguno and colleagues,95 with a mean age of 40 years, and 68% being male. Patient weight is substantially lower at 63 kg, although as mentioned above this only affects the drug costs. While these characteristics have been drawn from clinical trials conducted in relevant subgroups, the MS does not discuss how relevant these characteristics may be to the population of UK patients with chronic HCV, in general, or how relevant they may be to the UK population of patients to be re-treated following non-response to, or relapse following, prior peginterferon treatment or those with HCV/HIV co-infection.

In both analyses, patients enter the model in one of three states – with mild HCV (33%), moderate HCV (33%) or compensated cirrhosis (34%). Each of these is a treatment-eligible health state and the probability of EVR or SVR is assumed to be equal for each possible starting state.

Model/cost-effectiveness results

The MS reports total costs and QALYs for a cohort of 100 re-treated patients and 100 HCV/HIV co-infected patients. Both cohorts include genotype 1, 2, 3 and 4 patients. To estimate costs and outcomes for these cohorts of mixed genotypes, treatment efficacy estimates (SVR and, where relevant, EVR) for genotype subgroups were used to estimate response for each subgroup. The overall results for the cohort were then calculated as weighted totals (based on the proportion of the total cohort in each subgroup). The model results are also presented as an average cost and average QALYs per patient for the cohort including all genotypes and for subgroups of genotypes 1 and 4 and of genotypes 2 and 3. The distribution of patients across viral genotypes is based on the populations recruited to the clinical trials used to derive the efficacy data for the model. The MS does not discuss how generalisable these proportions may be to UK populations of UK patients with chronic HCV infection.

Table 27 reports the base-case results, including the ICER, from the Schering-Plough model92 for re-treatment of patients who did not respond or relapsed following previous interferon therapy and for HCV/HIV co-infected patients. Scatter plots showing the cost-effectiveness plane (incremental cost and incremental QALYs for peginterferon alfa-2b combination therapy) and CEACs are presented in a separate section of the MS reporting the results of the PSA.

The MS presents a further analysis for the cohort of re-treated patients, reporting separate analyses for previous relapsers, previous non-responders and previous treatment failures (although the definition of previous treatment failure is not very clear, and is described in the MS as referring to ‘patients who could not be classified as relapsers or non-responders due to missing data or other reasons’). The results for these subgroups are presented in Table 28 and show that previous non-responders have a lower QALY gain (and higher incremental cost) than previous treatment failures and relapsing patients.

The MS states that peginterferon alfa-2b in combination with ribavirin is cost-effective for adults with HCV/HIV co-infection and for patients whose previous treatment was unsuccessful. These conclusions draw on evidence from the base-case analyses presented above, from DSAs (where ICERs remained below £20,000 per QALY gained in the scenarios tested) and from PSAs where the probability for peginterferon alfa-2b being cost-effective, compared with no active treatment, for re-treating patients who did not respond or relapsed following previous interferon therapy was estimated at 95% at a willingness-to-pay threshold of £20,000 per QALY and the probability for peginterferon alfa-2b being cost-effective, compared with no active treatment, for HCV/HIV co-infected patients, was estimated at 98% at a willingness-to-pay threshold of £20,000 per QALY.

Outline appraisal of the cost-effectiveness analysis undertaken

The NICE reference case requirements (Schering-Plough) are shown in Table 29.

Outline review of modelling approach

Assessment of uncertainty

Schering-Plough have tested response to therapy (where the EVR and SVR are varied) and drug dosing requirements (varying patient weight and drug administration method) in their DSA. In addition, the disutility and distribution of disease severity and distribution of genotype at baseline were varied. Two scenario analyses have also been reported: one in which the re-treatment group is presented as non-responders and relapsed patients and a further scenario in which discounting is removed.

The ICERs for both the re-treated group and the HCV/HIV co-infection group were both sensitive to variation in the EVR and SVR, and to changes in patient weight. The proportion of patients achieving EVR and SVR in the re-treatment group were varied between the upper and lower CIs from the included studies. The ICER in this group then changed from £4387 in the base-case analysis to £4842, where EVR proportion was ‘low’, and to £4003, where this was ‘high’. The SVR was similarly varied, and the ICER changed to £5227 in the ‘low-value’ group and to £3615’ in the ‘high-value’ group.

The peginterferon alfa-2b arm of the Scotto and colleagues97 study was also used to calculate ICERs for re-treated patients. This resulted in a greatly increased ICER of £19,004 per QALY in the genotype 1 and 4 group, a decreased ICER of £2520 in the genotype 2 + 3 group, and £10,742 for all patients. The manufacturers state that this is owing to the re-treated patients in this study being previous non-responders.

Incremental cost-effectiveness ratios in the HCV/HIV co-infection group were sensitive to the SVR rate (the authors state that the EVR rate was unavailable in this group). Again, values for the sensitivity analysis were taken from the upper and lower CIs reported in the included study. The base-case ICER was £1077. In the ‘low-value’ group this increased to £4065 per QALY, and in the ‘high-value’ group, peginterferon alfa-2b was dominant. When the EVR and SVR values from the more recent Laguno and colleagues’ RCT94 were applied, the manufacturers report ICERs of £6140 per QALY in genotypes 1 and 4, £422 per QALY for genotypes 2 and 3, and £2311 for all genotypes. It is not clear if both EVR and SVR have been adjusted here.

Changes in the distribution of patients with different liver disease severity produced smaller variation in the ICERs in the re-treatment group. In the case of mild disease the percentage of patients decreases from 33% to 27%, for moderate disease this proportion decreased from 33% to 31%, and for compensated cirrhosis this proportion increased from 33% to 42%. This variation is quite small, and had the effect of decreasing the ICER to £3596 per QALY from £4387 per QALY.

A sensitivity analysis was performed on distribution of genotype at baseline. The treatment response of genotypes 1 and 4, and then 2 and 3, are applied to all patients. The treatment response of genotypes 1 and 4 applied to the entire cohort resulted in an ICER of £7176 per QALY, and that of genotypes 2 and 3 resulted in an ICER of £782, in the re-treatment group. In the HCV/HIV co-infected group, the ICERs became £1637 and £403. The ICERs are the same as those presented in the base-case analysis, and it is unclear what has been added to this analysis by the reporting of this scenario.

The first scenario analysis presented ICERs for the re-treatment ‘subgroups’: previous relapsers and non-responders to treatment. The base-case ICER for this group was £4387. For the ‘previous relapser’ group alone it was £2048 and for ‘previous non-responders’ alone in this scenario it was £7581, with the higher ICER thought by the authors to reflect the lower expected level of success in this group.

The second scenario analysis examined the effects of not discounting costs and outcomes. Where discounting is removed, the ICER is reduced to £1265 per QALY in the re-treatment group, and the intervention becomes dominant (more effective and less costly) in the HCV/HIV co-infection group.

Parameter uncertainty is also addressed in a PSA. The majority of parameters in the model are included in the PSA, including transition probabilities in the natural history model, health-state utilities, health-state costs, probability of discontinuing treatment as well as SVR and (where relevant) EVR. The choice of distribution applied to the parameters appears appropriate, using beta distributions for probabilities and utilities, and gamma distributions for costs. The electronic model appears to use an implementation of the Dirichlet distribution for sampling transition probabilities in the model that are competing risks (e.g. patients with compensated cirrhosis may remain in that state, may progress or may develop HCC), although this is not discussed in the submission. The written submission contains an appendix that lists the parameters included in the PSA, their mean value, SE and the choice of distribution, but not the parameterisation of the distribution.

The MS reports three PSAs for each patient group (re-treated and HCV/HIV co-infected patients), each based on 10,000 simulations. The first analysis applies to the overall cohort, followed by separate analysis for genotype subgroups (genotypes 1 and 4 and genotypes 2 and 3). Cost-effectiveness scatter plots are presented along with CEACs for each of these analyses. The MS also reports the probability of the intervention of interest being cost-effective at willingness-to-pay thresholds of £20,000 per QALY gained and at £30,000 per QALY gained. The presentation of the PSA appears generally to be in accordance with NICE methodological guidance68 but does not report mean costs and outcome for the PSAs.

The key source of heterogeneity in the modelled populations, in terms of response to treatment, has been taken into account through the presentation of separate analyses for viral genotype. The MS has not considered another important source of heterogeneity, in terms of response to treatment, which is the stage of disease at treatment. Where trials have analysed SVR by stage of disease they tend to indicate that response is lower in patients with cirrhosis.

Summary of general concerns

• The Schering-Plough model appears to underestimate the SVR in each analysis, as a result of applying an unnecessary adjustment for treatment discontinuation, but appears to overestimate the utility gain through treatment by not applying an adjustment for treatment discontinuation:

  1. – The observed SVR for a given patient population (e.g. 38% for HCV/HIV co-infected patients with genotype 1 or 4) is applied to the proportion of patients expected to be in that population (63% of HCV/HIV co-infected patients are assumed to be genotype 1 or 4 in a cohort of 100 patients, i.e. 63 people) – therefore the expected number of SVRs is 24. This value is then multiplied by the probability of not discontinuing treatment (probability of discontinuing is 0.1731, therefore probability of not discontinuing = 1 – 0.1731), which gives a value of 20 (the number of SVRs adjusted for discontinuation), resulting in an SVR rate of 31.42% (20/63). As the original SVR rate of 38% was based on the observed data reported in the RCT by Laguno and colleagues,95 adjusting by the discontinuation probability seems unnecessary.

• There is an implicit assumption that patients achieve an SVR immediately after treatment is initiated and therefore accrue health benefits on entering the model. It might be more reasonable to assume that transitions occur mid-cycle (essentially applying half-cycle adjustment). This would mean adjusting cycle lengths (currently annual) to cope with treatments that are significantly less than 52 weeks, or calculating a weighted combination of the utility for the initial state and the utility for the appropriate SVR state (weighted according to what proportion of the cycle is spent in the initial health state and what proportion in the SVR state).

• The model collapses the SVR state into one and therefore does not track whether patients have achieved SVR from mild HCV, moderate HCV or compensated cirrhosis. It applies the same health-state utility to patients achieving an SVR, irrespective of their stage of liver disease when treatment was initiated. This does not accord with utility data from the UK Mild Hepatitis C trial, which reported a lower mean utility for patients achieving SVR from moderate liver disease than those achieving SVR from mild liver disease.

• The model assumes that the SVR health-state cost is applied for all cycles the patient remains in the SVR state. This differs from the assumption applied in our previous assessment report,17 where it was assumed that the SVR cost applied only for the year following treatment response.

• The model appears to have underestimated the cost of ribavirin (tables 31 and 32 of the MS report weekly cost of ribavirin as £16.41 for re-treated patients and £13.13 for HCV/HIV co-infected patients). These are derived using an estimated average cost per 200-mg tablet of ribavirin of approximately £3.28. However, the figures used in the MS are the daily, not weekly, costs.

Additional analyses undertaken by SHTAC

The assessment group undertook additional analyses using the manufacturer’s model to address some of the concerns raised in the previous section. Table 30 reports the results of the additional analyses undertaken for HCV/HIV co-infected patients. Removing treatment discontinuation from the calculation of the SVR probability and applying only the SVR health-state cost in year after SVR occurs reduces the ICER – making treatment for genotype 2 + 3 patients dominant. In contrast, correcting the calculation of ribavirin costs and splitting the SVR state to apply utility values that take account of disease stage prior to SVR increase the ICER.

The same SVR was applied to all treated patients in the manufacturer’s model, regardless of stage of fibrosis. However, analyses of response to treatment, by stage of disease, typically suggest that treatment response is lower in patients with fibrosis. Ratios of the relative effectiveness of treatment for patients with fibrosis stages F2, F3 and F4 (derived using data reported in the MS for the EPIC3 study95) were used to examine the effect, on the cost-effectiveness results, of reducing the SVR for cirrhotic patients. This is labelled in Table 30 as ‘Adjust SVR for disease stage’.

Table 31 reports the results of the additional analyses undertaken for re-treated patients. Removing treatment discontinuation from the calculation of the SVR probability is less influential than in the analysis for HCV/HIV co-infected patients. Applying the SVR health-state cost in the year after SVR occurs reduces the ICER, while correcting the calculation of ribavirin costs and splitting the SVR state to apply utility values that take account of disease stage prior to SVR increase the ICER. Adjusting the SVR for disease stage has relatively little impact on the cost-effectiveness results. Overall, while these analyses suggest that the ICER for treating HCV/HIV co-infected patients with peginterferon alfa-2b combination therapy may be higher than in the manufacturer’s base case, they do not substantially alter the conclusions from the analysis.

Summary of manufacturers’ models, compared with SHTAC model from previous assessment report

Table 32 summarises the transition probabilities used in the manufacturers’ models and in our previous assessment of peginterferon alfa combination treatment for chronic HCV infection. 17 It is clear from the table that identical values have been used for the majority of transition probabilities modelling the natural history of progressive liver disease. These are primarily drawn from studies reported by Sweeting and colleagues,18 Wright and colleagues82 and Fattovich and colleagues. 98 The principal differences between the three models are that:

• Patients enter the Roche model with moderate HCV, so that the transition probability from mild-to-moderate disease is not relevant.

• The Schering-Plough model includes a small risk for non-cirrhotic patients (with moderate disease) developing HCC, based on a previously published economic evaluation by Bennett and colleagues. 76

• The Schering-Plough model uses a higher excess mortality risk for HCC than is applied in the other models, based on a previously published economic evaluation99 and cancer mortality statistics. 100

• The Schering-Plough model uses a slightly higher probability for developing HCC for patients with cirrhosis. The value in Table 32 is used in the electronic model, while a lower value of 0.014 (identical to that used by Roche and in our previous assessment) is reported in tables included in the main submission document. This discrepancy is not explained in the submission or the electronic model.

Sustained virological responses used in manufacturers’ models are shown below in Table 33.

Tables 34 and 35 report the health-state utility values applied in the three models and the impact on utility applied while patients are on treatment. The impact of structural assumptions in the models (particularly the inclusion of a single SVR state, which does not distinguish the stage of disease prior to SVR) and the selection of utility values applied to the SVR state have been discussed in the previous sections, appraising each of the manufacturer’s models separately. Table 34 shows that, in all but one case, identical utility values have been applied to the health states relating to more advanced liver disease, while there is considerable difference in the utility values applied to patients achieving an SVR [and, to a lesser extent, the HCV health state(s)].

Table 35 indicates that, although there are sizable differences in the utility values applied to the HCV and SVR health states, there is more agreement on the on-treatment utility reduction associated with peginterferon alfa and ribavirin. All three models have based their valuations on data from the UK Mild Hepatitis C Trial,82 which reported health-state valuations for treated and untreated patients by stage of disease (adopted by Roche and in our previous assessment) and an overall mean difference in EQ-5D utility score, for treated and control patients, at 12 weeks or 24 weeks following randomisation (adopted by Schering-Plough).

Table 36 summarises the health-state costs applied in the three models. The main differences between the three models relate to:

• Structural assumptions in the models (patients enter the Roche model with moderate HCV, so cost of mild HCV is not relevant, while both manufacturers collapse the SVR state and do not track the stage of disease prior to SVR).

• Sources of costs that have been inflated to current prices. Health-state costs in all three models are based on those reported for the UK Mild Hepatitis C Trial. 82 Roche have inflated health-state costs reported in our previous assessment (which had been inflated from 2002–3 to 2003–4 prices), whereas Schering-Plough inflated the original health-state costs reported for the trial. The discrepancies between the two sets of costs arises from slight adjustments that have been made to the HCHS Pay and Prices Index85 over time.

Effectiveness data

Table 38 reports the transition probabilities adopted in the natural history model for this economic evaluation. They represent the complete set of transition probabilities for the BSC comparator and are taken from our previous assessment report. 17

The transition probabilities from mild-to-moderate disease, and from moderate disease to compensated cirrhosis, were derived for the economic evaluation undertaken alongside the UK Mild Hepatitis C Trial82 and were based on a re-analysis of data from UK cross-sectional and longitudinal data sets. The remaining transition probabilities were taken from the literature on natural history and previous economic evaluations. 76,81,98,103 Targeted searches, undertaken as part of this assessment, did not identify new natural history evidence relating to progression or management of chronic HCV to update the model parameters.

Table 39 reports the treatment effects (proportion of patients achieving SVR) that have been applied, in the model, to estimate the effectiveness of peginterferon alfa and ribavirin combination therapy in the treatment strategies and patient subgroups being considered. The studies used to estimate the effectiveness of treatment have typically reported SVRs for all patients in the relevant subgroup and have not indicated the effect of stage of liver disease on response to treatment. For the base-case analyses we have assumed that the same SVR applies for patients with mild or moderate HCV, and for those patients with compensated cirrhosis. We examine the effect of cirrhosis, on reducing the response to treatment, in sensitivity analyses.

Sustained virological response estimates for patients receiving shortened courses of treatment are based on those used in our systematic review of clinical effectiveness (see Chapter 4, Assessment of clinical effectiveness). SVR estimates for patients co-infected with HCV/HIV were based on those reported from two recent systematic reviews of antiviral treatment in this patient group (further details can be found in Appendix 9). SVR estimates for patients re-treated following non-response to, or relapse from, a previous course of peginterferon alfa-2a were taken from the trial by Jensen and colleagues,88 supplemented with information from the Roche submission to NICE (see Appendix 9). SVRs for re-treatment with peginterferon alfa-2b were from the EPIC3 study,96 as summarised in the Schering-Plough submission to NICE (see Appendix 9).

Health-state values/utilities

A systematic search of the literature (see Appendix 2 for search strategy) and targeted searches did not find new utility data to update our model. In particular, the searches did not identify utility data that were specific to the patient populations within the scope of this assessment. As a result we have adopted the same utility values as for our previous assessment (Table 40). 17 These data are appropriate to the NICE reference case68 for measuring and valuing health benefits, in that the QoL measurements were undertaken using the EQ-5D in patients with chronic HCV recruited to the UK Mild HCV Trial,82 an observational study of patients with more severe liver disease conducted alongside the trial and a UK study of costs and outcomes following liver transplantation. 93 The QoL measurements were valued using a tariff derived in a general population. 84 While the use of these data has the advantage of consistency with those applied for the previous assessment,17 they are not specific to the patient populations in the scope of this assessment. It may be argued that values derived from HCV mono-infected patients may overestimate the heath–state utility for HCV/HIV co-infected patients or that values from treatment-naive patients (as in the UK Mild HCV Trial82) may not be representative of utilities for patients who have been previously treated.

Cost data

Discounting of future costs and benefits

A discount rate of 3.5% was applied to future costs and benefits, in line with current methodological guidance from NICE. 68 Discount rates of 0% (for both costs and benefits), 6% (for costs) and 1.5% (for benefits) were applied in the sensitivity analyses.

Presentation of results

We report findings on the cost-effectiveness of interventions based on analysis of a cohort of patients with age, sex and genotype characteristics as reported above (see Baseline cohort of adult chronic HCV patients). For HCV/HIV co-infected patients and re-treatment of patients who failed to respond to, or relapse from, prior peginterferon alfa therapy, the interventions assessed in this report are compared with BSC (i.e. without any form of interferon alfa therapy) as specified in the NICE scope. For patients who are eligible for shortened courses of peginterferon alfa, results are presented in comparison with the usual duration of treatment.

We report the results of these comparisons in terms of the incremental gain in QALYs and the incremental costs determined in the cohort analysis.

Assessment of uncertainty in the SHTAC analysis (sensitivity analysis)

Parameter uncertainty is addressed using PSA. Probability distributions are assigned to the point estimates used in the base-case analysis. The point estimates for state transitions and treatment effects are reported in Tables 38 and 39, for health-state utilities in Table 40 and for health-state costs in Table 43. Appendix 10 reports the variables included in the PSA, the form of distribution used for sampling and the parameters of the distribution.

Univariate sensitivity analysis is used to address particular areas of uncertainty in the model related to:

• model structure

• methodological assumptions

• transition probabilities, around which there is considerable uncertainty or which may be expected, a priori, to have disproportionate impact on study results.

The purpose of this analysis is to identify clearly the impact of this uncertainty and to test the robustness of the cost-effectiveness results to variation in structural assumptions and parameter inputs.

Peginterferon alfa-2a

Costs and outcomes modelled for patients who were eligible for shortened duration of treatment with peginterferon alfa-2a and ribavirin combination therapy are presented in Table 44 for genotype 1 patients and in Table 45 for patients with genotype 2 or 3. As it was not considered appropriate to conduct a meta-analysis of RCTs, we present separate results for each trial included in our systematic review of clinical effectiveness (with the exception of the RCT by Mangia and colleagues,52 which used both peginterferon alfa-2a and alfa-2b within the same trial; as the two drugs are considered pharmacologically different, we present cost-effectiveness estimates for peginterferon alfa-2a based on RCTs of alfa-2a, and peginterferon alfa-2b based on RCTs of alfa-2b). The comparator in each of the analyses is the standard duration of peginterferon alfa-2a combination therapy (48 weeks for genotype 1 patients and 24 weeks for patients with genotype 2 or 3). The tables report total costs (antiviral treatment and supportive care), health outcomes (in terms of life-years and QALYs) and the incremental cost-per-QALY ratios. Costs and health outcomes are discounted at 3.5%.

In both of the included trials of shortened treatment duration for genotype 1 patients, standard 48-week treatment was associated with an SVR of 100%. However, this high SVR is only applicable to the subgroup of patients who had baseline LVL (< 800,000 IU/ml in the trial by Liu and colleagues53 and < 400,000 IU/ml in the trial by Yu and colleagues, 200854 – the SPC for peginterferon alfa-2a42 defines LVL as ≤ 800,000 IU/ml at baseline) and who also demonstrate an RVR. Shortened treatment duration was associated with a slight reduction in SVR (to 94% in the trial by Liu and colleagues53 and 96% in the trial by Yu and colleagues54). Shorter duration of 24 weeks of treatment is associated with a reduction in total costs between £4800 and £5200. This is primarily due to the reduction in drug acquisition costs, although there is some additional reduction in cost of on-treatment monitoring. While the small reduction in SVR means that there are some additional costs associated with disease progression for the cohort of patients receiving shorter duration of treatment, these are not sufficient to offset the cost reduction associated with the shorter duration of treatment.

Given that the SVR is lower, and therefore there is a greater risk of progressive liver disease, there is a reduction in total QALYs between 0.08 and 0.14 (depending on trial) associated with shorter duration of treatment. This is not offset by the QoL impact of treatment-related adverse events estimated for the cohort of patients receiving shorter duration of treatment. As total costs and total QALYs are reduced in the cohort of patients receiving shorter duration of treatment, the ICER is positive but is located in the south-west (cost- and outcome-reducing) quadrant of the cost-effectiveness map rather than the more familiar north-east (cost-increasing and outcome-gaining) quadrant. This has implications for the interpretation of the results from the base-case (deterministic) analysis and from the PSA, including the interpretation of the CEACs.

In both of the included trials for genotype 2 or 3 patients, shortened treatment duration was associated with a higher SVR than was the case for standard treatment. Shorter duration of treatment is associated with a reduction in total costs between £2100 and £3150. This is primarily due to the reduction in drug acquisition costs and a reduction in the cost of on-treatment monitoring. Given that the SVR is higher for the shorter duration of treatment, there are small reductions in total cost associated with a reduced risk of disease progression for the cohort of patients receiving shorter duration of treatment. The higher SVR for the shorter duration of treatment also results in improvements in modelled outcomes associated with shorter duration of treatment so that the strategy of shortened treatment duration for this group of patients dominates standard duration treatment.

Deterministic sensitivity analysis

Table 46 reports the results of a DSA for genotype 1 patients who were eligible for shortened treatment duration, and Table 47 reports the results for genotype 2/3 patients. These are predominantly univariate sensitivity analyses – that is, varying one parameter at a time, from its base-case value, leaving all other variables unchanged. The table is divided to distinguish between analyses undertaken owing to structural uncertainties in the model, uncertainties over the composition of the baseline cohort and uncertainty over parameter values.

The DSA suggests that the results are robust to a change in structural assumptions (allowing spontaneous SVR from the mild chronic HCV state), the proportion of the baseline cohort that is male and the cost associated with the SVR health state. Reducing drug acquisition costs has the effect of reducing the cost-effectiveness of shortened treatment duration, as it reduces the cost saving between standard and shortened treatment duration while the outcome difference is unchanged.

The greatest variability in ICERs is associated with changes in two assumptions regarding baseline characteristics of the cohort of treatment-eligible patients. Increasing the mean age of patients at the start of the simulation up to 15 years leads to an approximate doubling of the ICER. This occurs because the QALY difference between the standard and shortened treatment duration reduces rapidly [from –0.14 at a starting age of 40 years to –0.08 at a starting age of 55 years (a reduction of 43%) using efficacy data from the RCT by Liu and colleagues53]. The difference in costs between standard and shortened treatment duration is less responsive to changes in starting age [from –£4807 at a starting age of 40 to –£5098 at a starting age of 55 (a reduction of 6%) using the same efficacy data]. Alternative assumptions regarding the stage of liver disease in patients starting treatment also has a large impact on the ICER, with shortened treatment duration being more cost-effective in patients with less severe disease than in those with cirrhosis. This arises because, all other things being equal, the higher the proportion of a cohort starting the simulation with cirrhosis, the greater the proportion that will progress to advanced liver disease. As a result, the penalty (in terms of a reduction in total QALYs) associated with a lower SVR for shortened treatment duration will be greater in cohorts that contain a higher proportion of patients with cirrhosis.

The pattern of results for genotype 2/3 patients, reported in Table 47, is similar to those for genotype 1 patients. The results are largely insensitive to changes in input parameters, other than baseline assumptions relating to age and stage of disease at start of treatment. Shortened treatment duration remains dominant in all the scenarios tested in Table 47, using efficacy data from either of the included trials.

As the included trials give contradictory results (with shortened duration less effective than standard duration for genotype 1 patients, but more effective for genotype 2/3 patients) and, in the case of genotype 2/3 patients, potentially counterintuitive results, we conducted an additional scenario analysis on the impact of the difference in SVR on the cost-effectiveness results, assuming that the SVR for shortened treatment duration is less than or equal to that for standard treatment duration (Table 48).

This suggests that shortened treatment duration may be a highly cost-effective option, where there is no difference (or a very small difference) in SVR between shortened and standard treatment duration, but as the SVR difference increases the cost reduction decreases and the QALY loss increases rapidly – particularly in the case of genotype 2/3 patients.

Probabilistic sensitivity analysis

In a PSA, where the probabilities of achieving SVR, health-state costs, health-state utility values and transition probabilities for the natural history parameters were sampled probabilistically, shortened duration of treatment with peginterferon alfa-2a and ribavirin combination therapy was generally associated with reduced QALYs. For genotype 1 patients incremental QALYs associated with shortened duration of treatment were negative for the majority of simulations – 95% of simulations using efficacy data (proportion of patients with SVR) from Liu and colleagues,53 and 99.5% of simulations using efficacy data from Yu and colleagues. 54 The opposite is true in the analysis of genotype 2 or 3 patients. Approximately 2% of simulations were associated with negative incremental QALYs, for genotype 2 patients, using efficacy data from Yu and colleagues,55 whereas none of the simulations resulted in negative incremental QALYs using efficacy data from von Wagner and colleagues. 56 Incremental costs associated with shortened duration of treatment were negative in all simulations – ranging from –£2500 to –£6000 for genotype 1 patients and from –£550 to –£5200 for genotype 2/3 patients. Table 49 reports summary information for the PSAs and Figures 3–6 show the scatter plots for each analysis, including 95% confidence ellipses.

FIGURE 3.

Cost-effectiveness plane for genotype 1 patients – incremental cost and incremental QALYs for shortened treatment duration using peginterferon alfa-2a and ribavirin combination therapy (24 vs 48 weeks of treatment) – efficacy from Liu and colleagues. 53

FIGURE 4.

Cost-effectiveness plane for genotype 1 patients – incremental cost and incremental QALYs for shortened treatment duration using peginterferon alfa-2a and ribavirin combination therapy (24 vs 48 weeks of treatment) – efficacy from Yu and colleagues. 54

FIGURE 5.

Cost-effectiveness plane for genotype 2 patients – incremental cost and incremental QALYs for shortened treatment duration using peginterferon alfa-2a and ribavirin combination therapy (16 vs 24 weeks of treatment) – efficacy from Yu and colleagues. 55

FIGURE 6.

Cost-effectiveness plane for genotypes 2/3 patients – incremental cost and incremental QALYs for shortened treatment duration using peginterferon alfa-2a and ribavirin combination therapy (16 vs 24 weeks of treatment) – efficacy from von Wagner and colleagues. 56

In this analysis, shortened duration of treatment using peginterferon alfa-2a and ribavirin combination therapy for genotype 1 patients had a probability of being cost-effective [compared with the standard duration (48 weeks) of treatment] of 83% at a willingness-to-pay threshold of £20,000 per QALY and 59% at a willingness-to-pay threshold of £30,000, using efficacy data from the trial reported by Liu and colleagues53 (Figure 7). The equivalent values using efficacy data from the trial reported by Yu and colleagues 200854 are 100% at willingness-to-pay thresholds of £20,000 and £30,000 per QALY.

FIGURE 7.

Cost-effectiveness acceptability curves for shortened treatment duration with peginterferon alfa-2a and ribavirin combination therapy (24 vs 48 weeks of treatment) for genotype 1 patients.

For patients with genotypes 2 and 3, the probability of being cost-effective [compared with the standard duration (24 weeks) of treatment] was 100% at willingness-to-pay thresholds of £20,000 and £30,000 per QALY, using efficacy data from either the trial reported by Yu and colleagues, 200755 or the trial by von Wagner and colleagues56 (Figure 8). This reflects the proportion of simulations located in the south-east quadrant of the cost-effectiveness map (where the intervention dominates the comparator) – see Figures 5 and 6.

FIGURE 8.

Cost-effectiveness acceptability curves for shortened treatment duration with peginterferon alfa-2a and ribavirin combination therapy (16 vs 24 weeks of treatment) for genotypes 2/3.

Deterministic sensitivity analysis

Table 51 reports the results of a DSA for genotype 1 patients who were eligible for shortened treatment duration with peginterferon alfa-2b and ribavirin combination therapy. The DSAs suggest that the results are generally insensitive to changes in structural assumptions and input parameter values. The greatest variability in ICERs is associated with changes in the mean age of patients at the start of the simulation and the initial distribution of patients across stages of liver disease.

As the included trial by Berg and colleagues59 gives a potentially counterintuitive result (with shortened treatment duration being more effective than standard duration), we conducted an additional scenario analysis on the impact of the difference in SVR on the cost-effectiveness results, assuming that the SVR for shortened treatment duration is less than or equal to that for standard treatment duration (Table 52).

This analysis suggests that shortened treatment duration may be a highly cost-effective option, where there is no difference (or a very small difference) in SVR between shortened and standard treatment duration. Where there is no difference in SVR, shortened duration of treatment dominates standard duration by reducing the utility loss associated with treatment.

Probabilistic sensitivity analysis

In a PSA, where the probabilities of achieving SVR, health-state costs, health-state utility values, and transition probabilities for the natural history parameters were sampled probabilistically, shortened duration of treatment with peginterferon alfa-2b and ribavirin combination therapy, for genotype 1 patients with baseline LVL and who achieve an RVR, is associated with reduced costs and increased QALYs in all simulations (using efficacy data from Berg and colleagues59). Table 53 reports summary information for the PSAs and Figure 9 shows the cost-effectiveness plane, including 95% confidence ellipses.

FIGURE 9.

Cost-effectiveness plane for genotype 1 patients – incremental cost and incremental quality-adjusted life-years for shortened treatment duration using peginterferon alfa-2b and ribavirin combination therapy (24 vs 48 weeks of treatment) – efficacy from Berg and colleagues. 59

In this analysis of shortened duration of treatment using peginterferon alfa-2b and ribavirin combination therapy for genotype 1 patients, all simulations were in the south-east quadrant of the cost-effectiveness map, where the comparator [in this case standard duration (48 weeks) of treatment] is dominated.

Peginterferon alfa-2a

Sustained virological responses for this patient population are taken from the RCT reported by Jensen and colleagues,88 which compared re-treatment with varying doses and duration of peginterferon alfa-2a in patients who had previously failed to respond to, or relapsed on, peginterferon or non-peginterferon alfa plus ribavirin. This trial was not included in our systematic review of clinical effectiveness, which specified, in line with the scope issued by NICE, that the comparator in trials of re-treated patients should be BSC (i.e. excluding active treatment with interferon alfa). For this analysis, in the absence of any relevant trial data, we assumed that the SVR for the cohort of re-treated patients receiving BSC would be zero. The assumed treatment duration for genotype 1 patients is 72 weeks, based on the SPC for peginterferon alfa-2a. 42 For genotype non-1 patients the treatment duration is 48 weeks (see Appendix 8).

Costs and outcomes modelled for re-treatment in patients previously treated with peginterferon alfa-2a and ribavirin combination therapy are presented in Table 54. This table reports total costs (antiviral treatment and supportive care), health outcomes (in terms of life-years and QALYs) and the incremental cost-per-QALYs ratios.

The impact of re-treating this group of patients is to improve the predicted outcome (by 0.31 and 0.59 QALYs for genotype 1 and genotype non-1, respectively) and to increase lifetime costs (by £16,130 and £6419 QALYs for genotype 1 and genotype non-1, respectively). The reduction in supportive care costs associated with disease progression in both groups of patients (genotype 1 and genotype non-1) is insufficient to fully offset the additional costs of antiviral treatment.

The cost-effectiveness results in Table 54 do not take account of patients withdrawing from treatment owing to adverse events, nor do they consider the impact of treatment stopping rules (e.g. ceasing treatment at 12 weeks in patients who do not demonstrate an EVR). Table 55 reports cost-effectiveness results for re-treated patients, allowing for patient withdrawals owing to adverse effects of treatment with peginterferon alfa and ribavirin combination therapy. This has a marginal impact on the cost-effectiveness results, with the ICER for patients with genotype 1 remaining high.

Table 56 reports cost-effectiveness results for re-treated patients, allowing for the adoption of early stopping rules whereby patients who do not demonstrate an EVR stop treatment at 12 weeks. This has a substantial impact on the cost-effectiveness results, reducing the increase in total costs for patients treated with peginterferon alfa and ribavirin combination therapy to between £1415 and £3398, depending on genotype grouping, while also increasing the QALY gain by approximately 0.06 QALYs. As a result the ICER for patients with genotype 1 falls to £9169.

The EVRs used in the analysis reported in Table 56 are taken from the Roche submission to NICE,104 as Jensen and colleagues88 do not report EVR separately for the genotype groupings used in this analysis. The interpretation of the data available in the MS is difficult, as the number of patients achieving SVR is not reported according to whether patients demonstrated an EVR, for each treatment arm. Rather, the submission reports only predictive values for patients achieving full viral suppression at week 12. The analysis in Table 56 assumes that all patients who achieve an SVR demonstrated an EVR.

Deterministic sensitivity analysis

Table 57 reports the results of a DSA for re-treatment using peginterferon alfa-2a and ribavirin combination therapy in previously treated patients. These are predominantly univariate sensitivity analyses, varying one parameter at a time from its base-case value, leaving all other variables unchanged.

The DSA suggests that the results are robust to a change in structural assumptions (allowing spontaneous SVR from the mild chronic HCV state), the proportion of the baseline cohort that is male and variation in early disease transition probabilities. Reducing drug acquisition costs has the effect of improving the cost-effectiveness of re-treatment, as would be expected, by reducing incremental costs while leaving incremental outcome unchanged.

The results are highly sensitive to two assumptions regarding baseline cohort characteristics. Increasing age at entry to the model is associated with a substantial increase in the ICER – the ICER value approximately doubles if age at entry is increased by 15 years from the base case. This arises as the QALY gain from re-treatment is reduced by approximately 43%, while incremental cost increases by 23%, for genotype 1 patients. The results also appear to be sensitive to the distribution of patients across liver disease stages, at entry to the model. Higher QALY gains are associated with more advanced disease stage, with lower incremental costs; however, in this analysis we have assumed the same SVR in patients with and without cirrhosis. Subsequent analyses suggest that the ICER is also sensitive to variation in the SVR applied for patients with cirrhosis at baseline.

Probabilistic sensitivity analysis

In a PSA where the probabilities of achieving EVR and SVR, health-state costs, health-state utility values and transition probabilities for the natural history parameters were sampled probabilistically, re-treatment using peginterferon alfa-2a and ribavirin combination therapy is associated with increased QALYs (with a range from 0.05 to 0.59 QALYs for genotype 1 patients, and from 0.05 to 2.94 QALYs for genotype non-1 patients), but for genotype 1 patients is typically also associated with increased costs when compared with BSC. Table 58 provides summary information and Figures 10 and 11 scatter plots that also show the 95% confidence ellipses. The incremental cost was negative in approximately 25% of simulations for genotype non-1 patients.

FIGURE 10.

Cost-effectiveness plane for genotype 1 – incremental cost and incremental QALYs for re-treatment using peginterferon alfa-2a and ribavirin combination therapy (applying early stopping rule based on early virological response).

FIGURE 11.

Cost-effectiveness plane for genotype non-1 – incremental cost and incremental QALYs for re-treatment using peginterferon alfa-2a and ribavirin combination therapy (applying early stopping rule based on early virological response).

In this analysis, re-treatment using peginterferon alfa and ribavirin combination therapy for genotype 1 patients has a probability of being cost-effective (compared with BSC) of 90% at a willingness-to-pay threshold of £20,000 per QALY and 98% at a willingness-to-pay threshold of £30,000 if a stopping rule based on EVR is adopted. If patients are treated for the full 72 weeks, regardless of EVR, the equivalent figures are 2% and 11% (Figure 12). For genotype non-1 patients, the probability of re-treatment using peginterferon alfa plus ribavirin being cost-effective (compared with BSC) was 96% at a willingness-to-pay threshold of £20,000 per QALY and 98% at a willingness-to-pay threshold of £30,000, when adopting the stopping rule based on EVR (Figure 13).

FIGURE 12.

Cost-effectiveness acceptability curves for re-treatment of genotype 1 patients with peginterferon alfa-2a, with and without stopping rules based on EVR.

FIGURE 13.

Cost-effectiveness acceptability curves for re-treatment of genotype non-1 patients with peginterferon alfa-2a, with and without stopping rules based on EVR.

Peginterferon alfa-2b

Sustained virological responses for this patient population are taken from the MS by Schering-Plough, which reported treatment outcomes for the EPIC3 study96 (a multicentre, non-randomised open-label uncontrolled study). This study did not meet the inclusion criteria for our systematic review of clinical effectiveness (see Appendix 8 for an explanation of the choice of clinical evidence in this patient group). The assumed treatment duration for all patients is 48 weeks, and the SVR for the cohort of patients receiving BSC is assumed to be zero.

Costs and outcomes modelled for re-treatment in patients previously treated with peginterferon alfa-2b and ribavirin combination therapy are presented in Table 59. This table reports total costs (antiviral treatment and supportive care), health outcomes (in terms of life-years and QALYs) and the incremental cost-per-QALY ratios.

The impact of re-treating this group of patients is to improve the predicted outcome (by 0.39 and 1.72 QALYs for genotypes 1 and 4 and genotypes 2 and 3, respectively) and to increase lifetime costs (by £9380 for genotypes 1 and 4). The reduction in supportive care costs associated with disease progression in genotype 2 + 3 patients, associated with re-treatment with peginterferon alfa-2b and ribavirin combination therapy, is sufficient to fully offset the additional costs of antiviral treatment. This is due to the high SVR reported for genotype 2 + 3 patients (58.3% overall and 56.8% in those demonstrating an EVR) reported for the EPIC3 study,96 in the MS.

The cost-effectiveness results in Table 59 do not take account of patients withdrawing from treatment owing to adverse events or consider the impact of treatment stopping rules (e.g. ceasing treatment at 12 weeks in patients who do not demonstrate an EVR). Table 60 reports cost-effectiveness results for re-treated patients, allowing for patient withdrawals owing to adverse effects of treatment with peginterferon alfa-2b and ribavirin combination therapy; this has a marginal impact on the cost-effectiveness results.

Table 61 reports cost-effectiveness results for re-treated patients, allowing for the adoption of early stopping rules whereby patients who do not demonstrate an EVR stop treatment at 12 weeks. This has a substantial impact on the cost-effectiveness results, reducing the increase in total costs for genotype 1 and 4 patients treated with peginterferon alfa-2b and ribavirin combination therapy to £3256. As a result the ICER for patients with genotype 1 falls to £7681.

Deterministic sensitivity analysis

Table 62 reports the results of a DSA for re-treatment using peginterferon alfa-2b and ribavirin combination therapy in previously treated patients. These are predominantly univariate sensitivity analyses – that is, varying one parameter at a time, from its base-case value, leaving all of the other variables unchanged. The DSA suggests that the results are robust to a change in structural assumptions (allowing spontaneous SVR from the mild chronic HCV state), the proportion of the baseline cohort that is male and transition probabilities for early disease states. Reducing drug acquisition costs has the effect of reducing the ICER, as might be expected, as it reduces the drug costs while the outcome difference is unchanged.

The greatest variability in ICERs is associated with changes in the age at which patients enter the model, the distribution of patients across disease stages and (to a lesser extent) response to treatment (SVR) for patients with cirrhosis. Increasing the mean age of patients at the start of the simulation up to 15 years leads to an approximate doubling of the ICER for genotype 1 and 4 patients and results in a positive, though low-value, ICER for genotype 2 and 3 patients. In both cases, the QALY gain with treatment is approximately halved. Similarly, alternative assumptions regarding the stage of liver disease in which patients enter the model has a large impact on the ICER, with less favourable results associated with patients being in the earlier (lower fibrosis) stages of disease. For genotype 2 and 3 patients the ICER becomes positive if all patients in the modelled cohorts have mild chronic HCV (rather than moderate chronic HCV or compensated cirrhosis).

Probabilistic sensitivity analysis

In a PSA, where the probabilities of achieving EVR and SVR, health-state costs, health-state utility values, and transition probabilities for the natural history parameters were sampled probabilistically, re-treatment using peginterferon alfa and ribavirin combination therapy is associated with increased QALYs (with a range from 0.08 to 0.80 QALYs for genotypes 1 and 4 and from 0.28 to 3.06 QALYs for patients with genotypes 2 and 3), but for genotype 1 and 4 patients is typically also associated with increased costs when compared with BSC (Table 63 provides summary information and Figures 14 and 15 scatter plots, which also show the 95% confidence ellipses). The incremental cost was negative in approximately 84% of simulations for genotype 2 and 3 patients.

FIGURE 14.

Cost-effectiveness plane for genotypes 1 and 4 – incremental cost and incremental QALYs for re-treatment using peginterferon alfa-2b and ribavirin combination therapy (applying early stopping rule based on EVR).

FIGURE 15.

Cost-effectiveness plane for genotypes 2 and 3 – incremental cost and incremental QALYs for re-treatment using peginterferon alfa-2b and ribavirin combination therapy (applying early stopping rule based on EVR).

In this analysis, re-treatment using peginterferon alfa-2b and ribavirin combination therapy for genotype 1 and 4 patients had a probability of being cost-effective (compared with BSC) of 99% at a willingness-to-pay threshold of £20,000 per QALY, and 100% at a willingness-to-pay threshold of £30,000 if a stopping rule based on EVR is adopted. If patients are treated for the full 48 weeks, regardless of EVR, the equivalent figures are 24% and 74% (Figure 16). For genotype 2 and 3 patients the probability of re-treatment using peginterferon alfa-2b and ribavirin being cost-effective (compared with BSC) was 100% at a willingness-to-pay threshold of £20,000 per QALY, and at a willingness-to-pay threshold of £30,000 when adopting the stopping rule based on EVR (Figure 17).

FIGURE 16.

Cost-effectiveness acceptability curves for re-treatment of genotype 1 and 4 patients with peginterferon alfa-2b, with and without stopping rules based on EVR.

FIGURE 17.

Cost-effectiveness acceptability curves for re-treatment of genotype 2 and 3 patients with peginterferon alfa-2b, with and without stopping rules based on EVR.

Peginterferon alfa-2a

Costs and outcomes modelled for patients co-infected with HCV/HIV receiving combination therapy with peginterferon alfa-2a plus ribavirin are presented in Table 64.

The impact of treating this group is to improve the predicted outcome (by 0.75 and 1.86 QALYs for genotypes 1 and 4 and genotypes 2 and 3, respectively) and to increase lifetime costs for patients with genotypes 1 and 4 (by £5932). However, in patients with genotypes 2 and 3 the modelled reduction in supportive care costs (in the peginterferon-treated cohort) offsets the additional costs of antiviral treatment; in this situation the strategy of providing antiviral treatment dominates.

The cost-effectiveness results in Table 64 do not take account of uncertainties regarding the potential impact of HIV co-infection on the natural history of HCV infection, overall mortality, utility gains from successful treatment or additional costs of on-treatment monitoring.

A published meta-analysis25 suggests that a RR for cirrhosis of 2.07 (95% CI 1.4 to 3.07) and a RR for decompensation of 6.14 (95% CI 2.86 to 13.20) in HCV/HIV co-infected patients compared with that in HCV mono-infected patients. Table 65 reports the cost-effectiveness results from the model when these RRs for liver disease progression are applied to the baseline risks in the natural history model. This suggests that treatment using peginterferon alfa-2a and ribavirin combination therapy will be more cost-effective in HCV/HIV co-infected patients, if the risks of fibrosis progression are greater than for mono-infected patients.

Table 66 reports the cost-effectiveness results from the model when the age-specific mortality risks are doubled for HCV/HIV co-infected patients. This would result in an age-specific life expectancy of 33.2 years at the age of 40 years for an HIV infected person (in the absence of chronic liver disease) compared with 39.8 years if the age-specific mortality risks for the general population are applied (as in the base-case analysis). This reduces lifetime costs and QALYs both for peginterferon treated and BSC cohorts. This suggests that treatment using peginterferon alfa-2a and ribavirin combination therapy will be less cost-effective in HCV/HIV co-infected patients, if mortality risk is greater than for mono-infected patients. However, while the incremental cost for peginterferon treatment increases and the QALY gain is reduced, with higher mortality risk for HCV/HIV co-infected patients, treatment with peginterferon still dominates BSC for genotype 2 and 3 patients.

Tables 67 and 68 report cost-effectiveness results from alternative assumptions on the utility gain for HCV/HIV co-infected patients who achieve an SVR. In the first case, the utility gain is assumed to be one-half of that reported for HCV mono-infected patients, and in the second case the utility gain is assumed to be zero. In both cases the QALY gain from treatment with peginterferon is reduced, indicating that treatment using peginterferon alfa-2a and ribavirin combination therapy will be less cost-effective in HCV/HIV co-infected patients, if utility gain from SVR is lower in HCV/HIV co-infected patients than in mono-infected patients.