Differential Hepatitis B Surface Antigen Glycan Isomer Kinetics Between Tenofovir Alafenamide and Entecavir: Hepatocellular Carcinoma Risk Stratification in Hepatitis B e Antigen‐Positive Patients

ABSTRACT

Nucleos(t)ide analog (NA) therapy suppresses hepatitis B virus (HBV) DNA, but residual hepatocellular carcinoma (HCC) risk persists. O‐glycosylated hepatitis B surface antigen glycan isomer (HBsAgGi) uniquely reflects virion burden. We evaluated the impact of tenofovir alafenamide (TAF) versus entecavir (ETV) on 48‐week HBsAgGi kinetics and its utility in stratifying HCC risk in 90 NA‐naïve patients (ETV: 73; TAF: 17). A favourable response was defined as a reduction or maintenance of low HBsAgGi levels. TAF independently predicted a favourable response compared with ETV (penalised odds ratio 3.60, p = 0.030). A significant interaction occurred between HBeAg status and HBsAgGi response in relation to HCC development (p = 0.038). In HBeAg‐positive patients, a poor response was associated with increased HCC risk (hazard ratio 7.34, 95% confidence interval [CI] 1.96–20.35, p = 0.001); however, the exact magnitude warrants cautious interpretation due to wide CIs. This association was absent in HBeAg‐negative cases. Adding HBsAgGi response to the aMAP score improved long‐term HCC prediction exclusively in the HBeAg‐positive cohort. As the assay targets genotype C‐specific O‐glycosylation, the findings cannot be generalised to genotypes A, B, or D. In conclusion, on‐treatment HBsAgGi response is a specific surrogate marker for stratifying residual HCC risk in HBeAg‐positive patients, and TAF is more effective than ETV in inducing these favourable kinetics.

Abbreviations

ALT

alanine aminotransferase

aMAP

age‐male‐ALBI‐platelets

AUC

area under the curve

CHB

chronic hepatitis B

CI

confidence interval

DCA

decision curve analysis

ETV

entecavir

FIB‐4

fibrosis‐4

HBcrAg

hepatitis B core‐related antigen

HBeAg

hepatitis B e antigen

HBsAgGi

hepatitis B surface antigen glycan isomer

HBV

hepatitis B virus

HCC

hepatocellular carcinoma

HR

hazard ratio

IDI

integrated discrimination improvement

IPTW

inverse probability of treatment weighting

LLOQ

lower limit of quantification

NA

nucleos(t)ide analog

NRI

net reclassification improvement

OR

odds ratio

qHBsAg

quantitative hepatitis B surface antigen

ROC

receiver operating characteristic

SMD

standardised mean difference

SVP

subviral particles

TAF

tenofovir alafenamide

Nucleos(t)ide analog (NA) therapy has revolutionised the management of chronic hepatitis B (CHB) by potently suppressing hepatitis B virus (HBV) DNA replication, improving liver histology, and significantly improving patient prognosis [1, 2, 3]. Entecavir (ETV) and tenofovir alafenamide (TAF) are the first‐line NAs recommended by major guidelines because of their high potency and high resistance barrier [4]. However, despite effective viral suppression, hepatocellular carcinoma (HCC) risk is not fully eliminated.

During NA therapy, serum HBV DNA is rapidly suppressed below the limit of quantification in most patients, losing its utility as a dynamic indicator of residual viral activity [4, 5]. Conversely, hepatitis B surface antigen (HBsAg) is continuously produced from integrated HBV DNA and covalently closed circular DNA (cccDNA) [6, 7]. However, most circulating HBsAg consists of non‐infectious subviral particles (SVPs) produced in vast excess relative to virions [8]. Because these SVPs show minimal treatment‐related changes [9], they obscure true virion kinetics in conventional assays. Therefore, a novel surrogate marker that more specifically reflects virion dynamics and residual HCC risk is urgently needed.

The PreS2 domain of middle HBsAg (M‐HBsAg) is known to be O‐glycosylated specifically in genotype C HBV virions, structurally distinguishing them from SVPs [10, 11]. Quantification of this O‐glycosylated HBsAg glycan isomer (HBsAgGi) enables assessment of viral kinetics distinct from SVP‐derived signals measured by conventional quantitative HBsAg (qHBsAg) assays. Previous studies have demonstrated the clinical utility of HBsAgGi, showing that serum HBsAgGi levels reflect HBV DNA and RNA virions and that baseline HBsAgGi is a strong independent predictor of HBsAg seroclearance in untreated patients [12, 13]. These findings suggest that HBsAgGi kinetics may better reflect the ‘quality’ of viral suppression than conventional markers.

Although both TAF and ETV are potent antiviral agents, their differential effects on HBsAgGi kinetics remain unknown. While HBsAgGi has been reported to decline during NA therapy [12], it is unclear whether NA choice influences this decline and whether early HBsAgGi response correlates with long‐term clinical outcomes, particularly HCC development. Given ongoing debate regarding the potential superiority of TAF over ETV in reducing HCC risk [14, 15], evaluating these drugs using this novel virion‐specific marker may provide new insights into optimising therapy.

This study aimed to evaluate the clinical utility of the 48‐week HBsAgGi response as a prognostic biomarker for identifying residual HCC risk. Secondarily, we compared the impact of TAF versus ETV on these kinetics.

1. Methods

This observational cohort study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Document S1).

1.1. Study Design and Patients

Between May 2005 and October 2018, 1042 patients with CHB presenting at Juntendo University Shizuoka Hospital (Japan) were screened for this retrospective study. Inclusion criteria were: (1) HBsAg positivity for > 6 months; (2) infection with HBV genotype C; (3) treatment‐naïve status for antiviral therapy; (4) availability of serum samples at baseline and week 48; and (5) initiation of antiviral therapy with either ETV (0.5 mg/day) or TAF (25 mg/day). The exclusion criteria were: (1) coinfection with human immunodeficiency virus or hepatitis C virus, (2) history or evidence of other chronic liver diseases (e.g., autoimmune hepatitis, primary biliary cholangitis, or hemochromatosis), (3) history of liver transplantation, (4) prophylactic NA therapy to prevent HBV reactivation, and (5) switching of NA agents during the observation period. Based on these criteria, 90 patients were included. The cohort consisted of 73 patients treated with ETV and 17 patients treated with TAF. The study protocol was approved by the ethics committee of the institution (E25‐0553) and adhered to the Declaration of Helsinki.

1.2. Laboratory Investigations

Baseline and week 48 laboratory data were collected. Serum HBV DNA levels were determined using a COBAS 6800/8800 system (Roche Diagnostics, Branchburg, NJ, USA), with a dynamic range of 1.0–9.0 log IU/mL. Serum HBsAg and hepatitis B core‐related antigen (HBcrAg; dynamic range, 2.1–7.1 log U/mL) levels were quantified using commercial chemiluminescent enzyme immunoassay kits. Serum hepatitis B e antigen (HBeAg) was detected using a commercial chemiluminescent enzyme immunoassay kit, and HBeAg positivity was defined as a level ≥ 1.0 cutoff index.

Serum HBsAgGi levels were measured in stored baseline and week 48 samples (−20°C) using a commercial ELISA kit (RCMG Inc., Tokyo, Japan), as previously described [13]. This assay utilises a monoclonal antibody targeting the O‐glycosylated PreS2 domain of middle HBsAg (M‐HBsAg), specific to HBV virions. Values were expressed as log ng/mL, with a lower limit of quantification (LLOQ) of 2.8 log ng/mL. The aMAP (age‐male‐ALBI‐platelets) score, a validated clinical indicator of HCC risk, was calculated from baseline age, sex, albumin‐bilirubin score, and platelet count. Patients were stratified into low (< 50), medium (50–60), and high (> 60) risk groups as previously described [16].

HBsAgGi levels were categorised as low (< 3.5 log ng/mL), middle (3.5–4.2 log ng/mL), or high (> 4.2 log ng/mL). The cutoff for the low category was determined based on the previous report, which identified < 3.5 log ng/mL as a favourable outcome [13]. The cutoff for the high category was defined as values in the top quartile of baseline HBsAgGi levels in the present cohort.

1.3. Definitions and Outcomes

The primary outcome was the HBsAgGi response at week 48. Given that the stable maintenance of low virion activity is considered as clinically important as achieving a categorical decline, a ‘Favourable response’ was defined as either a reduction in the HBsAgGi category (e.g., High to Middle, Middle to Low) or the maintenance of the ‘Low’ category at week 48. Conversely, a ‘Poor response’ was defined as either an increase in the HBsAgGi category or sustained ‘Middle’ or ‘High’ levels. The secondary outcome was the development of HCC. Surveillance for HCC was performed using ultrasonography, computed tomography, or magnetic resonance imaging every 3–6 months.

1.4. Statistical Analysis

For statistical analysis, values below the LLOQ or above the upper limit of quantification were treated as the corresponding limit values, unless otherwise specified. Continuous variables were compared using the Mann–Whitney U test, and categorical variables were compared using Fisher's exact test or the chi‐square test. Factors associated with favourable HBsAgGi response were analysed using univariate and multivariate logistic regression. To avoid overfitting due to the small sample size and limited events, particularly in the TAF group, Firth's penalised maximum likelihood logistic regression was used. Separate multivariate models were constructed. To reduce baseline imbalance between groups, inverse probability of treatment weighting (IPTW) with stabilised weights was performed. Propensity scores were estimated using age, baseline HBV DNA, qHBsAg, HBcrAg, and HBsAgGi. Covariate balance was assessed using standardised mean differences (SMD), with < 0.20 considered acceptable. The impact of TAF was re‐evaluated in the IPTW‐adjusted cohort.

HCC incidence was estimated using the Kaplan–Meier method and compared using the log‐rank test. To minimise immortal time bias, a landmark analysis was performed with a landmark time of week 48 and a 1‐year lag period; patients who developed HCC or were censored within this period were excluded from the analysis. Risk factors for HCC were assessed using Cox proportional hazards models. To address parameter instability and monotone likelihood issues due to the absence of HCC events in specific subgroups, Firth's penalised Cox regression was utilised for interaction and stratified analyses. Interaction robustness was validated using 1000 bootstrap resamples. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated. For the sensitivity analysis, the favourable response was stratified into ‘Maintained’ (persisting a low‐level HBsAgGi) and ‘Improved’ (achieving a reduction in the risk category).

Predictive performance of combining HBsAgGi response with the aMAP score was assessed using time‐dependent receiver operating characteristic (ROC) curves and the area under the curve (AUC). Incremental value was evaluated using integrated discrimination improvement (IDI) and net reclassification improvement (NRI). Decision curve analysis (DCA) assessed clinical net benefit versus the aMAP score. Given the limited sample size (n = 75), internal validation was performed using 1000 bootstrap resamples to estimate the 95% CIs for IDI and NRI, and to derive the median net benefit for DCA, ensuring robustness.

All statistical analyses were performed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA) and R software version 4.5.2. A two‐sided p‐value < 0.05 was considered statistically significant.

2. Results

2.1. Baseline Characteristics of the Study Population

Baseline characteristics of the 90 NA‐treated patients are summarised in Table 1. The cohort included 73 patients treated with ETV and 17 treated with TAF. Patients in the TAF group were significantly younger (median 41 vs. 60 years, p = 0.007) and had higher baseline HBV DNA, qHBsAg, and HBcrAg levels than those in the ETV group. Baseline HBsAgGi levels were also significantly higher in the TAF group (p = 0.024). The distribution of aMAP risk groups differed significantly between the two treatments (p = 0.035). No significant differences were observed in sex, HBeAg positivity, ALT levels, platelet counts, liver fibrosis markers (FIB‐4 index), or cirrhosis prevalence between the two groups.

TABLE 1.

Baseline characteristics of the study population.

Variables	All patients	ETV‐treated	TAF‐treated	p a (ETV vs. TAF)
	(n = 90)	(n = 73)	(n = 17)
Demographics
Age, years	58 (41–66)	60 (45–66)	41 (35–52)	0.007
Male sex, n (%)	58 (64.4)	48 (65.7)	10 (58.8)	0.591
Virological characteristics
HBeAg‐positive, n (%)	42 (46.6)	31 (42.5)	11 (64.7)	0.098
HBV DNA, log IU/mL	6.2 (3.5–7.6)	5.9 (3.3–7.2)	7.3 (5.5–8.4)	0.007
qHBsAg, log IU/mL	3.5 (3.1–4.0)	3.4 (3.1–3.8)	4.1 (3.2–4.7)	0.030
HBcrAg, log U/mL	6.2 (3.9–7.0)	5.6 (3.6–7.0)	7.0 (6.1–7.0)	0.010
HBsAgGi, log ng/mL	3.7 (3.3–4.2)	3.7 (3.3–4.0)	4.1 (3.5–5.5)	0.024
Clinical characteristics
ALT, U/L	47 (31–81)	46 (31–79)	47 (27–88)	0.967
Platelet count, ×104/μL	17.0 (11.0–21.8)	16.9 (11.2–21.9)	17.2 (8.8–22.3)	0.847
FIB‐4 index	2.21 (1.30–4.29)	2.30 (1.34–4.31)	1.74 (1.07–4.43)	0.389
aMAP score	57 (48–66)	58 (51–66)	49 (44–64)	0.188
aMAP risk group (Low/Medium/High)	26/35 / 29	17/32 / 24	9/3 / 5	0.035
Cirrhosis	19 (21.1)	14 (19.2)	5 (29.4)	0.342
Observation period
Follow‐up from baseline, year	8.4 (3.5–14.6)	11.4 (5.3–15.3)	3.5 (1.9–6.7)	< 0.001

Note: Data are presented as median (interquartile range) or number (percentage). Bold values indicate statistical significance (p < 0.05).

Abbreviations: ALT, alanine aminotransferase; aMAP, age‐male‐ALBI‐platelet; ETV, entecavir; FIB‐4, fibrosis‐4; HBcrAg, hepatitis B core‐related antigen; HBeAg, hepatitis B e antigen; HBsAgGi, hepatitis B surface antigen glycan isomer; HBV, hepatitis B virus; qHBsAg, quantitative hepatitis B surface antigen; TAF, tenofovir alafenamide.

^a Comparisons were performed between the ETV and TAF groups using the Mann–Whitney U test for continuous variables and the chi‐square test or Fisher's exact test for categorical variables.

2.2. Dynamics of HBsAgGi and the Impact of TAF Versus ETV

Among the 90 NA‐treated patients, the median HBsAgGi level was 3.7 (interquartile range [IQR], 3.3–4.2) log ng/mL at baseline and 3.7 (IQR, 3.5–3.9) log ng/mL at week 48, with no statistically significant difference (p = 0.308). Most patients (96.7%, 87/90) achieved or maintained HBV DNA suppression (< 3.3 log IU/mL) at week 48. However, as illustrated in the waterfall plot (Figure 1A), the quantitative changes in HBsAgGi from baseline to week 48 (ΔHBsAgGi) were highly heterogeneous. Although many patients exhibited a decline in HBsAgGi levels, these levels paradoxically increased in a subset of patients despite successful viral DNA suppression. Between groups, the magnitude of these quantitative changes differed significantly. The TAF group exhibited significantly greater reductions in both HBV DNA (p = 0.012) and HBsAgGi (ΔHBsAgGi, p = 0.009; Figure 1A, inset) than the ETV group. In contrast, the reductions in qHBsAg and HBcrAg did not differ significantly between the treatments (p = 0.057 and p = 0.193, respectively) (Table S1). Consistent with these findings, category shifts in HBsAgGi differed between treatments (Figure 1B). Notably, HBsAgGi levels paradoxically increased—shifting upward from the Low to the Middle category—in a substantial proportion of patients in the ETV group. Conversely, TAF treatment effectively prevented such upward shifts, with patients predominantly maintaining the Low category or exhibiting downward category shifts. Consequently, the TAF group achieved a significantly higher ‘Favourable response’ rate at week 48 compared to the ETV group (64.7% vs. 31.5%, p = 0.024).

FIGURE 1.

Differential dynamics of HBsAgGi levels between entecavir and tenofovir alafenamide treatments. (A) Individual changes in HBsAgGi levels from baseline to week 48 (ΔHBsAgGi). Blue and red bars in the waterfall plot indicate patients treated with ETV (n = 73) and TAF (n = 17), respectively. The inset box plot shows the distribution of ΔHBsAgGi between the two groups. p‐values were calculated using the Mann–Whitney U test. (B) Sankey diagrams illustrating categorical shifts in HBsAgGi levels over 48 weeks. Categories are defined as Low (< 3.5 log ng/mL), High (top quartile of baseline), and Middle. ETV, entecavir; HBsAgGi, hepatitis B surface antigen glycan isomer; TAF, tenofovir alafenamide.

2.3. Factors Associated With HBsAgGi Reduction and Favourable Response

Initial exploratory analysis using a correlation heatmap revealed that the change in HBsAgGi (ΔHBsAgGi) correlated with several baseline factors (Figure S1). To rigorously identify the independent predictors of quantitative HBsAgGi reduction, a multivariate linear regression analysis was performed (Table S2). After adjusting for age and baseline HBsAgGi levels (Model 1), TAF treatment (p = 0.041) and higher baseline HBsAgGi (p < 0.001) were independently associated with a greater reduction in HBsAgGi. Notably, further adjustment (Model 2) confirmed that baseline HBV DNA, qHBsAg, and HBeAg status did not independently affect ΔHBsAgGi, indicating that HBsAgGi kinetics are distinct from conventional viral markers.

We further evaluated the predictors of achieving a qualitative ‘Favourable response’ at week 48 using logistic regression analysis (Table 2). In the univariate analysis, TAF treatment and high baseline HBsAgGi levels were significantly associated with a favourable response. In the multivariate analysis, TAF treatment remained a strong and independent predictor of a favourable HBsAgGi response in both Model 1 (adjusted for baseline HBsAgGi and cirrhosis; penalised odds ratio [OR] 3.60, 95% CI 1.13–12.57, p = 0.030) and Model 2 (adjusted for age and cirrhosis; penalised OR 3.89, 95% CI 1.20–13.87, p = 0.023) (Table 2).

TABLE 2.

Factors associated with favourable HBsAgGi response at week 48.

Variables	Favourable response	Poor response	Univariate analysis		Multivariate analysis a
	n = 34	n = 56			Model 1 (HBsAgGi)		Model 2 (Age)
			OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
Age, per year	47 (39–64)	60 (45–66)	0.97 (0.94–1.00)	0.049	—	—	0.99 (0.95–1.02)	0.446
Male sex	20	38	0.68 (0.28–1.63)	0.385	—	—	—
Cirrhosis	4	15	0.40 (0.11–1.17)	0.096	0.33 (0.08–1.08)	0.068	0.35 (0.09–1.16)	0.088
HBeAg‐positive	19	23	1.79 (0.77–4.25)	0.176	—	—	—
HBsAgGi, log ng/mL	3.8 (3.4–5.3)	3.7 (3.3–3.8)	1.83 (1.09–3.18)	0.022	1.49 (0.86–2.63)	0.158	—	—
TAF treatment	11	6	3.80 (1.32–11.79)	0.013	3.60 (1.13–12.57)	0.030	3.89 (1.20–13.87)	0.023

Note: Bold values indicate statistical significance (p < 0.05).

Abbreviations: CI, confidence interval; ETV, entecavir; HBeAg, hepatitis B e antigen; HBsAgGi, hepatitis B surface antigen glycan isomer; OR, odds ratio; TAF, tenofovir alafenamide.

^a To account for the limited number of events and the small sample size of the TAF group, and to prevent overfitting and parameter instability, both univariate and multivariate analyses were performed using Firth's penalised maximum likelihood logistic regression. A favourable response was defined as a reduction in the HBsAgGi category (e.g., High to Middle, Middle to Low) or the maintenance of the ‘Low’ category at week 48.

To further validate these findings against the baseline imbalances between the two treatment groups, an IPTW analysis was performed. In this weighted pseudo‐population, TAF treatment remained significantly associated with a higher favourable response rate (IPTW‐adjusted OR 4.97, 95% CI 1.06–23.25, p = 0.041; Table S3), confirming the robustness of this categorical outcome.

2.4. Association Between Favourable HBsAgGi Respone and HCC Development

The potential of the HBsAgGi response at week 48 to predict the long‐term risk of HCC was evaluated using a landmark analysis with a 1‐year lag time to mitigate immortal time bias. After excluding patients who developed HCC or were censored before the end of this lag period, 75 patients from the NA‐treated cohort were eligible for the survival analysis. During the observation period from the landmark time point (median 8.0 years, IQR 4.6–11.5), a total of 12 patients developed HCC. Crucially, a formal interaction test revealed that the prognostic impact of the 48‐week HBsAgGi response on HCC development was significantly modified by the patient's baseline HBeAg status (p for interaction =0.038, Table 3). Consequently, survival analyses were stratified into HBeAg‐positive and HBeAg‐negative cohorts. In HBeAg‐positive patients, the HBsAgGi response powerfully stratified the long‐term HCC risk. The 10‐year cumulative incidence rate of HCC was markedly higher in the ‘Poor response’ group compared to the ‘Favourable response’ group (32.1% vs. 0.0%, p < 0.001; Figure 2, left panel). In the multivariate Cox proportional hazards model adjusted for the aMAP score risk group, a Poor response was identified as a strong and independent predictor of HCC development in this population (HR 7.34, 95% CI 1.96–20.35, p = 0.001; Table 3). In contrast, among HBeAg‐negative patients, the HBsAgGi response was not significantly associated with the incidence of HCC. The Kaplan–Meier curves for the Favourable and Poor response groups showed no significant separation (Figure 2, right panel), and the multivariate analysis confirmed no significant increase in HCC risk for the Poor response group (HR 0.49, 95% CI 0.06–2.72, p = 0.440; Table 3). These highly stratified results indicate that the on‐treatment HBsAgGi response serves as a specific and robust surrogate marker for residual HCC risk exclusively in HBeAg‐positive patients. In sensitivity analysis of the HBeAg‐positive cohort, the ‘Improved’ subgroup exhibited a robust reduction in HCC risk compared to the Poor response group (adjusted HR 0.19; 95% CI, 0.07–1.00) after adjusting for the aMAP score. The ‘Maintained’ subgroup did not show significant risk reduction. No significant differences were observed in the HBeAg‐negative cohort (Table S4). Furthermore, a competing risk analysis using the Fine‐Grey model was performed to account for death without HCC; however, as only one such competing event occurred in our cohort, the results were virtually identical to those of the standard Cox proportional hazards model.

TABLE 3.

Multivariate Cox proportional hazards analyses of factors associated with HCC development.

Cohort/variables	Adjusted HR (95% CI)	p
Overall Cohort (N = 75)
Model 1: Main effects
Poor Response	1.71 (0.46–11.72)	0.513
aMAP risk group	2.47 (1.32–6.60)	0.033
Model 2: Interaction analysis
Poor Response	0.58 (0.08–3.48)	0.570
HBeAg (Positive)	0.27 (0.07–2.14)	0.155
aMAP risk group	2.37 (1.09–8.44)	0.096
Poor Response × HBeAg	10.62 (1.17–108.39)	0.038
Model 3: Stratified by HBeAg status
HBeAg‐Positive Cohort (N = 36)
Poor Response	7.34 (1.96–20.35)	0.001
aMAP risk score	1.55 (0.57–6.37)	0.459
HBeAg‐Negative Cohort (N = 39)
Poor Response	0.49 (0.06–2.72)	0.440
aMAP risk group	3.88 (1.09–18.31)	0.057

Note: Multivariate Cox proportional hazards analysis was performed using a landmark approach at week 48 with a 1‐year lag time. To address the potential for parameter instability and bias due to the absence of HCC events in the HBeAg‐positive favourable response group (monotone likelihood), the interaction analysis and subsequent stratified models were evaluated using Firth's penalised likelihood Cox regression. The robust estimation of the interaction term was further validated using bootstrap resampling with 1000 iterations, yielding a consistent and significant interaction between HBsAgGi response and HBeAg status (p = 0.038). Bold values indicate statistical significance (p < 0.05).

Abbreviations: CI, confidence interval; HBeAg, hepatitis B e antigen; HBsAgGi, hepatitis B surface antigen glycan isomer; HCC, hepatocellular carcinoma; HR, hazard ratio.

FIGURE 2.

Cumulative incidence of HCC stratified by the 48‐week HBsAgGi response. Kaplan–Meier curves from the 48‐week landmark point are shown for HBeAg‐positive (left) and HBeAg‐negative (right) patients. Patients were stratified into Favourable (blue line) or Poor (red line) response groups based on their HBsAgGi kinetics from baseline to week 48. p‐values were calculated using the log‐rank test. HBeAg, hepatitis B e antigen; HBsAgGi, hepatitis B surface antigen glycan isomer; HCC, hepatocellular carcinoma.

2.5. Predictive Performance of the Integrated Model Combining aMAP and HBsAgGi Response

We evaluated whether incorporating the 48‐week HBsAgGi response could enhance the predictive accuracy of the established aMAP score. In the time‐dependent ROC analysis (Figure 3A, Table S5), the combined model incorporating the HBsAgGi response showed a trend toward higher AUC values compared with the aMAP score alone in both HBeAg‐positive and HBeAg‐negative patients. However, the 95% CI for the two models overlapped throughout the observation period in both groups, indicating no statistically significant difference in AUC at any specific time point.

FIGURE 3.

Predictive performance and clinical utility of combining HBsAgGi response with the aMAP score for HCC development. All metrics were validated using 1000 bootstrap resamples (n = 75). (A) Time‐dependent AUC. Although the combined model (cyan line) shows a trend toward higher AUC values than the aMAP score alone (orange line), 95% CIs (shaded areas) overlap in both cohorts. (B) Longitudinal IDI (left) and continuous NRI (right). The integrated model significantly improves individual risk reclassification exclusively in HBeAg‐positive patients (red lines). Filled circles indicate statistical significance (p < 0.05). (C) DCA at 5 and 10 years demonstrating the higher clinical net benefit of the combined model (cyan line) over the aMAP score alone (orange line) in HBeAg‐positive patients. aMAP, age‐male‐ALBI‐platelets; AUC, area under the curve; DCA, decision curve analysis; HBsAgGi, hepatitis B surface antigen glycan isomer; HCC, hepatocellular carcinoma; IDI, integrated discrimination improvement; NRI, net reclassification improvement.

Because ROC analysis often lacks the sensitivity to evaluate the added predictive value of a new factor when the base model already exhibits high performance—reflecting a ceiling effect—we further assessed risk reclassification using IDI and continuous NRI (Figure 3B, Table S6). In the HBeAg‐positive cohort, the addition of the HBsAgGi response significantly improved the model's predictive performance during the early to mid‐term follow‐up periods. Specifically, the integrated model demonstrated a significant IDI for HCC development from years 2 through 11 of follow‐up. Furthermore, a significant continuous NRI was observed from years 2 through 6. Crucially, DCA revealed that the integrated model provided a higher net clinical benefit than the aMAP score alone across a wide range of threshold probabilities in HBeAg‐positive patients (Figure 3C). In contrast, no significant IDI or NRI was observed at any time point in HBeAg‐negative patients, and the decision curves were nearly identical between the models. These results indicate that the on‐treatment HBsAgGi response provides significant added value for precise HCC risk stratification and clinical decision‐making, particularly in HBeAg‐positive patients.

3. Discussion

While recent studies have suggested an association between HBsAgGi levels and HCC risk [17, 18], this study is the first to demonstrate that the prognostic value of HBsAgGi kinetics is significantly modified by HBeAg status, specifically offering a powerful tool for risk stratification in HBeAg‐positive patients. We report three major observations. First, a significant interaction was observed between HBeAg status and the 48‐week HBsAgGi response in relation to HCC development (p = 0.038). However, due to the absence of HCC events in the HBeAg‐positive favourable response group, the CIs for the interaction term and the subsequent HR (HR 7.34) are extremely wide. Therefore, while the categorical association is statistically robust, the exact magnitude of this risk increase must be interpreted with caution. Second, TAF treatment demonstrated a superior capacity to induce a favourable HBsAgGi response compared to ETV (OR 4.29). Third, HBsAgGi levels paradoxically increased in a subset of patients despite successful HBV DNA suppression, particularly under ETV therapy. These results suggest that HBsAgGi kinetics unmask a ‘hidden’ quality of viral suppression that conventional markers fail to capture. The clinical utility of HBsAgGi stems from its unique ability to reflect the burden of virions, distinct from the overwhelming excess of SVPs. While standard qHBsAg assays primarily measure these non‐infectious SVPs, HBsAgGi specifically detects the O‐glycosylated PreS2 domain characteristic of HBV genome‐containing virions [10, 11]. Crucially, under NA therapy, serum HBsAgGi predominantly reflects RNA virions (pregenomic RNA‐containing particles) rather than mature DNA virions [12, 19, 20, 21], as the latter are effectively depleted by the inhibition of reverse transcription. In this context, a recent study by Stadelmayer et al. [22] using a novel 5′RACE method revealed that tenofovir treatment, while blocking DNA synthesis, can increase intracellular pgRNA levels in HBV‐infected hepatocytes. This phenomenon likely results from a feedback mechanism in which the blockade of reverse transcription leads to compensatory accumulation or diminished consumption of the pregenomic template. Aligned with this molecular evidence, we identified several cases in the present cohort where HBsAgGi levels—and thus the burden of these RNA‐containing virions—paradoxically increased despite ongoing NA therapy. This dissociation represents an ‘incomplete suppression’ of viral replication that likely contributes to residual hepatocarcinogenesis. The superior efficacy of TAF in preventing this paradoxical elevation may be attributed to its higher intracellular active metabolite concentrations [23], which could more effectively suppress the upstream production or packaging of these RNA‐containing particles compared to ETV.

While the potential superiority of TAF over ETV in preventing HCC has been the subject of ongoing debate [14, 15, 24], the limited follow‐up in our cohort precludes direct outcome comparisons. However, evaluating patients through the lens of HBsAgGi kinetics offers a biological explanation for heterogeneity in HCC risk: individual viral response. We observed that ‘Favourable responders’ and ‘Poor responders’ coexist within both treatment groups; some TAF‐treated patients exhibit a poor response, while many ETV‐treated patients achieve a favourable one. We hypothesise that in previous drug‐comparison studies, the inclusion of these mixed responder populations within each arm diluted the true statistical difference in HCC incidence. By stratifying patients based on the HBsAgGi response rather than drug type alone, our approach successfully unmasked the residual risk hidden by conventional markers. This suggests that the ‘depth’ of virion suppression varies significantly between individuals and drugs, and that HBsAgGi serves as a sensitive tool to capture and quantify this difference.

The most critical finding of this study is that the predictive power of HBsAgGi response is confined to HBeAg‐positive patients. In the HBeAg‐positive (replicative) phase, the HBsAgGi response serves as a sensitive indicator of the quality of viral suppression; specifically, a ‘Poor response’ represents the persistence or even paradoxical enhancement of the viral replication cycle up to the stage of RNA virion production. Such ongoing viral activity likely contributes directly to hepatocarcinogenesis, providing additive prognostic value beyond established risk factors [25, 26]. In contrast, among HBeAg‐negative patients, viral replication is generally decreased, and the primary drivers of hepatocarcinogenesis may shift from active viral replication to host‐related factors—such as advanced age and liver fibrosis—accumulated over the long course of infection [27], which are largely captured by the baseline aMAP score [16]. In this clinical stage, the ‘oncogenic weight’ of active replication may be surpassed by established host‐side liver damage, which likely diminishes the incremental benefit of virion‐specific markers, such as HBsAgGi. This biological divergence explains why the HBsAgGi response effectively stratified HCC risk in the HBeAg‐positive cohort (p = 0.025) but showed no significant separation of survival curves in HBeAg‐negative patients (p = 0.814). To further dissect this finding, our sensitivity analysis stratified the ‘Favourable’ response into ‘Maintained’ (maintaining a low‐level HBsAgGi) and ‘Improved’ (achieving a reduction in the risk category). Notably, in HBeAg‐positive patients, stepping down to a lower HBsAgGi risk category (‘Improvement’) strongly drove HCC prevention even after adjusting for the aMAP score, whereas maintaining a baseline low state did not show a distinct protective effect. This highlights that during the highly replicative phase, active therapeutic suppression of viral activity acts as a critical turning point. Conversely, the absence of distinction between these sub‐responses in HBeAg‐negative patients further suggests that viral suppression is generally stable in this group, and HCC risk is predominantly driven by cumulative host factors.

Our study further utilised advanced metrics—IDI, NRI, and DCA—to confirm the clinical added value of HBsAgGi. While the baseline aMAP score is undeniably robust, it is a static marker. In HBeAg‐positive patients, adding the 48‐week HBsAgGi response significantly improved the IDI from year 2 through year 11 and the NRI from year 2 through year 6. The diminishing significance after a decade may reflect the increasing impact of long‐term cccDNA depletion or immune shifts under prolonged therapy. However, the overall results, supported by DCA, highlight that HBsAgGi provides a critical ‘early window’ to adjust the long‐term risk trajectory, which baseline characteristics alone cannot fully capture. Based on these findings, incorporating HBsAgGi kinetics into clinical practice may offer a refined, personalised approach to CHB management. Patients identified as ‘Poor responders’ at week 48, particularly HBeAg‐positive patients on ETV therapy who show paradoxically increased kinetics, should be recognised as a high‐risk group. For such individuals, switching to TAF may be a rational therapeutic option to optimise virion dynamics and potentially minimise long‐term HCC risk.

This study has some limitations, including its retrospective nature and the relatively small sample size of the TAF group. A critical limitation of this study is the marked asymmetry in the follow‐up duration between the two treatment groups (median 11.4 years for ETV vs. 3.5 years for TAF). Because HCC development typically requires long‐term observation, this discrepancy precludes any meaningful direct comparison of long‐term HCC incidence between the two drugs. Therefore, our survival analyses strictly focused on the prognostic value of the biomarker itself rather than a drug‐level outcome comparison. Additionally, serum HBV RNA was not directly measured to confirm our hypothesis regarding RNA virions, although it was inferred from recent molecular studies. Furthermore, the O‐glycosylation of the PreS2 domain measured by the HBsAgGi assay is structurally specific to HBV genotype C [10, 11]. Therefore, our findings are strictly restricted to this population and cannot be generalised to patients infected with HBV genotypes A, B, or D, which collectively account for a major proportion of the global HBV burden. Notably, while the HBsAgGi assay is commercially available for research use internationally (RCMG Inc., Tokyo, Japan), it has not yet received regulatory approval for routine clinical diagnostics globally. Despite these limitations, our data provide compelling evidence that HBsAgGi is a valuable, specific surrogate marker for guiding treatment optimisation in HBeAg‐positive patients, paving the way for a ‘treat‐to‐target’ strategy in chronic hepatitis B.

In conclusion, the on‐treatment HBsAgGi response at 48 weeks is a robust and specific surrogate marker for stratifying residual HCC risk, particularly in HBeAg‐positive patients with CHB. Our findings highlight that a poor HBsAgGi response unmasks an ‘incomplete suppression’ of the viral replication cycle, which likely reflects the persistence of RNA virions—providing clinical insights that baseline risk scores and conventional DNA markers fail to capture. While TAF demonstrated a superior capacity to prevent paradoxical increases in HBsAgGi and induce a favourable response compared to ETV, monitoring these kinetics offers significant clinical benefit by improving long‐term risk stratification for up to a decade. Optimising the HBsAgGi response through tailored NA selection provides a novel ‘treat‐to‐target’ framework, enabling personalised interventions to minimise long‐term hepatocarcinogenesis in high‐risk replicative populations.

Author Contributions

Conceptualisation: Y.K. and T.G. Clinical data collection: Y.K., Y.I., H.N., M.Y., Y.K., R.O., Y.T., S.S., A.M., S.S., and Y.S. Data analysis: Y.K. and T.G. Writing: Y.K. and T.G. All authors have read and approved the final version of the manuscript.

Funding

This study was supported in part by research funding from JIMRO, Otsuka Pharmaceutical, and Taiho Pharma.

Ethics Statement

The study protocol was approved by the ethics committee of the institution and adhered to the Declaration of Helsinki.

Consent

Patient consent was obtained via an opt‐out form disseminated using the hospital's website.

Conflicts of Interest

T.G. has received honoraria from AbbVie and Gilead Sciences Inc., and research funding from AbbVie, JIMRO, and Takeda Pharmaceutical. The other authors declare no conflicts of interest for this article.

Supporting information

Data S1: STROBE statement checklist.

Table S1: Changes in viral markers from baseline to week 48 according to treatment group.

Table S2: Multivariate linear regression analysis of factors associated with ΔHBsAgGi.

Table S3: Baseline characteristics and HBsAgGi response before and after IPTW.

Table S4: Univariable and Multivariable Sensitivity Analyses of Favourable Response Subgroups on HCC Development, Stratified by Baseline HBeAg Status.

Table S5: Time‐dependent AUC comparison between aMAP and Combined model stratified by HBeAg status.

Table S6: Longitudinal improvement in predictive performance (IDI and continuous NRI) by adding HBsAgGi response to the aMAP score.

Figure S1: Spearman's correlation heatmap among baseline clinical parameters, virological markers, and the change in HBsAgGi at week 48. A correlation heatmap was generated to evaluate the associations between baseline factors and the quantitative change in HBsAgGi from baseline to week 48 (ΔHBsAgGi). The numbers in each cell represent Spearman's rank correlation coefficients (ρ). Darker/bold numbers indicate statistically significant correlations (p < 0.05), while lighter/faint numbers indicate non‐significant correlations. The colour gradient reflects the direction and strength of the correlation, ranging from −1 (strong negative correlation) to 1 (strong positive correlation). ALT, alanine aminotransferase; DNA, hepatitis B virus DNA; TAF, tenofovir alafenamide treatment; HBeAg, hepatitis B e antigen; FIB4, fibrosis‐4 index; HBsAgGi, hepatitis B surface antigen glycan isomer; qHBsAg, quantitative hepatitis B surface antigen.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.