Both tenofovir alafenamide (TAF) and entecavir (ETV) are currently recommended as the preferred first-line nucleos(t)ide analogue (NA) for chronic hepatitis B (CHB) 1. Emerging evidence suggests that they may differ in some aspects of clinical effectiveness, including virological response rates and post-withdrawal relapse patterns 2, 3. Whether the two agents result in different rates of functional cure, defined as sustained loss of hepatitis B surface antigen (HBsAg) with HBV DNA below detection limits, remains unclear. Given that HBsAg loss is a desirable treatment endpoint infrequently achieved with NA monotherapy 4, 5, clarifying this unresolved question is critical to inform clinical practice. In this issue, Kumada and colleagues address the question with a 39-centre Japanese cohort of 668 treatment-naïve CHB patients matched on propensity score (PS) 6. Using linear mixed-effects modelling, the authors found a steeper annual HBsAg decline with TAF than with ETV (−0.13 vs. −0.09 log IU/mL in the continuous-time model, p = 0.004). The difference was reportedly more pronounced in patients with younger age, absence of cirrhosis, or baseline HBsAg ≥ 3 log IU/mL. In contrast, the two NAs were similar in virological suppression, biochemical normalization, and hepatitis B core-related antigen (HBcrAg) trajectories. The study has clear strengths, including a large multicenter dataset, PS matching, longitudinal modelling, and exclusion of prior NA exposure. However, causal interpretation requires caution. Because treatment was not randomly allocated, residual confounding is an inherent concern. The PS calculation did not include several factors that could drive NA selection in routine clinical practice, such as age, renal function, or calendar year of treatment initiation 7; the significantly longer follow-up in ETV recipients may suggest incomplete exchangeability between the matched groups. The random-effects structure does not include a random slope, potentially under-capturing heterogeneity in HBsAg trajectories among individuals 8. Time-varying factors such as drug adherence and treatment switching are not addressed. To strengthen the evidence, future studies should be designed with functional cure as a primary endpoint, rather than relying on extrapolation of biomarker trajectories. Adequately powered cohorts followed for longer durations will be necessary. In view of the practical difficulties, data from randomized controlled trials are unlikely to be available, at least not in the near future. Advancing this field will thus require more robust study designs to address the inherent limitations of observational comparisons. Specifically, target-trial emulation with explicit time-zero anchoring, expanded baseline covariates, inverse-probability weighting, and consideration of informative censoring would more credibly approximate causal effects 9. Pre-specified interactions, rather than within-stratum significance tests, are needed to explore subpopulations with differential benefits 10. Finally, mechanistic studies to investigate why HBsAg and HBcrAg trajectories diverge under different NAs would complement clinical data and link surrogate signals to biological insights. In summary, Kumada et al. found that TAF was associated with faster quantitative HBsAg decline than ETV. While the study findings may provide useful information for patients with CHB, whether the association reflects a drug-specific effect, a meaningful surrogate signal, or residual confounding remains an open question. Comparative effectiveness of antiviral therapies for treatment-naïve CHB is an active research area that warrants more attention. Yao-Chun Hsu: conceptualization, methodology, writing – original draft, writing – review and editing. Ming-Lung Yu: conceptualization, methodology, supervision, project administration, writing – review and editing. The authors have nothing to report. This article is linked to Kumada et al. paper. To view this article, visit https://doi.org/10.1111/apt.70704. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.
Hsu et al. (2026) studied this question.