Liver Stiffness Measured by Vibration-Controlled Transient Elastography Predicts Hepatic Decompensation in Patients with Hepatocellular Carcinoma Receiving Systemic Treatments.

Introduction
Since the advent of atezolizumab plus bevacizumab (Atezo/Bev) for the treatment of advanced hepatocellular carcinoma (HCC), significant improvements in patient prognosis have been reported [1]. Currently, several systemic regimens, including sorafenib, lenvatinib, and Atezo/Bev, are approved as first-line treatments for advanced HCC [2–4]. In addition, the STRIDE regimen (durvalumab plus tremelimumab) has been introduced as another guideline-recommended first-line option, offering a VEGF-independent immunotherapy approach [5]. Given the increasing variety of therapeutic options, precision medicine tailored to individual tumor characteristics, liver function, and patient demographics has gained growing importance [6]. In this context, Atezo/Bev has been associated with a relatively high incidence of variceal bleeding (VB), a potentially life-threatening adverse event, making careful treatment selection essential for patients at risk [7, 8].
Recently, several studies have investigated the clinical implications of hepatic decompensation (HD) in the treatment of advanced HCC. One study analyzing individual patient data from the IMbrave150 trial evaluated survival outcomes in patients who experienced HD [9]. The results indicated that early HD was associated with a significantly higher risk of mortality. Similarly, Celsa et al. [10] reported that patients with HD prior to HCC progression was linked to poorer outcomes compared to patients with progression prior to HD, highlighting HD as a potential surrogate marker for outcomes in HCC.
Various methods have been developed to predict or assess HD. Clinical scoring systems, such as the Child-Pugh score (CPS), MELD score, ALBI score, and CLIF-C score, are commonly used to evaluate liver function and estimate the risk of HD [11–13]. Additionally, direct measurement of the hepatic venous pressure gradient has been shown to be highly predictive of decompensation in patients with advanced chronic liver disease, although its routine use is limited by its invasive nature [14]. Of note, liver stiffness measurement (LSM) using vibration-controlled transient elastography (VCTE) offers a noninvasive alternative and has demonstrated strong predictive performance for HD [15]. In this context, the Baveno VII consensus defined patients with an LSM ≥25 kPa as having a high likelihood of clinically significant portal hypertension (CSPH), placing them at increased risk of HD [16]. Based on the proposed LSM thresholds in the Baveno VII criteria, several studies have explored the diagnostic value of VCTE in HCC, particularly in predicting high-risk varices or HD, and have validated its applicability across various HCC clinical settings [17, 18].
While several studies have investigated the prognostic value of VCTE in patients with HCC, no studies to date have specifically evaluated its role in predicting HD following systemic treatment, including Atezo/Bev [19]. In this study, we aimed to assess the prognostic significance of VCTE in predicting HD among HCC patients receiving various systemic treatments, including Atezo/Bev, sorafenib, and lenvatinib.

Patients and Methods

Study Patients
This multicenter historical cohort study reviewed medical records of 396 patients with HCC who received first-line systemic treatment, including Atezo/Bev, sorafenib, and lenvatinib, between March 2009 and October 2024 at seven university-affiliated hospitals. Patients were eligible for inclusion if LSM by VCTE was available within 6 months prior to initiation of systemic treatment [20]. Additional inclusion criteria were (1) diagnosis of unresectable HCC, (2) Eastern Cooperative Oncology Group (ECOG) performance status of 0–1, and (3) at least one follow-up visit after treatment initiation. Exclusion criteria were (1) patients classified as Child-Pugh class C, (2) absence of valid VCTE assessment within the 6-month pre-treatment period, (3) LSM with an interquartile range-to-median ratio exceeding 30%, (4) alanine aminotransferase (ALT) levels more than five times the upper limit of normal as this may lead to falsely elevated LSM values [21], and (5) patients with significant right lobe liver masses in whom reliable VCTE measurement was technically infeasible due to anatomical distortion. In accordance with national treatment guidelines and reimbursement policy in Korea, all patients with HBV (hepatitis B virus)-related HCC and detectable HBV DNA were required to receive nucleos(t)ide analogue therapy, and prophylactic antiviral treatment was routinely administered even in cases with undetectable HBV DNA prior to starting systemic anticancer therapy to prevent viral reactivation. As a result, all HBV-positive patients in our cohort were receiving antiviral therapy at baseline. This study was approved by the Institutional Review Board of the Catholic University of Korea (approval number: XC23RADI0081) and was conducted in accordance with the Declaration of Helsinki.

Treatment Protocols and Response Evaluation
For Atezo/Bev, the standard dosing regimen from the IMbrave150 trial was used, consisting of intravenous administration of 1,200 mg atezolizumab plus 15 mg/kg bevacizumab every 3 weeks. Sorafenib was administered orally at a recommended dose of 800 mg/day. Lenvatinib was administered at 12 mg/day in patients weighing ≥60 kg and at 8 mg/day in those weighing <60 kg. Dose modifications were made at the treating physician’s discretion. Treatment continued until intolerable toxicity or radiologic disease progression.
Tumor status was evaluated every 6–8 weeks using computed tomography or magnetic resonance imaging. Treatment response was assessed according to the modified Response Evaluation Criteria in Solid Tumors (mRECIST) [22].

LSM and Patient Stratification
LSM was assessed using VCTE. All patients were instructed to fast for at least 8 h prior to the examination to ensure accuracy and minimize variability due to postprandial hepatic blood flow. Measurements were taken with patients lying in the supine position and the right arm positioned overhead to optimize intercostal access. Immediately before VCTE, an abdominal ultrasound was performed in all patients to identify and avoid tumor-involved areas and to localize the right hepatic lobe for optimal probe positioning [23]. The result was considered reliable if it consisted of ten valid measurements with a success rate exceeding 60%. A reliable measurement was defined as an interquartile range-to-median ratio <30% and was included for the analysis [24]. In accordance with the Baveno VII consensus, which provides criteria for the noninvasive diagnosis of CSPH, a liver stiffness threshold of 25 kPa was adopted. Patients were subsequently categorized into two groups: a low LSM group (<25 kPa), indicating a low probability of CSPH, and a high LSM group (≥25 kPa), consistent with the presence of CSPH [16].

Definition of HD
HD was defined as the new onset or worsening of ascites, HE of any grade (per the West Haven criteria), or VB, in accordance with international guidelines [25]. In patients with pre-existing ascites, decompensation was defined as a documented worsening of ascites’ grade, confirmed independently by two researchers. To account for the potential influence of tumor progression on HD, only events occurring prior to radiologic tumor progression were considered liver-specific decompensation events. In cases where tumor progression occurred before or concurrently with HD, these events were not classified as decompensation, based on the rationale that tumor burden itself could contribute to clinical deterioration. This approach was adopted to isolate liver-specific events and minimize bias due to cancer-related clinical deterioration, as previously described in studies assessing progression-free HD [9]. To evaluate the risk of VB, upper gastrointestinal endoscopy (UGIE) performed prior to the initiation of Atezo/Bev was reviewed. High-risk varices were defined as varices of grade ≥2 or those exhibiting red color signs [26].

Study Endpoints
The primary endpoint was the development of HD within 12 months following the initiation of systemic treatment. Secondary endpoints included the incidence of VB events during the same 12-month period, overall survival (OS), and progression-free survival (PFS). Patients lost to follow-up or alive at the end of the study period were censored. Objective response rate (ORR) was calculated as the proportion of patients achieving a complete response or partial response.

Statistical Analysis
All statistical analyses were conducted using R software (version 4.3.3; R Foundation for Statistical Computing, Vienna, Austria; http://cran.r-project.org; accessed on May 1, 2025). Continuous variables are presented as means ± standard deviations and were compared using the Student’s t test. Categorical variables were compared using the chi-square test. Survival outcomes were analyzed using the Kaplan-Meier method and compared using the log-rank test. Propensity score matching (PSM) was performed to minimize baseline differences between the tyrosine kinase inhibitor (TKI)-treated group and the Atezo/Bev-treated group. Matching was based on key prognostic variables, including age, tumor burden, tumor marker levels, and liver function parameters, using a 1:1 nearest-neighbor algorithm with a caliper width of 0.05. Cox proportional hazards models were employed to identify factors associated with HD. Variables with statistical significance in univariate analysis were included in the multivariate model. Time-dependent area under the receiver operating characteristic curve (AUROC) was calculated to evaluate the predictive performance of LSM for cumulative HD. Harrell’s concordance index (C-index) was also assessed to examine the predictive performance of LSM over time. To address potential temporal heterogeneity arising from the introduction of Atezo/Bev in clinical practice, an additional sensitivity analysis was performed by stratifying the study population into two eras based on the year when Atezo/Bev was approved for reimbursement by the Korean National Health Insurance Service in 2022: the pre-Atezo/Bev era (before 2022) and the post-Atezo/Bev era (2022 onward). Outcomes were compared between these eras to assess the consistency of the associations across treatment periods. A two-sided p value <0.05 was considered statistically significant.

Results

Baseline Characteristics
A total of 396 patients were included in the study analysis (online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000550440). Among them, 237 patients were categorized as the low LSM group, while 159 patients were categorized as the high LSM group (Table 1). Although hepatitis B virus was the most common etiology of HCC in both groups, the high LSM group had a significantly higher proportion of hepatitis C virus- and alcohol-related HCC compared to the low LSM group (p = 0.022). The laboratory findings revealed that high LSM group had lower platelet count compared to the low LSM group (p = 0.032). In terms of liver function, the low LSM group had a significantly higher proportion of patients with Child-Pugh class A compared to the high LSM group (p = 0.002). Portal vein tumor thrombosis (PVTT), including high-grade PVTT (VP3/4), was more prevalent in the high LSM group (p = 0.004 and p = 0.001, respectively), and tumor size was larger in the high LSM group (p = 0.002). Regarding treatment modalities, 176 patients (44.4%) received Atezo/Bev, 175 (44.2%) received sorafenib, and 45 (11.4%) received lenvatinib. The distribution of treatment modalities did not significantly differ between the two groups (p = 0.546).

Factors Associated with HD
Cox regression was conducted to identify clinical factors associated with HD in the study population (Table 2). In the univariate analysis, several variables were significantly associated with increased risk of HD. These included treatment with Atezo/Bev compared to TKI (hazard ratio [HR], 1.75; 95% confidence interval [CI], 1.15–2.68; p = 0.010), ECOG PS 1 (HR 1.61; 95% CI, 1.05–2.46; p = 0.028), CPS 5 (HR, 0.23; 95% CI, 0.14–0.36; p < 0.001), LSM ≥25 kPa (HR, 3.00; 95% CI, 1.95–4.63; p < 0.001), maximum intrahepatic tumor size greater than 5 cm (HR, 2.55; 95% CI, 1.65–3.94; p < 0.001), multiple intrahepatic tumors (HR, 2.41; 95% CI, 1.55–3.75; p < 0.001), PVTT (HR, 3.14; 95% CI, 2.06–4.79; p < 0.001), and extrahepatic metastasis (HR, 0.59; 95% CI, 0.39–0.90; p = 0.014). Multivariate analysis identified four variables that remained independently associated with HD: CPS 5 (HR, 0.30; 95% CI, 0.18–0.48; p < 0.001), LSM ≥25 kPa (HR, 2.13; 95% CI, 1.36–3.32; p < 0.001), multiple intrahepatic tumors (HR, 1.68; 95% CI, 1.05–2.66; p = 0.029), and high-grade PVTT (HR, 1.99; 95% CI, 1.24–3.19; p = 0.004), suggesting high LSM and tumor-related factors are significantly related to the HD in HCC patients receiving systemic treatments. Regarding VB risk, univariate analysis identified Atezo/Bev versus TKI, CPS 5, LSM ≥25 kPa, multiple intrahepatic tumors, and high-grade PVTT as significant predictors of VB risk, while in the multivariate model only Atezo/Bev versus TKI and CPS 5 remained independently associated (online suppl. Table S1).

Incident HD and Survival Outcomes according to the LSM
Study outcomes including the incidence of HD, VB, OS, and PFS and ORR were compared between the low and high LSM groups among the overall cohort (n = 396, Fig. 1). During 12 months after the treatment initiation, a total of 88 patients experienced HD prior to disease progression. The 6-month cumulative incidence of HD was 44.0% in the high LSM group and 14.2% in the low LSM group. Overall, the high LSM group had a significantly increased risk of HD compared to the low LSM group (HR, 3.00; 95% CI, 1.95–4.63; p < 0.001) (Fig. 1a). VB occurred in 21 patients following treatment. The high LSM group again demonstrated an increased risk of VB compared to the low LSM group (HR 2.34; 95% CI, 1.01–5.56; p = 0.048) (Fig. 1b). Median OS was 15.2 months in the low LSM group and 9.5 months in the high LSM group, with a trend toward worse outcomes observed in the high LSM group (HR 1.27; 95% CI, 0.99–1.65; p = 0.065) (Fig. 1c). However, PFS and ORR did not differ between the two groups (Fig. 1d and e), suggesting that LSM measured by VCTE could be a predictor of HD and OS but not of treatment responses.
A sensitivity analysis stratified by treatment era (pre-Atezo/Bev vs. post-Atezo/Bev) was also performed. In the post-Atezo/Bev era, patients in the high LSM group exhibited a significantly higher risk of HD than those in the low LSM group (HR, 4.64; 95% CI, 2.54–8.48; p < 0.001) (online suppl. Fig. S2A). In the pre-Atezo/Bev era, a similar trend toward increased HD risk was observed in the high LSM group, although it did not reach statistical significance (HR, 1.70; 95% CI, 0.88–3.29; p = 0.111) (online suppl. Fig. S2B).

Subgroup Analysis on Patients Receiving Atezo/Bev
Further comparisons were conducted between patients treated with Atezo/Bev and those treated with TKIs (online suppl. Fig. S3). The comparison of baseline characteristics between Atezo/Bev and TKIs treated patients is presented in online suppl. Table S2. The Atezo/Bev group demonstrated superior OS (HR, 0.72; 95% CI, 0.54–0.95; p = 0.021) and PFS (HR, 0.75; 95% CI, 0.60–0.94; p = 0.012) compared to the TKI group. However, the Atezo/Bev group was also associated with a significantly higher incidence of HD (HR, 1.75; 95% CI, 1.15–2.68; p = 0.009) and VB (HR, 4.58; 95% CI, 1.68–12.51; p < 0.001) compared to the TKI group. Following PSM, which was performed to minimize baseline differences between the TKI and Atezo/Bev groups (online suppl. Table S3), the Atezo/Bev group consistently demonstrated a higher incidence of HD (HR, 1.80; 95% CI, 1.01–3.20; p = 0.045) and maintained superior OS compared with the TKI group (HR, 0.65; 95% CI, 0.45–0.94; p = 0.023) (online suppl. Fig. S4).
Next, HD and survival outcomes were further analyzed in the subgroup of patients treated with Atezo/Bev (n = 176). Consistent with the findings in the overall cohort, the high LSM group (n = 76) demonstrated a significantly increased risk of HD compared to the low LSM group (HR, 4.92; 95% CI, 2.68–9.02; p < 0.001, Fig. 2a). The incidence of VB was also significantly higher in the high LSM group compared to the low LSM group (HR, 2.87; 95% CI, 1.03–7.94; p = 0.034, Fig. 2b). In terms of survival outcomes, the high LSM group showed significantly worse OS (HR, 1.65; 95% CI, 1.02–2.66; p = 0.038) and PFS (HR, 1.45; 95% CI, 1.01–2.08; p = 0.045) compared to the low LSM group (Fig. 2c and d). ORR did not differ significantly between the groups, with rates of 22.4% in the high LSM group and 33.0% in the low LSM group (p = 0.121, Fig. 2e).
Additionally, clinical outcomes in the Atezo/Bev group were compared according to the presence of high-risk varices measured by UGIE, assessed prior to treatment (online suppl. Fig. S5). Among the 129 patients evaluated, 26 had high-risk varices. As a result, there was no significant difference in VB risk according to the high-risk varices (HR, 2.01; 95% CI, 0.75–6.45; p = 0.140), which further supports the similar or superior predictive value of LSM by VCTE for VB.

Different Clinical Outcomes between Atezo/Bev and TKIs according to LSM
Next, stratified analysis was performed to evaluate differential treatment outcomes between Atezo/Bev and TKIs according to LSM level (Fig. 3). In the low LSM group, the HD risk did not differ significantly between Atezo/Bev and TKIs (HR, 0.96; 95% CI, 0.48–1.91; p = 0.338). In contrast, in the high LSM group, Atezo/Bev was associated with a significantly higher risk of HD compared to TKIs (HR, 2.79; 95% CI, 1.59–4.90; p < 0.001). Similarly, with regard to VB, there was no statistically significant difference between treatment groups in the low LSM cohort (HR, 2.93; 95% CI, 0.73–11.71; p = 0.129), whereas Atezo/Bev was associated with a significantly higher bleeding risk compared to TKIs in the high LSM group (HR, 6.59; 95% CI, 1.44–30.18; p = 0.015). In terms of survival outcomes, Atezo/Bev demonstrated superior outcomes in the low LSM group for both OS (HR, 0.67; 95% CI, 0.46–0.97; p = 0.032) and PFS (HR, 0.66; 95% CI, 0.49–0.88; p = 0.005). However, in the high LSM group, no significant differences in OS or PFS were observed between the two treatment groups. Additionally, in the PSM cohort, consistent trends were observed. Among patients with high LSM, Atezo/Bev was associated with a significantly increased risk of HD compared with TKIs (HR, 2.80; 95% CI, 1.31–6.01; p = 0.008), whereas in the low LSM group, the incidence of HD remained comparable between the two treatments (HR, 0.85; 95% CI, 0.33–2.20; p = 0.737) (online suppl. Fig. S6). These findings suggest that Atezo/Bev might provide superior survival outcomes in patients with low baseline LSM, whereas patients with high baseline LSM may be at increased risk of HD and VB when treated with Atezo/Bev compared to TKIs.

Predictive Performance of LSM for HD
The predictive performance of LSM assessed by VCTE for HD and OS was further evaluated (online suppl. Table S4). For the overall cohort, the time-dependent AUROC for predicting 3-, 6-, and 12-month HD was 0.703, 0.730, and 0.758, respectively (C-index = 0.687). The predictive performance for HD was generally higher among patients treated with Atezo/Bev, with 3-, 6-, and 12-month AUROCs of 0.764, 0.803, and 0.840, respectively (C-index = 0.748), compared to those treated with TKIs. Regarding OS, the time-dependent AUROCs for predicting 6-, 12-, 18-, and 24-month survival in the entire cohort were 0.589, 0.615, 0.621, and 0.646, respectively (C-index = 0.566). In the Atezo/Bev group, the corresponding AUROCs were 0.575, 0.670, 0.647, and 0.670 (C-index = 0.565).

Risk Stratification Based on Predictors of HD
To enhance the predictive accuracy of HD based on LSM, four independent risk factors identified through multivariate Cox regression analysis – CPS 5, LSM ≥25 kPa, presence of multiple intrahepatic tumors, and high-grade PVTT – were incorporated into a risk scoring system. The risk score was calculated by summing the corresponding beta coefficients for each variable as follows: the score = 0.756 (if LSM ≥25 kPa) – 1.218 (if CPS = 5) + 0.516 (if multiple intrahepatic tumors are present) + 0.688 (if high-grade PVTT is present). The time-dependent AUROC for predicting HD at 12 months using this score was 0.832 (95% CI: 0.760–0.904, Fig. 4a). The optimal cutoff value, determined by Youden’s index, was zero. Based on this threshold, patients were stratified into low- and high-risk groups. Clinical outcomes were subsequently compared between the high-risk and low-risk groups defined by the cutoff score (Fig. 4b–e). The high-risk group showed a significantly increased risk of HD compared to the low-risk group (HR, 6.76; 95% CI, 3.81–11.99; p < 0.001, Fig. 4b). The incidence of VB was also markedly higher in the high-risk group (HR, 7.35; 95% CI, 2.16–25.02; p < 0.001, Fig. 4c). Consistent findings were observed for OS, with the high-risk group demonstrating significantly poorer survival compared to the low-risk group (HR, 1.87; 95% CI, 1.44–2.41; p < 0.001, Fig. 4d). PFS was also significantly shorter in the high-risk group compared to the low-risk group (HR, 1.40; 95% CI, 1.12–1.73; p = 0.003, Fig. 4e).

Risk-Stratified Analysis of HD between Atezo/Bev and TKIs
Comparative analyses of HD and VB between Atezo/Bev and TKIs were conducted in both low- and high-risk groups based on the scoring system with five independent risk factors (online suppl. Fig. S7). In the low-risk group, the risk of HD did not significantly differ between Atezo/Bev and TKIs (HR, 1.18, 95% CI, 0.41–3.37; p = 0.756, online suppl. Fig. S7A). In contrast, the risk of VB significantly differed between the two treatment groups (p = 0.030, online suppl. Fig. S7B). In the high-risk group, Atezo/Bev was associated with an increased risk of HD compared to TKIs (HR, 1.90, 95% CI, 1.19–3.03; p = 0.006, online suppl. Fig. S7C). Furthermore, Atezo/Bev demonstrated a significantly increased risk of VB compared to the TKIs (HR, 3.43, 95% CI, 1.22–9.64; p = 0.013), further supporting the potential utility of risk stratification as a biomarker to guide systemic treatment selection in advanced HCC (online suppl. Fig. S7D).

Discussion
In this study, we comprehensively evaluated clinical outcomes, including HD, in patients with advanced HCC undergoing systemic treatments, stratified by LSM obtained via VCTE. An LSM threshold of 25 kPa effectively predicted the risk of HD, with patients in the ≥25 kPa group demonstrating a HR of 2.13 compared to those with LSM <25 kPa. Patients with LSM ≥25 kPa also exhibited an increased risk of VB. This discriminative performance of the 25 kPa threshold was consistently observed in the subgroup treated with Atezo/Bev, in whom LSM ≥25 kPa was associated with significantly higher incidences of HD and VB, as well as inferior OS and PFS. To our knowledge, this is the first study to evaluate the prognostic utility of VCTE-derived LSM in predicting HD and related clinical outcomes in advanced HCC patients receiving systemic treatments including Atezo/Bev.
HD represents a critical event in the management of advanced HCC, with accumulating evidence suggesting that it is a key determinant of both survival and treatment continuity [9, 10, 27]. Previous studies have highlighted that the development of HD, whether in the form of ascites, HE, or VB, can significantly alter treatment plans and is often associated with increased mortality [9]. Unlike tumor progression, which can often be managed by switching to other systemic treatments, HD typically limits further treatment options due to impaired liver function, making it a critical inflection point in the clinical course of HCC [28]. A recent study further emphasized that HD leading to permanent treatment discontinuation is associated with significantly worse outcomes, highlighting its clinical impact [27]. In this context, accurate prediction and risk stratification of HD have become essential for optimizing therapeutic strategies in advanced HCC.
LSM using VCTE has emerged as a robust, noninvasive tool to assess liver fibrosis and portal hypertension [29]. Although extensively validated in non-HCC populations for predicting the onset of HD and long-term prognosis, its application in HCC has primarily focused on predicting tumor development [19, 20, 30]. A recent study proposed a noninvasive clinical model for identifying high-risk varices in unresectable HCC but notably did not incorporate VCTE [31]. In this context, our study is the first to directly evaluate the utility of VCTE-derived LSM in predicting HD following systemic treatment in advanced HCC. Given its inclusion in multiple international practice guidelines and its proven correlation with portal hypertension severity, VCTE represents a practical and scalable tool [16, 32]. Our findings confirm that an LSM threshold of 25 kPa – a criterion adopted from Baveno VII – effectively stratifies patients by HD risk, underscoring its potential role in clinical decision-making for patients undergoing systemic treatment for HCC.
Interestingly, when study cohorts were divided into two groups using 25 kPa thresholds, the survival benefit of Atezo/Bev was attenuated in the high LSM group compared to TKIs. In terms of OS, Atezo/Bev demonstrated superiority over TKIs in the low LSM group but showed no such benefit in the high LSM group. Furthermore, in the high LSM group, Atezo/Bev was associated with a significantly increased risk of both HD and VB, whereas no significant differences were observed between treatment groups in the low LSM group. This disparity could be explained by the synergistic effects of the anti-VEGF effect of bevacizumab [7, 33, 34], and the advanced portal hypertension, leading to heightened risk of VB, as well as poorer outcomes with Atezo/Bev. Additionally, patients in the high LSM group, who already have compromised hepatic reserve, may be more susceptible to HD and further worsening of hepatic function following the immune activation by atezolizumab [35]. Collectively, these findings suggest that LSM may serve as a valuable biomarker to inform the selection of systemic treatment between Atezo/Bev and TKIs in patients with advanced HCC.
Given concerns that high-risk varices may adversely impact outcomes with Atezo/Bev, we conducted a subgroup analysis of patients who had undergone UGIE prior to treatment. While patients with high-risk varices indeed demonstrated higher rates of HD and VB, predictive value of VCTE was superior or similar to UGIE. In addition to the VCTE, our multivariate analysis also identified high-grade PVTT as an independent predictor of HD. This is in line with previous literature suggesting that PVTT exacerbates intrahepatic circulatory compromise, thereby increasing portal pressure and hepatic vulnerability [33, 36]. Furthermore, tumor burden, reflected by large intrahepatic tumor size and the presence of multiple intrahepatic lesions, was also associated with an increased risk of HD, in agreement with prior reports [37]. These findings reinforce that noninvasive assessments including VCTE and tumor-related factors rather than endoscopic assessments might be useful in pre-treatment risk stratification.
Building on these observations, we developed an LSM-based risk score that incorporates four independent predictors of HD: CPS 5, LSM ≥25 kPa, presence of multiple intrahepatic tumors, and high-grade PVTT. The resulting stratification successfully distinguished patients at high risk for HD, VB, and poor survival outcomes in our study cohort. Although larger study is warranted, our findings suggest that the LSM-based scoring system could serve as a valuable noninvasive tool to personalize systemic treatment decisions and predict post-treatment complications, especially in the Atezo/Bev treatment.
There are some limitations that need to be acknowledged. First, the retrospective nature of the study may introduce selection bias and limit the ability to control unmeasured confounders, such as clinician judgment in treatment selection or supportive care decisions that may affect outcomes. Selection bias may be present in determining which patients received VCTE or a particular systemic therapy. Information bias could arise from inconsistent documentation or definition of HD events over the long study period. While multivariate analysis was performed, it cannot account for unmeasured confounding variables that may have influenced treatment decisions and outcomes. Third, given the long study period (2009–2024), potential temporal heterogeneity may have arisen due to evolving treatment options and advances in supportive care. To address this concern, we conducted a sensitivity analysis stratified by treatment era (pre-Atezo/Bev vs. post-Atezo/Bev), which showed generally consistent associations between LSM and HD across both eras, thereby helping to alleviate, at least in part, concerns regarding temporal heterogeneity. Fourth, the study cohort did not include patients treated with the durvalumab/tremelimumab (STRIDE) regimen, which was approved in Korea only in May 2024 and was not yet reimbursed during our study period. Therefore, our comparison was limited to Atezo/Bev and TKIs, warranting further studies to elucidate the hepatic safety of STRIDE relative to Atezo/Bev. Finally, the study was conducted only in Korea, which is hepatitis B endemic region, thereby limiting the generalizability to the other parts of worlds where other etiologies for HCC predominate.
In conclusion, our study demonstrates that LSM measured by VCTE is a significant predictor of HD in advanced HCC patients receiving systemic treatments. VCTE-based risk stratification can help identify patients at higher risk for adverse outcomes and may inform systemic treatment selection, particularly in deciding between Atezo/Bev and TKIs. Future prospective studies are warranted to validate these findings and to integrate VCTE-based tools into clinical workflows.

Statement of Ethics
This study was approved by the Institutional Review Board of the Catholic University of Korea (Approval No. XC23RADI0081). Informed consent was waived due to the retrospective nature of the study. Informed consent was waived due to the retrospective nature of the study and was approved by the Institutional Review Board of the Catholic University of Korea (Approval No. XC23RADI0081).

Conflict of Interest Statement
The authors have no conflicting financial interests.

Funding Sources
This research was supported by the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (Grant No. RS-2024-00406716 to J.W.H.). The funding source had no role in the study design; data collection, analysis, or interpretation; manuscript conception, planning, or writing; or in the decision to submit the manuscript for publication.

Author Contributions
Study concept and design, data analysis and interpretation, manuscript writing, and conceptualization and methodology: Jaejun Lee and Ji Won Han; data collection: Jaejun Lee, Keungmo Yang, Hee Sun Cho, Hyun Yang, Ji Hoon Kim, Heechul Nam, Hae Lim Lee, Sung Won Lee, Hee Yeon Kim, Ahlim Lee, Do Seon Song, Seok Hwan Kim, Myeong Jun Song, Soon Kyu Lee, and Ji Won Han; supervision: Pil Soo Sung, Jung Hyun Kwon, Jeong Won Jang, Seung Kew Yoon, Si Hyun Bae, Chang Wook Kim, Soon Woo Nam, and Ji Won Han; and final approval of the version to be published: all authors.