Incidence, Factors, and Patient-Level Data for Spontaneous HBsAg Seroclearance: A Cohort Study of 11,264 Patients

INTRODUCTION: Spontaneous hepatitis B surface antigen (HBsAg) seroclearance, the functional cure of hepatitis B infection, occurs rarely. Prior original studies are limited by insufficient sample size and/or follow-up, and recent meta-analyses are limited by inclusion of only study-level data and lack of adjustment for confounders to investigate HBsAg seroclearance rates in most relevant subgroups. Using a cohort with detailed individual patient data, we estimated spontaneous HBsAg seroclearance rates through patient and virologic characteristics. METHODS: We analyzed 11,264 untreated patients with chronic hepatitis B with serial HBsAg data from 4 North American and 8 Asian Pacific centers, with 1,393 patients with HBsAg seroclearance (≥2 undetectable HBsAg ≥6 months apart) during 106,192 person-years. The annual seroclearance rate with detailed categorization by infection phase, further stratified by hepatitis B e antigen (HBeAg) status, sex, age, and quantitative HBsAg (qHBsAg), was performed. RESULTS: The annual seroclearance rate was 1.31% (95% confidence interval: 1.25–1.38) and over 7% in immune inactive patients aged ≥55 years and with qHBsAg <100 IU/mL. The 5-, 10-, 15-, and 20-year cumulative rates were 4.74%, 10.72%, 18.80%, and 24.79%, respectively. On multivariable analysis, male (adjusted hazard ratio [aHR] = 1.66), older age (41–55 years: aHR = 1.16; >55 years: aHR = 1.21), negative HBeAg (aHR = 6.34), and genotype C (aHR = 1.82) predicted higher seroclearance rates, as did lower hepatitis B virus DNA and lower qHBsAg (P < 0.05 for all), and inactive carrier state. DISCUSSION: The spontaneous annual HBsAg seroclearance rate was 1.31%, but varied from close to zero to about 5% among most chronic hepatitis B subgroups, with older, male, HBeAg-negative, and genotype C patients with lower alanine aminotransferase and hepatitis B virus DNA, and qHBsAg independently associated with higher rates (see Visual Abstract, Supplementary Digital Content 2, http://links.lww.com/CTG/A367).


INTRODUCTION
Chronic hepatitis B (CHB) affects about 257 million people worldwide (1) and can lead to cirrhosis and hepatocellular carcinoma (HCC) (2,3). Persons who have achieved hepatitis B surface antigen (HBsAg) seroclearance, the functional cure, had a significantly lower risk of HCC than those without, even in the setting of established cirrhosis (4)(5)(6). However, only a small proportion of patients will eventually experience HBsAg seroclearance spontaneously or with currently available antiviral therapies (7)(8)(9)(10)(11)(12). Thus, major efforts are underway to identify new strategies and drugs that can yield higher seroclearance rates.
As the rate of spontaneous HBsAg seroclearance can vary by host and viral factors (13), it is important to inform stakeholders of detailed subgroup data for the plan and design of therapeutic trials of drugs that aim to significantly increase the rate of HBsAg seroclearance compared with the expected spontaneous rate. Because of the low rate, previous studies are limited by inadequate sample sizes and follow-up periods. In 2 recent meta-analyses, we and others have provided pooled estimates for the HBsAg seroclearance rates (12,13). However, the results, especially for subgroup analyses, were limited by a number of factors. First, there was severe heterogeneity among the studies, and adequate adjustment for confounders was not possible without individual patient-level data. Second, because of a lack of primary data, subgroup analyses were not possible for several clinically relevant variables, such as age and infection phase. Third, without individual patient data, subgroup analyses were possible for only 1 patient or 1 viral factor at a time, such as male vs female or high vs low hepatitis B virus (HBV) DNA. Therefore, to fill in current knowledge gaps, we used data from a large collaborative cohort of never-treated patients not only to determine the annual and cumulative incidence rates of spontaneous HBsAg seroclearance but also to investigate independent host and viral factors associated with spontaneous HBsAg seroclearance by detailed subgroup analyses and multivariable regressions.

Data source
The current study cohort comprised patients from 4 North American and 8 Asian Pacific study centers. Data were collected using a standardized data collection form with unified variable definition at each center. Deidentified data were transmitted to the data monitoring and analysis center at Stanford University. The study was approved by the Institutional Review Board at Stanford University and at each center.

Inclusion and exclusion criteria
The study included patients with CHB (18 years or older at baseline, not receiving HBV treatment at baseline and throughout study follow-up, and HBsAg and/or HBV DNA positive for at least 6 months), who were enrolled at each study center with serial laboratory data (including HBsAg, quantitative HBsAg [qHBsAg], hepatitis B e antigen [HBeAg], liver chemistries, and HBV DNA).
We excluded patients with any of the following: (i) hepatitis C virus, hepatitis D virus, or HIV coinfection, (ii) prior HCC, (iii) medical history of solid organ or bone marrow transplantation, immunosuppression including chemotherapy and biologics, (iv) other known liver diseases (autoimmune hepatitis, hemochromatosis, Wilson disease, and alpha-1 antitrypsin deficiency), and (v) known malignancy within 5 years of study entry date.

Study variable definitions
Infection phase was defined as per the American Association for the Study of Liver Diseases Guidance (see Table 1, Supplementary Digital Content 1, http://links.lww.com/CTG/A366) (14). The primary outcome of this study was confirmed spontaneous HBsAg seroclearance, defined as having $2 negative HBsAg results $6 months apart in the absence of any antiviral treatment. The secondary outcome was to examine the determinants of spontaneous HBsAg seroclearance. Follow-up was from the first date patients were evaluated at the study centers for CHB. Patients were censored once they started treatment, with HBsAg seroclearance, death, or the end of follow-up as the study end point, whichever came first.

Statistical analysis
We estimated the follow-up period in person-years by multiplying the duration of follow-up by the number of patients. We calculated the annual HBsAg seroclearance rate by dividing the number of confirmed HBsAg seroclearance events by the total follow-up period. Kaplan-Meier analysis was used to estimate the 5-, 10-, 15-, and 20-year cumulative HBsAg seroclearance rates. Log-rank tests were used to determine differences in cumulative incidence rates. To avert bias caused by the small sample size, we omitted the data from subgroups with ,2 events or ,20 patients. In addition, we performed separate subgroup analyses for genotype B and C patients with further stratification by age, sex, and qHBsAg.
Given that the observations were derived in clusters from each study site, we used Cox models with frailty to evaluate the HBsAg seroclearance rate (15) to allow us to model the heterogeneity across the cohorts and to account for the potential random effects at study level. Thus, a mixed-effects model was adopted, with random effects set for cohorts and fixed effects for individual patients' characteristics. We performed multivariable Cox regression to adjust for clinically relevant covariates. Variables with .10% missing data were not included in the main model. The assumption of the Cox regression was tested. For variables with violation, we added a time interaction of that variable into the  model. We also assessed goodness of fit of the model by estimating the Harrell C statistics and determined the E values of variables to assess the magnitude that an unmeasured confounder would have to explain the association between the corresponding variable and spontaneous HBsAg seroclearance (16). Two-sided P values were calculated, with the threshold of significance set at 0.05. All statistical analyses were performed using R software (3.5.1).

RESULTS
The study analysis included 11,264 eligible patients. The study cohort was predominately male (60.1%), Asian/Pacific Islanders/ Polynesians (98.1%), noncirrhotic (97.0%), aged 55 years or younger (82.8%), and HBeAg negative (82.0%) at baseline (Table 1). Patients seen at health care centers accounted for approximately two-thirds of the study population. Patients who achieved HBsAg seroclearance during follow-up period were compared with those remained positive for HBsAg.
To further assess the association between infection phase and spontaneous HBsAg seroclearance, we adjusted for sex, age, ethnicity, and study setting and showed that immune active, eAg1, immune active, eAg2, and immune tolerant were significantly associated with lower probability of spontaneous HBsAg seroclearance (see Table 8, Supplementary Digital Content 1, http://links.lww.com/CTG/A366).

DISCUSSION
In this large, multinational cohort study that included 11,264 untreated patients with CHB, the annual rate of spontaneous HBsAg seroclearance was 1.31%. However, both males and females aged $55 years with immune inactive phase and qHBsAg ,100 IU/mL presented an annual incidence rate higher than 7%. Overall, the 10-and 20-year cumulative incidence rates were 10.72% and 24.79%. On multivariable Cox regression, independent predictors of spontaneous seroclearance were male, 40 years or older, HBeAg negative, has genotype C infection compared with genotype B, and low levels of HBV DNA and qHBsAg. Compared with immune-inactive patients, immune-active (HBeAg-negative), immune-active (HBeAg-positive), and immune-tolerant patients were about 45%, 75%, and 90% less likely to experience spontaneous HBsAg seroclearance, respectively.
Our findings are unique for our large and diverse study population and the long duration of follow-up. Our recent metaanalysis of HBsAg seroclearance in both treated and untreated patients reported that the pooled annual rate of HBsAg seroclearance was 1.02%, and the cumulative incidence rates were 4.03% at 5 years, 8.16% at 10 years, and 17.99% at 15 years (13). Categorized by treatment history, the annual incidence rates in treated and untreated patients were 0.82% and 1.31%. In this study, using individual patient data of untreated patients with HBV, we were able to fill in these gaps and provide data on spontaneous HBsAg seroclearance rates for a comprehensive range of relevant clinical subgroups after controlling for key confounders. As a result, although there was no difference in sex and HBV genotype in our earlier meta-analysis, males and genotype C patients were found in the current study to be 70% and 80% more likely to achieve spontaneous HBsAg seroclearance compared with females and genotype B patients, respectively. On the other hand, while consistent with finding in our recent metaanalysis, which reported higher chance of spontaneous HBsAg seroclearance for HBeAg-negative compared to HBeAg-positive patients, the adjusted hazard ratio for spontaneous HBsAg seroclearance in the current study for HBeAg-negative patients was 6.34 (compared with only 1.74 times from unadjusted metaanalytic results of mixed treated and untreated patients). The current study also showed that the immune active phase was associated with lower HBsAg seroclearance compared with the  immune inactive phase, and this is likely the reason for the lack of difference in HBsAg seroclearance rates between treated and untreated patients seen in the recent meta-analysis because treated patients were most likely immune-active patients, and untreated patients were more likely to be immune inactive.
The finding that HBV viral markers, such as lower HBV DNA and qHBsAg, confer a higher probability of spontaneous HBsAg seroclearance in a dose-dependent manner was supported by previous literature (10,17,18). This association was also manifested in our subgroup analysis stratified by infection phase, sex, age, qHBsAg level, and multivariable Cox regression.
Observations regarding the association between genotype and HBsAg seroclearance have been conflicting. One prior case-control study showed that patients with CHB with genotype B infection were more likely to achieve HBsAg seroclearance than those with genotype C infection (19). However, that study included less than 50 patients in each genotype subgroup. A larger study with longitudinal follow-up indicated a nonsignificant association between genotype and HBsAg seroclearance in patients with negative HBeAg (20). Another cohort study by Tseng et al. showed that HBeAg-negative patients with genotype C patients had a higher lifetime cumulative incidence of HBsAg seroclearance (21). In the current study with 3,934 patients with genotype B infection and 2,025 with genotype C, after adjusting for relevant demographic and clinical variables, genotype C was independently and significantly associated with a higher probability of achieving a spontaneous HBsAg seroclearance rate than genotype B. Previous report has indicated differences in geographic distribution and clinical outcomes of genotype C subtype B1 and B2 (formerly Ba and Bj) (22). B1 subtype was associated with a higher rate of basal core promoter mutation and a higher risk of developing advanced liver disease at younger age (22,23). However, we did not have subgenotype data to include in this study.
This study has several strengths. To our knowledge, this is the largest study on the incidence rates of spontaneous HBsAg seroclearance using individual patient-level data with longterm follow-up that also included both health care center and population-based data and from both North America and Asia Pacific. The large sample size and comprehensive set of variables enabled detailed subgroup analyses providing HBsAg seroclearance data for a whole host of specific relevant subgroups.
This study also has limitations. First, there were missing data on variables such as quantitative HBsAg and HBV DNA, precluding the inclusion of these variables into the main multivariable model. However, we did evaluate additional models using the same variables as in the main predictor model, plus these variables in the subsets of patients with available data (8,273 patients for HBV DNA and 7,011 for qHBsAg). Second, the study population may not represent the general HBV population, as two-thirds of the patients were from health care centers. To overcome this limitation, we performed separate analysis for health care-based and population-based cohorts, which showed higher rates of HBsAg seroclearance in the latter. In addition, our findings may not be generalizable to patients infected with genotypes other than B and C due to the small sample size. This study also included very few non-Asian/Pacific Islanders/Polynesians (1.9%). However, the majority of CHB cases in the world are Asian/Pacific Islanders/Polynesians. Third, the number of patients with immune-active HBeAgpositive CHB and immune-tolerant CHB was small, limiting further subgroup analyses in these 2 subgroups. Fourth, the seroclearance rates derived from certain subgroup such as immune inactive carriers may not be helpful in planning trials, which aim at patients with untreated immune active disease. However, the findings are helpful in understanding the natural history of HBV infection. Finally, follow-up testing after confirmation of HBsAg seroclearance was limited, and we could not calculate the rate of HBsAg seroreversion. Thus, we could not address the durability of HBsAg seroclearance.
Using patient-level data from a large cohort of 11,264 untreated patients with CHB from 12 study sites in 7 countries/ regions, we found that the overall annual spontaneous HBsAg seroclearance rate was low at 1.31%. Over 20 years, the cumulative HBsAg seroclearance rate increased to 25%, with rates ranging from 3.01% to 60.25% in different subgroups, highlighting the need to consider patient demographics and clinical phenotype in the management of patients with CHB and in planning future HBV cure clinical trials.

Study Highlights WHAT IS KNOWN
3 Spontaneous HBsAg seroclearance is a key milestone of the natural history of HBV infection.
3 Previous studies are limited by the insufficient sample size and/or follow-up, and recent meta-analyses are limited by study-level data and lack of adjustment for confounders.