Telbivudine for the treatment of chronic hepatitis B

The manufacturer's decision problem specified telbivudine monotherapy as the intervention of interest in a population of adults with compensated liver disease and active chronic hepatitis B (that is, evidence of viral replication and active liver inflammation and/or fibrosis). The decision problem considered people with HBeAg-positive and HBeAg-negative chronic hepatitis B disease as separate subgroups. The manufacturer did not consider telbivudine in combination with other antiviral treatments, arguing that there was not enough evidence and that combination therapy was not within the current marketing authorisation for telbivudine. The comparator specified by the manufacturer was lamivudine as first-line oral antiviral treatment in HBeAg-positive and HBeAg-negative disease. The health outcomes considered were seroconversion rates of hepatitis B e antigen (HBeAg); virological response (a reduction in hepatitis B virus [HBV]) DNA); histological improvement (in inflammation and fibrosis); biochemical changes (for example, reduction in serum ALT); and development of viral resistance to treatment.

The manufacturer's submission presented evidence on the clinical effectiveness of telbivudine monotherapy based on the GLOBE trial, which was a randomised, double-blind trial comparing the efficacy, safety and tolerability of telbivudine (600 mg/day) with lamivudine (100 mg/day) for 104 weeks. In total, 1,367 patients were recruited, of whom 921 had HBeAg-positive chronic hepatitis B (458 in the telbivudine arm and 463 in the lamivudine arm) and 446 had HBeAg-negative chronic hepatitis B (222 in the telbivudine arm and 224 in the lamivudine arm). Patients were recruited from 20 countries and were nucleoside-naive. Patients were randomised in a one-to-one ratio to receive telbivudine or lamivudine (each with matching placebo for blinding purposes) once daily as oral tablets. The primary endpoint was therapeutic response, which was defined as suppression of HBV DNA to less than 5 log₁₀ copies/ml plus either clearance of detectable HBeAg or serum ALT normalisation.

In patients with HBeAg-positive disease, there was a statistically significantly higher therapeutic response rate in the telbivudine group (63.3%) compared with the lamivudine group (48.2%) at week 104. That is an absolute difference of 15.1 percentage points (95% confidence interval [CI] 8.6 to 21.6, p<0.0001). The mean reduction in baseline HBV DNA level was statistically significantly greater in the telbivudine group (‑5.74 log₁₀ copies/ml) compared with the lamivudine group (‑4.42 log₁₀ copies/ml) at week 104 in patients with HBeAg-positive disease (p<0.0001). In patients with HBeAg-negative disease, there was also a statistically significantly higher therapeutic response rate in the telbivudine group (77.5%) compared with the lamivudine group (66.1%). That is an absolute difference of 11.4 percentage points (95% CI 2.9 to 19.9, p=0.0069). A statistically significant reduction in HBV DNA levels was also observed in patients with HBeAg-negative disease: the mean reduction in HBV DNA levels from baseline at week 104 in the telbivudine group was ‑5.00 log₁₀ copies/ml compared with ‑4.17 log₁₀ copies/ml in the lamivudine group (p=0.0002).

In patients with HBeAg-positive disease, the proportion of patients with HBV DNA undetectable by polymerase chain reaction (PCR) assay at week 104 was statistically significantly higher in the telbivudine arm (55.6%) compared with the lamivudine arm (38.5%; p<0.0001). Virological breakthrough (defined in the trial protocol as an increase in HBV DNA to greater than or equal to 5 log₁₀ copies/ml on two consecutive occasions in patients who had previously achieved post-baseline virological response) was statistically significantly lower in the telbivudine arm (23.3%) than in the lamivudine arm (37.1%; p<0.0001). Virological breakthrough (defined as >1 log₁₀ above nadir) was statistically significantly lower in the telbivudine arm (28.6%) than in the lamivudine arm (45.5%). HBV resistance (as defined in the trial protocol) was statistically significantly lower in the telbivudine group (21.7%) compared with the lamivudine group (34.1%; p<0.0001).

Similar treatment effects were observed in patients with HBeAg-negative disease. The proportion of patients with HBV DNA undetectable by PCR was statistically significantly higher in the telbivudine arm (82.0%) compared with the lamivudine arm (56.7%; p<0.0001). Virological breakthrough (as defined in the trial protocol) was statistically significantly lower in the telbivudine arm (8.4%) than in the lamivudine arm (19.7%; p=0.0013). Virological breakthrough (defined as >1 log₁₀ above nadir) was statistically significantly lower in the telbivudine arm (12.2%) than in the lamivudine arm (30.4%). HBV resistance was statistically significantly lower in the telbivudine arm (8.4%) than in the lamivudine arm (20.2%; p=0.0008).

The ERG considered that on the whole the manufacturer's submission was an unbiased estimate of the anti-viral treatment effects of telbivudine. However, the ERG suggested that although the results from the GLOBE trial were statistically significant, the clinical significance of the results was open to question. On the basis of the proportion of patients who discontinued treatment because of disease progression or lack of efficacy (0.8% versus 2.6% for telbivudine and lamivudine, respectively), there is an absolute difference of approximately 2 percentage points between telbivudine and lamivudine. According to the ERG, although virological breakthrough (defined as >1 log₁₀ above nadir) at 104 weeks in patients with HBeAg-positive disease was lower in the telbivudine arm (28.6%) than in the lamivudine arm (45.5%), it was still clinically high. In addition, it is not clear if the GLOBE trial was powered to detect differences in subgroups of race or serum ALT levels. Over-representation of HBeAg-positive patients in the trial may have affected the statistical validity of the results in the HBeAg-negative disease group. The ERG noted that effects of treatment on health-related quality of life were not measured in the GLOBE trial.

The manufacturer's submission presented an analysis of the cost effectiveness of telbivudine in patients with chronic hepatitis B whose serum ALT levels are more than or equal to twice the upper normal limit. Two Markov state-transition models were provided in the manufacturer's submission: a seroconversion model (applicable to only HBeAg-positive disease) and a viral load model (applicable to both HBeAg-positive and HBeAg-negative disease). Both models used a lifetime horizon. The seroconversion model evaluated the following treatment sequences and best supportive care (BSC): lamivudine only (that is, no further antiviral treatment if resistance develops), lamivudine followed by adefovir dipivoxil (as 'salvage therapy' if resistance develops), telbivudine only, adefovir dipivoxil only, adefovir dipivoxil followed by lamivudine, adefovir dipivoxil followed by telbivudine, and telbivudine followed by adefovir dipivoxil. In the viral load model, the only comparator considered was lamivudine.

The viral load model submitted by the manufacturer assumed that patients entered the model in the chronic hepatitis state without cirrhosis. Health states associated with disease progression were divided by serum ALT and viral load levels, resulting in a large number of possible health states. Consequently the data available from the GLOBE trial to populate the viral load model were sparse. In an attempt to deal with this, the manufacturer used values of 0.0 and 0.5 (which they referred to as 'non-informative priors') to correct for the probabilities of health-state transitions for which there were one or more zero observations and no data available.

The results of the economic analysis were presented as incremental costs per quality-adjusted life year (QALY) gained for telbivudine relative to lamivudine in the viral load model. In the seroconversion model, a comparison between a set of treatment algorithms relative to BSC was made. The manufacturer's main submission did not report on univariate sensitivity analyses and the base-case results were taken from probabilistic sensitivity analysis. After the ERG identified errors in the manufacturer's original viral load model, the manufacturer presented amended base-case analyses.

The base-case analysis of the viral load model (based on probabilistic sensitivity analysis) comparing telbivudine with lamivudine and assuming a 'non-informative prior' of 0.0 produced an incremental cost-effectiveness ratio (ICER) of £15,377 per additional QALY gained for HBeAg-positive disease; the corresponding ICER with a 'non-informative' prior of 0.5 was £8,542 per additional QALY gained. For HBeAg-negative disease, the ICER for a comparison of telbivudine with lamivudine with a 'non-informative prior' of 0.0 was £20,256 per additional QALY gained. The corresponding ICER with a 'non-informative prior' of 0.5 was £27,801 per additional QALY gained.

Deterministic base-case analyses (requested from the manufacturer) of the viral load model comparing telbivudine with lamivudine, with a 'non-informative prior' of 0.0, produced an ICER of £12,278 per additional QALY gained for HBeAg-positive disease. The corresponding ICER, with a 'non-informative prior' of 0.5, was £8,669 per additional QALY gained. For HBeAg-negative disease, the ICER for a comparison of telbivudine with lamivudine was £20,383 per additional QALY gained with a 'non-informative prior' of 0.0; the corresponding ICER, with a 'non-informative prior' of 0.5, was £57,419 per additional QALY gained.

The manufacturer's economic analysis based on the seroconversion model (HBeAg-positive disease only) gave an ICER of £13,193 per additional QALY gained (95% CI £7,788 to £25,194) for a comparison of telbivudine alone (followed by BSC if appropriate) with BSC alone. A comparison of telbivudine followed by adefovir dipivoxil and then BSC against BSC alone gave an ICER of £15,684 per additional QALY gained (95% CI £9,491 to £28,151). Adefovir dipivoxil followed by telbivudine and then BSC compared with BSC alone gave an ICER of £18,388 per additional QALY gained (95% CI £11,707 to £30,357). Adefovir dipivoxil followed by lamivudine and then BSC compared with BSC alone gave an ICER of £17,398 per additional QALY gained (95% CI £11,063 to £28,322).

The ERG considered that the seroconversion model structure used to assess the cost effectiveness of telbivudine was consistent with methods adopted in previous technology appraisals in chronic hepatitis B. However, the ERG identified a number of issues and uncertainties relating to the economic evidence presented by the manufacturer. It noted the economic models presented in the manufacturer's submission contained insufficient discussion of uncertainty; in particular, no univariate sensitivity analyses were presented in the main body of the submission for either model. Although the submitted viral load model included a worksheet that contained univariate sensitivity analysis, these results were not discussed in the submission itself. The ERG noted that there was no explanation of the results of the univariate sensitivity analysis, or of the rationale for the choice of variables included or excluded. Also, no explanation of the choice of variable ranges was given. Consequently it was not clear what the key drivers of the economic model were. In addition, there was no detailed discussion about the probabilistic sensitivity analysis conducted.

The ERG noted that evidence on the clinical and cost effectiveness of adefovir dipivoxil was not adequately identified. No attempts were made to justify or investigate the assumptions made about the clinical and cost effectiveness of adefovir dipivoxil. The ERG further noted that entecavir was not included as a comparator in the original submission from the manufacturer; it did not consider that methodological concerns about indirect comparisons were an adequate reason for not including this comparator. In addition, little account was taken of entecavir's possibly better resistance profile compared with telbivudine. Alternative approaches to populating the viral load model were not considered; in particular, the possibility of developing statistical risk models to address the sparsity of observed data from the GLOBE trial. Impacts of the so-called 'non-informative priors' on the economic results could not be adequately assessed by the ERG.

The ERG noted discrepancies between the calibration factors in the risk equations used for the compensated cirrhosis and hepatocellular carcinoma states in the original and resubmitted viral load models, and the factors listed in the appendices to the manufacturer's submission. The ERG also noted that an excessive reliance on visual basic coding made it unclear which parameters had or had not been included in the economic analyses. Further, the ERG noted that the manufacturer's submission did not discuss the power of the GLOBE trial to detect statistically significant effects of treatment in the subgroup of patients with serum ALT levels greater than or equal to twice the upper limit of normal. Data used to populate the economic models were taken from this subgroup of patients. No information was provided on the baseline characteristics of this subgroup of patients. The ERG stated that there is real uncertainty about the completeness of data (from the Globe study) used to populate the model and that the key clinical-effectiveness data in the economic model could not be critically appraised. The ERG noted in its conclusions that sensitivity analyses undertaken by the ERG were able to address a limited number of the concerns raised above.

The ERG carried out scenario analyses on the viral load model (with a 'non-informative prior' of 0.0) using non-constant age-specific utilities, increasing the proportion of cirrhotic patients at treatment initiation to 15% and applying model calibration factors (for risk of advanced liver disease). The cumulative effects of varying these parameters for HBeAg-positive disease gave an ICER of £16,100 per additional QALY gained. The corresponding ICER for HBeAg-negative disease was £26,200 per additional QALY gained.

The ERG conducted exploratory scenario analyses on the seroconversion model:

• assuming no treatment with telbivudine for people with decompensated liver disease,

• removing treatment-resistant patients from the denominators used to calculate transition probabilities for HBeAg seroconversion,

• increasing the proportion of cirrhotic patients at the start of treatment to 15%, and

• assuming treated people with cirrhosis seroconvert at the same rate as people with treated non-cirrhotic chronic hepatitis B.
  
  The cumulative effects of varying the first three parameters gave an ICER of £20,200 per additional QALY gained for telbivudine followed by adefovir compared with lamivudine followed by adefovir in the HBeAg-positive group. Adding the last assumption results in an ICER of £8,400 per additional QALY gained for the same comparison. The cumulative effects of varying the first three parameters gave an ICER of £22,500 per additional QALY gained for telbivudine alone compared with lamivudine alone. Adding the last assumption results in an ICER of £10,800 per additional QALY gained for the same comparison.

The ERG conducted a probabilistic sensitivity analysis using the viral load model with a 'non-informative prior' of 0.0 only. It replaced constant health-state utilities with non-constant age-specific utilities and applied the model calibration factors for risk of advanced liver disease listed in the appendices to the manufacturer's submission. This reduced the probability of telbivudine being cost effective for any given willingness to pay (cost-effectiveness) threshold when compared with lamivudine. For the HBeAg-positive group, the probabilities that telbivudine was cost effective at willingness to pay thresholds of £20,000 and £30,000 per additional QALY gained were 0.53 and 0.82, respectively. For the HBeAg-negative group, the probabilities of telbivudine being cost effective at willingness to pay thresholds of £20,000 and £30,000 per additional QALY gained were 0.01 and 0.54, respectively. The ERG also conducted a probabilistic sensitivity analysis using the seroconversion model, and the results differed from the manufacturer's analysis: in particular, lamivudine is optimal over a wider range of willingness to pay, with lamivudine followed by adefovir being optimal over a cost-effectiveness threshold range of £22,000 to £24,000 per additional QALY, whereas telbivudine was the optimal strategy over this range in the manufacturer's probabilistic sensitivity analysis. At higher cost-effectiveness thresholds (greater than £25,000 per QALY gained), telbivudine followed by adefovir remained the optimal strategy.

Full details of all the evidence are in the manufacturer's submission and the ERG report.