Recombinant interferon-2b was recently approved by the U.S. Food and Drug Administration (FDA) for the treatment of chronic hepatitis B [20], but clinical trials show that only a proportion of treated patients lose viral serologic markers (such as HBV DNA, hepatitis B virus e antigen [HBeAg], or eventually hepatitis B virus surface antigen [HBsAg]) [21-23]. Information on more definitive clinical end points, such as a decline in the incidence and mortality of cirrhosis or hepatocellular carcinoma, must await the results of long-term follow-up during the next 20 to 30 years. Interferon therapy is now available for persons with chronic hepatitis B but is expensive and commonly associated with short-term adverse effects. In our study, we projected the expected clinical and economic outcomes of patients with chronic hepatitis B and estimated the cost-effectiveness of interferon therapy.

Methods

Top Methods Results Discussion Author & Article Info References

Decision Analytic Model

We determined whether interferon should be given to patients with HBeAg-positive chronic hepatitis B who did not have cirrhosis (as determined by biopsy specimens) by using a decision analytic model to predict the likely clinical and economic outcomes for two cohorts of patients with chronic hepatitis. One cohort received a 16-week course of interferon and the other received standard care (treatment for cirrhosis-related signs and symptoms using diuretics, lactulose, and prophylactic antibiotics).

Using a decision analysis software program, Decision Maker 7.0 (Pratt Medical Group, Boston, Massachusetts), we did a Markov computer simulation [24, 25] to estimate prognosis in a hypothetical cohort of patients. The Markov simulation is a modeling technique in which hypothetical patients are followed over time as they move among states of health. We defined each state by serologic viral markers and clinical descriptors (Figure 1). The likelihood and timing of adverse or beneficial events determine prognosis. All patients began the simulation having chronic hepatitis and both HBeAg and HBsAg. After each simulated year, some patients remained in the same state of health, whereas others improved (for example, became HBeAg negative) or developed complications (such as compensated cirrhosis). The likelihood of an improved or worsened state of health was specified by a set of probabilities (Table 1). The simulation continued until no patient remained alive. To estimate the average survival or life expectancy of each cohort, we recorded the number of patients in each state of health at the end of each simulated year; each member of the cohort who remained alive contributed 1 person-year to the total survival of the cohort.

View larger version (24K): [in this window] [in a new window]   Figure 1. States of health in the decision model. Each circle represents a state of health. The represents the potential improved or worsened health status that might occur during a year. For example, a patient with chronic hepatitis may have a change in serologic markers or may develop cirrhosis. If the health status does not change, then patients remain in the same state of health (arrows not shown). The signifies that patients in any state of health may develop hepatocellular carcinoma or die. Dotted arrows represent transitions that occur at a low frequency. HBeAg = hepatitis B e antigen; HBsAg = hepatitis B surface antigen.

Because the quality of life in some states of health may be preferable to that in other states (for example, having asymptomatic chronic hepatitis may be preferable to having decompensated cirrhosis), we calculated the quality-adjusted life expectancy for each cohort by adding less than 1 full person-year for members in these less desirable states. As explained below, we based these adjustments on the judgments of senior clinicians familiar with both liver disease and interferon therapy. For example, we diminished the quality of life for patients with hepatocellular carcinoma by approximately 50%; in other words, in each year in which they were alive and had hepatocellular carcinoma, they contributed only 6 quality-adjusted months to the cohort's quality-adjusted survival. We also diminished the quality of life during the 16-week course of interferon therapy by approximately 13%. Because techniques for quality adjustment remain controversial and are not standardized, we report both life expectancy and quality-adjusted life expectancy for each strategy. We based our economic projections on the costs associated with caring for patients in each state of health each year. In accordance with standard principles of economic analyses [26-28], we discounted both future costs and future life-years by 5% per year to reflect the higher value of spending a dollar now as opposed to a year from now.

Although we recognized the importance of other clinical descriptors of disease activity (such as serum enzyme levels [29, 30] and histologic evidence of inflammatory activity [31, 32]) in patients with chronic hepatitis, sufficiently detailed data about prognosis based on combinations of these descriptors along with serologic viral markers and liver function do not currently exist. Consequently, we modeled only the nine states of health shown in Figure 1. The serologic viral marker events we modeled included loss of HBeAg or HBsAg and relapse or reactivation of HBeAg. Clinical events included the progression of chronic hepatitis to compensated and then to decompensated cirrhosis; the development of hepatocellular carcinoma; and death.

Data Sources

Probabilities and Utilities

We did a MEDLINE search and examined the bibliographies of all identified articles. For studies in which data were sparse or missing, we obtained a consensus opinion from an expert panel (see Acknowledgments), using a modified Delphi method.

Costs

We obtained actual variable hospital costs (the cost to treat one additional patient with the same disease) from the Clinical Cost Manager (Transition Systems I, Boston, Massachusetts) accounting system at New England Medical Center. We approximated physician fees for inpatient and outpatient care by multiplying physician charges by the mean reimbursement-to-charge ratio for internal medicine services (0.51). Laboratory costs were obtained from actual variable costs. Drug costs reflect the wholesale drug price. The expert panel estimated the frequency of hospitalizations, outpatient visits, and laboratory tests, as well as which drugs would probably be used for each health state.

Assumptions

1. We considered the prognosis of patients who did not lose HBeAg within 1 year despite receiving interferon therapy to be identical to that of untreated patients.

2. We considered loss of HBsAg to be permanent [33]. Although HBV DNA may persist (as measured by the polymerase chain reaction assay in serum or liver after the loss of HBsAg [34, 35]), we considered patients with chronic hepatitis who lose both HBeAg and HBsAg to be effectively "cured," reflecting the natural extension of the asymptomatic healthy carrier who has become HBsAg negative [36]. On the basis of the expert panel's recommendations, we assumed that these patients have a risk for developing compensated cirrhosis and hepatocellular carcinoma that is one hundredth that of patients who remain HBsAg positive.

3. We did not consider unusual complications from chronic hepatitis B (such as glomerulonephritis, polyarteritis nodosa, cryoglobulinemia, or vasculitis [37]), some of which may be ameliorated by interferon therapy [38].

4. We did not include screening for hepatocellular carcinoma in our analysis [39]. Although one study [40] in Asian persons suggested that screening might be advantageous, the benefits may not apply to other groups [41]. An Italian study [42] failed to find any benefit from ultrasonography or -fetoprotein screening tests.

5. We did not consider a second course of interferon therapy in nonresponders or patients who became positive again (reactivation or relapse) for HBeAg (although such patients often respond to a second course of interferon therapy [20]) or in patients with cirrhosis [37, 43-45] because few studies have been done for these settings and no consensus has been reached [43, 46].

6. We estimated the benefit of interferon from studies with interferon-2b, which appears to be more effective than interferon- [47-49]. Two studies [50] suggest that interferon-2a results in a higher prevalence of anti-interferon antibody than interferon-2b. Other therapies, including combination or sequential therapy with interferon or priming with steroid therapy, have not improved response rates [51, 52].

7. We considered loss of HBeAg as equivalent to loss of HBV DNA as a marker of viral replication. We did not consider the presence of the precore mutant (HBeAg negative) HBV, which is more likely to be resistant to therapy but is rare in the United States [53-55].

8. We did not consider liver transplantation because of the frequent recurrence of HBV infection and progressive hepatitis in persons who receive transplanted livers [56-59] and because only a small proportion of patients with decompensated cirrhosis could be served by the currently limited supply of donor organs [60].

9. We did not consider concomitant infection with hepatitis C virus or hepatitis D (delta) virus.

10. We used age-specific standard life-tables to reflect the underlying background likelihood of dying from causes occurring in the general population [61]. In the base-case analysis, the age at inception was assumed to be 35 years.

Sensitivity Analysis

Published studies and expert opinion vary in their estimates of the values used in our analysis. In the sensitivity analysis, we analyzed the effects of these alternative assumptions in the decision model by varying the value of each parameter over a wide range to determine whether the optimal decision changed.

Summary of the Data

Natural History of Chronic Hepatitis B

Our meta-analysis of nine randomized, controlled clinical trials of interferon-2b in 552 patients suggests that 9.1% of untreated patients will become HBeAg negative in the first year (see Appendix). Most studies have found that the annual probability that untreated patients will become HBeAg negative is between 9% and 14% [19, 62-66], but some uncontrolled studies suggest a much lower (0% to 3%) [67, 68] or higher (19% to 20%) [69, 70] annual probability. We assumed that after the first year, the annual probability of becoming negative for HBeAg was 10%.

Our meta-analysis suggests that 1.7% of all untreated patients will become HBsAg negative in the first year after detection and that 18.8% of untreated patients who lose HBeAg will become HBsAg negative in the first year after detection (see Appendix). Except for one study [62] in which none of 21 patients lost HBsAg over 3.6 years, most studies [19, 33, 67, 69-75] suggest that beyond the first year after detection, the annual probability of becoming negative for HBsAg ranges from 0.5% to 2.3% among untreated patients. In all of these studies, however, either the HBeAg status was unknown or some of the patients were known to be positive. Because we assumed that patients must first lose HBeAg before becoming negative for HBsAg, these studies underestimate the likelihood of becoming negative for HBsAg among patients who are HBeAg negative. In the only two published studies examining patients after the spontaneous loss of HBeAg, the annual probability of becoming HBsAg negative was 2.3% [76] and 2.9% [6], respectively. Beyond the first year after detection, we assumed that the annual probability of becoming HBsAg negative was 2.9%. The probability of relapse in the first year after becoming HBeAg negative ranges from 3% to 28% [50, 64-66, 77]. We used a probability of 7% in the first year and an annual probability of 2.9% during subsequent years [6].

Studies [70, 78-80] suggest that the annual probability of developing cirrhosis ranges from 0.4% to 14.2%. Using Cox regression analysis, Fattovich and colleagues [78] found that persons with HBV DNA had an odds ratio of 13.1 for developing cirrhosis. On the basis of their regression analysis for patients older than 30 years of age, we estimated that the annual probability of developing cirrhosis was 1.0% for patients who had HBsAg and 12.1% for patients who had HBeAg (see Appendix).

Studies [81-84] suggest that the annual probability that patients with cirrhosis will decompensate ranges from 3.8% to 9.5%. In a multicenter study [84] involving 366 patients followed for a mean of 6 years, neither the presence of HBV DNA nor that of HBeAg predicted the development of decompensated cirrhosis by Cox regression analysis. On the basis of this latter study [84], we used an annual probability of 5.6%. Once decompensated cirrhosis has developed, the disease-specific excess mortality rate (the residual mortality rate after adjusting for general population mortality or all-cause age-, sex-, and race-specific mortality and other causes of death explicitly represented in the model) for these patients is 0.39 per year, regardless of HBeAg status [44].

The annual probability of developing hepatocellular carcinoma among patients with chronic hepatitis ranges from 0.2% to 0.7% [80]. Among patients with cirrhosis, the annual probability of developing hepatocellular carcinoma ranges from 0.2% to 7.8% [33, 42, 82-86]. Although Beasley [33] found that the presence of HBeAg increased the risk for hepatocellular carcinoma, a more recent study [84] found that this risk was not affected by HBeAg status. Among persons who are positive for HBsAg, the annual incidence of hepatocellular carcinoma is 0.5% and 2.4% for those with chronic hepatitis and cirrhosis, respectively [87]. Hepatocellular carcinoma is resistant to radiation and chemotherapy [41] but may sometimes be resectable [87, 88]. According to the most recent report [89] from the Surveillance, Epidemiology and End Results (SEER) Program, the 5-year relative survival for persons with liver cancer is 5% to 6%, yielding a disease-specific excess mortality rate of 0.56 per year.

Effect of Interferon-2b

Our meta-analysis suggests that 45.6% of patients receiving interferon-2b and 9.1% of controls will become HBeAg negative in the first year after treatment (see Appendix) (an absolute difference of 36.5%). High serum aminotransferase levels, a low HBV DNA level, a history of acute icteric hepatitis, female sex, non-Asian origin, absence of antibody to the human immunodeficiency virus, and even chronic active hepatitis (compared with chronic persistent hepatitis or minimal disease determined using liver biopsy specimens) have been reported to increase the likelihood of response [43, 90-92]. The natural history of nonresponders has not been clearly defined in published studies, in part because many of these patients have been enrolled in new drug trials [93]. We assumed that nonresponders had the same natural history as untreated persons.

Our meta-analysis suggests that 7.7% of all patients receiving interferon-2b and 1.7% of controls will become HBsAg negative in the first year after treatment (an absolute difference of 6.0%) (see Appendix). It also suggests that 37.2% of patients treated with interferon who become HBeAg negative and 18.8% of controls will become HBsAg negative (an absolute difference of 18.4%). Because significant heterogeneity was noted among study results (see Appendix), we biased our model against interferon by assuming that interferon does not increase the likelihood of becoming negative for HBsAg among patients who lose HBeAg (see footnote in Table 1. Two long-term studies suggest that patients becoming negative for HBeAg in the first year after interferon treatment have an increased rate of HBsAg loss beyond the first year of 2.8 times [76] (95% CI, 0.3 to 27) by our logistic regression and 5.7 times [6] (CI, 1.8 to 17.5) that occurring after spontaneous HBeAg loss. But, again, to bias our baseline analysis against interferon, we assumed that interferon did not increase the rate of HBsAg loss after the first year. We analyzed the effect of a higher rate of HBsAg loss (see Appendix).

Although studies [94] in Asian persons suggest a much higher rate of relapse of HBeAg after interferon therapy, studies in Western populations suggest relapse rates between 3% and 14% in the first year after therapy and no relapses thereafter [6, 65, 77]. We assumed that the probability of relapse was 7% in the first year, but, to bias our analysis against interferon, we assumed that the probability was 2.9% in each subsequent year, even among patients who received interferon. We analyzed the effects of assuming that no late relapses occurred after the first year (see Appendix).

Nearly all patients who receive interferon develop flu-like symptoms [20, 95]. The dose of interferon is reduced in about 35% of patients, and therapy with the drug is discontinued prematurely in about 5% because of fatigue, emotional lability, alopecia, autoimmune disease, and bone marrow suppression [20, 95-97]. We assumed that the randomized trials used in our meta-analysis reflected intention to treat and therefore included any premature discontinuation of interferon therapy discussed in their reported results. To continue our bias against interferon, we did not decrease the cost of interferon therapy to reflect any discontinuation of treatment due to side effects.

Quality of Life

Preliminary data with the Sickness Impact Profile [98] suggested that, at the end of follow-up, patients with chronic hepatitis B infection who were treated with interferon had improved quality of life, whereas the scores of the untreated controls did not change. Our expert panel assessed their own utilities for being in each health state using the standard reference gamble (balancing a risk for dying in the near term against the benefit of being in a state of health with a better quality of life) and the time-tradeoff technique (balancing longer survival at a lower quality of life against shorter survival in a state of health with a better quality of life). We averaged the two results to calculate the quality-of-life adjustments shown in Table 1. We assumed that the relation between long-term quality-of-life adjustments was multiplicative (for example, the quality of life for a patient with compensated cirrhosis [0.92] and HBeAg [0.95] was 0.92 x 0.95, or 0.87).

Costs

Using the wholesale drug price of interferon to reflect societal costs (excluding charges that include an arbitrary profit margin) and assuming a treatment course of 10 million units three times per week for 16 weeks, we estimated the cost of interferon therapy to be $4595 for the drug itself plus $975 for office visits and laboratory fees [99-101], for a total cost of $5570. We also analyzed the effects of higher treatment costs (see Appendix). Table 1 summarizes the data used in our decision model.

Results

Top Methods Results Discussion Author & Article Info References

Baseline Analysis

Table 2 summarizes the results of our analysis for a 35-year-old man with chronic hepatitis who is positive for both HBeAg and HBsAg. Compared with standard care, interferon would substantially increase both life expectancy and quality-adjusted life expectancy; it would also decrease costs. Therefore, interferon dominates the standard care strategy by improving outcomes and saving resources.

A rough, "back-of-the-envelope" calculation is provided in Table 3 for two cohorts of 1000 patients. Neither life expectancies nor expenses are discounted. For each cohort, the results are predicated on the distribution of serologic viral markers at the end of the first year. For example, based on the data shown in Table 1, 9.1% of patients receiving standard care would become HBeAg negative in the first year and, among those, 18.8% would also become HBsAg negative in the first year. Of 1000 patients, 91 would become HBeAg negative; 74 of the 91 would become HBeAg negative alone and 17 would become negative for both HBeAg and HBsAg. The survival numbers are identical for patients who have the same serologic pattern regardless of treatment, but more patients treated with interferon have serologic patterns that confer a more favorable prognosis. Thus, interferon treatment is projected to save about $6.6 million and 3100 years of life for every 1000 patients treated.

Figure 2 shows the cumulative lifetime probability of developing liver complications from chronic hepatitis B infection. We estimated that interferon-2b would reduce the absolute cumulative lifetime incidence of cirrhosis, decompensated cirrhosis, and hepatocellular carcinoma by 13%, 9%, and 4%, respectively. Thus, approximately 7 patients would need treatment to prevent 1 case of cirrhosis, 11 patients would need treatment to prevent 1 case of decompensated cirrhosis, and 26 patients would need treatment to prevent 1 case of hepatocellular carcinoma. Our calculated 10-year risk for cirrhosis with standard care—48%—is similar to the 50% to 60% risk reported by Fattovich and colleagues [78]. Our estimated lifetime risk for decompensated cirrhosis or hepatocellular carcinoma for patients who are positive for HBeAg is about 64%, slightly higher than the 40% to 50% risk estimated by Beasley [102]. Of course, some of Beasley's patients may not have been positive for HBeAg, whereas we assumed that all of our patients were positive for HBeAg.

View larger version (19K): [in this window] [in a new window]   Figure 2. Cumulative probability of developing cirrhosis, decompensated cirrhosis, or hepatocellular carcinoma with and without interferon therapy. The lines represent the cumulative probabilities of developing these conditions.

Sensitivity Analysis

We used sensitivity analysis to examine the stability of our results by varying each parameter over a wide range of alternative values (Table 1). The range was determined by using previously reported values, 95% CIs for meta-analytic results, resource utilization and quality adjustments, and total charge or national reimbursement data. For the range of values listed in Table 1, interferon remained the preferred strategy.

For example, if the annual probability of developing compensated cirrhosis was one tenth the baseline likelihood (resulting in a 10-year cumulative risk of 7% compared with 48% using standard care at baseline), interferon would nonetheless increase quality-adjusted life expectancy by 1.1 years (compared with 3.4 years in the base-case analysis). Some experts might argue that by using data primarily from Asian persons, we have overestimated the annual probability of developing hepatocellular carcinoma. Even if the annual probability of developing hepatocellular carcinoma was one thirtieth the baseline likelihood (lifetime cumulative risk of 1% compared with 24% using standard care at baseline), interferon would still increase quality-adjusted life expectancy by 3.1 years. If the annual probability of developing cirrhosis and hepatocellular carcinoma was low (one tenth and one thirtieth the baseline rates, respectively) and if the quality of life for patients with either HBeAg or HBsAg positivity was not diminished, then interferon would still increase quality-adjusted life expectancy by 8 months. Finally, if loss of HBeAg alone did not confer any benefit in reducing the annual probability of developing cirrhosis or hepatocellular carcinoma, would interferon still be beneficial? Our model suggests that it would still increase quality-adjusted life expectancy by 2.0 years at a marginal cost-effectiveness ratio of $980 per discounted quality-adjusted life-year (DQALY) gained.

Figure 3 shows the effect of changing the probability of becoming negative for HBeAg within the first year after interferon treatment. The benefit of interferon approaches zero as the probability approaches that occurring with standard care (9.1%). Even at the lower limit of the 95% CI for the risk difference between interferon and standard care (37.8% – 14.1% or 23.7%, that is, a 32.8% probability of becoming negative for HBeAg in the first year compared with the baseline probability of 45.6%), interferon would increase quality-adjusted life expectancy by 2.2 years and would cost $2318 (not discounted) less than the cost with standard care. The discounted marginal cost-effectiveness ratio for interferon at that lower limit of the CI would be only $840 per DQALY gained. Although the undiscounted costs are lower, the discounted lifetime costs of interferon exceed those for standard care because the expense of interferon occurs in the near term, whereas the gains from cost savings and survival occur far in the future.

View larger version (16K): [in this window] [in a new window]   Figure 3. Sensitivity analysis varying the probability of becoming negative for hepatitis B e antigen in the first year after interferon therapy.Top. The two lines in the top panel represent the quality-adjusted life expectancy afforded by each therapy. The preferred strategy corresponds to the higher line. Bottom. The sharply curving line represents the marginal cost-effectiveness of interferon in discounted thousands of dollars per discounted quality-adjusted life-year gained (K$/DQALY). The denotes the baseline value for the probability of becoming negative for hepatitis B e antigen (HBeAg) (45.6%) in the first year after interferon therapy, and the horizontal bar represents the 95% CIs. As the probability of becoming HBeAg negative after interferon therapy approaches the probability occurring with standard care (9.1%), the benefit of interferon approaches zero and the marginal cost-effectiveness ratio approaches infinity.

Finally, we changed several assumptions at the same time to magnify any effects. If we biased all the values in favor of standard care Table 1, the decision became essentially a "toss-up" [103], although interferon remained slightly preferred (14 days difference). However, if the risk for developing either cirrhosis or hepatocellular carcinoma was returned to our baseline assumptions, then interferon would be preferred by between 4 and 16 quality-adjusted months. If hepatitis B and its complications were associated with no costs but if a course of interferon continued to cost $5570, interferon would have a marginal cost-effectiveness ratio of $4900 per DQALY gained.

Discussion

Top Methods Results Discussion Author & Article Info References

No long-term randomized trial of interferon therapy in patients with chronic hepatitis B with definitive health state end points (such as survival, cirrhosis, or hepatocellular carcinoma) exists. Using natural history data and the results of available randomized studies, we translated the reported changes in serologic viral markers into differential risks for developing cirrhosis or hepatocellular carcinoma. Despite the limited effect of interferon on the serologic viral markers, our analysis strongly supports using interferon: It should both prolong survival and diminish lifetime costs. Our analysis shows that the marginal cost-effectiveness ratio for interferon does not exceed $12 000 per gain in quality-adjusted life-year despite variation of all parameters and structural assumptions used in our analysis. If we restricted our analysis to a 10-year time horizon, thus limiting any benefits to that time frame, interferon would cost about $3600 per gain in DQALY. Thus, even if the benefit of interferon beyond 10 years was zero, it would still not be "expensive" compared with many therapies in common use in 1995.

Because we restricted our meta-analysis to studies that used interferon-2b and because we included additional published trials [95, 104-108], our results differ from those of previously published studies [22, 23]. With their lower estimates [22, 23] of responsiveness, the marginal cost-effectiveness ratio would still fall below $9000 per gain in DQALY (results available from authors on request). Our decision model differs from previous analyses [1, 3, 4, 109, 110] of the prognosis of hepatitis B by including serologic viral markers as states of health that differentiate subsequent prognosis. Consequently, some probabilities differ from those used in previously described decision models. Although other investigators assigned an excess mortality to patients with chronic hepatitis or cirrhosis, we instead explicitly modeled separate states of health for patients with decompensated cirrhosis or hepatocellular carcinoma and assigned substantially higher mortality rates to those states.

The cost estimates used in our analysis are based on a societal perspective and differ from those used in previous analyses [1, 3, 4, 109]. Previous investigators relied on charge data or mean payments and may have selected study groups with high resource utilization by basing their analysis on longitudinal hospital discharge data. We used actual variable cost data and a survey questionnaire given to our expert panel. If we had used the higher charge figures for each health state, our analysis would have shown interferon to have an even greater economic benefit.

Our results are based on projections of the effects of interferon on serologic markers and do not represent trials showing benefits on hard clinical end points. Our results do not necessarily apply to patients who have HBeAg-negative hepatitis or to those with symptomatic cirrhosis in whom the benefits of interferon are less clear. Our cost-effectiveness analysis assesses the long-term effects of interferon using data from clinical trials that only show the effects of interferon on serologic viral markers. Showing the benefits of interferon on mortality will require decades of follow-up. When decisions must be made before long-term results are available, decision analytic models can use intermediate outcomes to project long-term benefits. Even in the absence of a long-term trial, the available data strongly support the routine use of interferon-2b until the results of a clinical trial prove otherwise. In fact, interferon-2b therapy for chronic hepatitis B infection is not expensive; rather, it can save lives and societal resources.

Appendix: Meta-Analysis, Probability of Developing Cirrhosis, Cost Estimates, and Decision Analysis

Meta-Analysis

Methods

Using MEDLINE and examining the bibliographies of the identified references, we searched for all randomized trials involving interferon-2b (which, as of December 1993, was the only type of interferon approved for treatment of chronic hepatitis B infection). We excluded pediatric studies [95] and earlier studies whose findings have been superseded by later results. We excluded patients receiving less than the FDA-approved total course of 480 million units of interferon-2b [91, 104, 111, 112], patients crossing over from a control group to a treatment group [91, 107, 113], and one study [111] in which all patients were HBV DNA positive but were HBeAg negative [111]. To consider variations between and within studies, we calculated odds ratios and rate differences using the conservative DerSimonian and Laird method [114, 115]. A chi-square test was used for heterogeneity but, because the chi-square test is insensitive [115-117], we defined P < 0.1 as statistically significant for testing heterogeneity. However, we considered P < 0.05 as statistically significant for testing treatment results.

Results

Appendix Table 1 summarizes the characteristics of patients from the nine studies used in the meta-analysis. All patients had HBsAg, HBeAg, and HBV DNA with the exception of 2 patients (one each in the treatment and control groups) in the study by Muller and colleagues [112]. After combining nine studies with 552 patients (91,95,104-108,112,113), we found that interferon-2b therapy had a pooled odds ratio of 7.4 (CI, 3.9 to 14.0; P <0.001) for causing disappearance of HBeAg. Interferon therapy increased the absolute rate of causing disappearance of HBeAg by 36.5% (CI, 23.7% to 49.2%; P < 0.001), from 9.1% to 45.6% (the chi-square test for heterogeneity was not significant).

Analyzing five studies [91, 95, 107, 108, 113] with 385 patients, we found that patients receiving interferon-2b therapy had a pooled odds ratio of 6.9 (CI, 3.3 to 14.6; P < 0.001) for becoming HBV DNA negative. Interferon therapy increased the absolute rate of becoming HBV DNA negative by 32.6% (CI, 17.7% to 47.6%; P < 0.001), an increase from 7.9% to 40.5% (the chi-square test for heterogeneity was not significant).

Examining eight studies [91, 95, 104-108, 113] with 501 patients, we found that patients receiving interferon-2b therapy had a pooled odds ratio of 3.2 (CI, 1.2 to 8.2; P = 0.016) for becoming HBsAg negative. Interferon therapy increased the absolute rate of becoming HBsAg negative by 6.0% (CI, 2.8% to 9.3%; P < 0.001), an increase from 1.7% to 7.7%. The chi-square test for heterogeneity yielded a P value of less than 0.025, suggesting significant heterogeneity among the study results.

Examining seven studies [91, 95, 104, 105, 107, 108, 113] with 127 patients, we found that interferon increased the absolute rate of becoming HBsAg negative after the loss of HBeAg by 18.4% (CI, 8.3% to 28.5%; P < 0.001), an increase from 18.8% to 37.2% (the chi-square test for heterogeneity was not significant).

Probability of Developing Cirrhosis

We applied a previously developed technique [118] to the Cox proportional-hazards model developed by Fattovich and colleagues [78]. Because we considered only adults and did not consider histologic results, we simplified their model to estimate the hazard rate for developing cirrhosis stratified by HBeAg status (assuming that HBeAg status approximated HBV DNA status). To calculate the overall hazard rate, we used the average incidence of each variable in the study by Fattovich and colleagues (that is, 88% had persistent HBV DNA and 46% were 30 years of age or older). For patients who were HBeAg negative, the relative risk was 0.18; for patients who were HBeAg positive, the relative risk was 2.39; both of these risks were used to modify the average hazard rate in the study by Fattovich and colleagues (21 events in 388.5 patient-years of follow-up or 0.054 per year). Assuming a constant hazard rate, we converted to an annual probability of 12.1% for patients with HBeAg and 1.0% for patients with HBsAg.

Cost Estimates

Appendix Table 2 contains the components of the cost estimates shown in Table 1.

Decision Analysis

Cost-effectiveness Analysis*

We determined the effect of varying our assumptions on the marginal cost-effectiveness ratios (Appendix Table 3). For persons who were 70 years of age, interferon had a marginal cost-effectiveness ratio of $2800 per DQALY gained. For persons who were 35 years of age and who had known compensated cirrhosis, interferon had a ratio of $79 500 per DQALY gained.

What if all patients who initially lose HBsAg with interferon therapy then have a return of HBsAg (reactivation or relapse)? Our results suggest that interferon would still increase quality-adjusted life expectancy by 2.7 years and would be cost saving because a greater proportion of patients would be HBeAg negative. What if all patients treated with interferon who lose HBeAg but not HBsAg within the first year have a fivefold higher risk (35%) of HBeAg reactivation than that which occurs with standard care (7%)? Our model suggests that interferon would increase quality-adjusted life expectancy by 2.5 years at a marginal cost-effectiveness of $22 per DQALY gained. Finally, we analyzed the possibility that patients who lose HBeAg after interferon therapy would lose it eventually anyway but that the rate of loss would be accelerated by interferon. If this were the case, the patients who were not responsive to interferon might then lose HBeAg at a rate lower than that observed within the standard care group beyond the first year of therapy. If that annual probability of becoming HBeAg negative beyond the first year decreases to less than 0.02% in these unresponsive patients, then standard therapy with its 10% annual probability would be preferred. After 10 years, this would result in only 28% of patients having HBeAg-positive chronic hepatitis or compensated cirrhosis after receiving standard therapy compared with 43% of patients receiving interferon.

Relaxing Deliberate Biases against Interferon

Finally, our baseline analysis was biased against interferon by excluding some other benefits suggested by limited data in the medical literature. If patients becoming HBeAg negative in the first year after interferon lose HBsAg at the annual probability of 15.3% beyond the first year after therapy (as previously reported) and if they do not have reactivation of HBeAg beyond the 7% risk in the first year after therapy, then the benefit of interferon increases to 5.0 years and 5.7 quality-adjusted life years (compared with 3.1 years and 3.4 quality-adjusted life years in our base case). To do this analysis, we added two states of health to our baseline model with nine states of health to represent patients treated with interferon who have chronic hepatitis or compensated cirrhosis and who lose HBeAg but not HBsAg in the first year after therapy. If we combine these assumptions with the values biasing against standard care (by implying a poorer prognosis with chronic hepatitis B) in Table 1, interferon increases quality-adjusted life expectancy to 20.1 years compared with 7.6 years using standard care, a benefit of 12.5 quality-adjusted life years.