Response to Letter to the Editor Regarding ‘Practical Considerations for <scp>MRI</scp> ‐Derived Retreatment Response Scores in Viable <scp>HCC</scp> ’

We thank the authors for their thoughtful comments on our study and for highlighting important issues related to endpoint timing, treatment heterogeneity and statistical modelling, which are all relevant to the clinical interpretation of imaging-based prediction models [1]. First, regarding the timing of retreatment and response assessment, we acknowledge that in routine practice the intervals between the 1-month MRI, retreatment, and follow-up imaging may vary due to liver function, toxicity and physician preference. In our study, response was assessed on the approximately 6-month follow-up MRI, which represents a routine surveillance time point in real-world clinical practice at the participating centres. We did not adjust for the exact timing of retreatment or response assessment, and some heterogeneity therefore exists. Our aim was to predict retreatment outcomes under real-world clinical pathways rather than under strictly standardised schedules. Nevertheless, variability in assessment timing may influence radiologic outcomes and represents a limitation of our retrospective design. Future prospective studies with predefined time windows or adjustment for timing variables are needed to better isolate tumour sensitivity to retreatment. Second, we agree that treatment heterogeneity is a potential source of confounding. Retreatment modality was included in our multivariable models, and the VITAL-P score remained an independent predictor of response after adjustment, suggesting that its predictive value is not solely driven by treatment selection. Although the derivation cohort included different retreatment strategies, the external validation cohort consisted only of patients receiving locoregional therapy. This setting aligns with one intended clinical use of the score, risk stratification among candidates for further locoregional retreatment after detection of viable lesions. We agree, however, that causal conclusions regarding optimal treatment selection cannot be drawn, and that prospective trials stratified by VITAL-P risk are required to establish its role as a treatment-guiding biomarker. Third, we acknowledge that in the lesion-level analysis, multiple lesions from the same patient were treated as independent observations to increase sample size and support feature selection. This approach does not account for within-patient correlation and may underestimate uncertainty, which is a limitation of our analysis. Future studies will apply clustered data methods, such as mixed-effects models or generalised estimating equations, and perform patient-level sensitivity analyses to further validate the robustness of the imaging predictors. In summary, we appreciate the authors' constructive comments, which raise important methodological considerations for clinical translation. While we adjusted for treatment modality and performed external validation, limitations inherent to retrospective data remain. Prospective studies with standardised protocols will be essential to further define the clinical role of the VITAL-P score in guiding retreatment strategies for patients with hepatocellular carcinoma. The authors declare no conflicts of interest. This article is linked to Huang and Yu papers. To view this article, visit https://doi.org/10.1111/liv.70528. The authors have nothing to report.