In-depth analysis of AFP trajectories in HCC: bridging gaps in survival data and clinical applicability

Dear Editor, We commend the authors of “Comparison of the efficacy and safety of TACE-HAIC-MTTs-ICIs and TACE-MTTs-ICIs in the hepatocellular carcinoma: a prognostic analysis based on the dynamic changes of serum AFP” for their innovative investigation into dynamic AFP trajectories within complex hepatocellular carcinoma (HCC) treatment paradigms[1]. The authors’ thorough analysis of AFP dynamics, a promising biomarker for predicting survival outcomes in HCC, is noteworthy. The identification of six dynamic AFP trajectories and their correlation with overall survival (OS) provides valuable insights into how AFP changes can inform clinical decision-making. Furthermore, the study’s robust methodological approach, including the use of inverse probability of treatment weighting (IPTW) and the Joint Latent Class Model (JLCM), enhances the study’s credibility. While this work advances our understanding of biomarker-driven prognostication, we believe several methodological aspects merit clarification to ensure the robustness of the conclusions. The manuscript followed the TITAN guidelines[2]. First, the lack of a significant survival advantage between the TACE-HAIC-MI and TACE-MI groups, despite the theoretical benefits of adding hepatic arterial infusion chemotherapy (HAIC) to the treatment regimen, raises questions about the efficacy of the more complex treatment approach. Although the study provides valuable insights into the role of HAIC in combination therapies, it would be beneficial to further explore whether specific patient subgroups—particularly those with vascular invasion[3]—might still benefit from the additional complexity of TACE-HAIC-MI. We recommend conducting sensitivity analyses using stratification by key confounders (e.g., vascular invasion) to validate the nonsignificant survival outcomes. Second, significant pretreatment disparities persisted between cohorts. For instance, vascular invasion was present in 66.9% of the TACE-HAIC-MI group compared to 42% in the TACE-MI group, and BCLC stage C was present in 81.2% vs. 55.4% of the respective groups (both P<0.001). Despite the use of IPTW, further validation is required. Additionally, the smaller sample size of the TACE-HAIC-MI cohort (n = 154 vs. 305) raises concerns regarding statistical power, particularly for subgroup analyses. Third, while the quadruplet regimen demonstrated a significantly higher complete response (CR) rate (7.79% vs. 1.31%, P = 0.001), it showed no corresponding OS benefit (P = 0.432). This discordance warrants further exploration to determine whether the CR advantage translates into secondary resectability or delayed progression. Potential mechanisms, such as differential resistance patterns or tumor genomic profiling, should also be explored. To strengthen the manuscript’s impact, we suggest performing landmark analyses to reconcile the CR-OS discordance. Fourth, the six-class JLCM model, while statistically sophisticated, faces practical limitations: (1) small subgroup sizes: Class 3 (favorable responders; n ≈ 30, 6.54%) exhibited wide OS confidence intervals (35.87–NA months), reducing the reliability of survival estimates for this subgroup. (2) Overfitting risk: The extreme hazard ratio for Class 5 (OS HR = 45.28, CI: 17.23–119.03) likely reflects model instability due to its small sample size (n = 55). (3) Validation gap: Claims of superiority over established scores (C-index 0.787 vs. mHAP-III’s 0.649) require external validation, and a defined clinical action timeline (e.g., “Trajectory assignment at Week 12”) is necessary to improve its clinical applicability. Therefore, we recommend defining a clinically actionable window for trajectory classification (e.g., 8–12 weeks post-treatment) and comparing the JLCM model against dynamic multi-biomarker scores, such as AFP combined with PIVKA-II. Additionally, while AFP trajectories may hold promise for prognostic stratification, further exploration into integrating dynamic AFP changes into multi-modal prediction models that combine clinical and genetic data could offer more precise guidance in individualizing treatment plans for HCC patients. Fifth, the conclusion of “acceptable safety” overlooks critical disparities in adverse events (AEs). For example, the TACE-HAIC-MI group experienced higher rates of Grade ≥3 electrolyte imbalances (13% vs. 4%, P < 0.001) and INR elevation (5% vs. 0%, P < 0.001), while the TACE-MI group showed increased rates of transaminitis (AST elevation: 84% vs. 73%, P = 0.013). However, the manuscript lacks data on AE management, such as dose reductions, treatment discontinuation rates, and their impact on therapeutic efficacy. To improve safety reporting, we suggest detailing AE mitigation protocols (e.g., hydration strategies for HAIC-related toxicity) and disclosing treatment discontinuation rates due to AEs. In conclusion, this study makes a pioneering contribution by integrating dynamic biomarkers with emerging therapies for HCC. However, the conclusions regarding treatment equivalence require further validation. Addressing the concerns raised in this letter will help solidify the study’s findings and enhance its impact on personalized HCC management. We congratulate the authors for their comprehensive and methodologically sound research and encourage further exploration into patient-specific factors that might influence treatment outcomes, as well as the potential for integrating AFP trajectories into broader predictive models.