It seems logical that the termination of viral replication should decrease the risk for developing progressive liver disease. In the past 15 years, investigators have therefore targeted for treatment the subgroup of patients with chronic HBV infection with ongoing viral replication, as indicated by the presence of hepatitis B e antigen (HBeAg) and HBV DNA in serum Figure 1 [19]. Many antiviral and immunomodulatory drugs have been tried without proven efficacy. Alpha-interferons in particular have been studied extensively. The preparations of -interferon used were either lymphoblastoid, leukocyte, or recombinant (recombinant -interferon 2a, 2b, and 2c). Early studies showed that a thrice-weekly schedule produced a response similar to that seen with daily dosing but caused fewer side effects [20], that treatment for less than 3 months was ineffective [21], and that recurrences of viral replication were observed after cessation of interferon [22]. Hence, most current trials use a thrice-weekly dosing schedule for at least 3 months with a follow-up period of at least 6 months.

View larger version (24K): [in this window] [in a new window]   Figure 1. Natural history of chronic hepatitis B acquired in childhood. The initial phase of immune tolerance is followed by the phase of immune clearance of hepatitis B virus (HBV) and finally the cessation of HBV replication. The transition from the replicative to the nonreplicative phase may be rapid and smooth (——) or prolonged and fluctuating (—-). CPH = chronic persistent hepatitis; NSRH = nonspecific reactive hepatitis; CLH = chronic lobular hepatitis; CAH = chronic active hepatitis; HCC = hepatocellular carcinoma. From Lok and Lai [19], with permission.

Long-term follow-up studies to assess sustained viral eradication are now being published, but the sample sizes are small and the duration of follow-up is still less than 10 years [23]. No study has assessed the effect of interferon therapy on the clinically relevant end points of liver failure, hepatocellular carcinoma, and death as the progression of this disease evolves over decades. Such a study would require large numbers of patients followed for long periods. All studies to date have assessed efficacy using intermediate outcomes such as loss of hepatitis B surface antigen (HBsAg), HBeAg, and HBV DNA. Most studies have failed to show statistically significant treatment effects, but favorable trends have frequently been observed. We therefore conducted a meta-analysis of randomized controlled trials to examine the efficacy of interferon treatment in patients with chronic HBV infection as determined by the loss of HBsAg, HBeAg, and HBV DNA.

Methods

Top Methods Results Discussion Author & Article Info References

Literature Search

The meta-analytic methods used in this study conform to those previously described [24, 25]. The English-language medical literature was searched for randomized controlled trials of -interferon therapy for chronic HBV infection. Trials were identified by a computer-aided search of the National Library of Medicine MEDLINE database from January 1966 to June 1992 using the terms "exp INTERFERON TYPE I" and "HEPATITIS B," by manual review of the references of trials and review articles, and by personal communication with colleagues who had coordinated such studies.

Inclusion and Exclusion Criteria

All prospective randomized controlled trials of interferon for the treatment of patients with chronic HBV infection, the results of which were published as articles in peer-reviewed journals, were included. For the purpose of this study, patients were defined as having chronic HBV infection with active viral replication if they had HBsAg and HBeAg in serum for more than 6 months and a pretreatment liver biopsy result consistent with chronic HBV infection. Duplicate reports, as well as studies reported solely in abstracts, letters, and preliminary reports, were excluded. Other predetermined criteria excluded trials that enrolled only children (younger than 18 years), trials that treated predominantly HBeAg-negative patients or patients who were coinfected with delta hepatitis, and trials that had treatment periods of less than 3 months or had follow-up periods of less than 6 months. Trials that used interferon in combination with another immunomodulatory or antiviral agent were excluded; however, if these trials included an arm in which interferon alone was compared with a control group, the data from that arm were included.

Outcome Analysis

As noted above, intermediate outcomes such as loss of HBsAg, HBeAg, and HBV DNA were evaluated as primary outcomes (see Figure 1). Secondary outcomes examined included the development of antibodies to HBsAg [anti-HBs] and to HBeAg (anti-HBe), normalization of serum alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels, and improvement in histologic indices. Our analyses were based on intent to treat; thus, all patients who were lost to follow-up were considered to be treatment failures.

Subgroup analyses were selected prospectively and were based on factors that have been hypothesized to affect the outcome of interferon therapy [26-29]. These factors include gender, sexual preference, coinfection with human immunodeficiency virus (HIV), pretreatment AST and ALT levels, degree of inflammatory activity on pretreatment biopsy specimens, ethnic origin, duration of therapy and follow-up, preparation of -interferon, and dose of interferon used. We arbitrarily defined a dosage of more than 15 MU/m² per week or its equivalent as high-dose treatment. Variables were analyzed within trials whenever possible; if sufficient data were not available, variables were analyzed across trials. Subgroup analyses based on duration of disease, previous history of acute icteric hepatitis, and pretreatment HBV DNA levels were not done because sufficient data were not available. The P values were not adjusted for multiple subgroup analyses.

Data Abstraction

Data abstraction was done by two independent readers. Results were subsequently compared, and differences were settled by consensus. Raw data on treatment and control modalities and outcomes in each trial were recorded on standardized forms. Contamination, cointervention, patient compliance, and the number of patients who dropped out or who were lost to follow-up were also noted. The data form for each study was sent to the respective investigators for confirmation of raw data and for completion of relevant missing data. A second form was sent if a reply was not obtained within 2 months. Seven of 16 trial investigators responded. Only published data were used for the analyses of primary and secondary outcomes, but data from personal communication were used, when available, for subgroup analyses.

Quality Assessment

Assessment of study quality was done using the quality assessment system of Chalmers and colleagues [30] as modified by L'Abbe and associates [25]. Two independent appraisers, not involved with data abstraction, evaluated only the Materials and Methods and Results sections of each study and were blinded with respect to the study authors, journals of publication, and specific treatment modalities used. The appraisers subsequently met to resolve any disagreements on an item-by-item basis. A quality score, which ranged from 0.0 to 1.0 points, was generated for each trial based on the number of predetermined criteria that were satisfied (supplied on request). The effect of study quality was assessed using the quality scores in sensitivity analyses.

Statistical Analysis

The appropriateness of statistically combining selected randomized controlled trials was assessed by evaluating the comparability of trial protocols, the clinical features of included patients, and outcome measures. An underlying assumption in combining individual study results to obtain a better estimate of the "true" treatment effect was that the differences in the results among the studies were due to sampling variation alone. To test this assumption, characteristics of treatment and control groups in each individual trial were graphed and visually inspected for imbalances in absolute and proportionate terms [25]. Chi-square tests of heterogeneity were also done [31].

All outcome data except paired biopsy results from each study were pooled using difference in proportions (DP) (the proportion of outcome of interest in the treated group [P_t] minus the proportion of outcome of interest in the control group [P_c]):

DP = P_t – P_c

A positive value favored the treatment arm, and a negative number favored the control arm. The individual confidence intervals (CIs) were calculated using a simple normal approximation [30]. Primary outcome data were also expressed as the number of patients who need to be treated to have one outcome of interest (NNT), where NNT is equal to the inverse of the difference in proportions:

NNT = 1/DP

and the 95% CIs for NNT are the reciprocals of the 95% CI for DP [32, 33]. Data were pooled, and confidence intervals were calculated using the method of DerSimonian and Laird [31] with pooled study variance estimates. The P values for the pooled treatment effects were then calculated using the means and variances in a z-test. Our statistical method of analysis used differences in proportions instead of the log odds scale [24] or odds ratios as the primary analysis because of the presence of a large number of zero outcomes in the control groups. To confirm our results, the Freeman-Tukey transformation [34] and a nonparametric sign test [35, 36] were also done.

Paired biopsy results were assigned a numeric score (better = 1, same = 0, worse = –1).Data were analyzed using ordinal logistic regression, and the results were expressed as an odds ratio:

P_t

—————-

1 –P_t

odds ratio =—————-

P_c

—————-

1 –P_c

with the corresponding 95% CI [37].

Subgroup analyses were done within trials whenever the variables of interest were included in both the treatment and control arms, and the outcomes were reported separately. These intratrial comparisons were done for dose, HIV status, gender, and pretreatment AST and ALT values. All other subgroup analyses were done across trials. The difference in subgroup response was expressed as the difference between differences in proportions (DDP):

DDP = Dpa – DPb

where a and b were variables within a subgroup. The 95% CIs were calculated using the same statistical methods outlined above. The P values for analyses within trials were calculated using the pooled difference between differences in proportions weighted by their variances, and the P values for analyses across trials were calculated using the means and variances from each subgroup's pooled results in a z-test. The Freeman-Tukey transformation was used to confirm our results.

Sensitivity analyses were conducted to assess whether the results were robust. One analysis sequentially eliminated studies with lower quality scores using the method outlined by Detsky and colleagues [38] and used the method of DerSimonian and Laird to repeatedly analyze the remaining pooled data. Other analyses were also conducted to exclude trials in which HBV DNA was not assessed, trials with patients who tested positive for anti-HBe and antibodies to hepatitis D virus (anti-HDV), trials that did not test for anti-HDV, trials that included children (younger than 18 years), and trials with contaminations or cointerventions.

Results

Top Methods Results Discussion Author & Article Info References

Literature Review

Forty published prospective, randomized, controlled trials of -interferon treatment of chronic HBV infection, excluding abstracts, were identified in the English-language medical literature [39-78]. Twenty-five of these reports were excluded based on predetermined criteria (Table 1). Patients in one study were stratified by pretreatment ALT levels before randomization [78]. This study was treated as two separate trials because separate results were presented for both strata. Therefore, 16 trials formed the basis of our meta-analysis [64-78].

Quality of Studies

The overall rating of studies ranged from 0.55 to 1.00 points. Methodologic limitations in the published reports included poor delineation of patients excluded from the trial (62%), incomplete description of randomization (56%), incomplete description of the therapeutic regimen for the control group (44%), failure to blind outcome assessors with respect to treatment received (44%), possible use of inappropriate statistical methods (75%), failure to report CIs or to do post-hoc power calculations if the trial was negative (100%), and failure to specify prospective sample size justification (94%). We emphasize that many of these perceived shortcomings may be due to problems in reporting rather than to deficiencies in the design, conduct, or analysis of the trials.

Comparability of Trials

Eight hundred thirty-seven HBsAg- and HBeAg-positive patients from eight different countries were included. Eight studies clearly stated that they excluded patients with decompensated liver disease. Most patients tested negative for delta hepatitis, with the exceptions noted in Table 2. Overall, 689 of 837 (82%) patients were men, and 55 of 837 (6.6%) were coinfected with HIV.

Pretreatment variables, such as gender, sexual preference, and liver biopsy results were comparable between treatment and control groups within each study. Pretreatment AST and ALT levels and HBV DNA levels were comparable within all but two studies: one in which the control group had a higher median baseline ALT [70] and another in which HBV DNA levels were higher in the control group [74]. Five studies included HIV-positive patients [64, 66-69]: Testing for HIV was retrospective in three [64, 66, 67], and patients were not stratified by HIV status in the remaining two. The result was an imbalance with a higher proportion of HIV-positive patients in the control group than in the treatment group. This imbalance did not reach statistical significance within each individual study and had an overall P value of borderline significance (P = 0.05) when studies were aggregated.

The study protocols of the 16 trials included in our meta-analysis are shown in Appendix Table 1. Interferon was administered as subcutaneous or intramuscular injections on a daily to twice-weekly schedule. The dosages used ranged from a low of 7 MU per week to a high of 30 MU/m² per week for 3 to 6 months. The duration of follow-up ranged from 6 to 12 months after therapy.

All studies reported the loss of HBeAg as a primary end point (Table 3). All but three studies reported changes in HBsAg status [66, 71, 73]. Wide variations in the rates of HBeAg clearance were reported, ranging from 9.5% to 69.7% in the treated groups and from 0% to 38.7% in the control groups. Despite this variability, all studies showed a trend favoring treatment. Fifteen of 16 trials monitored changes in serum HBV DNA [64-76, 78]. Five studies showed statistically significant results for loss of viral replication as defined by loss of both HBV DNA and HBeAg [65, 67-70]. Only two showed statistically significant results for loss of HBsAg [67, 70].

The chi-square test for heterogeneity was not statistically significant for any of the primary end points. The P values were 0.53, 0.89, and 0.95 for loss of HBsAg, HBeAg, and HBV DNA, respectively. Furthermore, visual inspection of graphs showing differences in proportions in absolute Figure 2 and proportionate terms (not shown) suggested little, if any, heterogeneity.

View larger version (20K): [in this window] [in a new window]   Figure 2. Randomized controlled trials of -interferon treatment for patients with chronic hepatitis B infection and hepatitis B e antigen. The pooled and individual differences in proportions and their 95% CIs are shown for loss of hepatitis B surface antigen (HBsAg) (left) and loss of hepatitis B e antigen (HBeAg) (right). Values greater than 0 indicate a beneficial treatment effect.

Effect of Alpha-Interferon on Markers of Disease Activity

Interferon treatment had a favorable and statistically significant effect on all three primary end points when data were pooled by the method of DerSimonian and Laird. Analysis of data from 837 patients (498 treated and 339 control) followed for 6 to 12 months after therapy showed the following differences in proportions: for loss of HBsAg, 0.06; for loss of HBeAg, 0.21; and for loss of HBV DNA, 0.20 (Table 4). In other words, loss of the HBV carrier state occurred 6% more often than the natural seroconversion seen in controls (7.8% compared with 1.8%), and the loss of viral replication occurred approximately 20% more often than in controls (33% compared with 12% for loss of HBeAg and 37% compared with 17% for loss of HBV DNA) if patients were treated with interferon for 3 to 6 months and followed for 6 to 12 months. Therefore, one would need to treat 18 patients to observe a loss of HBsAg in 1 patient and would need to treat 5 patients to observe loss of HBeAg and loss of HBV DNA in 1 patient.

Aggregate data showed that the differences in proportions for the secondary outcomes (the development of anti-HBs and anti-HBe and the normalization of ALT) were also statistically significant. The odds ratio for paired liver biopsies of 2.09 (CI, 0.96 to 4.55; P = 0.052) also favored interferon treatment. Only four trials, however, compared and reported paired biopsy results in both treatment and control groups [69, 70, 75, 77], and, when data were aggregated, only 55 of 104 patients (52.9%) in the treatment group and only 35 of 99 patients (35.4%) in the control group had both pretreatment as well as post-treatment biopsies. These results should therefore be interpreted with caution.

Subgroup Analyses

Study Protocol

Subgroup analysis was done to assess dose effect. Five trials reported results on our primary end points stratified by dose and were thus analyzed [64, 65, 68, 74, 77]. Statistically significant differences in proportions (DP) were noted for the loss of HBeAg 0.22 (CI, 0.10 to 0.35; P = 0.0004) and the loss of HBV DNA 0.14 (CI, 0.00 to 0.28; P = 0.04), when comparing patients treated with high-dose interferon (defined as > 15 MU/m² per week or its equivalent) to patients treated with low-dose interferon (Figure 3).

View larger version (16K): [in this window] [in a new window]   Figure 3. The effect of high-dose compared with low-dose interferon therapy in patients with chronic hepatitis B and hepatitis B e antigen. The pooled and individual differences in proportions and their 95% CIs are shown for loss of hepatitis B surface antigen (HBsAg) (left) and loss of hepatitis B e antigen (HBeAg) (right) for studies that randomized patients to different doses of interferon. A high dose was arbitrarily defined as greater than 15 MU/m2 per week (> 5 MU/m2 thrice weekly).

Aggregation of studies by duration of therapy showed significant losses of HBeAg and HBV DNA in the studies that treated patients for 3, 4, or 6 months (P < 0.0002). The loss of HBsAg was significant only for studies in which interferon was used for 4 or 6 months (P < 0.03). No significant differences were seen, however, when studies using 3 months were compared with those using 6 months of therapy.

Aggregation of studies by the duration of follow-up after cessation of therapy showed that interferon had a statistically significant effect on the loss of HBeAg and HBV DNA at 6 months (P < 0.0006), 8 or 9 months (P < 0.006), and 12 months (P = 0.0001). The loss of HBsAg was significant only in studies with a follow-up period of 12 months (P = 0.0001). When comparing these follow-up periods, we noted a higher proportion of treated patients with the loss of HBsAg and HBeAg in the 12-month compared with the 6-month follow-up group (P = 0.01 and 0.053, respectively).

Studies were also aggregated according to three of the preparations of interferon used (lymphoblastoid interferon and recombinant -interferon 2a and 2b). Each individual preparation was found to induce a statistically significant loss of HBeAg (P = 0.0001, 0.0002, and 0.0001, respectively) and loss of HBV DNA (P = 0.0001, 0.007, 0.0002, respectively). The loss of HBsAg was statistically significant only in studies using lymphoblastoid interferon (P = 0.002) and recombinant -interferon 2b (P = 0.01). A significant difference between differences in proportions (DDP) for the loss of HBsAg (12%; CI, 1% to 24%) was also seen in studies using lymphoblastoid interferon compared with studies using recombinant -interferon 2a (P = 0.03).

Patient Characteristics

Because data for the loss of HBeAg were most complete for all other subgroups, this variable was used as the outcome for further analyses.

Thirteen trials reported response for HIV-negative patients in both treatment and control groups. All favored treatment in HIV-negative patients, and the pooled difference in proportions was 0.20 (CI, 0.13 to 0.26; P = 0.0001). Data for HIV-positive patients were less definitive. Five trials [64, 66-69] included HIV-positive patients, but only two [68, 69] reported response by HIV status in both treatment and control groups. When aggregating these two trials, the difference in proportions for HIV-positive patients was 0.0 (CI, –0.16 to 0.16) for the loss of HBeAg. We were unable to rule out a treatment effect in HIV-positive patients because of the lack of statistical power, but the difference between differences (DDP) in HIV-negative and HIV-positive patients of 0.38 (CI, 0.06 to 0.70) was statistically significant (P = 0.02).

Women were included in 13 trials representing 17.7% (148 of 837) of the patients studied, but response by gender in both treatment and control groups was available in only 7 trials [64, 66, 70, 73-75, 77]. Because HIV-positive status occurred only in male patients in the 16 trials, it acted as a confounding factor that might cause the response rate in men to appear lower. We therefore eliminated the two trials that included HIV-positive patients [64, 66] before gender subgroup analysis. Aggregation of data from the remaining five trials showed that interferon had a significant effect on the loss of HBeAg in both men and women. When comparing men and women, we found a trend favoring response in women, but this trend was not statistically significant.

Four studies that excluded patients with normal AST and ALT values found, on retrospective analysis, that patients who lost HBeAg had significantly higher baseline AST and ALT levels compared with nonresponders [65, 66, 69, 71]. Our meta-analysis was only able to assess the effect of pretreatment AST and ALT levels based on the two studies in which sufficient data were available. One study reported results stratified by pretreatment AST and ALT levels [78]; data were available from another study by personal communication [74]. Aggregating results from both studies, we were unable to show that pretreatment AST and ALT levels affected outcomes.

Chinese patients were previously thought to have a poorer response to interferon treatment when compared with white patients [26-28]. Two studies of Chinese patients were analyzed [74, 78]. Individually, they showed a trend toward the loss of HBeAg in patients treated with interferon. When data were aggregated, statistical significance was achieved for the loss of HBeAg: The difference in proportions between treatment and control groups was 0.12 (CI, 0.02 to 0.23; P = 0.01). When comparing these two studies to the remaining 13 studies, we found a trend toward a lower response in Chinese patients (CI, –0.24 to 0.004; P = 0.053). Two of the remaining 13 studies, however, included a total of seven Chinese patients [64, 68]. Seven studies did not enroll any Chinese patients, and four studies did not specify the ethnic origins of their patients: two from the United States [65, 66], one from the United Kingdom [67], and one from Spain [76].

Other subgroup analyses, such as those by sexual preference and pretreatment liver biopsy results, could not be done because the response rate for each subgroup was not reported consistently for both the treatment and control groups.

Reanalysis of Outcomes

Because HIV status was found to be a significant factor affecting response, outcomes were re-examined using the studies that enrolled only HIV-negative patients. The absolute difference in proportions between the treatment and control groups for loss of the following were HBsAg, 0.04 (0.006 to 0.08, P = 0.02); HBeAg, 0.19 (0.12 to 0.26, P = 0.0001); and HBV DNA, 0.17 (0.10 to 0.24, P = 0.0001) when patients were treated with 3 to 6 months of interferon and followed for 6 to 12 months.

All primary and secondary outcomes and subgroup analyses were reanalyzed using the Freeman-Tukey transformation. Two results with borderline significance by the method of DerSimonian and Laird (lymphoblastoid compared with recombinant interferon- 2a for the loss of HBsAg and 12 months compared with 6 months of follow-up) were not statistically significant by the Freeman-Tukey transformation, but the secondary outcome of developing anti-HBs remained statistically significant (P = 0.005).

None of the 16 published trials favored the control group for any of the primary or secondary outcomes. To account for publication bias, the nonparametric sign test was used to determine that five unpublished trials favoring the control group for the loss of HBeAg or the loss of HBV DNA and two unpublished trials favoring the control group for the loss of HBsAg would have to exist for our meta-analysis to lose statistical significance at P < 0.05 (Appendix Table 2).

Cointervention and Contamination

Cointervention occurred in one study [66] in which lymphacytopheresis was done for immunologic testing in some patients and was thought to adversely affect response.

Another study was contaminated by six patients who crossed over from the control group to one of two treatment arms [78]. One was a prednisone withdrawal arm, which was excluded from our analysis. Our meta-analysis was therefore known to be contaminated by no more than six patients.

Sensitivity Analyses

Sensitivity analyses sequentially eliminating trials with low quality scores were not found to have a statistically significant effect on the loss of HBsAg or HBV DNA (Figure 4). A smaller treatment effect was observed on the loss of HBeAg in studies with higher quality scores (P = 0.02). A widening of the pooled CIs was observed as trials were discarded, but no thresholds were observed.

View larger version (24K): [in this window] [in a new window]   Figure 4. Sensitivity analyses by quality scores. The effect of quality scores on the pooled confidence intervals is shown for loss of hepatitis B surface antigen (HBsAg) (left) and loss of hepatitis B e antigen (HBeAg) (right). The trials of lower quality were sequentially eliminated.

Further analyses were done to exclude trials with cointervention [66] and contamination [78], trials that enrolled anti-HDV-positive patients [64, 75] or did not check for anti-HDV status [66, 68], trials that enrolled children (age <18 years) [75], trials that enrolled anti-HBe-positive patients [73], and trials that did not measure HBV DNA [77]. No significant differences in outcome were observed for any of these analyses.

Adverse Effects

Quantitation of the rate and severity of adverse effects was difficult because of the lack of uniform reporting criteria. The most common side effects were "flu-like" symptoms of fever, chills, myalgias, arthralgias, headaches, and fatigue. These symptoms were most marked during the initial week of therapy and were more severe with higher doses. Leukopenia and thrombocytopenia occurred frequently. Other common adverse effects included alopecia, weight loss, and depression. Exacerbation of liver disease with elevations in serum AST and ALT levels was seen in most responders to interferon just before seroconversion. Seroconversion hepatitis was usually asymptomatic, but decompensation of liver disease with jaundice and liver failure was reported in three trials [64, 67, 77].

At least 20% of treated patients required dose reduction, and 5% required early termination of treatment because of adverse reactions. Three of 498 patients died during treatment (1 of liver failure, 1 of sudden death, and 1 of stroke), and 1 of 339 in the control group died (variceal bleed) over the same period. All these deaths occurred in one study [77]. Using the Fisher exact test, we found no statistically significant difference in mortality between the treatment and control groups, but the power of this test was low.

Discussion

Top Methods Results Discussion Author & Article Info References

Our meta-analysis examined 16 randomized, controlled trials in which -interferon was used to treat patients with chronic HBV infection (HBsAg-positive) with compensated liver disease and active viral replication (HBeAg- and HBV DNA-positive). Because the follow-up in these trials ranged from 6 to 12 months after therapy, only intermediate outcomes such as loss of HBsAg, HBeAg, and HBV DNA could be evaluated. Only 2 of 16 trials were able to show statistical significance for the loss of HBsAg [67, 70]. Our aggregate data showed that interferon had a statistically significant effect on all three major end points. The absolute difference for loss of the HBV carrier state (indicated by loss of HBsAg) was 6% higher than the natural seroconversion rate of 1.8% seen in controls, and the absolute difference for loss of active viral replication was approximately 20% above the spontaneous loss seen in the controls (33% compared with 12% for loss of HBeAg and 37% compared with 17% for loss of HBV DNA) over a follow-up period of 6 to 12 months. The long-term effect of interferon treatment on morbidity and mortality could not be assessed, but recently published 7-year follow-up data indicate that interferon treatment is associated with a sustained effect on loss of viral replication, and the clearance rate of HBsAg was 5.7 times the spontaneous seroconversion rate [22].

Natural history studies show that seroconversion from HBeAg to anti-HBe is preceded by loss of HBV DNA [79, 80]. This loss of HBV DNA is usually accompanied by an initial rise of AST and ALT levels, followed by normalization of these liver enzymes and subsequent inactivation of disease [81, 82]. The presence of HBsAg may persist for years after the loss of HBeAg [22, 83]. Our meta-analysis showed that the development of anti-HBs, anti-HBe, and the normalization of AST and ALT levels in the treatment group were all statistically significant, confirming the inactivation of disease.

Wide variations in loss of HBeAg were seen in the treated (9.5% to 69.7%) and the control (0% to 38.7%) groups among the 16 trials. Despite differences in the populations studied and the treatment protocols used, an absolute treatment effect 21% higher than that of spontaneous seroconversion was seen in HIV-negative patients. The response in HIV-positive patients was significantly poorer than the response in HIV-negative patients. This result confirms the results of previous analyses that showed HIV coinfection was a marker of poor response to interferon therapy [26, 27, 29, 84]. The degree of immune deficiency may be a confounding factor. A recently published study showed that chemotherapy was associated with enhanced viral replication and reactivation in HBV carriers [85].

Previous reviews have suggested that interferon treatment may not be effective in Chinese patients [26-28]. Factors such as genetic differences, longer duration of infection, and immune tolerance to HBV infection have been postulated; however, our analysis showed that interferon had a treatment effect on the loss of HBeAg in patients reported in the two studies from Hong Kong. When comparing the two studies of Chinese patients to those of white patients, we found a trend toward a lower loss of HBeAg in Chinese patients. Caution should be used when interpreting such results because multiple confounding variables are intrinsic to a comparison between studies.

The optimal dose of interferon and the optimal duration of treatment have been controversial. Our meta-analysis found that a high dose of interferon was more effective than a low dose, but we were unable to show that treatment for 6 months was more effective than treatment for 3 months. Unfortunately, higher doses were associated with an increase in the frequency and severity of adverse effects. As many as 20% of all treated patients required dose reductions, and 5% required early termination of treatment due to adverse reactions. Furthermore, high-dose regimens are much more expensive. Further analyses are required to determine the optimal dose and duration of interferon therapy.

Although most patients with chronic HBV infection are asymptomatic, their initial presentation to medical attention may be related to the nonspecific symptoms that accompany the exacerbation of liver disease just before spontaneous seroconversion [86, 87]. It is unclear whether this group of patients will benefit from treatment. The studies evaluated here used a run-in period of 6 to 12 months during which HBeAg and HBV DNA were examined repeatedly to confirm the presence of sustained viral replication before initiation of interferon therapy. Patients who lose HBV DNA during this observation period are likely to seroconvert spontaneously from HBeAg to anti-HBe [79, 80]. One study did not measure HBV DNA levels, which is a costly and not universally available test, but monitored HBeAg positivity for 6 months before the initiation of therapy to minimize the inclusion of spontaneous seroconverters. This study obtained a statistically significant result for loss of HBeAg [77]. Thus, the strategy of observing the continued presence of HBeAg for 6 months may be an acceptable alternative to HBV DNA testing to identify patients who warrant therapy.

Multiple statistical methods can be used in doing a meta-analysis. The choice of scale of treatment effect and the transformation of outcomes are crucial when zero outcomes exist. The magnitude of response, CIs, and P values may vary with the different statistical methods, but a simple nonparametric sign test can be used to rule out a zero or negative treatment effect for each of our outcomes (see Appendix Table 2. Using this confirmatory method, our primary outcomes of loss of HBsAg, HBeAg, and HBV DNA remained statistically significant [P < 0.005]. We did not adjust for multiple comparisons in our P values; therefore, results with borderline significance (such as the development of anti-HBs, lymphoblastoid interferon compared with recombinant interferon 2a for the loss of HBsAg, 12 months compared with 6 months of follow-up, and the effect of quality score on the loss of HBeAg) should be interpreted with caution.

In summary, we have shown that -interferon is effective for the treatment of patients with chronic HBV infection and compensated liver disease who are HBs Ag-, HBeAg-, and HBV DNA-positive and HIV-negative in the setting of clinical trials. Higher doses of interferon have been shown to be more effective but are associated with more adverse effects. It remains too early to determine whether the use of interferon will significantly reduce the complications of liver failure, hepatocellular carcinoma, and death. It may therefore be premature to recommend the widespread use of interferon. A concerted effort should be made to form a registry of treated patients to determine whether viral replication has been permanently eradicated and whether the risks for developing the long-term sequelae of this disease have truly been reduced.