Distinct seromarkers predict different milestones of chronic hepatitis B progression

Authors


  • Supported by the Department of Health, Taiwan; Bristol-Myers Squibb Co., USA; Roche Diagnostics, Switzerland; Academia Sinica, Taiwan; and the National Health Research Institutes (NHRI-EX98-9806PI), Taiwan.

  • Potential conflict of interest: Dr. Batrla-Utermann is an employee of and holds stock in Roche Diagnostics. The other authors report no potential conflict of interest. The sponsors of this study had no role in study design, data collection, data analysis, data interpretation, or writing of this report. Moreover, the corresponding author of this article had full access to all data in the study and had final responsibility for the decision to submit the results of this study for publication.

Abstract

Spontaneous seroclearance of hepatitis B e antigen (HBeAg) and hepatitis B virus (HBV) DNA undetectability are important milestones of chronic hepatitis B and major treatment endpoints of antiviral therapy. This study investigated the role of serum hepatitis B surface antigen (HBsAg) levels and established models for predicting HBeAg seroclearance and HBV DNA undetectability. A total of 2,139 HBsAg-seropositive, anti-HCV-seronegative, and treatment-naïve participants without liver cirrhosis at study entry were included. Spontaneous HBeAg seroclearance and HBV DNA undetectability were analyzed in 431 HBeAg-seropositive participants and 1,708 HBeAg-seronegative participants, respectively. Regression coefficients of predictors in Cox proportional hazard models were converted into integer scores for predicting seroclearance and predictive accuracy was assessed with time-dependent receiver operating characteristic (ROC) curves. The HBV DNA level was the most important predictor of HBeAg seroclearance but serum HBsAg level was the most significant predictor of HBV DNA undetectability. Compared to individuals with HBsAg levels ≥10,000 IU/mL, the multivariate-adjusted rate ratio (95% confidence interval) of HBV DNA undetectability was 1.20 (0.62-2.30), 2.49 (1.31-4.75), and 6.08 (3.19-11.61) for those with serum HBsAg levels of 1,000-9,999, 100-999, and <100 IU/mL, respectively. The area under the ROC curve (AUROC) of the prediction models for predicting the 5- and 10-year probabilities of HBeAg seroclearance and HBV DNA undetectability were 0.85 (0.80-0.90) and 0.78 (0.73-0.83) for HBeAg seroclearance, and 0.77 (0.72-0.82) and 0.73 (0.70-0.76) for HBV DNA undetectability. Conclusion: Prediction models incorporating important host and virus factors can predict HBeAg seroclearance and HBV DNA undetectability. Serum HBsAg levels rather than HBV DNA is the most important predictor of spontaneous HBV DNA undetectability. Serum HBsAg levels should be monitored in the management of patients with HBeAg-seronegative chronic hepatitis B. (Hepatology 2014;60:77–86)

Abbreviations
ALT

alanine aminotransferase

AUROC

area under the ROC

CHB

chronic hepatitis B

HBsAg

hepatitis B surface antigen

HCC

hepatocellular carcinoma

ROC

receiver operating characteristic

Chronic hepatitis B (CHB) is a global public health challenge due to its widespread distribution and potential to progress to severe clinical outcomes such as cirrhosis, hepatocellular carcinoma, and even death.[1] Approximately 350-400 million people are chronically infected, resulting in 0.5-1.2 million deaths per year.[1, 2] The natural history of CHB progression is typically defined by three phases: immune tolerance phase, immune clearance phase, and residual phase.[3] These phases are characterized by changes and interactions among several key biomarkers including serum alanine aminotransferase (ALT), hepatitis B surface antigen (HBsAg), hepatitis B e antigen (HBeAg), antibody against HBeAg (anti-HBe), and serum hepatitis B virus DNA (HBV DNA).[2, 4, 5]

The REVEAL-HBV study has shown that serum HBV DNA levels are the predominant drivers of CHB progression, showing an increasing risk of hepatocellular carcinoma (HCC), cirrhosis and liver death, and a decreasing rate of HBsAg seroclearance with increasing viral load.[3, 6-9] Seropositivity of HBeAg, a marker of active viral replication, is associated with a high risk of HCC.[10] The seroclearance of HBeAg and undetectability of HBV DNA are two major milestones in the natural history of CHB and in current clinical management guidelines.[3, 11-13] A recent study of CHB patients with high viral loads (≥10,000 copies/mL) found that lower serum HBV DNA levels were the strongest predictor of the spontaneous seroclearance of HBeAg and undetectability of HBV DNA, even after adjustment for age, gender, serum ALT levels, and HBV precore 1896 mutation.[4]

Quantification of serum HBsAg has been assessed for its role in the natural history of CHB.[14, 15] Significant correlations between levels of HBsAg in serum and cccDNA in hepatocytes, the transcriptional template of HBV, have been documented.[15-17] Quantitative serum HBsAg level has also been suggested as a promising new seromarker for the immunological response to CHB therapy as well as for the progression of CHB.[14-16, 18] Rapid on-treatment declines in HBsAg levels have been found to predict sustained virological response to therapy as well as subsequent HBsAg loss.[19, 20] In addition, a score-based prediction model was also recently developed that can accurately predict HBsAg seroclearance.[21]

However, it is still unclear whether serum HBsAg levels play a role in predicting spontaneous HBeAg seroclearance and HBV DNA undetectability in the natural history of CHB. The aims of this analysis of the REVEAL-HBV cohort data are to investigate the importance and predictability of serum HBsAg levels in spontaneous HBeAg seroclearance and HBV DNA undetectability, and to develop models for the prediction of both outcomes.

Materials and Methods

Study Cohort

From 1991 to 1992, 23,820 individuals aged 30-65 years from seven townships across Taiwan were enrolled. Each participant consented to a questionnaire interview, physical examination, and blood sample collection for serological testing and biochemical assays at study entry and follow-up examinations. Consents included follow-up through medical record review and computerized data linkages with national health insurance, cancer registry, and death certification databases. There were 4,155 HBsAg-seropositive participants in the study cohort. Participants were excluded if they had no adequate baseline serum samples for HBV DNA and quantitative HBsAg testing (n = 572), were seropositive for antibodies against hepatitis C virus (anti-HCV), or presented with cirrhosis detected by abdominal ultrasonography at study entry (n = 230), had baseline serum HBV DNA levels <300 copies/mL or no follow-up serum samples for HBV DNA testing (n = 1,214). A total of 2,139 participants were included in this analysis. A subset of individuals seropositive for HBeAg at study entry (n = 431) was used for the analysis of HBeAg seroclearance. In this study cohort, almost all individuals spontaneously serocleared HBeAg before their serum HBV DNA levels became undetectable. Thus, HBV DNA undetectability was analyzed in individuals seronegative for HBeAg (n = 1,708). This study was approved by the Institutional Review Board of the College of Public Health, National Taiwan University, Taipei, Taiwan.

Data Collection and Laboratory Methods

All participants were interviewed with a structured questionnaire by trained public health nurses at enrollment. Information was collected on sociodemographic characteristics, family history of major diseases, as well as dietary and lifestyle habits. Blood collection and health examinations were performed at study entry and follow-ups conducted every 6-12 months. Tests on serum HBsAg, HBeAg, Anti-HCV serostatus, and serum HBV DNA and ALT levels were performed using commercial kits: HBsAg and HBeAg by radioimmunoassay (Abbott Laboratories, North Chicago, IL), anti-HCV by enzyme immunoassay using second-generation test kits (Abbott Laboratories), ALT by serum chemistry autoanalyzer (model 736; Hitachi, Tokyo, Japan) using commercial reagents, serum HBV DNA levels by polymerase chain reaction (COBAS Amplicor; Roche Diagnostics, Indianapolis, IN) for baseline samples, and by real-time polymerase chain reaction (COBAS TaqMan; Roche Diagnostics) for follow-up samples. These laboratory procedures have been described previously.[3, 7] Serum HBsAg levels at study entry were quantified using the Elecsys HBsAg II Quant assay (Roche Diagnostics, Mannheim, Germany), which has been certified at a lower limit of detection of 0.05 IU/mL.

Ascertainment of Outcomes

Spontaneous HBeAg seroclearance and HBV DNA undetectability were the two outcomes of interest in this study. Patients were tested for HBeAg serostatus and HBV DNA levels at follow-up examinations. Seroclearance and undetectability cases were defined as the first instance in which an individual tested seronegative for HBeAg or had undetectable serum HBV DNA levels (<100 copies/mL) and remained HBeAg seronegative or HBV DNA undetectable at subsequent tests throughout the remainder of follow-up. This date was designated as the date of spontaneous seroclearance or of HBV DNA undetectability.

Statistical Analysis

Person-years of follow-up were calculated as the time from enrollment to the date of HBeAg seroclearance / HBV DNA undetectability, the date at death, the date at last-follow-up, or June 30, 2004, whichever came first. Both serum HBV DNA (copies/mL) and HBsAg (IU/mL) levels were logarithmically transformed for analysis. Differences in HBsAg levels between groups were tested using the Mann-Whitney U test. The Kaplan-Meier method was used to examine the cumulative incidence of HBeAg seroclearance and HBV DNA undetectability. Cox proportional hazards models were used to analyze the multivariate-adjusted rate ratios (with 95% confidence intervals [CI]) of seroclearance and undetectability. Variables that were previously established as important factors in the natural history of CHB, statistically significant (P < 0.05) in univariate analyses, and that are clinically applicable factors were included in multivariate analyses. Smoking was ultimately excluded from the final HBV DNA undetectability model, as including it did not affect the predictive accuracy of the model.

Development and Validation of Prediction Models

Prediction models were developed using a well-known score-based method.[22] Cox proportional hazards models were used to estimate regression coefficients for predictors included in each prediction model. Coefficients were converted into integer scores by dividing each regression coefficient by the corresponding standardizing coefficient for each model (female gender for HBeAg seroclearance, and age >60 years for HBV DNA undetectability), then rounding each quotient to the nearest integer. Predictive accuracy was assessed by time-dependent receiver operating characteristic (ROC) curves and examining the area under the ROC (AUROC) for the 5-year and 10-year prediction of HBeAg seroclearance and HBV DNA undetectability.[23] Confidence intervals for the AUROCs were constructed using the bootstrap method with 1,000 replications. Statistical significance was determined by two-tailed tests (P < 0.05). Statistical analyses were performed with SAS software (v. 9.2; SAS Institute, Cary, NC).

A separate internal validation was performed by allocating each subset of patients in a 2:1 ratio into development and validation groups. The development group was used to reestablish regression coefficients for each predictor, and the validation group was used to independently validate the developed model using ROC curves and the Hosmer-Lemeshow test for goodness-of-fit. As sample sizes were small using this method, the main findings are still presented using 431 individuals for HBeAg seroclearance, and 1,708 individuals for HBV DNA undetectability.

Results

Baseline Characteristics

Among the 2,139 individuals included in this analysis, the majority were male (65%), HBeAg-seronegative (80%), and had serum ALT levels <45 U/L (93%), no habit of alcohol drinking (88%), and serum HBsAg levels >1000 IU/mL (53%). Among the 431 HBeAg-seropositive individuals included in the analysis of HBeAg seroclearance, the majority were male (65%) and had serum ALT levels <45 U/L (84%), serum HBV DNA levels ≥108 copies/mL (66%), and serum HBsAg levels >10,000 IU/mL (57%).

Predictors of Spontaneous HBeAg Seroclearance

Among 431 HBeAg-seropositive participants with baseline HBV DNA levels ≥10,000 copies/mL and 3036.4 person-years (average 7.05 years) of follow-up, HBeAg seroclearance occurred in 187 participants, giving an incidence rate of 615.9 per 10,000 person-years. Baseline serum HBsAg levels were significantly lower in those who reached HBeAg seroclearance with a median (interquartile range [IQR]) of 3.85 (1.20) logs compared to 4.61 (1.27) logs in those who did not reach HBeAg seroclearance (P < 0.001) (Table 1). Only serum HBV DNA levels remained as a significant predictor of HBeAg seroclearance in multivariate analyses. Compared to participants with HBV DNA levels ≥108 copies/mL, the multivariate-adjusted rate ratio (95% CI) of HBeAg seroclearance was 1.91 (1.09-3.35) and 3.47 (1.84-6.56) for those with HBV DNA levels of 106-<108, and 104-<106 copies/mL, respectively (Table 1). Gender, serum ALT level, precore mutation, and HBV genotype were also significant predictors of HBeAg seroclearance. ROC curve analysis using the above predictive factors confirmed that HBsAg levels did not add to the predictability of HBeAg seroclearance. The 5- and 10-year AUROCs for HBeAg seroclearance remained the same with and without serum HBsAg levels in the regression models, at 0.86 for 5 years and 0.79 for 10 years.

Table 1. Incidence and Predictors of HBeAg Seroclearance by Baseline Characteristics
Baseline CharacteristicsN (%)HBeAg Seroclearance CasesPerson-Years of Follow-upIR (per 1000 PY)Crude Rate Ratio (95% CI)P ValueMultivariate Adjusted Rate Ratio (95% CI)P Value
  1. a

    Data missing for 2 individuals.

  2. b

    Data missing for 3 individuals.

  3. c

    Only available for patients with HBV DNA ≥10,000 copies/mL.

  4. d

    Data missing for 32 individuals.

Age, years        
30-39228 (53)1131578.271.601.0 1.0 
40-49114 (26)44809.754.340.73 (0.52-1.04)0.080.83 (0.55-1.25)0.38
50-5976 (18)26576.345.120.58 (0.38-0.89)0.010.89 (0.56-1.44)0.65
≥6013 (3)472.155.450.89 (0.33-2.42)0.821.84 (0.66-5.10)0.24
Gender        
Male281 (65)1362174.462.551.0 1.0 
Female150 (35)51861.959.170.96 (0.70-1.33)0.811.93 (1.29-2.89).001
Serum ALT (U/L)        
<45362 (84)1512603.458.001.0 1.0 
≥4569 (16)36432.983.161.47 (1.02-2.11)0.042.06 (1.35-3.14)<.001
Smokinga        
No287 (67)1211955.361.881.0   
Yes142 (33)661055.162.561.05 (0.78-1.42)0.76Not included 
Alcohol consumptionb        
No375 (87)1622626.461.681.0   
Yes53 (13)25379.665.851.08 (0.71-1.65)0.72Not included 
Precore mutantc        
Wildtype288 (79)1132208.651.161.0 1.0 
Mixed type27 (7)15168.389.151.97 (1.15-3.39)0.011.26 (0.70-2.25)0.44
1896 G/A51 (14)24241.195.272.37 (1.52-3.69)<.0011.66 (1.03-2.68)0.04
Basal core promoterc        
Wildtype197 (54)771326.658.041.0   
Mixed type84 (23)37572.264.671.12 (0.76-1.66)0.57  
Mutant84 (23)38713.553.260.90 (0.61-1.32)0.59Not included 
Genotyped        
B or B+C16192882.7104.231.0 1.0 
C238791967.040.160.33 (0.24-0.44)<.0010.33 (0.22-0.48)<.001
Serum HBsAg level (iu/mL)        
≥ 100,00049 (11)13413.931.411.0 1.0 
10,000-99,999198 (46)691509.845.701.54 (0.85-2.79)0.151.11 (0.60-2.04)0.74
< 10,000184 (43)1051112.794.373.51 (1.97-6.26)<.0011.07 (0.51-2.25)0.86
Serum HBV DNA level (copies/mL)        
≥ 108283 (66)952215.542.881.0 1.0 
106-<10887 (20)50545.991.592.45 (1.73-3.45)<.0011.91 (1.09-3.35)0.02
104 - 10661 (14)42274.9152.794.53 (3.13-6.55)<.0013.47 (1.84-6.56)<.001

Predictors of HBV DNA Undetectability

HBV DNA undetectability was analyzed in 1,708 HBeAg-seronegative participants detectable HBV DNA. During 13,547.3 person-years of follow-up, HBV DNA undetectability occurred in 408 participants, giving an incidence rate of 301.16 per 10,000 person-years. (Table 2). Baseline serum HBsAg and HBV DNA levels were significantly lower in participants with HBV DNA undetectability with a median (IQR) of 2.36 (1.34) and 3.73 (1.44) logs, respectively, compared to 2.99 (0.98) and 3.98 (1.53) logs in those who did not reach HBV DNA undetectability (P < 0.001 and P = 0.02, respectively). The cumulative incidence of HBV DNA undetectability for baseline serum HBsAg levels of ≥10,000, 1,000-9,999, 100-999, and <100 IU/ml were 33.3%, 30.2%, 41.6%, and 75.1%, respectively. Only serum HBsAg levels remained as a significant predictor of HBV DNA undetectability in multivariate analyses. Compared to baseline serum HBsAg levels ≥10,000 IU/mL, the multivariate adjusted rate ratio (95% CI) of HBV DNA undetectability was 1.20 (0.62-2.30), 2.49 (1.31-4.75), and 6.08 (3.19-11.61) for those with serum HBsAg levels of 1,000-9,999, 100-999, and <100 IU/mL, respectively (Table 2). The addition of genotype did not change results and thus was excluded from the final model, as much genotype data were missing.

Table 2. Incidence and Predictors of HBV DNA Undetectability by Baseline Characteristics
 HBeAg Seronegatives (N = 1,708)  
 N (%)HBV DNA Seroclearance CasesPerson-Years of Follow-upIR (per 1000 PY)Crude Rate Ratio (95% CI)P ValueMultivariate Adjusted Rate Ratio (95% CI)P Value
  1. a

    Data missing for 1 participant.

  2. b

    Data missing for 2 participants.

  3. c

    Data missing for 811 participants.

  4. d

    Data missing for 831 participants.

  5. e

    Data missing for 245 participants.

  6. f

    Rate ratio for ≥60 years vs. <60 years.

  7. g

    Third group represents those with HBV DNA 300 - < 105 copies/mL.

Age, years        
30-39519 (30)1214403.627.481.0 1.35 (1.00-1.82)f0.05
40-49509 (30)1184030.929.271.09 (0.85-1.41)0.49  
50-59519 (30)1183993.929.551.10 (0.85-1.41)0.48  
≥60161 (10)511118.945.581.81 (1.31-2.52)<.001  
Gender        
Male1117 (65)29910276.929.091.0 1.0 
Female591 (35)1093270.433.331.15 (0.92-1.44)0.211.35 (1.08-1.69)0.01
Serum ALT (U/L)        
<451632 (96)38512884.729.881.0 Not included 
≥4576 (4)23662.734.711.19 (0.78-1.82)0.41  
Smokinga        
No1137 (67)2398522.128.051.0 Not included 
Yes570 (33)1695016.733.711.24 (1.02-1.51)0.04  
Alcohol Consumptionb        
No1505 (88)34411823.729.091.0 Not included 
Yes201 (12)631703.036.991.29 (0.99-1.69)0.06  
Precore Mutantc        
Wild Type187 (21)391354.228.791.0 Not included 
Mixed Type82 (9)14602.423.240.75 (0.52-1.08)0.12  
1896 G/A628 (70)1155062.322.720.78 (0.42-1.44)0.42  
Basal Core Promoterd        
Wild Type523 (60)934011.923.181.0 Not included 
Mixed Type101 (12)16761.721.011.05 (0.74-1.48)0.80  
Mutant253 (28)492012.924.340.89 (0.52-1.51)0.67  
Genotypee        
B or B+C1056 (72)1888217.922.881.0 Not included 
C407 (28)1233471.335.431.52 (1.21-1.91)<.001  
Serum HBsAg level (iu/ml)        
≥ 10,00067 (4)10660.515.141.0 1.0 
1000-9999685 (40)966006.415.981.11 (0.58-2.14)0.751.20 (0.62-2.30)0.58
100-999624 (37)1484754.831.132.38 (1.25-4.51).0082.49 (1.31-4.75).006
<100332 (19)1542125.772.455.81 (3.06-11.03)<.0016.08 (3.19-11.61)<.001
Serum HBV DNA level (copies/ml)        
≥ 106105 (6)12768.315.621.0 1.0 
105-<106229 (13)401943.420.581.24 (0.65-2.36)0.521.15 (0.60-2.20)0.66
104-<105483 (28)1243541.635.012.17 (1.20-3.92)0.011.34 (0.75-2.39)g0.32
300 - <104891 (47)2327294.131.811.83 (1.03-3.28)0.04

In the stratification analysis by combinations of serum levels of both HBV DNA and HBsAg as shown in Table 3, HBsAg levels were the main predictors of HBV DNA undetectability. Even at the same serum HBsAg levels, serum HBV DNA levels were not significantly associated with rates of HBV DNA undetectability. ROC curve analysis using the multivariate regression models shown in Table 2 confirmed that serum HBV DNA levels did not add to the predictability of HBV DNA undetectability. The 5- and 10-year AUROCs for HBV DNA undetectability remained the same or similar (P > 0.05) when serum HBV DNA levels were added to the models, with AUROCs of 0.77 with and without HBV DNA levels at 5 years, and 0.73 versus 0.72 with and without serum HBV DNA levels at 10 years.

Table 3. Incidence Rates and Rate Ratios of HBV DNA Undetectability According to Both HBV DNA and HBsAg Levels
 N (%)HBV DNA Seroclearance CasesPerson-Years of Follow-upIR (per 1000 PY)Crude Rate Ratio (95% CI)P ValueMultivariate Adjusted Rate Ratio (95% CI)aP Value
  1. a

    Adjusted for age and gender.

Serum HBV DNA (copies/mL) / HBsAg levels (IU/mL)        
≥106/any level105 (6)12768.315.621.0 1.0 
105- 106/≥1000125 (7)191140.216.670.93 (0.45-1.92)0.840.96 (0.46-1.98)0.90
300-105/≥1000560 (33)785003.515.590.86 (0.47-1.57)0.620.82 (0.44-1.50)0.51
105- 106/100-99985 (5)11670.216.411.06 (0.47-2.40)0.891.09 (0.48-2.46)0.84
300-105/100-999508 (30)1353876.634.822.10 (1.16-3.79)0.012.04 (1.13-3.68)0.02
105- 106/<10019 (1)10133.175.144.86 (2.10-11.26)<.0015.08 (2.19-11.78)<.001
300-105/<100306 (18)1431955.673.124.68 (2.60-8.43)<.0014.66 (2.58-8.41)<.001

Long-Term Tracking of HBsAg Levels Prior to HBV DNA Undetectability

Among those with persistently detectable HBV DNA levels throughout follow-up, serum HBsAg levels remained very stable at or around 3 log IU/mL during follow-up (Fig. 1A). However, among those with HBV DNA undetectability, serum HBsAg levels prior to undetectability showed an overall steady slow decrease in median HBsAg levels (Fig. 1B).

Figure 1.

(A) Serum HBsAg levels prior to last follow-up in HBeAg seronegative patients who did not reach HBV DNA undetectability. (B) Serum HBsAg levels prior to HBV DNA undetectability in HBeAg seronegative patients.

Prediction Models for HBeAg Seroclearance and HBV DNA Undetectability

Predictive factors that were significant in multivariate analyses were included into prediction models for both HBeAg seroclearance and HBV DNA undetectability (Tables 4 and 5). Variables such as the precore mutation in Table 4 and age in Table 5 were combined into binary variables for easier clinical use and according to similarities in incidence rates shown in Tables 1 and 2. Corresponding regression coefficients and estimated predictive scores for each cofactor included in the models is also shown. Total scores ranged from 0-7 for the HBeAg seroclearance model and from 0-8 for the HBV DNA undetectability model.

Table 4. Regression Coefficients and Associated Prediction Scores for HBeAg Undetectability
 Multivariate Adjusted Rate Ratio (95% CI)Regression CoefficientScore
Gender   
Male1.0 0
Female1.92 (1.29-2.85)0.651221
Serum ALT (U/L)   
<451.0 0
≥452.11 (1.40-3.18)0.745451
Precore mutant   
Wildtype1.0 0
1896 G/A + mixed type1.47 (0.99-2.21)0.388301
Genotype   
C1.0 0
B or B+C3.06 (2.11-4.44)1.117582
Serum HBV DNA level (copies/mL)   
≥ 1081.0 0
106-<1081.89 (1.28-2.78)0.636811
104 - 1063.27 (2.01-5.32)1.185402
Table 5. Regression Coefficients and Associated Prediction Scores for HBV DNA Undetectability
 Multivariate Adjusted Rate Ratio (95% CI)Regression CoefficientScore
Age (years)   
< 601.0 0
≥ 601.35 (1.00-1.81)0.298171
Gender   
Male1.0 0
Female1.37 (1.10-1.72)0.316631
Serum HBsAg level (iu/mL)   
≥ 10,0001.0 0
1000-99991.18 (0.61-2.27)0.163961
100-9992.49 (1.31-4.74)0.911653
<1006.18 (3.24-11.79)1.821756

Predictive accuracy was measured using the AUROC. The AUROCs for predicting the 5- and 10-year probability of HBeAg seroclearance were 0.85 (0.80-0.90) and 0.78 (0.73-0.83), while the AUROCs for predicting the 5- and 10-year probability of HBV DNA undetectability were 0.77 (0.72-0.82) and 0.73 (0.70-0.76) (Fig. 2A,B). Using the developed models, predicted 5- and 10-year probabilities of HBeAg seroclearance and HBV DNA undetectability according to each possible total score are shown in the nomograms in Fig. 3A,B. The 5- and 10-year probabilities ranged from 0.08-0.72 and 0.23-0.98 for HBeAg seroclearance, and from 0.04-0.36 and 0.14-0.80 for HBV DNA undetectability.

Figure 2.

(A) ROC curves for 5- and 10-year prediction of HBeAg seroclearance in HBeAg seropositive patients. (B) ROC curves for 5- and 10-year prediction of HBV DNA undetectability in HBeAg seronegative patients.

Figure 3.

(A) Nomograms for the 5- and 10-year predicted probabilities of HBeAg seroclearance. (B) Nomograms for the 5- and 10-year predicted probabilities of HBV DNA undetectability.

Internal Validation of Prediction Models

When the models were reestablished and internally validated using 244 and 122 patients for the model predicting HBeAg seroclearance, and 1,139 and 569 patients for the model predicting HBV DNA undetectability, AUROCs in the development groups were still very similar from the AUROCs developed using the entire subset. Internal validation of the HBeAg seroclearance model with 122 patients was more imprecise due to small sample size, but still showed 5- and 10-year AUROCs of 0.75 and 0.73. Internal validation of the HBV DNA undetectability model with 569 patients showed 5- and 10-year AUROCs of 0.80 and 0.71.

Discussion

A recent study from the REVEAL-HBV group found that decreasing HBV DNA levels was the most important predictor of both HBeAg seroclearance and HBV DNA undetectability.[4] However, the previous study did not consider serum HBsAg levels, and only included individuals with viral loads ≥10,000 copies/mL. This study included all patients with detectable viral loads and is the first study to examine whether quantitative serum HBsAg level is an independent predictor of HBeAg seroclearance and HBV DNA undetectability.

Our results show that serum HBV DNA levels were the most important predictor of the first milestone of CHB progression, HBeAg seroclearance; serum HBsAg levels were not a significant predictor after adjustment for HBV DNA levels. We created a score-based prediction model for HBeAg seroclearance that could, with relatively high predictive accuracy, predict 5- and 10-year probabilities of HBeAg seroclearance using several important predictors. For HBeAg-seropositive individuals with high viral loads, it seems that the most vital goal for clinical management is to encourage host-mediated suppression of serum HBV DNA levels. Future studies should seek additional predictors of HBeAg seroclearance.

The second milestone of CHB progression, HBV DNA undetectability, was examined only in HBeAg-seronegatives, as almost all individuals serocleared HBeAg prior to HBV DNA undetectability. We found a significant biological gradient of decreasing serum HBsAg levels with increasing rates of HBV DNA undetectability. Notably, serum HBV DNA level was no longer a significant predictor of HBV DNA undetectability after taking serum HBsAg level into consideration.

Moreover, in predictability analyses, serum HBsAg levels did not increase predictability for HBeAg seroclearance when added to serum HBV DNA levels, emphasizing HBV DNA as the most important biomarker to monitor in HBeAg seropositives. On the other hand, serum HBV DNA levels could not increase predictability for HBV DNA undetectability when added to serum HBsAg levels, indicating HBsAg as the most significant predictor of HBV DNA undetectability. The developed predictive scoring system could still predict HBV DNA undetectability with fair accuracy. The findings suggest that distinct seromarkers predict different milestones of CHB, and that serum levels of HBV DNA and HBsAg are recommended for monitoring the clinical status of HBeAg-seropositives and HBeAg-seronegatives, respectively. For clinical management of HBeAg-seronegative patients, it may be important to monitor serum HBsAg levels. However, it is important to note that HBV DNA monitoring cannot be replaced. Although this study has shown that HBV DNA levels cannot predict viral load clearance, previous studies have shown that monitoring HBV viral loads are still essential for tracking progress of the disease, and for the prediction of other long-term serological and clinical outcomes such as HBsAg seroclearance, liver cirrhosis, and HCC.[24]

Previous studies have shown that HBsAg levels are relatively stable compared to serum HBV DNA levels in CHB patients.[25, 26] The one-time measurement of serum HBsAg level may reflect a long-term immune response to HBV, while a one-time measurement of serum HBV DNA level may not reflect fluctuations in viral replication. Our previous study has also shown that serum HBsAg levels have good predictability at low viral loads.[21] Thus, serum HBsAg levels may reflect the translation of mRNA produced from cccDNA and may be an excellent marker of immune response.[17, 18] Whereas a decline in serum HBV DNA level directly reflects a reduction in viral replication, a decline in serum HBsAg level may represent a reduced translation of mRNA and a balanced interaction between HBV and the host immune response, which determines the chance to lower viral loads.

The use of HBsAg levels can help to further refine risk categories and improve treatment of patients.[11] As viral load suppression is a main goal of antiviral therapy, the results of this study show that quantitative HBsAg monitoring can help clinicians to assess the probability of HBV DNA suppression and its impact on clinical treatment. A previous study showed that on-treatment monitoring of HBsAg in patients with HBeAg-negative CHB treated with peginterferon alpha-2a could identify patients who achieved sustained response 6 months posttreatment.[20] Eighty-eight percent of patients who achieved an end-of-treatment HBsAg level <10 IU/mL achieved HBV DNA suppression to <400 copies/mL 6 months posttreatment, whereas zero patients who had an end-of-treatment HBsAg ≥5,000 IU/mL were able to reach HBV DNA suppression. Patients with low end-of-treatment HBsAg levels were also much more likely to achieve HBsAg seroclearance 3 years posttreatment.[20] As this study was unable to take long-term trajectories of HBsAg into account, future studies should further examine the role of longitudinal HBsAg measurements in predicting serological outcomes such as HBV DNA undetectability and HBsAg seroclearance.

Moreover, elucidating the determinants of HBV DNA undetectability is clinically significant, as it has implications for outcomes such as HBsAg seroclearance and the development of HCC. Previous studies have shown that in addition to HBsAg levels, lowering HBV DNA levels is still an important determinant for future HBsAg seroclearance.[21, 27] However, both outcomes in this study are still intermediate milestones that may predict the occurrence of hard clinical endpoints. In a recent study, it was also found that among HBeAg seronegatives with detectable HBV DNA levels, reaching HBV DNA seroclearance during follow-up could reduce future HCC risk by almost 4-fold, emphasizing the importance of suppressing viral loads.[28] Other studies in the REVEAL-HBV cohort have also found the importance of HBeAg and HBsAg seroclearance and HBV DNA undetectability in reducing mortality, HCC, and liver cirrhosis incidence.[3, 6, 7, 24, 28, 29]

Interestingly, we also found that genotype B carriers were more likely to clear HBeAg, while genotype C carriers were more likely to reach undetectable HBV DNA levels. A previous study found higher HBeAg secretion in genotype C carriers, compared to genotype B, thus leading to earlier HBeAg seroclearance among genotype B carriers. On the other hand, they also found less secretion of HBsAg in genotype C carriers compared to genotype B carriers.[30] As this study showed that HBsAg levels are the strongest predictor of HBV DNA undetectability, it would thus be plausible that genotype C carriers would have higher rates of undetectability. However, the biological mechanisms and clinical significance of these results should be further clarified.

Lastly, the score-based prediction model has been widely used by the Framingham Heart Study to estimate the risk of coronary heart disease.[22] The predictive scoring models developed in this study not only incorporate other well-known predictors of HBeAg seroclearance and HBV DNA undetectability, but also capture a profile-based probability distribution for spontaneous HBeAg seroclearance and HBV DNA undetectability that can be used to refine clinical management of patients with CHB. A simple score-based prediction model can potentially eliminate the need for complicated calculations in clinical settings, assisting clinicians in providing a personalized consultation to each patient.

In this study, internal validation was not presented as the main study findings, as the subsetted validation groups sometimes had small sample sizes resulting in less precise analyses, and also reduced the number of patients available to develop the model in the first place. However, when the models were redeveloped using further subsetted development groups, internal validation still showed adequate model fit and AUROCs that were close to those developed using all patients. More important, however, further external validation of both models is needed in the future.

Other studies have proposed cutoff points of serum HBsAg or HBV DNA levels for the prediction of various outcomes. However, cutoff points have varied. It is thus important to note that the probability of outcomes such as HBeAg seroclearance or HBV DNA undetectability is a continuum that varies with different host and virus factors present in each patient. Therefore, a reverse biological gradient should be emphasized; lowering HBV DNA confers higher rates of HBeAg seroclearance, and lowering HBsAg levels confers higher rates of HBV DNA undetectability. While predictive accuracy was relatively high for HBeAg seroclearance, the AUROCs were moderately high at 0.77 and 0.73 for 5- and 10-year prediction of HBV DNA undetectability. There may be missing predictive factors that have yet to be used. Future studies should address the possible role of host immune and genetic factors, which may also play an important role and contribute to increasing the predictive accuracy of not only HBV DNA undetectability, but HBeAg seroclearance as well. With higher predictive accuracies, score-based clinical tools can be modified to better help clinicians predict these outcomes.

There are some limitations to be noted. This study only included baseline serum HBsAg levels, so the long-term changes in serum HBsAg levels and their effects on HBV DNA undetectability must be further elucidated with repeated measurements. Participants in this study were previously untreated chronic HBV carriers. Therefore, the applicability of these results to patients undergoing treatment is unclear, and should be clarified. Lastly, participants in this study were between 30 and 65 years old and infected exclusively with genotype B and/or C. These results should be validated in younger carriers, or in other genotypes.

In conclusion, HBV DNA levels are the strongest determinant of HBeAg seroclearance, and should be closely monitored in HBeAg-seropositive patients. In HBeAg-seronegative patients, serum HBsAg levels are strong determinants of HBV DNA undetectability, and monitoring of HBsAg levels is a clinically important procedure among chronically infected HBeAg-seronegative genotype B and/or C carriers.

Acknowledgment

Other members of R.E.V.E.A.L.-HBV Study Group are as follows: National Taiwan University Hospital: C. Y. Hsieh, H.S. Lee, P. M. Yang, C. H. Chen, J. D. Chen, S. P. Huang, C. F. Jan. National Taiwan University: T. H. H. Chen. National Defense Medical Center: C. A. Sun. Taipei City Psychiatric Center: M. H. Wu. Tzu Chi University: S. Y. Chen. Shin Kong Wu Ho-Su Memorial Hospital: K. E. Chu. Huhsi Health Center, Penghu County: S. C. Ho, T. G. Lu. Provincial Penghu Hospital:W. P.Wu, T. Y. Ou. Sanchi Health Center, Taipei County: C. G. Lin. Provincial Chutung Hospital: K. C. Shih. Provincial Potzu Hospital: W. S. Chung, C. Li. Kaohsu Health Center, Pingtung County: C. C. Chen. Paihsa Health Center, Penghu County: W. C. How.

Author Contributions

Dr. Chien-Jen Chen had full access to all of the data in the study and takes responsibility for the integrity of the data as well as the accuracy of the data analysis. Study concept and design: C.-J.C., J.L., H.-I.Y. Acquisition of data: J.L., H.-I.Y., M.-H.L., C.-L.J., S.-N.L., L.-Y.W., S.-L.Y., C.K.C, C.-J.C. Analysis and interpretation of data: J.L., H.-I.Y., C.-J.C. Drafting of the article: J.L. Critical revision of the article for important intellectual content: J.L., H.-I.Y., M.-H.L., R.B.-U., C.-L.J., S.-N.L., L.-Y.W., S.-L.Y., U.H.I., C.K.H., C.-J.C. Obtained funding: C.-J.C. Administrative, technical, or material support: J.L., H.-I.Y., M.-H.L., R.B.-U., C.-L.J., S.-N.L., L.-Y.W., S.-L.Y., U.H.I., C.K.H., C.-J.C. Study supervision: C.-J.C.

Ancillary