We sincerely appreciate the thoughtful letter from Dr. Shi and colleagues and are grateful for their careful appraisal of our study evaluating a steatosis-related polygenic risk score (PRS) for hepatocellular carcinoma (HCC) risk following sustained virological response (SVR) to direct-acting antiviral therapy 1. In our study, we demonstrated that a higher PRS-5 was independently associated with an increased risk of HCC after SVR. Dr. Shi et al. raised several important methodological considerations, regarding the interpretation of spline-based hazard ratios in the presence of competing risks, the robustness and clinical validity of the proposed high-risk threshold (PRS-5 ≥ 0.66), and the potential influence of differential surveillance intensity on observed associations 2. We are grateful for the opportunity to clarify these issues. First, restricted cubic splines within the Cox framework were employed to characterise the continuous association between PRS and HCC risk and to explore potential non-linearity, rather than to estimate cumulative incidence. For this exploratory objective, cause-specific hazard modelling was selected because it evaluates the exposure-event relationship independently of competing events. Importantly, all formal risk estimation and clinical inference were derived from Fine-Grey subdistribution hazard models, which accounted for competing risks and constituted our primary inferential framework. The proposed threshold was further supported by complementary decile-based sensitivity analyses, which demonstrated that the excess risk was predominantly concentrated in the highest PRS decile. Together, these findings reinforce the internal consistency of the identified cut-point. Second, although percentile-based cut-offs are frequently used for within-cohort stratification in genetic association studies, we agree that they require cautious interpretation 3. The 90th percentile threshold was not derived from outcome-driven optimization procedures (e.g., receiver operating characteristic-based selection), which may inflate type I error or overfit limited event data. We acknowledge that the modest number of HCC events (38 overall, including 8 in the highest decile) limits the precision of any single cut-point estimate. Formal resampling-based internal validation or optimism correction was not performed because our objective was etiologic risk characterisation rather than development of a prediction model or implementation-ready clinical score 4. Moreover, in rare-event settings with competing risks, resampling-based cut-point estimation may itself yield unstable thresholds sensitive to minor redistribution of events. Robustness was therefore assessed through multiple complementary analytic strategies, including three progressively adjusted multivariable models and concordant findings across prespecified subgroup analyses. These consistent results support the stability of the observed risk gradient while underscoring the need for external validation. Finally, differential surveillance alone is unlikely to account for the observed relationship. Given the aggressive biology of HCC, variations in surveillance intensity are more likely to influence stage at diagnosis than overall incidence, particularly in the context of long-term follow-up 5. While current surveillance guidelines are primarily guided by fibrosis severity 6, 7, PRS-5 remained independently associated with HCC risk after fibrosis adjustment, consistent across subgroup analyses. Anyway, we fully concur that intensive surveillance for high-risk patients is essential for early detection and improved survival outcomes 8, 9. We thank Dr. Shi and colleagues for the opportunity to clarify these important methodological considerations. Yu-Sheng Lin: conceptualization, writing – original draft. Yun-Yu Chen: conceptualization, writing – review and editing. Teng-Yu Lee: conceptualization, writing – original draft, writing – review and editing, supervision. The authors' declarations of personal and financial interests are unchanged from those in the original article 1. Teng-Yu Lee: research support from Gilead, Merck and Roche diagnostics; consultant of BMS, Gilead, and AstraZeneca and speaker of Abbvie, BMS, Eisai, Gilead, Roche and AstraZeneca. This article is linked to Lin et al. papers. To view these articles, visit https://doi.org/10.1111/apt.70566 and https://doi.org/10.1111/apt.70595. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.
Lin et al. (Sat,) studied this question.