Introduction

Chronic hepatitis B (CHB) comprises liver inflammation and damage caused by the hepatitis B virus (HBV), which can lead to cirrhosis and hepatocellular carcinoma 1. Eight drugs are now licensed in the UK for use in CHB: the nucleosides entecavir (ETV), lamivudine (LAM) and telbivudine (LdT); the nucleotides adefovir dipivoxil (ADV) and tenofovir disoproxil fumarate (TDF); and peginterferon-alpha-2a, interferon-alpha-2a and interferon-alpha-2b 2.

Although numerous trials have evaluated the efficacy of the three newest agents (ETV, LdT, and TDF) compared with LAM or ADV 3-10, no head-to-head randomized controlled trials (RCTs) have yet compared ETV, LdT, and TDF directly. Estimates of relative efficacy are required to inform clinical decisions about patient management, national and local treatment guidelines, and economic evaluations assessing the cost-effectiveness of the different treatment options.

A previous meta-analysis pooled the results of studies evaluating ADV, ETV, LAM, and placebo 11 and concluded that ETV was superior to both ADV and LAM. However, the analysis was primarily based on the absolute effects observed in individual arms of different trials 11 and therefore did not take account of the relative treatment effects observed in each study. Furthermore, the analysis did not assess the efficacy of LdT or TDF or include studies published since 2004.

Standard meta-analytical techniques evaluate the relative efficacy of one treatment compared with a single comparator 12. However, mixed treatment comparison (MTC) methods have recently been developed that estimate the relative efficacy of any number of different treatments by taking account of the entire network of RCT evidence 12-15. MTC is also known as network meta-analysis and multiple-treatments meta-analysis and is now recommended by the National Institute for Health and Clinical Excellence (NICE) 16.

We set out to use MTC to assess the relative efficacy of the following nucleos(t)ides in the treatment of CHB in terms of achieving viral suppression and seroconversion: ADV; ETV; LAM; LdT; TDF; placebo; and combinations of these drugs.

Methods Systematic Review

MEDLINE/PubMed and the Cochrane library were searched on August 31, 2007; additional studies were identified through manufacturer/conference websites and citations within published review articles.

The systematic review used broad, prespecified inclusion criteria to identify all relevant studies evaluating nucleos(t)ides in the treatment of CHB; these inclusion criteria are available on request. The wider review included nonrandomized studies as well as RCTs on diverse populations in order to provide data inputs for a model-based economic evaluation 17 as well as identifying RCTs for meta-analysis.

Studies identified in the systematic review were considered for meta-analysis if they met the inclusion criteria shown in Table 1. Meta-analyses were planned on four patient subgroups: nucleos(t)ide-naive patients with 1) hepatitis B e antigen (HBeAg)-positive or 2) HBeAg-negative CHB; and patients who were resistant/refractory to one or more nucleos(t)ides at baseline with 3) HBeAg-positive or 4) HBeAg-negative CHB. However, this article focuses on the results of analyses on HBeAg-positive nucleos(t)ide-naive patients.

Table 1. Summary of inclusion and exclusion criteria for the meta-analysis Study design Only randomized controlled trials were included. Subjects Studies were only included if:  The entire study population was chronically infected with HBV, or if results for a subgroup of patients with CHB were reported separately.  <50% of patients had HIV coinfection, because HIV coinfected patients are likely to have a different probability of response than mono-infected patients 1, 61. In practice however, all trials meeting inclusion criteria excluded patients with HIV. Studies were excluded if they exclusively recruited:  Children (<18 years), because nucleos(t)ides are licensed only for adults 2.  Pregnant women, pre-, post- or peri-transplant patients or patients with decompensated cirrhosis, cancer, or inactive liver disease. Intervention Studies were only included if they evaluated ≥1 of the following treatments:  245 mg/day tenofovir disoproxil (as fumarate), which is equivalent to 300 mg/day tenofovir disoproxil fumarate (TDF)  10 mg/day adefovir dipivoxil (ADV)  0.5 mg/day or 1 mg/day entecavir (ETV)  600 mg/day telbivudine (LdT)  100 mg/day lamivudine (LAM)  Any combination of these treatments   The doses considered comprise all those licensed in the UK for CHB. Study arms evaluating interferon-alpha or peginterferon-alpha, unlicensed treatments/doses (other than placebo) or sequential use of several treatments within the same 12-month period were excluded from the analysis. Any study that had <2 treatment arms after exclusion of any arms using interferons or unlicensed therapies was also excluded from the meta-analysis. Interferons were excluded because they were outside the scope of the analysis and as the outcomes of interferon therapy are most appropriately assessed 24 weeks after discontinuation of therapy, which is not the case for nucleos(t)ides. Furthermore, the efficacy of interferons is affected by race, age, sex, viral genotype, and baseline viral load and ALT 1, 62, whereas nucleos(t)ides suppress HBV DNA effectively across patient groups 1, 62; subsequently, including interferons would have introduced additional heterogeneity not applicable to nucleos(t)ides. Additionally, consideration of the wide range of formulations, doses and dosing frequencies or durations that are available for interferons would have complicated the analysis substantially. Outcome measures Trials were only included if they reported ≥1 of the below outcomes after 40–72 weeks of therapy:  Percentage/number of patients with HBV DNA levels below a threshold of 1000 copies/ml or less. Studies exclusively using higher thresholds were excluded as these are generally based on older assays, such as bDNA or hybrid capture techniques (LLQ 1.4 × 105–0.7 × 106 copies/ml) rather than PCR techniques (LLQ 20–1000 copies/ml) 63-65.  Percentage/number of patients with HBeAg seroconversion or loss. Date of publication No limits on date of publication were used, although searches for studies on adefovir and lamivudine were limited to those published since July 1, 2004 as earlier studies were identified from previous systematic reviews 66, 67. Language of publication Inclusion was not limited to published trials. Language limited inclusion only if no English translation was available from the British Library. Subgroups considered Trials were only included in the meta-analysis of nucleos(t)ide-naive HBeAg-positive patients if (a) >66.7% of patients were HBeAg-positive or if results for an HBeAg-positive subgroup were reported separately and (b) <33.3% of patients were resistant to, failed to respond to or relapsed after initially responding to treatment with lamivudine or 1 of the nucleos(t)ides considered in the trial. Trials on 3 other patient populations were considered for meta-analysis, although results are not presented in this article:  Predominantly (>66.7%) HBeAg-negative and <33.3% treatment-refractory.  Predominantly (>66.7%) HBeAg-positive and >66.7% lamivudine-refractory.  Predominantly (>66.7%) HBeAg-negative and >66.7% lamivudine-refractory. CHB, chronic hepatitis B; HBeAg, hepatitis B e antigen; HBV, hepatitis B virus; HIV, human immunodeficiency virus; LLQ, lower limit of quantification; PCR, polymerase chain reaction. Data Extraction

Data were extracted from journal articles, study reports, conference abstracts, summaries of product characteristics and (where available) clinical study reports. Data were extracted by one researcher and all data used in the meta-analysis were checked against the original sources by two other researchers. Any typos or discrepancies between the extracted data and the source material were resolved through reference to the original sources.

Data were extracted on the number/proportion of patients with HBeAg loss, HBeAg seroconversion, and HBV DNA levels undetectable by polymerase chain reaction (PCR) after 1 year of treatment (range: 40–72 weeks). In cases where no available data source reported the number of patients meeting a particular end point, patient numbers were calculated from percentages and were not rounded. All outcomes were recorded as reported by the trial, whether intent-to-treat or per protocol in order to maximize the amount of data available for the meta-analysis; however, all but two trials included in the analysis of HBeAg-positive CHB presented intent-to-treat data (Table 2).

Table 2. Methodology and baseline characteristics of the trials included in the meta-analysis Study Population HIV+ pts excluded? Blinding ITT/PP analysis HBV DNA assay (HBV DNA threshold used) Time of outcome assessment Intervention (no. pts randomized) % (n/N) HBeAg +ve Mean baseline HBV DNA (log10 c/ml) % (n/N) pts HBV DNA <300 c/ml by year 1 % (n/N) HBeAg+ pts who HBeAg seroconverted by year 1 BEHoLD AI463022 3, 68-70 Treatment-naive Yes DB Included all pts randomized and treated COBAS (300 c/ml) 48 weeks ETV 0.5 mg/day (n = 357) 97% (348/357) 9.62 67% (236/354) 21% (74/348) LAM 100 mg/day (n = 358) 98% (351/358) 9.69 36% (129/355) 18% (64/351) EARLY 4, 71 Antiviral naive Yes OL NS NS (300 c/ml) 48 weeks ETV 0.5 mg/day (n = 35) 94% (33/35) 10.26 58% (19.1/33)† 15% (5/33) ADV 10 mg/day (n = 34) 94% (32/34) 9.88 19% (6.1/32)† 21 (7/32) Study 018 6, 72‡ No prior treatment with nucleos(t)ides Yes OL ITT COBAS Amplicor® (300 c/ml) 52 weeks ADV 10 mg/day (n = 44) 100% (44/44) 9.98 40% (17.6/44)† 19% (8.4/44)† LdT 600 mg/day (n = 45) 100% (45/45) 9.57 60% (27.0/45)† 28% (12.6/45)† GLOBE 33, 73, 74* Nucleoside naive Yes DB ITT COBAS Amplicor® (300 c/ml) 52 weeks LdT 600 mg/day (n = 458) 100% (458/458) 9.50 60% (274.8/458)† 23% (99.4/432)† LAM 100 mg/day (n = 463) 100% (463/463) 9.50 40% (185.1/463)† 22% (97.2/442)† Study 103 9, 21 Pts with >12 weeks prior nucleos(t)ide therapy excluded Yes DB ITT COBAS TaqMan (300 c/ml) 52 weeks TDF 300 mg/day (n = 176) 100% (176/176) 8.64 74% (130/176) 21% (32/153) ADV 10 mg/day (n = 90) 100% (90/90) 8.88 12% (11/90) 17.5% (14/80) AHLT 34‡ No antiviral/immune modulating therapy within 3 months Yes DB ITT N/A 52 weeks PLA (n = 73) 99% (72/73) 7.30 — 4% (3/70) LAM 100 mg/day (n = 143) 100% (143/143) 7.25 — 16% (22/140) Marcellin 2003 18‡ Pts with >12 weeks prior nucleos(t)ide therapy excluded Yes DB Pts who received ≥1 dose study drug Roche Amplicor® (400 c/ml) 48 weeks ADV 10 mg/day (n = 172) 100% (172/172) 8.25 21% (36/171) 12% (20/171) PLA/day (n = 170) 100% (170/170) 8.12 0% (0/167) 5.5% (9/161) Sung 2003 36 Treatment-naive Yes NS ITT COBAS (200 c/ml) 52 weeks LAM 100 mg/day + PLA (n = 57) 95% (54/57) 9.17 41% (23/56) 27% (15/54) LAM 100 mg/day + ADV (n = 55) 96% (53/55) 8.84 39% (21/54) 32% (17/53) Dienstag 1999 35 No previous antiviral treatment Yes DB Modified ITT N/A 52 weeks PLA (n = 72) 100% (72/72) 7.20 — 6% (4/69) LAM 100 mg/day (n = 71) 100% (71/71) 7.46 — 17% (11/63) Lai 2005 8, 75, 76 Nucleos(t)ide-naive Yes DB ITT COBAS Amplicor® (200 c/ml) 52 weeks LdT 400 mg/day (n = 22) 100% (22/22) 8.9 61% (26.8/44)†§ 31% (13.6/44)†§ LdT 600 mg/day (n = 22) 100% (22/22) 9 LdT 400 mg/day + LAM 100 mg/day (n = 21) 100% (21/21) 9.3 49% (20.1/41)†§ 15% (6.2/41)†§ LdT 600 mg/day + LAM 100 mg/day (n = 20) 100% (20/20) 9.7 LAM 100 mg/day (n = 19) 100% (19/19) 9.3 32% (6.1/19)† 22% (4.2/19)† ETV-023 5, 77 Nucleos(t)ide-naive Yes DB ITT Roche Amplicor® (300 c/ml) 48 weeks ETV 0.5 mg/day (n = 258) 87% (225/258) 8.56 76% (196.1/258)† 15% (33.8/225)† LAM 100 mg/day (n = 261) 85% (221/261) 8.56 43% (112.2/261)† 18% (39.8/221)† Schiff 2003 37‡ Nonresponders to IFN Yes DB Pts with baseline and 1 postbaseline visit PCR (750 c/ml) 52 weeks LAM 100 mg/day (n = 119) 97% (116/119) 7.50 34% (34/99) 18% (19/108) PLA (n = 56) 96% (54/56) 7.35 17% (8/47) 13% (7/53) II_NUC20905 38* HDV coinfection. No treatment with IFN or antivirals within 6 months Yes DB ITT N/A 52 weeks PLA (n = 11) 9% (1/11)* NS — 0% (0/1) LAM 100 mg/day (n = 20) 5% (1/20)* NS — 100% (1/1) * Meta-analyses included only data for the subgroup of patients who were HBeAg-positive. † The number of patients responding to treatment was calculated from a percentage as patient numbers were not reported. ‡ A third study arm that did not meet inclusion criteria is omitted from this table and was not included in the meta-analysis. § For this study, outcome data pooled across both LdT doses was used in the meta-analysis, because data specific to each dose were not reported. ADV, adefovir; c/ml, copies [of HBV DNA] per milliliter; DB, double-blind; ETV, entecavir; FTC, emtricitabine; HBV, hepatitis B virus; HDV, hepatitis D virus; HIV, human immunodeficiency virus; IFN, interferon-alpha; ITT, intent-to-treat; LAM, lamivudine; LdT, telbivudine; N/A, not applicable; NS, not stated; OL, open-label; PCR, polymerase chain reaction (noncommercial assay with a lower limit of quantification of 750 c/ml); PP, per protocol or as treated; TDF, tenofovir disoproxil fumarate. Allowance for Missing Data

Meta-analyses were conducted on two outcomes: the probability of HBeAg seroconversion and the probability of achieving HBV DNA < 300 copies/ml after 1 year's treatment. These outcomes were chosen to provide transition probabilities for a model-based economic evaluation 17. A threshold of 300 copies/ml was used because this is the threshold most commonly reported in trials using PCR. Because a wide range of HBV DNA thresholds have been used in the literature and as some studies report HBeAg loss but not HBeAg seroconversion, statistical transformations were used to estimate these parameters from data on closely related outcome measures so that all available data could be meta-analyzed. Combining data on different outcomes without making appropriate adjustments could introduce bias into the analysis; in particular, because two studies observed that no placebo-treated patients achieved HBV DNA < 400 copies/ml 18, 19, combining data on different thresholds without adjusting for this is likely to have biased analyses in favor of TDF and ADV.

Data on the number of patients with HBeAg loss were converted into estimates of the number undergoing HBeAg seroconversion by assuming that 92% of patients losing HBeAg will also undergo HBeAg seroconversion within the same year, based on data extracted from the three largest trials reporting both measures, where the ratio of seroconversion to loss ranged from 88% to 100% 3, 20, 21.

Data on the proportion of patients in the two main TDF trials 9, 10, 21, 22 who had HBV DNA below one of four different thresholds (169, 300, 400, and 1000 copies/ml) were used to estimate a statistical model to predict the number of patients with HBV DNA levels < 300 copies/ml from the number of patients meeting other HBV DNA targets (details available on request). The model fitting the data best (R2 = 0.7554) was

(1)

where

• 

ThresholdX = threshold HBV DNA level (in copies/ml) for which data are available; if data on several thresholds are available, the threshold closest to ThresholdY should be used;

• 

nX = the number of patients in this trial with HBV DNA below ThresholdX;

• 

nY = the number of patients in this trial with HBV DNA below ThresholdY;

• 

N = total number of patients with HBV DNA measurements;

• 

ThresholdY = threshold HBV DNA level of interest (in copies/ml).

This equation was validated against data from the BEHOLD AI463022 trial 3 and was used to estimate the number of patients with HBV DNA < 300 copies/ml for use in the meta-analysis in cases where trials used other thresholds. In cases where the equation predicted negative numbers (for example if no patients had HBV DNA < 400 copies/ml), it was assumed that no patients would have HBV DNA < 300 copies/ml. Equation 1 may also be used for other purposes where it is necessary to convert between HBV DNA thresholds between 169 and 1000 copies/ml.

Statistical Methods

The techniques of Bayesian MTC meta-analysis 12, 13, 15 were used to assess the relative efficacy of the nucleos(t)ides. MTC techniques allow all evidence to be taken into account, whether direct (from head-to-head RCTs) or indirect (calculated from trials with common comparators). These methods use the principles underpinning standard indirect comparisons 23, 24 and are based on relative treatment effects (in this case the natural log of the odds ratio, log-OR) in order to preserve trial randomization and minimize bias.

MTC was conducted using WinBUGS Version 1.4 (MRC Biostatistics Unit, Cambridge, UK) 25, which uses Bayesian Markov chain Monte Carlo Gibbs sampling methods to fit user-defined models. We used statistical models for fixed and random-effects MTC analyses that were developed by Bristol University and allow for trials with up to three treatment arms 13, 14, 26, 27. The code used is available at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Fidler.asp.

As with all meta-analyses, MTC may be conducted using either fixed or random-effects models. Random-effects models allow for the possibility that the true treatment effect may differ between trials 12. We used random-effects models when there was evidence of heterogeneity or when fixed-effects models fitted the data poorly. Model fit was assessed based on residual deviance 28 and deviance information criteria (DIC) 25.

LAM was used as the baseline drug because it has been evaluated in the largest number of RCTs. The absolute probability of responding to LAM was based on the crude average probability of response across the LAM arms of all trials evaluating this treatment; the probability of responding to other treatments was calculated from this figure and the log-ORs for each treatment relative to LAM.

Differences between treatments were considered significantly significant at the 0.05 level if the 95% credible (Bayesian probability) interval (CrI) for the log-OR did not cross zero. We also calculated the probability that each treatment was best based on the proportion of Markov chain iterations in which that treatment had the highest probability of viral suppression/HBeAg seroconversion. All P-values represent Bayesian P-values. Treatment effects based on head-to-head trials were estimated using pair wise frequentist meta-analyses conducted in Stata Version 10.0 (StataCorp, College Station, TX) using the metan command; the same methods were used to calculate the odds ratio for viral suppression in HBeAg-negative patients compared with HBeAg-positive patients.

The impact of using fixed and random-effects models, alternative priors, adding/removing trials and making no adjustment for HBV DNA threshold was evaluated in sensitivity analyses.

In common with previous research 13, 27, 29, 30, non-informative prior distributions were used for all treatment effects and the odds of responding to LAM to ensure that the results were primarily driven by the data; see code for more details: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Fidler.asp. However, sensitivity analyses suggested that the posterior estimates of the uncertainty around treatment effects (but not the posterior means) were sensitive to the priors used. Informative half-normal priors 31 were therefore used for the between-studies SD in order to allow this external data to help inform the between-studies SD; these distributions were based on a meta-analysis of interferon trials identified in a published systematic review 32. The methods used to estimate these priors are described in more detail at: http://www.ispor.org/Publications/value/ViHsupplementary/ViH13i8_Fidler.asp.

Although reasonable convergence was achieved after a burn-in of 100,000 simulations, between 500,000 and 925,000 burn-in simulations were conducted for the main analyses to ensure that robust results were generated. For all analyses, results were based on a further 20,000 sampled simulations of two chains using different initial values.

Additional details of the methods used, data on model fit or convergence and results of sensitivity analyses are available from the authors on request.

Results Studies Identified

Of the 1272 publications identified through electronic searches, 77 RCTs and 46 nonrandomized studies met the inclusion criteria for the wider systematic review (Fig. 1); details available from the authors on request. For the purposes of this analysis, the HBeAg-positive and HBeAg-negative subgroups of the GLOBE trial comparing LdT and LAM 33 were counted as two separate studies.

Flow diagram showing study identification for the systematic review. Abbreviations: CHB, chronic hepatitis B; HBeAg, hepatitis B e antigen; LAM, lamivudine; RCT, randomized controlled trial. *For the purposes of this analysis, the HBeAg-positive and HBeAg-negative subgroups of the GLOBE trial comparing LdT and LAM 33 were counted as two separate studies. †These trials were identified by two previous systematic reviews conducted for the National Institute for Health and Clinical Excellence (NICE) appraisal of ADV and peginterferon alpha-2a