Adefovir dipivoxil and pegylated interferon alpha for the treatment of chronic hepatitis B: an updated systematic review and economic evaluation

Proportion of patients achieving an HBV DNA response – ADV studies

Table 5 presents the proportion of patients achieving a defined threshold of HBV DNA response in the two trials. In the trial by Zeng and colleagues,19 the proportion of patients whose HBV DNA level dropped below 10⁵ copies/ml reached 67% by week 52 in patients treated continuously with ADV [the adefovir–adefovir–adefovir (AAA) group]. For the group who commenced open-label ADV after 12 weeks of placebo [the placebo–adefovir–adefovir (PAA) group], the proportion reached 70% at week 52. For the group who initially received ADV and were then re-randomised to placebo at week 40 [the adefovir–adefovir–placebo (AAP) group], the proportion fell to 11% at week 52, having previously reached 59%. Zeng and colleagues19 also reported the proportion of patients whose HBV DNA was undetectable, defined as HBV DNA < 300 copies/ml. At week 52, the proportions were similar in the AAA and the PAA groups (28% and 30% respectively). In the AAP group, the proportion fell from 59% at week 40 to 11% by week 52, following the withdrawal of ADV. No statistical tests were reported for any of these comparisons.

In the trial by Rapti and colleagues,18 the proportion of patients with HBV DNA ≤ 1000 copies/ml after 24 months of therapy was higher in the group treated with ADV and LAM than in the group treated with ADV monotherapy (82.6% versus 75% respectively). However, this difference did not reach statistical significance.

Table 6 presents HBV DNA response rates for all 125 patients in the long-term open-label study. 20 (Note that results after 240 weeks apply only to the ADV–ADV group as the placebo–ADV group commenced ADV only after week 48.)

In the ‘ITT missing data = failure for resistance or HCC’ analysis (i.e. missing values from patients who left the study for other reasons were excluded), the proportion of patients whose HBV DNA levels were < 1000 copies/ml peaked after 96 weeks, after which they gradually fell to 67%. In the ‘ITT missing data = failure’ analysis, the proportion peaked after 144 weeks, and fell to 53% after 240 weeks.

In summary, a greater proportion of ADV-treated patients experienced an HBV DNA response relative to comparators, although this was not confirmed statistically. The proportion of responders was generally maintained during long-term treatment.

Proportion of patients achieving an HBV DNA response – PEG-α-2b studies

Table 7 shows the proportion of patients achieving an HBV DNA response as reported in four of the five studies. Response was measured by reductions in HBV DNA levels to a given threshold. Each of the studies of PEG-α-2b used differing thresholds of HBV DNA response. These proportions also vary across the studies.

Chan and colleagues23 report HBV DNA response for three groups of patients, receiving staggered regimens of PEG-α-2b and LAM. At week 52, Group A had a higher proportion of patients (44%) with negative HBV DNA than either Group B (22%) or Group C (10%). Neither of these differences was statistically significant (p = 0.62 and p = 0.24). At end of treatment (week 104) Group B had the highest proportion of patients (56%) with negative HBV DNA, compared with Group A (33%) and Group C (40%). Baseline HBV DNA had been significantly lower in this group. At end of follow-up (week 128), Group A had a slightly higher proportion of patients reaching ‘undetectable’ HBV DNA of 22%, compared with Group B at 11% and Group C at 20%. Statistical significance was not given for these results. Definitions of ‘negative’ and ‘undetectable’ are not reported, and so it is unclear how these differ.

The trial by Janssen and colleagues26 compares a group receiving PEG-α-2b and LAM (Group A), and a group receiving PEG-α-2b and placebo (Group B). At end of treatment (week 52) the proportion of patients in Group A reaching < 200,000 copies/ml was 74%, compared with 29% in Group B (p < 0.0001). At end of follow-up this had fallen in Group A to 32% and in Group B to 27% (p = 0.44). Thirty-three per cent of patients in Group A at end of treatment reached < 400 copies/ml, compared with 10% in Group B (p < 0.0001). This, again, fell by end of follow-up to 9% in Group A and 7% in Group B (p = 0.43).

In the trial by Kaymakoglu and colleagues,25 a higher proportion of patients receiving PEG-α-2b and LAM achieved HBV DNA < 4 pg/ml (picograms per millilitre; in this paper defined as the ‘lower limit of detection’) at end of treatment (week 48) than of those receiving PEG-α-2b alone: 79% versus 63% respectively. At end of follow-up, the proportions reaching 400 copies/ml were similar for both groups: 24% versus 26%. However, none of these differences was statistically significant.

Zhao and colleagues27 compared patients receiving PEG-α-2b (Group A) with those taking IFN-α (Group B). A higher proportion of patients in Group A (29.6%) reached < 5 log₁₀ copies/ml at end of follow-up (week 48), compared with 19.1% in Group B (p = 0.06). In both groups, 12.2% reached < 3 log₁₀ copies/ml (p = 1.00).

In summary, the results of these studies show that there was no significant difference between concurrent and staggered commencement of PEG-α-2b and LAM; and no consistent statistically significant differences between the combination of PEG-α-2b and LAM versus PEG-α-2b monotherapy, or between PEG-α-2b and IFN-α.

Changes in HBV DNA levels – ADV studies

Table 8 reports the median changes in HBV DNA from baseline in the trial by Zeng and colleagues. 27

At week 12 there was a statistically significant difference between the patients randomised to ADV (the AAA group) and those randomised to placebo (the PAA group) (p < 0.001). At week 40, median reductions in HBV DNA appeared to be similar for all three groups (all three had been receiving open-label ADV since week 12). At week 52, median HBV DNA reductions appeared to be similar for the patients who had received ADV continuously (the AAA group) and those who had switched to ADV from placebo after week 12 (the PAA group) (–4.5 and –5.0 log₁₀ copies/ml respectively). However, those who had received ADV until week 40 and were then re-randomised to placebo had a smaller reduction (–0.2 log₁₀ copies/ml).

Changes in HBV DNA levels – PEG-α-2b studies

Changes in HBV DNA levels are shown in Table 9.

Chan and colleagues (2007)23 report the median log HBV DNA reduction from baseline at weeks 4, 8, 52 and 104. The median difference between Group A and Group B was 3.59 (95% CI 1.49–5.65, p < 0.0001) at week 4 (at this stage, because of the staggered regimen, Group A was receiving PEG-α-2b and LAM, Group B was receiving PEG-α-2b monotherapy and Group C received LAM monotherapy). At week 52, the point at which the primary outcome was measured, the difference between Groups A and B was significant (6.38 versus 3.43, p = 0.030), but the difference between Groups A and C was not (p = 0.06). By end of treatment, week 104, there were no significant differences in median log HBV DNA reduction between any of the groups.

In the trial by Chan and colleagues (2005),24 there were greater reductions in HBV DNA for combination treatment than for monotherapy. The median difference between groups at end of treatment was reported as 1.24 copies/ml (95% CI 0.78–1.66); however, no statistical tests were reported for this outcome.

The HBV DNA change from baseline in Janssen and colleagues26 was estimated by the reviewers from a figure in the paper, and showed a similar reduction in mean log HBV DNA copies/ml of 2.3 for Group A and 2.2 for Group B at end of follow-up.

Zhao and colleagues27 reported the HBV DNA mean reduction from baseline, log₁₀ copies/ml at end of treatment and end of follow-up. The difference was significant at week 24 (end of treatment) (Group A 2.22 versus Group B 1.66, p = 0.03), but again was not statistically significant by end of follow-up (p = 0.34). At end of follow-up, the ‘mean reduction’ in Group A was –1.4 ± 2.2, and in Group B it was –1.1 ± 2.1, indicating that overall there had been an increase in the HBV DNA level in both groups from baseline.

In summary, the results of the trials show that PEG-α-2b is generally associated with greater reductions in HBV DNA levels than are comparators. This was the case when PEG-α-2b was added to LAM, or vice versa, and for PEG-α-2b versus IFN. However, there were no consistent statistically significant differences between treatments. The results suggested a greater reduction in HBV DNA for those who received concurrent commencement of PEG-α-2b and LAM, but by the end of treatment there were no statistically significant differences between this and the staggered regimens.

ALT normalisation – ADV studies

Table 10 reports the proportion of patients with normal ALT levels in two of the studies.

Four of the trials reported changes in HBV DNA. In the trial by Zeng and colleagues,19 the proportion of patients with normal ALT appeared similar at week 12 in the groups randomised to receive ADV (42% and 44% in the AAA and AAP groups respectively). There was a statistically significant difference between the AAA and AAP groups combined compared with the group randomised to placebo (the PAA group). The proportion of responders in the AAA and AAP groups remained similar to each other at week 40 after 28 weeks of open-label ADV (73% and 74% respectively). At week 40, in the group randomised to placebo in the first 12 weeks and who had subsequently received open-label ADV for 28 weeks (the PAA group), the proportion of ALT responders was slightly lower than in the other two groups (65%). At week 52, the proportion of responders in the PAA group had increased to 69%, while the proportion of responders of those who had received ADV continuously (the AAA group) had increased to 79%. The proportion of responders fell to 21% in the group re-randomised at week 40 to placebo (the AAP group). No statistical tests were reported for these comparisons.

In the trial by Rapti and colleagues, there was a higher proportion of responders in the group who received ADV and LAM compared with the group that received ADV monotherapy (91% versus 72.7% respectively). However, the difference was not statistically significant.

Table 11 presents HBV DNA response rates for all 125 patients in the long-term open-label study. 20 (Note that results after 240 weeks apply only to the ADV–ADV group as the placebo–ADV group commenced ADV only after week 48.)

The proportion of patients with normal ALT values declined from 75% after 48 weeks to 69% and 59% at week 240 for the ‘ITT missing data = failure for resistance or HCC’ and ‘ITT missing data = failure’ analyses respectively.

In summary, a greater proportion of ADV-treated patients experienced ALT normalisation relative to comparators, although this was not always statistically significant. The proportion of ALT responders was generally maintained during long-term treatment.

ALT normalisation – PEG-α-2b studies

Table 12 reports the proportions of patients with ALT normalisation in the PEG-α-2b studies.

All five of the PEG-α-2b trials reported results of the proportion of patients with ALT normalisation. 23–27 All reported ALT normalisation at end of follow-up, and four also reported results at end of treatment. 23–26 None stated the ‘normalisation’ threshold used.

In the trial by Chan and colleagues (2007),23 three groups of patients on staggered regimens of PEG-α-2b and LAM had similar rates of ALT normalisation at end of treatment [Group A 9 (100%), Group B 9 (100%), Group C 9 (90%)]. At end of follow-up these numbers had decreased to Group A 4/9 (44%), Group B 5/9 (56%) and Group C 4/9 (40%). There were no statistical tests reported.

In the study by Chan and colleagues (2005),24 a slightly higher proportion (90%) of Group A (PEG-α-2b and LAM), had normal ALT levels compared with 78% of Group B (LAM alone) at end of treatment. This was similar at follow-up (Group A 50% versus Group B 30%). No statistical tests were reported for these differences.

The trial by Janssen and colleagues26 found a significant difference between Group A (PEG-α-2b and LAM) and Group B (PEG-α-2b and placebo) in ALT normalisation at end of treatment (51% and 34% respectively, p = 0.005). By end of follow-up, the difference between groups was no longer significant (35% and 32% respectively, p = 0.60).

There were no significant differences in the proportions of patients reaching normal ALT levels at end of treatment or end of follow-up in the trial by Kaymakoglu and colleagues. 25 The proportions of patients with normal ALT at these points were slightly higher in Group B (PEG-α-2b and LAM) (p > 0.05). Again, these proportions had decreased between end of treatment and end of follow-up, e.g. 53% of patients in Group A (PEG-α-2b) at end of treatment and 42% at end of follow-up.

The proportion of patients reaching ALT normalisation in the trial by Zhao and colleagues27 was very similar at end of follow-up for the two study groups (33.9% in Group A receiving PEG-α-2b versus 34.8% in Group B receiving IFN-α; p = 0.93).

In summary, the proportion of patients with normalised ALT tended to be greater for the combination of PEG-α-2b and LAM compared with either as monotherapy. The proportions were generally similar when comparing different commencement regimens of PEG and LAM, or PEG-α-2b with IFN-α. Rates of normalisation usually fell between end of treatment and follow-up. Few statistically significant differences between treatments were reported.

Changes in ALT levels – ADV studies

Table 13 presents median ALT levels in the two ADV trials which reported this outcome.

In the trial by Zeng and colleagues,19 median ALT was similar at week 12 in the groups randomised to receive ADV (1.1 × ULN, in both the AAA and AAP groups). Median ALT remained similar in the AAA and AAP groups at week 40, after 28 weeks of open-label ADV (0.7 and 0.9 × ULN respectively). At week 40, in the group randomised to placebo in the first 12 weeks and who had subsequently received open-label ADV for 28 weeks (the PAA group), median ALT was similar to the other two groups (0.9 × ULN). At week 52, median ALT levels remained similar in the PAA and AAA groups, but increased to 3.0 × ULN in the group who were re-randomised to placebo at week 40 (the AAP group). No statistical tests were reported for these comparisons.

In the trial by Rapti and colleagues,18 median ALT fell to around 24 IU/l in both treatment groups at 24 months of treatment, with no statistically significant difference between groups.

Changes in ALT levels – PEG-α-2b studies

Only one of the PEG-α-2b studies reported this outcome (Table 14). In the trial by Chan and colleagues,24 medial ALT fell to around 60–70 IU/l at 24 weeks’ post-treatment follow-up.

HBeAg loss/seroconversion – ADV studies

Table 15 presents rates of HBeAg loss and seroconversion for the Zeng and colleagues trial. 19 (This outcome was not reported in the study by Rapti and colleagues,18 and was not applicable to the study of HBeAg-negative patients in the study by Hadziyannis and colleagues. 20)

At week 52, rates of HBeAg loss and seroconversion were highest for patients randomised to placebo in the first 12 weeks and who then received open-label ADV until week 52 (the PAA group). The authors attribute this to the six cases of HBeAg loss in this group within the first 12 weeks (actual date not recorded) which, it is suggested, represent spontaneous cases. At week 40, 16 of the 114 patients (14%) in the AAP group lost HBeAg. Following re-randomisation to placebo at week 40, nine of them regained HBeAg between week 40 and week 52. The higher rate of seroconversion in the PAA group at week 52 compared with the AAA group is again attributed by the study authors to the six patients who spontaneously seroconverted on placebo in the first 12 weeks. The authors also acknowledged that the seroconversion rate in the AAA group was lower than achieved in other trials. In summary, the results of this study suggest that HBeAg seroconversion rates are generally maintained with continued ADV treatment, and are reduced when treatment is withdrawn.

HBeAg loss/seroconversion – PEG-α-2b studies

Table 16 reports the results for HBeAg loss and seroconversion in four of the PEG-α-2b studies. This outcome was not applicable in the trial of HBeAg-negative patients by Kaymakoglu and colleagues. 25 All four trials reports HBeAg seroconversion at end of follow-up; three of these report rates at end of treatment.

Chan and colleagues (2007)23 compared three groups who received staggered regimens of PEG-α-2b and LAM as described above. At end of treatment (week 104) 56% of patients in Group A, 33% of patients in Group B and 60% of patients in Group C had seroconverted. At week 128, these figures remained the same in Groups A and B and had fallen in Group C to 40%. None of the between-group differences were statistically significant.

Chan and colleagues (2005)24 report the rate of seroconversion at end of treatment in Group A receiving PEG-α-2b and LAM (8 weeks of PEG-α-2b only, followed by 24 weeks of combination and then 28 weeks of LAM only) compared with Group B receiving LAM alone. In Group A, 60% of patients had seroconverted during the last 28 weeks of LAM therapy; in Group B this figure was 28%. No statistical tests were reported for these results.

Janssen and colleagues26 reported HBeAg loss and HBeAg seroconversion at end of treatment (week 52) and end of follow-up (week 78). At week 52, 44% of patients in Group A (PEG-α-2b and LAM) lost HBeAg compared with 29% in Group B (PEG-α-2b and placebo) (p = 0.01). Also at this time, 25% of patients in Group A and 22% of patients in Group B had seroconverted (p = 0.52). At week 78, 35% of patients in Group A and 36% of patients in Group B had lost HBeAg (p = 0.91). At this time, 38 patients (29%) in Group A and 29 patients (29%) in Group B had seroconverted (p = 0.92).

Zhao and colleagues27 reported that the difference between groups in the rate of HBeAg loss at end of treatment (week 24) was not significant: Group A (PEG-α-2b monotherapy) 22.6% versus Group B (IFN-α monotherapy) 17.4% (actual p-value not reported). At end of follow-up (week 48) this number had decreased in Group B (i.e. some of the patients re-acquired HBeAg; Group A 24.4% versus Group B 13.9%) and the difference was statistically significant (p = 0.04). HBeAg seroconversion rates at end of follow-up were 21.7% in Group A and 13.9% in Group B (p = 0.92).

In summary, the results for HBeAg seroconversion and loss were mixed. In one trial, higher rates of HBeAg seroconversion were reported for the combination of PEG-α-2b and LAM compared with LAM monotherapy, although these were not confirmed statistically. In another trial, HBeAg seroconversion rates were similar for PEG-α-2b and LAM compared with PEG-α-2b monotherapy. There were higher rates of seroconversion for PEG-α-2b compared with IFN-α, but the difference was not statistically significant. There was also no significant difference between staggered commencement regimens of PEG-α-2b and LAM.

HBsAg loss/seroconversion – ADV studies

This outcome was reported in only one of the three ADV studies, the long-term open-label follow-up study by Hadziyannis and colleagues. 20

Six patients (5%) had HBsAg loss after a median of 196 weeks (range 20–260) of ADV. Five of these patients developed antibody to hepatitis B surface antigen (anti-HBs) at their last measurement.

HBsAg loss/seroconversion – PEG-α-2b studies

Table 17 reports HBsAg loss or seroconversion as reported in three of the PEG-α-2b studies. 24,26,27

Chan and colleagues24 reported that five patients in Group A (staggered regimen of PEG-α-2b and LAM) and seven in Group B (LAM alone) underwent HBsAg clearance. No statistical test was reported.

Janssen and colleagues26 reported results for HBsAg loss and seroconversion at both end of treatment (week 52) and end of follow-up (week 78). The results show very similar, small proportions of patients from each group undergoing HBsAg loss or seroconversion. For example, HBsAg loss at end of treatment had occurred in nine patients (7%) in Group A and in seven (5%) in Group B (p = 0.54). By end of follow-up this was nine (7%) in both groups (p = 0.92).

In the trial by Zhao and colleagues,27 none of the patients in Group A (receiving PEG-α-2b alone) had experienced HBsAg seroconversion, compared with two (1.7%) in Group B (receiving IFN therapy alone). This difference was not statistically significant (p = 0.50).

In summary, rates of HBsAg loss/seroconversion were comparatively low (< 15%) and there were no consistent statistically significant differences between treatment groups. However, statistical differences are less likely to be reported for a relatively rare outcome.

Liver histological response – PEG-α-2b studies

Liver histological response, as reported in three of the PEG-α-2b studies,23,24,26 is shown in Table 19. One of these trials reported using both the Ishak and the Knodell classification systems,24 another used the Ishak system26 and the third trial did not state which system was used. 23

Chan and colleagues (2007)23 reported no significant differences across their three groups receiving staggered regimens of PEG-α-2b and LAM (described above), although group B had a slightly lower proportion of patients for each change: six (75%) for both necroinflammation improvement and fibrosis change, compared with eight (89%) in Group A for both results, and eight (89%) and nine (100%) in Group C for necroinflammation and fibrosis change respectively.

In the trial by Chan and colleagues (2005),24 the results across groups were again broadly similar, for both improvements and worsening in the necroinflammatory score; e.g. four (10%) in Group A (PEG-α-2b) versus four (9%) in Group B (LAM) for a ≥ 2-point decrease in this score. For both improvement and worsening in fibrosis scores, Group A had the highest proportion of patients: six (15%) versus four (9%) for a ≥ 2-point decrease, for example. No p-values are reported for these results.

Janssen and colleagues26 reported improvement, no change and deterioration for both fibrosis and inflammation. A higher proportion in Group A (PEG-α-2b and LAM) improved their fibrosis scores: 33% versus 22% in Group B (PEG-α-2b and placebo). A higher proportion in Group B experienced no change, and the proportions were equal for deterioration (38% in each group). There was no statistically significant difference between groups for improvement or no change versus worsening (p = 0.22).

This pattern was reversed for the inflammation results, with Group B having a higher proportion of patients with improvement. The groups were again equal for the proportion that deteriorated. Again, the differences between groups for improvement and no change versus worsening were not statistically significant. The authors of this study advise caution with these results as post-treatment biopsies were optional and selection bias may have occurred. 26

A follow-up paper to this study29 reported the improvements in necroinflammatory and fibrosis scores (mean ± SD, range) and these are presented in Table 19. Results show that Group B had a smaller reduction in necroinflammatory score than Group A (–1.5 and –1.7 respectively). While the improvements in these scores within groups were significant, no p-value is given for the difference between groups. This is repeated for the changes in fibrosis score, where again improvements within the groups were significant, but the difference between groups in improvements was not (p = 0.59).

In summary, there were mixed results for liver histology. In some instances, differences between treatments in necroinflammation and fibrosis favoured PEG-α-2b and LAM combination therapy, while in other instances, PEG-α-2b or LAM monotherapy appeared more favourable. Where statistical tests were reported there were no significant differences between treatments.

Combined outcomes – ADV studies

None of the ADV studies reported combined outcome measures.

Combined outcomes – PEG-α-2b studies

Table 20 reports results for the two PEG-α-2b trials that reported combined outcome measures. In the trial by Chan and colleagues,24 a virological response was defined as HBeAg seroconversion, detection of antibody to HBeAg and HBV DNA level < 500,000 copies/ml and normalisation of ALT. In a follow-up study,28 sustained virological response (SVR) was defined as persistent HBeAg loss and HBV DNA < 100,000 copies/ml from treatment cessation until the end of follow-up (up to 124 weeks). In the Zhao and colleagues trial,27 a ‘sustained combined response’ was defined as serum HBV DNA level < 10⁵ copies/ml, HBeAg loss and normal ALT levels.

In the study by Chan and colleagues,24 60% of patients taking the combination of PEG-α-2b and LAM (Group A) achieved the virological response at week 52, compared with 28% in the LAM group (Group B) (p = 0.001). Results are also given for SVR at both follow-up and long-term follow-up. A higher proportion of patients in Group A (18/50, 36%) achieved an SVR at 24 weeks’ follow-up than of those in Group B (7/50, 14%) (p = 0.011). At end of long-term follow-up, these numbers decreased in Groups A and B (29% and 8% respectively). No statistical test is reported for this difference.

In the trial by Zhao and colleagues,27 a slightly higher proportion of patients receiving PEG-α-2b (17.4%) achieved the sustained combined response than of those receiving IFN-α (10.4%). This difference was not statistically significant.

Viral resistance – ADV studies

Table 21 presents rates of viral resistance in two of the ADV studies.

No participants in the trial by Zeng and colleagues19 developed a resistance to ADV during the course of the study. Those included in analysis for resistance were patients with an increase in serum HBV DNA of at least 1 log₁₀ copies/ml while on ADV from their lowest point during treatment and therefore had isolates analysed for the presence of ADV associated-mutations at week 52. These totalled 45 in the study, and are distributed among treatment groups as shown Table 21.

In the trial by Rapti and colleagues,18 a higher proportion of patients (21%) developed resistance in the group receiving ADV alone, compared with 0% of patients in the groups receiving ADV and LAM combination therapy. This difference was statistically significant (p = 0.0182).

Table 22 presents HBV DNA response rates for all 125 patients in the long-term open-label study. 20 (Note that results after 240 weeks apply only to the ADV–ADV group as the placebo–ADV group only commenced ADV after week 48.)

The incidence of three definitions of resistance was measured: (1) ADV resistance mutations (N236T or A181V); (2) ADV-resistant mutations with HBV DNA increased from nadir by at least 1 log₁₀ copies/ml (confirmed or last measurement) or never suppressed to less than 3 log₁₀ copies/ml (‘virological resistance’); and (3) ADV resistance mutations with virological resistance and ALT elevations (ALT greater than ULN after normalising ALT; ‘clinical resistance’). At 240 weeks of ADV treatment, the cumulative probabilities of ADV resistance were 29%, 20% and 11% for the three definitions of resistance respectively.

In summary, rates of ADV reported in these studies were relatively low and tended to remain so over long-term treatment.

Viral resistance – PEG-α-2b studies

Table 23 presents rates of LAM resistance rates in three of the four PEG-α-2b trials that included LAM. 23–26

Chan and colleagues (2007)23 reported low rates of LAM resistance. In Group C there were two patients with LAM resistance, compared with one patient in each of Groups A and B at week 104 (end of treatment).

In the trial by Chan and colleagues (2005),24 a higher proportion of patients exhibited LAM resistance in the group receiving LAM monotherapy than in the group receiving PEG-α-2b and LAM (40% versus 21%). No statistical test was reported for these results.

In the trial by Janssen and colleagues,26 14 patients (11%) in Group A (PEG-α-2b and LAM) exhibited LAM resistance at the end of treatment. The comparison group in this trial did not receive LAM. Seven of the 14 had previously been treated with LAM and had a mutant from the start of therapy.

In summary, the addition of LAM to PEG-α-2b was associated with lower rates of LAM resistance, although this was not confirmed statistically.

Adverse events – ADV studies

Table 24 reports dose discontinuations, reductions and incidence of serious adverse events in the ADV studies.

Zeng and colleagues19 reported that two patients in the AAA group and one patient in the AAP group discontinued the study drug because of adverse events. The authors state that adverse events were similar in nature and severity between treatment groups. No patients died during the trial. The authors stated that adverse events rarely occurred at a frequency of 5% or greater in any treatment group. The incidence of CHB-related adverse events was reported, with 9% in the AAP group, compared with < 1% in the AAA and PAA groups. The authors report that these events all occurred after patients in the AAP group were randomised to placebo at week 40.

There were no dose discontinuations in the trial by Rapti and colleagues,18 but two patients in the group receiving ADV and LAM had their ADV dose reduced, compared with zero reductions in the ADV monotherapy group. Three patients experienced HCC during the course of the trial, all in the ADV and LAM group. There was no statistically significant difference between groups (p = 0.545). Only serious adverse events were reported.

In summary, the incidence of adverse events was low and generally similar between treatment groups.

Adverse events – PEG-α-2b studies

Table 25 reports dose discontinuations, reductions and incidence of serious adverse events in the PEG-α-2b studies.

Chan and colleagues (2007)23 compared three groups receiving staggered regimens of PEG-α-2b and LAM as described above. Two patients in both Group A and Group C had their dose of PEG-α-2b halved owing to neutropenia (0.7–0.9 × 10⁹/ml) at doses 6–19. While no patients in Group B had a dose reduction, one experienced a serious adverse event (hysterectomy for menorraghia). The authors state that this was unrelated to the study drug, and the study medication was uninterrupted. No patients in this study died. Each of the groups in this trial experienced generally similar numbers of adverse events; however, Group B (receiving PEG-α-2b for 8 weeks prior to commencing LAM) experienced the fewest adverse events. The largest apparent difference was in upper respiratory tract symptoms, where one incident occurred in Group B, compared with five (50%) in Group A and seven (70%) in Group C (not shown in Table 25; for more information refer to Appendix 1).

In the trial by Chan and colleagues (2005),24 Group A (PEG-α-2b and LAM) had a larger proportion than Group B (LAM monotherapy) of patients discontinuing the study drug (4 versus 0), dose reduction (5 versus 0) and serious adverse events (4 versus 0). The authors report that most adverse events were transient and related to the use of PEG-α-2b. No patient died or required liver transplantation. Reduction of the PEG-α-2b dose to 50 µg/week (if body weight > 65 kg) or 1.0 µg/kg/week if < 65kg) was due to anaemia (n = 1), neutropenia (n = 3) and/or thrombocytopenia (n = 4). PEG-α-2b was discontinued in all four cases of serious adverse events, which were: bipolar disorder, pulmonary tuberculosis, thyrotoxicosis and severe local reaction at injection sites. Three patients continued with LAM through to week 60, while the fourth patient withdrew from the study and was considered a treatment failure.

The most common adverse events in Groups A and B were upper respiratory tract symptoms, which included cough, running nose and sore throat. Group A had a higher proportion of patients experiencing adverse events across all types, and all of these differences between the two groups were statistically significant, apart from vomiting and diarrhoea, weight loss > 10% and abdominal discomfort (not shown in Table 25, for more information refer to Appendix 1).

A follow-up publication30 of the trial by Janssen and colleagues26 reported that a slightly higher percentage of patients in Group A (PEG-α-2b and LAM) than Group B (PEG-α-2b and placebo) had the dose of PEG-α-2b reduced for any adverse event: 37 (54%) and 32 (47%) respectively. There was no significant difference between treatment groups, for these and for all side effects. The authors of this study reported discontinuations of PEG-α-2b across the two groups of patients (28, 9%) but not how these were distributed between groups. Neutropenia was the most common reason for dose reduction in this trial (n = 36, 52%). The most common reason for early discontinuation was local reaction (n = 10, 36%). There were no dose reductions of LAM or placebo. Fifty per cent of the dose reductions occurred within the first 10 weeks, with numbers of dose reductions decreasing thereafter and only two reported after week 32, when the scheduled dose reduction took place. Discontinuation of therapy was reported more frequently before the scheduled dose reduction of PEG-α-2b at week 32.

There were 33 serious adverse events in the trial by Janssen and colleagues;26 the authors state that 17 (53%) were probably related to therapy, and that all were reversible after treatment had stopped. The frequency of all side effects is reported as not being statistically significant between groups.

Kaymakoglu and colleagues25 reported that no patients from either treatment group had their dose reduced or discontinued for any adverse event. Of the adverse events experienced, 71% were of flu-like symptoms. The authors do not comment on severity or likelihood of relation to the study drug.

The study drug was discontinued in four patients in the group receiving IFN-α, and in no patients in the group receiving PEG-α-2b in the trial by Zhao and colleagues. 27 There were no dose reductions in either group. The authors report that 75% of patients in each group experienced ‘various forms’ of drug-related adverse events, the most common of which, again, were flu-like symptoms and fever.

In summary, there were mixed findings for adverse events. In some trials, the incidence of events and dose discontinuations was generally similar between treatments. In at least one trial, there was a higher incidence of events and discontinuations for PEG-α-2b and LAM compared with LAM. Common adverse events included flu-like symptoms and upper respiratory tract infections.

Details of economic evaluations based on the Roche model

As noted earlier, two economic evaluations (Veenstra and colleagues44 and Sullivan and colleagues45) use the same model to evaluate PEG-α-2a versus LAM in HBeAg-positive patients. This model uses the structure and some of the transition probabilities presented in 2005 in the Roche submission to NICE for their appraisal of PEG-α-2a. The model evaluated a 48-week course of PEG-α-2a versus comparators IFN-α, LAM, ADV and best supportive care. (For a fuller description see our previous report. 12) However, unlike the Roche model, economic evaluations reported in Veenstra and colleagues44 and Sullivan and colleagues45 apply exclusively to an HBeAg-positive population.

The 48-week outcomes of the RCT of PEG-α-2a versus LAM reported by Lau and colleagues50 provided short-term clinical effectiveness data for the base-case analysis in all three of these models. Long-term clinical effectiveness data (rates of seroconversion, relapse and LAM resistance) were taken from previously published studies (Liaw and colleagues,51 Leung and colleagues52 and Lok and colleagues53). The Roche model estimated cost-effectiveness of a 48-week course of PEG-α-2a versus two LAM treatment alternatives: treatment for 48 weeks and for 4 years. Veenstra and colleagues44 estimated cost-effectiveness of a 48-week course of PEG-α-2a versus up to 4 years of LAM treatment (i.e. patients who do not achieve a sustained seroconversion after 48 weeks continue LAM treatment for up to 4 years or until they achieve seroconversion). Sullivan and colleagues45 assumed that a 48-week course duration is applied to both PEG-α-2a and the comparator, LAM.

Demographic and clinical characteristics of the hypothetical cohort of patients in Veenstra and colleagues44 and Sullivan and colleagues45 mirrored the baseline characteristics of patients enrolled in the RCT reported by Lau and colleagues. 50 Of note, 87% of patients in the modelled cohort of CHB patients were Asian. This population may not be representative of the general population in England and Wales, which may limit generalisability of the outcomes of the cost-effectiveness analysis.

All three studies use a Markov model consisting of the following health states (CHB, HBeAg seroconversion, compensated cirrhosis, decompensated cirrhosis, HCC, liver transplantation, post-liver transplantation and death). A health state not included in these studies is HBsAg seroconversion, a state which has been included in other economic evaluations (e.g. Crowley42 and Crowley and colleagues43). The outcomes are expressed as incremental cost per quality-adjusted life-year (QALY) in these studies. Sensitivity analysis is performed in all three studies.

As in the model presented in the Roche submission to NICE, Veenstra and colleagues44 and Sullivan and colleagues45 did not include the short-term effect of antiviral therapy on progression to compensated cirrhosis, such as that estimated in recent economic evaluations of LAM (Orlewska,41 Crowley42 and Crowley and colleagues43 The base-case analyses in Veenstra and colleagues44 and Sullivan and colleagues45 did not include the effect of LAM resistance. Drug resistance was explored in the scenario analysis reported in Veenstra and colleagues44 but not in Sullivan and colleagues45 In the base-case analysis of these models, it was assumed that by taking HBeAg seroconversion rates from long-term follow-up (which show reducing denominators over time), some of the effects of drug resistance, as indicated by reduced seroconversion rates, will have been captured.

Drug acquisition costs were taken from the most recent (at the time of writing) British National Formulary in Veenstra and colleagues44 and in the Roche model, and from the 2004 Taiwan Fee Schedule for Medical Service in Sullivan and colleagues45

As stated in our previous review,12 in the Roche model the health-state costs were developed by means of a combination of methods, including assumption, bottom-up costing using protocols based on expert opinion and extrapolation from costs developed for previous submissions. These costs were not adjusted for the differences in the intensity of medical management between the treatment groups. We previously12 also noted that the assumption that the HBeAg seroconverted health state has zero costs and does not correspond with current clinical guidelines that suggest that seroconverted patients should be reviewed every 6–12 months, during which time their serological status/HBV DNA should be assessed and a screen for HCC should be undertaken.

In the recent publication by Veenstra and colleagues,44 estimates of the costs of management of patients in different health states were taken directly from the economic evaluation in our previous report. 12 Health-state costs used in our previous economic evaluation were estimated specifically for the assessment. The costs were a combination of values from published cost estimates for the progressive stages of liver disease and estimates based on treatment protocols developed with expert advisors to the project. Unit costs for health-care resources were obtained from the finance department at Southampton University Hospitals Trust.

Sullivan and colleagues45 obtained cost estimates for the disease states of CHB and compensated and decompensated cirrhosis by applying the 2004 Taiwan Fee Schedule for Medical Service unit costs to the resource use reported by treating clinicians (no further details are provided). Cost estimates for HCC were taken from the published literature (Wang and Kowdley54). Liver transplantation and post-transplantation costs were obtained from the Chang Gung Memorial Hospital in 2004.

Veenstra and colleagues44 and the Roche submission used the same approach to estimating utility values. This was based principally on values reported by Wong and colleagues. 8 Sullivan and colleagues45 used higher estimates of utility in four health states (seroconversion, CHB, compensated cirrhosis and HCC), based on Pwu and Chan49 and Bennett and colleagues55 (see Table 28 below).

Sullivan and colleagues45 concluded that treatment with PEG-α-2a compared with LAM results in higher total cost but longer quality-adjusted life expectancy, yielding an ICER of NTD381,000 or US$12,000 at 2004 prices). As in the Roche submission, Veenstra and colleagues44 confirmed that PEG-α-2a is associated with higher discounted total health-care cost but also with additional discounted QALYs compared with long-term (up to 4 years) LAM treatment. The estimated ICER of £10,444 is almost twice as high as the ICER of £5948 reported in the Roche original submission. The difference most likely relates to the different method of cost estimation and the difference in the population cohort.

Details of economic evaluations based on the Kanwal and colleagues38 model

The original model was published in 2005 (Kanwal and colleagues38) and assessed in our earlier report. 12 Kanwal and colleagues concluded that in the base-case analysis (with 55% of patients being HBeAg negative at baseline), neither LAM nor ADV monotherapy is cost-effective in chronic HBV infection. However, depending on financial restrictions, either IFN or a hybrid strategy that reserves ADV as a salvage therapy only for LAM-resistant patients may be cost-effective. The objective of the more recent publication of Kanwal and colleagues46 was to estimate cost-effectiveness of alternative therapies in the subgroup of the CHB population with cirrhosis, and, in particular, to test whether the newer and more expensive agents such as ADV and entecavir become cost-effective in this subgroup. The structure and transition probabilities in the model reported by Kanwal and colleagues38 were adjusted for the subgroup of CHB patients with cirrhosis.

At baseline, 50% of patients in the cohort have compensated cirrhosis and 50% have decompensated cirrhosis, and in each treatment arm separate transition probabilities are assigned to patients in the compensated and decompensated cirrhosis groups as they progress through the stages of the disease. The baseline ratio of patients with compensated versus decompensated cirrhosis was tested in the sensitivity analysis.

As noted earlier, in the study by Kanwal and colleagues,46 the best outcome in the subgroup of cirrhotic patients is either to remain in the compensated cirrhosis stage or to revert from decompensated to compensated cirrhosis as a result of treatment or spontaneously. Patients reverting to compensated cirrhosis were eligible to decompensate a second time. The rate of subsequent decompensation was higher than the initial rate. Hepatocellular carcinoma could develop at any stage and all patients with decompensated cirrhosis or HCC were eligible for a liver transplant.

The study evaluated cost-effectiveness of six strategies in treatment of cirrhosis in CHB patients:

• strategy 1: no pharmacological treatment of chronic HBV (‘do nothing’ strategy)

• strategy 2: LAM monotherapy 100 mg once daily for an indefinite period

• strategy 3: ADV monotherapy 10 mg once daily for an indefinite period

• strategy 4: LAM with crossover to ADV on development of resistance (‘ADV salvage’ strategy)

• strategy 5: entecavir monotherapy 0.5 mg once daily for an indefinite period

• strategy 6: LAM with crossover to entecavir on development of resistance (‘entecavir salvage’ strategy).

The first four strategies are relevant to the scope of this report.

Unlike the 2005 study by Kanwal and colleagues,38 the 2006 study46 did not include a treatment strategy based on IFN-α, as this is not approved for patients with decompensated cirrhosis. In both studies, the perspective is of a US third-party payer.

In both studies38,46 estimates of costs of health-care resources were obtained by (1) calculating direct cost estimates by multiplying unit prices for the drugs and medical services by the estimated use of these resources in natural units, and (2) combining these costs with other cost estimates (e.g. costs of complications) obtained from the literature. In particular, in Kanwal and colleagues,46 costs of physician services and procedures were obtained from the 2005 American Medical Association Current Procedural Terminology codebook and the 2005 Medicare Fee Schedule. Pharmaceutical costs were obtained from the average wholesale prices (AWPs) listed in the 2006 Red Book. Cost estimates for cirrhosis and related health states were obtained from a published study of detailed, itemised inpatient and outpatient direct costs incurred by patients with cirrhosis (Bennett and colleagues55). While the model first presented in the Roche submission has only one health state corresponding to decompensated cirrhosis, the structure of the model in Kanwal and colleagues38,46 differentiates between different types of decompensation (i.e. variceal haemorrhage, ascites and encephalopathy), which allowed for a more precise estimation of the associated costs in the first and subsequent years.

In both studies,38,46 transition probabilities were obtained from a systematic review of the literature. Appendix 5 compares transition probabilities used in studies assessed in our previous report12 and those in the present report (the study by Buti and colleagues47 is not included in this appendix, as explained below).

The model reported in Kanwal and colleagues38,46 used utility values obtained from the literature that reported utilities for chronic liver disease associated with hepatitis C. Kanwal and colleagues argue that both hepatitis C and hepatitis B lead to cirrhosis and related complications and there is no a priori reason to believe that the quality of life decrements assigned to the corresponding health states would depend on the underlying aetiology. (The same approach was used in the economic evaluation conducted in our previous report. 12 Table 28 compares utility values used in studies assessed in our previous report12 and in the present report, with the exception of the study by Buti and colleagues47).

The result of economic modelling in Kanwal and colleagues (2006)46 indicated that:

• LAM monotherapy is dominated.

• The ICER of ADV monotherapy versus ‘doing nothing’ is $19,731($14,342–$24,224) at 2005 prices.

• The ICER of ADV monotherapy versus entecavir monotherapy is $25,626 ($19,637–$31,184) at 2005 prices.

• ADV salvage strategy is dominated.

• Entecavir salvage strategy is dominated.

The sensitivity analysis showed that the model outcomes were sensitive to the cost of ADV and entecavir; the annual rate of progression from compensated to decompensated cirrhosis with LAM resistance; the annual rate of progression from compensated to decompensated cirrhosis with entecavir (no resistance); the annual rate of progression from compensated to decompensated cirrhosis with ADV (no resistance); and the annual rate of progression from compensated to decompensated cirrhosis with ADV resistance. For example, if the incidence of progression from compensated to decompensated cirrhosis in LAM-resistant patients is less than the threshold of 3.5% (8% in the base-case analysis), than LAM becomes cost-effective. The results were robust with respect to the baseline ratio of patients with compensated versus decompensated cirrhosis.

Details of economic evaluation reported by Buti and colleagues47

Buti and colleagues47 estimate cost-effectiveness of a 4-year LAM with ADV as a salvage therapy strategy for LAM-resistant patients compared with ADV monotherapy.

In the decision tree model, different health states in two treatment groups are assigned to describe patient progression:

• In the LAM arm, these are: receiving LAM treatment with response; continuing LAM treatment with response; developing resistance to LAM, followed by receiving ADV as a salvage therapy; receiving ADV treatment with response; and developing resistance and no response to ADV treatment, in which case no other active treatment is received.

• In the ADV arm, these are: receiving ADV treatment with response; continuing ADV treatment with response; and developing resistance to ADV, in which case no other active treatment is received.

Other health states that characterise disease progression (e.g. compensated and decompensated cirrhosis, HCC, liver transplant) are not explicitly included in the model. Nevertheless, a certain proportion of patients who do not receive treatment are assumed to develop a ‘decompensated CHB’, which includes cirrhosis, hepatic encephalopathy, varicose haemorrhage, ascites and hepatocarcinoma, and is associated with the aggregated ‘cost of decompensation’ of €172.50. Buti and colleagues47 do not provide a clear explanation of either the proportion of patients with decompensation or the monetary value of health-care resources associated with treatment of decompensated CHB. In particular, it is not clear what proportion of patients (if any) start at the compensated cirrhosis state from which decompensated cirrhosis is later developed. It does not appear that a systematic review of the clinical evidence used in the model was undertaken. The probability of response, non-response and resistance seem to have been derived from averaging the response rate across a few selected studies, including non-randomised observational studies (see Appendix 4 for details). It assumed that patients receiving an active treatment, including those with compensated cirrhosis at baseline, do not develop decompensated CHB. This assumption is not consistent with assumptions used in other economic evaluations. 12,38,44–46

Although the systematic review by Sun and colleagues48 assessed the economic evaluation reported in Buti and colleagues47 as being of high quality in comparison with the other models discussed above, it is characterised by a number of shortcomings, in addition to the issues outlined earlier. It has a short time horizon of just 4 years rather than a lifetime horizon, as is appropriate in a chronic disease. A discounting factor is applied only to costs and not to the outcomes. The outcome of cost-effectiveness analysis is expressed in terms of additional cost per patient with response [defined as decrease of serum HBV DNA to undetectable levels by polymerase chain reaction (PCR) assay] instead of the conventional incremental cost per incremental QALY. Another methodological shortcoming is that clinical effectiveness data used in the two treatment groups come from different clinical trials and may therefore involve patient populations with different baseline characteristics. These shortcomings may potentially introduce a bias to the cost-effectiveness estimates, which may compromise the outcomes of the cost-effectiveness analysis.

Buti and colleagues47 estimated an incremental cost of ADV per additional patient with response at €27,872 at 2003 prices. Notwithstanding the shortcomings listed above, the results expressed in units other than QALYs renders the study outcomes of limited use for decision making in the area of allocating the limited health-care resources across the treatment alternatives.

Comparison of estimates of health-related quality of life

Table 29 compares health-state utilities used in different economic evaluations of antiviral treatment for patients with CHB. For completeness, the methodologically robust studies that used utility weights and were assessed in our previous report12 are included, along with the economic evaluations identified in our update search.

Economic evaluations by Wong and colleagues,8 Crowley,42 Crowley and colleagues43 and Dusheiko and Roberts39 applied health-state utility estimates derived from clinicians’ opinion rather than from patients’ preferences. These estimates are characterised by a large variation. Some of these estimates were subsequently reused in our previous model,12 and in models by Kanwal and colleagues38,46 Veenstra and colleagues44 and Sullivan and colleagues,45 along with utility weights elicited from patients with hepatitis C (see Appendix 4 for details). The utility values elicited by Levy and colleagues56 from the HBV population generally fall within this broad range of estimates used in different economic evaluations.

Applying alternative utility sets – base case

Table 35 reports total cost, discounted life expectancy and discounted QALYs for the overall cohort of patients with HBeAg-positive and HBeAg-negative CHB modelled in our previous report, using updated assumptions on health-state utility (utility set 1), updated costs and applying a discount rate of 3.5% for both costs and benefits. Incremental cost-effectiveness ratios are substantially higher than for the base case reported in our previous report (see Table 30). Total costs are between 23% and 45% higher, while discounted QALYs are 27–28% lower. However, much of this difference arises from the change in discount rates, rather than from changes in utility weights or inflating costs to current prices. For example, for IFN-α the same analysis, but using discount rates of 6% for costs and 1.5% for QALYs, yields total costs of £13,768, total QALYs of 16.96 and an ICER of £6981 relative to best supportive care, which is broadly comparable to the ICER of £5994 from our previous report.

The cost-effectiveness frontier in Figure 2 shows that allowable interventions (in cost-effectiveness terms) are the same as for the analysis based on our previous report (shown in Figure 1). These are: IFN-α followed by LAM (ICER = £8552 per QALY gained relative to best supportive care), PEG-α-2a followed by LAM (ICER = £12,801 relative to IFN-α followed by LAM) and PEG-α-2a followed by LAM followed by ADV (ICER = £26,379 relative to PEG-α-2a followed by LAM).

FIGURE 2.

Incremental cost-effectiveness and cost-effectiveness frontier, applying utility set 1.

Again, much of the difference with the results based on our previous model (reported earlier in this chapter) is due to changes in discount rate. As an example, ICERs for these strategies, discounted at 6% for costs and 1.5% for outcomes are: IFN-α followed by LAM (ICER = £5367 per QALY gained relative to best supportive care), PEG-α-2a followed by LAM (ICER = £8192 relative to IFN-α followed by LAM) and PEG-α-2a followed by LAM followed by ADV (ICER = £12,171 relative to PEG-α-2a followed by LAM).

Results for separate cohorts of HBeAg-positive and HBeAg-negative patients are reported in Appendix 7.

Applying alternative utility sets – deterministic sensitivity analysis

A series of one-way sensitivity analyses was conducted using the updated model, based on the range of values and sources of uncertainty reported in sensitivity analyses in our previous review. 12 These are reported in Table 36. To simplify the presentation and interpretation of the cost-effectiveness estimates in the sensitivity analyses, we report ICERs only for the optimal strategies in each analysis (using the methods for identifying the cost-effectiveness frontier and for excluding strategies using the principle of extended dominance, as described for Figure 1.

In this sensitivity analysis, the selection of optimal strategies is generally robust to changes in structural assumptions, baseline characteristics and parameter values. As with the base-case analysis, the optimal treatment sequence was generally identified as IFN-α or PEG-α-2a, followed by LAM, reserving ADV as a salvage strategy for patients who develop LAM resistance. The estimated ICERs are also generally robust to changes applied in the sensitivity analysis. However, changes in some key assumptions produce less favourable cost-effectiveness estimates than the base case adopted for this analysis:

• Assuming that patients with compensated cirrhosis cannot achieve HBeAg seroconversion produces a significantly less favourable ICER for the treatment strategy containing ADV as salvage for patients who develop LAM resistance. In contrast, the ICERs for IFN-α or PEG-α-2a followed by LAM are largely insensitive to this changed structural assumption.

• Reducing the proportion of the baseline cohort that has HBeAg-positive CHB also produces a less favourable ICER for the strategy including ADV (relative to PEG-α-2a followed by LAM), while the cost-effectiveness of PEG-α-2a followed by LAM improves (relative to IFN-α followed by LAM).

• Cost-effectiveness estimates for all treatment strategies are less favourable with increasing patient age at start of treatment. QALY gains from interventions are reduced by 15–20%, whereas incremental costs are reduced by 2–6%.

• Reducing the utility gain from seroconversion (to either no utility gain or using the lower value of 0.04 used in our previous report) gives a less favourable ICER than for the base case.

Reducing drug costs for PEG-α-2a leads to the elimination of ‘IFN-α followed by LAM’ from the sequence of optimal strategies and leads to an improvement in the cost-effectiveness of PEG-α-2a followed by LAM (relative to best supportive care).

Applying alternative discount rates (6% for costs and 1.5% for outcomes, as in our previous report, or 0% for both costs and outcomes) produces more favourable ICERs than the base case – reducing the ICER for PEG-α-2a followed by LAM with ADV as salvage from £26,379 to £12,171 (relative to PEG-α-2a followed by LAM).

Applying alternative utility sets – probabilistic sensitivity analysis

A probabilistic sensitivity analysis was undertaken using utility set 1 (see Table 31), updated costs (see Tables 32 and 33) and a discount rate of 3.5% for both costs and outcomes. The utilities were sampled from beta distributions with parameters calculated using the method of moments63 (using the reported mean values and standard errors derived from 95% CIs reported for UK infected patients by Levy and colleagues56) (see Appendix 6 for full details). Health-state costs were sampled from gamma distributions with parameters calculated using the method of moments (see Appendix 6 for full details).

Table 37 reports the mean cost and QALYs (with percentile-based 95% CIs) and the ICERs for the sequential strategies from the probabilistic sensitivity analysis. The mean discounted QALYs are close to those in the deterministic base-case analysis. However, the mean costs are around £1500 lower for each strategy.

Figure 3 shows the cost-effectiveness acceptability curves (CEACs) for all interventions included in the analysis of sequential treatment strategies. As with our previous report, this suggests that IFN-α-2a or PEG-α-2a followed by LAM would be the optimal strategy at lower threshold values of willingness to pay, but as the threshold increases the sequential treatment strategy including ADV salvage is increasingly likely to be the optimal intervention.

FIGURE 3.

Cost-effectiveness acceptability curves for sequential treatment strategies in overall cohort of patients with HBeAg-positive and HBeAg-negative CHB.

In contrast with our previous report,12 this strategy (interferon followed by LAM with ADV as salvage) becomes optimal only at the upper range of ICERs conventionally deemed as cost-effective from an NHS decision-making perspective. This is reinforced by Figure 4 which shows the cost-effectiveness acceptability frontier64 for the analysis based on our previous report (Figure 4a) and using the updated model (Figure 4b). The cost-effectiveness acceptability frontier comprises those portions of the CEAC where interventions are deemed optimal (using the maximum net benefit criterion) over a range of willingness-to-pay values. This clearly illustrates that interferon alpha followed by LAM is optimal, using the updated model, over a wider range of willingness-to-pay values (£9000–£12,000 for IFN-α followed by LAM and £13,000–£26,000 for PEG-α followed by LAM) than in the analysis based on the previous report (£5000–£6500 for conventional IFN followed by LAM and £7000–£11,500 for PEG-α followed by LAM). As discussed earlier, this arises largely as a result of the change in discount rates applied in the updated model.

FIGURE 4.

Cost-effectiveness acceptability frontier based on analysis in original assessment report and on updated model. (a) Cost-effectiveness acceptability frontier based on analysis in original assessment report. (b) Cost-effectiveness acceptability frontier based on updated model.

Cost-effectiveness of PEG-α-2b – base case

Table 38 reports total cost, discounted life expectancy and discounted QALYs for a cohort of patients with HBeAg-positive CHB receiving 24 months of PEG-α-2b compared with 24 months of IFN-α-2b (based on HBeAg seroconversion rates reported by Zhao and colleagues27). This is based on the clinical effectiveness review (see Chapter 3).

The results suggest that PEG-α-2b is a cost-effective alternative to IFN-α-2b for the treatment of patients with CHB. In the absence of a comparison with best supportive care, this analysis implicitly assumes that IFN-α-2b is a cost-effective option and a current standard of care. Supportive care has not been included in this analysis as the trial reported by Zhao and colleagues27 did not include a placebo or no treatment arm. No trials comparing IFN-α-2b with placebo or no treatment were included in the clinical effectiveness review reported in Chapter 3.

Cost-effectiveness of PEG-α-2b – deterministic sensitivity analysis

Table 39 reports a series of one-way sensitivity analyses based on the range of values and sources of uncertainty reported in sensitivity analyses in our previous review. 12

The ICERs are generally robust to changes in structural assumptions, baseline characteristics and parameter values. However, changes in some key assumptions produce less favourable cost-effectiveness estimates than the base case adopted for this analysis:

• Cost-effectiveness estimates for all treatment strategies are less favourable with increasing patient age at start of treatment. QALY gains from interventions are reduced by 5–12%, whereas incremental costs reduce by less than 1%.

• Reducing the utility gain from HBeAg seroconversion (to either no utility gain or using the lower value of 0.04 used in our previous report) gives a less favourable ICER than for the base case.

Reductions in drug costs for PEG-α-2b and the use of alternative discount rates are associated with more favourable cost-effectiveness estimates in the sensitivity analysis. Reducing drug costs for PEG-α-2b by 20% leads to a 4% reduction in total costs associated with PEG-α-2b, reducing the ICER to £5906. Using discount rates that applied at the time we conducted our previous review (6% for costs and 1.5% for outcomes), the ICER reduces to £6225. Applying zero discount rates, the ICER reduces to £4218.

In all analyses the ICER is below the threshold usually taken to define cost-effectiveness from an NHS decision-making perspective.

Cost-effectiveness of PEG-α-2b – probabilistic sensitivity analysis

A probabilistic sensitivity analysis was undertaken using utility set 1 (see Table 31), updated costs (see Tables 32 and 33) and a discount rate of 3.5% for both costs and outcomes. The utilities were sampled from beta distributions with parameters calculated using the method of moments63 (using the reported mean values and standard errors derived from 95% CIs reported for UK infected patients by Levy and colleagues56) (see Appendix 6 for full details). Health-state costs were sampled from gamma distributions with parameters calculated using the method of moments (see Appendix 6 for full details).

Table 40 reports the mean discounted cost and mean discounted QALYs (with percentile-based 95% CIs) for IFN-α-2b and PEG-α-2b from the probabilistic evaluation of the model. Table 40 also reports the ICER for PEG-α-2b compared with IFN-α-2b, based on the mean discounted cost and mean discounted QALYs. The results from the probabilistic evaluation of the model are similar to those in the deterministic base-case analysis.

Figure 5 shows the CEACs for PEG-α-2b for patients with HBeAg-positive CHB. This suggests that PEG-α-2b is likely to be a cost-effective option for the treatment of HBeAg-positive CHB, in comparison with IFN-α-2b.

FIGURE 5.

Cost-effectiveness acceptability curve for PEG-α-2b in patients with HBeAg-positive CHB.

In the probabilistic sensitivity analysis PEG-α-2b had a probability of being cost-effective (compared with IFN-α-2b) of 79% at a willingness-to-pay threshold of £20,000 per QALY, and 86% at a willingness-to-pay threshold of £30,000 per QALY.

Quality assessment for RCTs (quality criteria – CRD report 4)

Quality assessment for RCTs (quality criteria – CRD report 4)