Tumor protrusion assessment and clinical integration for enhanced mortality risk prediction in spontaneously ruptured hepatocellular carcinoma

Abstract

BACKGROUND

Spontaneously ruptured hepatocellular carcinoma (rHCC) carries high early mortality following emergency hemostatic procedures like transarterial embolization (TAE) or transarterial chemoembolization (TACE). Existing prognostic models lack accuracy, particularly by omitting morphological risk factors such as tumor protrusion.

AIM

To develop and validate a novel scoring system that integrates quantitative tumor protrusion with clinical parameters to improve 30-day mortality prediction.

METHODS

This retrospective cohort study included consecutive patients with rHCC treated with emergency TAE/TACE between January 2007 and December 2024. Three quantitative protrusion metrics (percentage, maximum diameter, and area) were compared for predictive performance. Inter-observer reliability was assessed using intraclass correlation coefficients. Multivariable logistic regression was used to identify independent predictors of 30-day mortality. A weighted scoring system was derived from regression coefficients and internally validated using bootstrap resampling (1000 iterations). The model’s discriminative ability was compared to the model for end-stage liver disease and Child-Pugh scores using the area under the receiver operating characteristic curve (AUC) and DeLong’s test.

RESULTS

Eighty-nine patients were included. Protrusion area (PA) exhibited the highest discriminative power for 30-day mortality (AUC = 0.618) and excellent inter-observer reliability (intraclass correlation coefficients = 0.981). Multivariable analysis identified three independent predictors: PA ≥ 8.7 cm², total bilirubin ≥ 2.5 mg/dL, and albumin ≤ 3.0 g/dL. The derived PA-bilirubin-albumin (PBA) score stratified patients into low-, intermediate-, and high-risk groups with 30-day mortality rates of 8.6%, 40.0%, and 81.8%, respectively (P < 0.001). The PBA score demonstrated excellent and robust discrimination, confirmed by bootstrap validation (optimism-corrected AUC = 0.814). The PBA score significantly outperformed both the model for end-stage liver disease score (AUC = 0.672, P = 0.005) and Child-Pugh score (AUC = 0.594, P = 0.001).

CONCLUSION

By integrating PA with bilirubin and albumin, the PBA score is a novel, validated prognostic tool specifically for rHCC. It provides superior risk stratification over existing scores and identifies high-risk patients for whom emergency TAE/TACE is unlikely to provide clinical benefit.

Key Words: Hepatocellular carcinoma; Tumor rupture; Transarterial chemoembolization; Prognosis; Risk score

Core Tip: Spontaneously ruptured hepatocellular carcinoma carries a high 30-day mortality risk. Current assessment of tumor protrusion remains largely subjective, and existing prognostic models often lack precision in acute settings. This study identifies “protrusion area” (PA) as a superior predictor of early mortality compared to other metrics. We developed the PA-bilirubin-albumin score, which significantly outperforms conventional scoring systems (area under the receiver operating characteristic curve = 0.813). By stratifying patients into three distinct risk groups, the PA-bilirubin-albumin score provides clinicians with a robust, objective tool for rapid bedside decision-making, helping to identify patients who may benefit more from palliative care than from futile aggressive interventions.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a major global cancer problem. It is the sixth most common type of cancer in the world and the third most common cause of death from cancer[1,2]. The disease burden is notably significant in Asian populations, characterized by heightened incidence rates of HCC, which are indicative of regional epidemiological trends shaped by endemic hepatitis B infection[3].

Spontaneously ruptured HCC (rHCC) represents a severe complication that adversely affects patient outcomes. As the third most frequent cause of HCC-related death following tumor progression and hepatic failure, rHCC demonstrates marked geographic variation in incidence that ranges from < 3% in Western populations to 10%-26% in Asian and African cohorts[2-5]. This life-threatening condition is linked to significant mortality with immediate 30-day fatality rates of 17%-71%, and overall mortality is often above 50%[2,5-7]. The considerable variation in outcomes underscores the critical need for reliable prognostic stratification tools to guide clinical decision-making[6].

Current therapeutic methods for rHCC include conservative therapy, transarterial embolization (TAE), and surgical intervention[2]. TAE has been the favored first strategy for hemodynamically unstable patients, providing fast hemostatic control with high technical success rates (53%-100%) and clinical efficacy around 94%[2-5,7]. However, clinical outcomes remain inconsistent and difficult to predict[8]. Established prognostic factors such as Child-Pugh classification, hemodynamic status at presentation, tumor distribution pattern, and portal vein involvement significantly affect treatment success; however, current risk stratification models are insufficient for routine clinical application[2,4,6,7,9,10].

Tumor morphology, particularly the degree of protrusion beyond the hepatic capsule, has been recognized as a crucial factor influencing rupture risk in HCC[11]. Lesions that extend much beyond the liver surface lack the protective buffering effect provided by the surrounding parenchyma, making them susceptible to mechanical stress and ischemic injury[7,11]. This anatomical tendency is exacerbated by treatment-related conditions such as capsular necrosis subsequent to transarterial chemoembolization (TACE)[11].

The literature on tumor protrusion quantification is subjective and inconsistent[2,3,11]. Examples of descriptive criteria, such as tumor protrusion from the hepatic surface or a peripherally located tumor with a contour bulge, are imprecise and unreliable[2,3,11]. Recent standardization efforts include classification systems based on tumor extension beyond the liver capsule[11]. Nevertheless, it is still unclear whether a measurement technique is best to predict clinical outcomes, whether it be area-based evaluations, absolute diameter measures, or percentage-based ratios[3,11]. The lack of standardized, objective measurement techniques hinders prognostic accuracy and treatment optimization, and the comparative effectiveness of different protrusion assessment methodologies in predicting early mortality has not been systematically evaluated.

This study aimed to systematically compare three quantitative protrusion assessment methodologies (percentage-based diameter ratios, absolute diameter measurements, and area measurements) for 30-day mortality prediction in rHCC, and to develop and internally validate an integrated scoring system incorporating the optimal morphological metric with clinical parameters.

MATERIALS AND METHODS

Study population

This retrospective cohort study was approved by the Institutional Review Board and adhered to the principles of the Declaration of Helsinki. We examined the medical records of all patients who received TAE or TACE for the treatment of spontaneously rHCC from January 2007 to December 2024. This study encompassed consecutive patients who satisfied the following inclusion criteria: Adults aged 18 years or older with spontaneously rHCC who had TAE or TACE as the main treatment and possessed high-quality pre-procedural imaging appropriate for protrusion measurement analysis.

The American Association for the Study of Liver Diseases established imaging criteria for the diagnosis of HCC[12]. These criteria require the demonstration of arterial-phase hyperenhancement followed by portal or delayed-phase washout in lesions measuring at least 1 cm in diameter within a cirrhotic liver, or histopathological confirmation when imaging was inconclusive[12]. A comprehensive multidisciplinary evaluation that integrated clinical presentation, laboratory findings, and advanced imaging characteristics was used to establish the diagnosis of rHCC. Acute abdominal pain, hemodynamic instability, or blood-stained ascites were among the clinical manifestations indicative of HCC rupture during diagnostic paracentesis. The radiological confirmation of tumor rupture was achieved through contrast-enhanced computed tomography (CT) or magnetic resonance imaging (MRI), which demonstrated one or more of the following definitive signs: Active contrast media extravasation from the tumor into the peritoneal cavity, tumor protrusion beyond the hepatic surface with associated hemoperitoneum, focal discontinuity or disruption of the tumor surface contour with adjacent hemorrhagic fluid collection, or the presence of an enucleation sign indicating tumor extrusion from the liver parenchyma[13].

Patients were systematically excluded from analysis if they presented with any of the following criteria: Recent history of percutaneous liver biopsy within 30 days prior to rupture, inadequate or poor-quality imaging precluding accurate protrusion measurement, incomplete clinical documentation or medical records, recurrent tumors with previous treatment history, or patients who underwent emergency curative hepatectomy or surgical hemostasis instead of TAE/TACE. Of the initial cohort of patients treated for rHCC during the study period, 89 patients satisfied the inclusion-exclusion criteria and constituted the study population.

The primary outcome was 30-day mortality, defined as death from any cause occurring within 30 days of the beginning of the TAE/TACE operation. Patients were followed via medical records and the national death registry when available. Since the study was retrospective, the Institutional Review Board waived further approval for the research.

Technique of TAE/TACE

All transarterial procedures were performed by experienced interventional radiologists via a standardized right femoral arterial approach under local anesthesia. Pre-procedural planning utilized contrast-enhanced CT or MRI to localize the ruptured tumor based on perihepatic hematoma, active contrast extravasation, and disruption of the HCC contour. Following arterial access, comprehensive hepatic angiography was performed using a 5-Fr catheter to evaluate vascular anatomy, identify feeding arteries, and detect active bleeding sites when present.

The extent of embolization was tailored to the clinical scenario and imaging findings. Selective or subsegmental embolization was performed when the tumor was solitary and the bleeding source was clearly delineated. In multifocal cases where the exact ruptured lesion could not be localized on angiography, segmental or lobar embolization was performed based on imaging interpretation. Embolic agents were selected by the operating radiologist based on clinical judgment and included absorbable gelatin sponge (Gelfoam), polyvinyl alcohol particles (355-500 μm) (Contour, Boston Scientific, MA, United States), iodized oil (Lipiodol Ultra Fluide, Guerbet, Paris, France) combined with chemotherapeutic agents including doxorubicin hydrochloride or mitomycin C, or combinations thereof.

All procedures were performed with standard radiation protection measures and contrast allergy precautions. Technical success was defined as complete or near-complete stasis in the target feeding arteries confirmed by post-embolization angiography. Clinical success was defined as hemodynamic stability (systolic blood pressure > 90 mmHg without vasopressors) and hemoglobin stabilization (increase or < 2 g/dL decrease from post-transfusion baseline) maintained for 48 hours without evidence of ongoing bleeding.

Data collection

Medical records were systematically reviewed to evaluate patient characteristics, laboratory parameters, imaging findings, and 30-day outcomes. The clinical variables collected were age at presentation, sex, underlying liver disease etiology, including hepatitis B and C status, presenting symptoms and vital signs at admission, blood product transfusion requirements, relevant medical comorbidities, Child-Pugh classification, Barcelona Clinic Liver Cancer staging[12], and treatment outcomes. Complete blood count with hemoglobin levels, comprehensive metabolic panel including serum bilirubin and albumin concentrations, serum creatinine, international normalized ratio, and alpha-fetoprotein levels were all laboratory parameters obtained within 24 hours of admission and prior to intervention.

Cross-sectional imaging studies and angiographic findings were systematically analyzed to evaluate tumor characteristics, including number, size, morphological type, anatomical location, and distribution. Also evaluated were the degree of hepatic capsule disruption, presence and extent of active contrast extravasation, and portal vein involvement. Tumor morphology was categorized based on established criteria into single nodular, multiple nodular, or infiltrative patterns. Portal vein thrombosis was systematically classified as absent, segmental (affecting subsegmental or segmental branches), lobar (involving the right or left portal vein), or involving the main portal vein. Procedural reports from TAE/TACE operations were thoroughly evaluated to record technical details such as the selection of embolic agents and procedural complications that included access site issues, contrast-induced nephropathy, and post-embolization syndrome.

Image evaluation

All pre-procedural CT or MRI images were reviewed using a picture archiving and communication system to assess tumor protrusion characteristics. Three distinct quantitative assessment methods were employed: Percentage-based diameter ratios, absolute diameter measurements, and absolute area measurements (Figure 1). All measurements utilized portal venous phase images (70-80 seconds post-contrast), demonstrating maximum tumor protrusion across orthogonal planes.

Figure 1 Schematic illustration of the three quantitative tumor protrusion assessment methods. Protrusion area represents the absolute area of the tumor extending beyond the expected hepatic contour (A). Protrusion percentage is calculated as the ratio of the maximum protruding tumor diameter (B) to the maximum total tumor diameter (C). Protrusion diameter is the absolute maximum diameter of the protruding tumor portion (D). PA: Protrusion area; PP: Protrusion percentage; PD: Protrusion diameter.

Prior to independent measurements, both radiologists jointly established a standardized approach for hepatic contour extrapolation through consensus review of representative cases. The extrapolated hepatic contour reference line was a smooth curve reconstruction over protruding areas that ignored cirrhosis-related surface imperfections. To further standardize contour reconstruction and minimize inter-observer variability, the following criteria were applied: (1) The reference contour followed the outer border of non-protruding hepatic parenchyma; (2) Cirrhosis-related surface nodularities were smoothed by interpolating between adjacent inflection points; and (3) All measurements were performed on portal venous phase images at the level of maximum protrusion across orthogonal planes.

Protrusion percentage (PP) was determined by the ratio of the protruding tumor diameter to the total tumor diameter. Total tumor diameter was quantified as the maximum measurement of the overall tumor mass, whereas the protruding diameter indicated the maximum measurement going beyond the consensus-defined projected hepatic contour. PP was determined using electronic calipers with soft tissue window parameters (width 350-400 HU, level 40-60 HU) and represented as a percentage between 0% and 100%. Protrusion diameter (PD) was measured as the absolute maximum diameter of tumor extending beyond the extrapolated hepatic contour, expressed in centimeters, using the same consensus-established hepatic reference line and measurement protocols as defined for PP assessment. Protrusion area (PA) was calculated as the absolute area (cm²) of tumor extending beyond the extrapolated hepatic contour. Tumor boundaries were manually contoured with semi-automated area calculation software to determine the extrahepatic tumor area expressed in cm² by applying the consensus-defined contour extrapolation methodology.

In patients with multiple HCC lesions, the ruptured tumor was identified based on imaging evidence of active bleeding, surface discontinuity, or closest proximity to the hemoperitoneum. All measurements were performed specifically on the ruptured lesion rather than the largest tumor.

Following the initial consensus training phase, all measurements were performed independently by two radiologists with greater than 5 years of hepatobiliary imaging experience, blinded to clinical outcomes and each other’s assessments. Inter-observer agreement was evaluated using intraclass correlation coefficients (ICCs) for all three measurement methods. Final measurements for analysis were calculated as the mean values between the two radiologists for each protrusion assessment method.

Statistical analysis

Statistical analysis was performed using R statistical software version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria) with significance set at P < 0.05. Categorical variables were presented as frequencies and percentages and continuous variables as median with interquartile range (IQR). Baseline characteristics were compared using Pearson χ² or Fisher’s exact test for categorical variables and Mann-Whitney U test for continuous variables.

The primary endpoint was 30-day mortality following TAE/TACE. Receiver operating characteristic (ROC) curve analysis assessed the discriminative performance of the three protrusion assessment methods: PP, diameter, and the area based on manual tracing of the protruding portion on axial CT images. For each approach, the AUC with 95% confidence interval (CI) was determined. Cutting points for the protrusion measurements were established as follows: Youden’s index from ROC analysis of the current dataset determined PA and diameter cut-points, whereas the 50% cut-point for PP was based on earlier literature[11]. Pairwise discriminative ability comparisons used DeLong’s test for paired ROC curves. The protrusion measures were examined for inter-observer agreement using ICCs, where ≥ 0.75 indicated acceptable reliability. Correlation analysis between the three measurement methods was performed using Spearman correlation coefficients.

Univariate logistic regression identified prognostic factors for 30-day mortality. Variables with P < 0.05 were entered into multivariable logistic regression using backward elimination to identify independent predictors. The most effective protrusion metric was chosen for the final predictive model based on AUC values and statistical significance.

The β coefficients from the final multivariable logistic regression analysis were multiplied by 3 and rounded to the nearest integer to obtain clinically practical point values for bedside calculation of the scoring system. Patients were stratified into three risk groups based on score distribution and mortality outcomes. To avoid overoptimistic results due to model overfitting, internal validation was performed using bootstrap resampling with 1000 iterations. Bootstrap validation assessed model stability, bias correction, and CIs for performance metrics. The discriminative performance of our score was compared with established liver function scoring systems [Child-Pugh classification and model for end-stage liver disease (MELD) score] using AUC comparison and DeLong’s test.

Overall survival was analyzed using the Kaplan-Meier method with the log-rank test for group comparisons. Cox proportional hazards regression was performed to calculate hazard ratios for survival outcomes. Pairwise survival comparisons between score groups were conducted using log-rank test. All statistical tests were two-sided, and P values of < 0.05 were considered statistically significant.

RESULTS

Patient characteristics

The baseline characteristics of 89 patients with spontaneous rHCC are summarized in Table 1. The cohort comprised predominantly men (n = 73, 82%) with a median age of 59.0 years (IQR: 51.7-66.1). Hepatitis B virus was more common than hepatitis C virus [n = 40 (45%) vs n = 18 (20%); P = 0.006]. Most patients had preserved liver function: Child-Pugh class A (n = 47, 53%) and class B (n = 39, 44%). According to Barcelona Clinic Liver Cancer staging, 46 patients (52%) were classified as stage B and 35 patients (39%) as stage C. Hypovolemic shock was present at admission in 36 patients (40%), which reflected the severity of hemorrhage. The median size of the ruptured tumor was 7.8 cm (IQR: 5.9-11.1), and multiple tumors were present in 62 patients (70%). Portal vein thrombosis was identified in 33 patients (37%), with main portal vein involvement in 5 cases. Extrahepatic metastasis was documented in 13 patients (15%).

Table 1 Baseline characteristics of the study participants (n = 89).

Variables	n (%)/median (IQR)
Age (years)	59.0 (51.7-66.1)
Sex (men/women)	73 (82)/16 (18)
Hepatitis B virus (positive/negative)	40 (45)/49 (55)
Hepatitis C virus (positive/negative)	18 (20)/71 (80)
Shock on admission (yes/no)	36 (40)/53 (60)
Time from imaging diagnosis to TAE/TACE (hours)	12.1 (5.2-72.0)
Biochemical parameters
Hemoglobin (g/dL)	9.9 (8.1-11.5)
Platelet count (× 103/μL)	167 (119-253)
Albumin (g/dL)	3.2 (2.8-3.5)
Total bilirubin (mg/dL)	1.22 (0.77-2.04)
AST (U/L)	102 (58-181)
ALT (U/L)	55 (33-100)
ALP (U/L)	122 (76-196)
INR	1.27 (1.16-1.44)
Creatinine (mg/dL)	0.95 (0.72-1.22)
Alpha-fetoprotein (ng/mL)	226 (22-14327)
Alpha-fetoprotein (< 200/≥ 200 ng/mL)	41 (46)/48 (54)
Imaging parameters
Tumor number (single/multiple)	27 (30)/62 (70)
Ruptured tumor size (cm)	7.8 (5.9-11.1)
Ruptured tumor location (caudate/left/right)	1 (1)/20 (23)/68 (76)
Portal vein thrombosis (present/absent)	33 (37)/56 (63)
Extrahepatic metastasis (present/absent)	13 (15)/76 (85)
Child-Pugh class (A/B/C)	47 (53)/39 (44)/3 (3)
BCLC stage (A/B/C)	8 (9)/46 (52)/35 (39)
Embolic agents
Gelatin sponge	72 (81)
Gelatin sponge + polyvinyl alcohol	2 (2)
Lipiodol and gelatin sponge	15 (17)
Protrusion assessment
Protrusion area	15.4 (7.5-31.4)
Protrusion diameter	3.0 (1.7-4.7)
Protrusion percentage	40% (26%-58%)
Clinical outcome
30-day mortality	22 (25)
Overall mortality	83 (93)
Overall survival	5.03 (3.12-6.24)

IQR: Interquartile range; TAE: Transarterial embolization; TACE: Transarterial chemoembolization; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; ALP: Alkaline phosphatase; INR: International normalized ratio; BCLC: Barcelona Clinic Liver Cancer.

Protrusion assessment and inter-observer agreement

Three protrusion measurement methods demonstrated different median values: PA 15.4 cm² (IQR: 7.5-31.4), PD 3.0 cm (IQR: 1.7-4.7), and PP 40% (IQR: 26%-58%). Inter-observer agreement was excellent for all three methods with ICCs of 0.981 (95%CI: 0.971-0.987) for PA, 0.955 (95%CI: 0.932-0.970) for PD, and 0.954 (95%CI: 0.930-0.969) for PP. Individual ROC analysis revealed limited discriminative ability to predict 30-day mortality, with AUC values of 0.618 (95%CI: 0.493-0.744) for PA, 0.581 (95%CI: 0.453-0.709) for PD, and 0.567 (95%CI: 0.431-0.702) for PP. Using Youden’s index, optimal cut-points were determined: 8.7 cm² for PA (sensitivity 90.9%, specificity 38.8%) and 1.6 cm for PD (sensitivity 95.5%, specificity 26.9%).

Emergency TAE/TACE procedures and clinical outcomes

All 89 patients underwent emergency TAE or TACE procedures with 100% technical success. Clinical success was achieved in 73 patients (82%) while three patients (3%) required re-embolization within 48 hours. The median time from imaging diagnosis to embolization was 12.1 hours (IQR: 5.2-72.0). Active contrast extravasation was visualized in 18 patients (20%). Gelatin sponge was the primary embolic agent used in 81% of patients.

Post-embolization syndrome occurred in 35 patients (39%), with no cases of liver abscess, biloma, cholecystitis, or pancreatitis. The 30-day mortality rate was 25% (22/89), which was primarily due to tumor rebleeding (n = 12, 55%), liver failure (n = 8, 36%), and hepatorenal syndrome (n = 2, 9%). Overall mortality was 93% (83/89) with a median survival of 5.03 months (IQR: 3.12-6.24).

Predictors of 30-day mortality

Multivariable logistic regression analysis identified three independent predictors of 30-day mortality: Albumin ≤ 3.0 g/dL [adjusted odds ratio (OR) = 4.21, 95%CI: 1.39-14.12, P = 0.014], total bilirubin ≥ 2.5 mg/dL (adjusted OR = 5.48, 95%CI: 1.43-24.24, P = 0.016), and PA ≥ 8.7 cm² (adjusted OR = 7.34, 95%CI: 1.66-55.26, P = 0.020) (Table 2). Among the protrusion measurement methods, only PA demonstrated a significant association with mortality in both univariate (OR = 6.34, P = 0.018) and multivariable analyses.

Table 2 Univariate and multivariable logistic regression analysis for predictors of 30-day mortality.

Variable	Reference	Univariate analysis		Multivariate analysis
		Odds ratio (95%CI)	P value	Adjusted odds ratio (95%CI)	P value
Age ≥ 65 years	< 65 years	0.56 (0.15-1.74)	0.348
Sex (men)	Women	1.52 (0.43-7.17)	0.543
Hepatitis B positive	Negative	0.62 (0.22-1.66)	0.353
Hepatitis C positive	Negative	3.26 (1.07-9.86)	0.035
Presence of shock	Absence	1.79 (0.67-4.80)	0.240
Hemoglobin ≤ 10 g/dL	> 10 g/dL	2.75 (1.00-8.45)	0.060
Platelet count ≤ 100000/μL	> 100000/μL	0.51 (0.07-2.11)	0.406
Albumin ≤ 3.0 g/dL	> 3.0 g/dL	5.10 (1.83-15.88)	0.003	4.21 (1.39-14.12)	0.014
Total bilirubin ≥ 2.5 mg/dL	< 2.5 mg/dL	4.21 (1.34-13.48)	0.013	5.48 (1.43-24.24)	0.016
AST ≥ 100 U/L	< 100 U/L	6.67 (2.21-25.06)	0.002
ALT ≥ 100 U/L	< 100 U/L	2.16 (0.74-6.16)	0.150
ALP ≥ 150 U/L	< 150 U/L	3.83 (1.42-10.96)	0.009
INR ≥ 1.2	< 1.2	2.44 (0.85-8.13)	0.115
Creatinine ≥ 1.5 mg/dL	< 1.5 mg/dL	1.50 (0.42-4.76)	0.506
Alpha-fetoprotein ≥ 200 ng/mL	< 200 ng/mL	1.70 (0.64-4.75)	0.295
Multiple tumor nodules	Single	3.53 (1.07-16.14)	0.060
Tumor size ≥ 5 cm	< 5 cm	5.06 (0.92-94.67)	0.130
Portal vein thrombosis: Present	Absent	2.63 (0.99-7.19)	0.055
Child-Pugh class B and C	Class A	1.89 (0.72-5.17)	0.201
BCLC stage C	Stage A and B	3.16 (1.18-8.81)	0.023
Procedure type: TAE	TACE	5.55 (1.01-103.65)	0.108
Protrusion area ≥ 8.7 cm2	< 8.7 cm2	6.34 (1.66-41.81)	0.018	7.34 (1.66-55.26)	0.020
Protrusion diameter ≥ 1.6 cm	< 1.6 cm	7.71 (1.44-143.24)	0.054
Protrusion percentage ≥ 50%	< 50%	0.84 (0.29-2.28)	0.733

CI: Confidence interval; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; ALP: Alkaline phosphatase; INR: International normalized ratio; BCLC: Barcelona Clinic Liver Cancer; TAE: Transarterial embolization; TACE: Transarterial chemoembolization.

Development and performance of the PA-bilirubin-albumin scoring system

Based on the multivariable regression model, a prognostic scoring system incorporating the three independent predictors was developed (Table 3). The regression coefficients (β values) for PA ≥ 8.7 cm², total bilirubin ≥ 2.5 mg/dL, and albumin ≤ 3.0 g/dL were converted to weighted scores of 6 points, 5 points, and 4 points, respectively. The resulting PA-bilirubin-albumin (PBA) score ranges from 0 point to 15 points. The PBA scoring system achieved excellent discriminative performance with an AUC of 0.813 (95%CI: 0.707-0.919), which represented a substantial improvement of 0.195 over the best individual protrusion measurement. Bootstrap validation with 1000 iterations confirmed model robustness, with the bootstrap-corrected AUC (0.814, 95%CI: 0.695-0.912) virtually identical to the apparent AUC (0.813), indicating negligible overfitting.

Table 3 β coefficient and corresponding rounded risk score from results of the multivariable logistic regression analysis (n = 89).

Prognostic factors	β coefficient	Adjusted odds ratio (95%CI)	P value	Weighted risk score
Protrusion area ≥ 8.7 cm2	1.9937	7.34 (1.66-55.26)	0.020	6
Total bilirubin ≥ 2.5 mg/dL	1.7020	5.48 (1.43-24.24)	0.016	5
Albumin ≤ 3.0 g/dL	1.4367	4.21 (1.39-14.12)	0.014	4

CI: Confidence interval.

Risk stratification and survival analysis

Using empirically derived thresholds, patients were stratified into three risk groups with significantly different outcomes (Table 4). The low-risk group (PBA score ≤ 8, n = 58, 65%) had 8.6% 30-day mortality and a median survival of 6.18 months. The intermediate-risk group (PBA score 9-10, n = 20, 23%) had 40.0% mortality and a median survival of 2.12 months. The high-risk group (PBA score ≥ 11, n = 11, 12%) had 81.8% mortality and a median survival of 0.33 months. Kaplan-Meier survival analysis demonstrated statistically significant differences between the risk groups (log-rank test: P < 0.001) with the most pronounced difference between low-risk and high-risk groups (P < 0.001) (Figure 2).

Figure 2 Kaplan-Meier survival curves. Results demonstrate overall survival stratified by protrusion area-bilirubin-albumin score risk groups in patients with spontaneously ruptured hepatocellular carcinoma treated with emergency transarterial embolization/transarterial chemoembolization. PBA: Protrusion area-bilirubin-albumin.

Table 4 Summary of the 30-day mortality rates of different risk groups stratified according to the protrusion area-bilirubin-albumin score.

Risk group	Score	n (%)	30-day mortality (95%CI)	Median survival (months)
Low	≤ 8	58 (65)	8.6% (3.2%-18.1%)	6.18 (5.03-9.53)
Intermediate	9-10	20 (23)	40.0% (19.1%-63.9%)	2.12 (0.39-7.52)
High	≥ 11	11 (12)	81.8% (48.2%-97.7%)	0.33 (0.16-NA)

CI: Confidence interval; NA: Not available.

Comparative performance analysis

Figure 3 illustrates the ROC curve comparison of the PBA, MELD, and Child-Pugh scoring systems for 30-day mortality prediction. The PBA score demonstrated superior discriminative performance (AUC = 0.813, 95%CI: 0.707-0.919) compared to the MELD score (AUC = 0.672, 95%CI: 0.547-0.797) and Child-Pugh score (AUC = 0.594, 95%CI: 0.468-0.721). Statistical comparison using the DeLong’s test revealed significantly superior performance of the PBA score compared to both established systems (PBA vs MELD: P = 0.005; PBA vs Child-Pugh: P = 0.001) with AUC improvements of 0.141 and 0.219, respectively. The MELD and Child-Pugh scores showed no significant difference (P = 0.275).

Figure 3 Receiver operating characteristic curve. A comparison of protrusion area-bilirubin-albumin, model for end-stage liver disease, and Child-Pugh scores for 30-day mortality prediction. PBA: Protrusion area-bilirubin-albumin; MELD: Model for end-stage liver disease; AUC: Area under the receiver operating characteristic curve.

DISCUSSION

In this study, we developed and validated the PBA score, which is a novel predictive tool that combines clinical indicators of liver function, such as albumin and total bilirubin, with the morphological feature of tumor PA. Our results showed that protrusion measurements by themselves had little predictive value for 30-day mortality (AUC for PA = 0.618). However, incorporating these measurements into the PBA score significantly improved the predictive accuracy (AUC = 0.813). Importantly, the PBA score performed significantly better than established prognostic models, such as the Child-Pugh classification (P = 0.001) and MELD score (P = 0.005). Since the 30-day mortality rates ranged from 8.6% in the low-risk group to a substantially elevated 81.8% in the high-risk group, the score’s clinical relevance is highlighted by its capacity to stratify patients into distinct risk categories.

To our knowledge, this is the first systematic comparison of quantitative protrusion techniques in rHCC. The two-dimensional surface of the tumor, which extends beyond the protective liver parenchyma, is captured by PA, which is why it is preferable to one-dimensional metrics[7,11,14]. Interestingly, PP was not a predictor for 30-day mortality (OR = 0.84, P = 0.733), which is contrary to previous studies[11]. This discrepancy reflects fundamental differences in clinical objectives. Previous studies predicted rupture occurrence post-TACE in intact tumors. Our study examined post-rupture mortality, which is a distinctly different pathophysiological process. More importantly, percentage-based measurements can mislead across tumor sizes. A small tumor with a high PP may expose less absolute surface area than a large tumor with a lower percentage protrusion. Our PA assessment measures the absolute exposed surface area, which directly reflects the hemostatic difficulty during TAE/TACE treatments because larger exposed surfaces require more embolization and complete devascularization is difficult to achieve. The pathogenesis of rupture is directly linked by this anatomical measurement since a larger exposed surface area is more susceptible to ischemic injury and mechanical stress[2,14]. Additionally, the tumor’s intrinsic fragility, which is characterized by abnormal microvasculature and a thin capsule, exacerbates this external vulnerability[15-18]. These features are susceptible to rupture as a result of minor external trauma or increased intratumoral pressure. Our results quantitatively validate this concept by demonstrating that a PA of approximately 8.7 cm² is a substantial independent predictor of 30-day mortality (OR = 7.34). Nevertheless, the standalone predictive accuracy of PA was modest (AUC = 0.618) despite its exceptional inter-observer reliability (ICC = 0.981), which made it the most robust morphological predictor. This limitation emphasizes that mortality in rHCC is multifactorial and is influenced not only by tumor anatomy but also significantly by the underlying hepatic function[6,19-21].

The PBA score is based on the essential clinical principle that post-embolization survival relies primarily on achieving hemostasis and the remaining hepatic reserve to endure the acute insult of significant hemorrhage and shock[21-23]. The score accurately reflects these two drivers of outcome by integrating anatomical vulnerability (PA) with functional resilience (total bilirubin and albumin). Elevated total bilirubin signifies decreased hepatic excretory function and an increased risk of liver failure, whereas hypoalbuminemia indicates diminished synthetic ability and serves as a proven independent predictor of early mortality[22-24]. The specific cutoffs of total bilirubin ≥ 2.5 mg/dL and albumin ≤ 3.0 g/dL were adopted from thresholds previously validated in spontaneous rHCC, where both values independently predicted 30-day mortality, reflecting reduced physiological tolerance to acute hemorrhage[6]. The resultant score had exceptional discriminative ability (AUC = 0.813) that was validated as strong and not a consequence of statistical overfitting, which was shown by rigorous bootstrap validation, with the corrected AUC virtually identical to the apparent AUC, confirming negligible overfitting.

The PBA score’s superior performance underscores the inadequacy of general liver function assessments (MELD, Child-Pugh) in acute rHCC. The conventional scores were developed for chronic liver disease and do not include variables that account for the pathophysiology of acute hemorrhage or tumor-specific factors[6,10,25,26]. This limitation was evident in our results, which confirmed that risk stratification in this emergency setting requires a disease-specific tool. Previous tools for rHCC often relied on subjective assessments; however, our study is the first to systematically compare objective, quantitative ways to measure tumor protrusion. The PBA score is therefore unique because it combines the best of these objective measurements with standard liver function tests to create a more reliable basis for clinical decision-making.

In practice, the PBA score facilitates a concrete, evidence-based approach to clinical management stratified by risk. In high-risk patients (PBA score ≥ 11) whose 30-day mortality is exceptionally high at 81.8%, the utility of aggressive intervention is highly questionable, and a shift towards palliative-intent goals of care should be strongly considered[6,10,27]. Intermediate-risk patients (PBA score 9-10) require intensified surveillance and a tailored treatment strategy optimally established through a multidisciplinary team approach to carefully assess the risks and benefits of TAE/TACE[28]. Conversely, low-risk patients (PBA score ≤ 8) have a favorable prognosis with a death rate of 8.6%; therefore, the application of standard TAE/TACE procedures is recommended. The PBA score also serves as an effective bedside instrument to facilitate rapid risk evaluation using accessible data. The objectivity of the PBA score enhances the clarity of interactions with patients and families on collaborative decision-making. Ultimately, the application of this score enables a shift from standardized protocols to individualized treatment that guarantees patients receive care personalized to their specific risk while optimizing resource allocation[10].

This study offers a number of notable strengths, such as its novel systematic comparison of protrusion metrics, robust internal validation using bootstrap resampling, and high inter-observer reliability (ICC > 0.95). However, it is important to recognize the main limitations. The retrospective, single-center design with a sample size of only 89 patients represents a key methodological limitation. The small sample size resulted in wide CIs for some variables, which may compromise the precision and generalizability of the model. Multi-center external validation with a larger and more diverse cohort is therefore essential to confirm the robustness and broad applicability of the PBA score. Additionally, diagnostic performance was not stratified by sex due to unequal group sizes between men and women, which would have resulted in insufficient statistical power for meaningful subgroup comparisons. Future studies should stratify analyses by sex, as diagnostic performance and optimal cutoff values may differ between men and women. The prolonged 17-year data collection period includes possible temporal bias, as the advancement of embolization procedures and supporting care during this time is an unmeasured confounder. Furthermore, measurement of the PA can be subjective, particularly in cirrhotic livers, due to the reliance on hepatic contour extrapolation. Although our consensus training protocol guaranteed high inter-observer reliability, this may not accurately represent real-world variability in the absence of such specific training. The absence of external validation is the most significant limitation. Therefore, future large-scale, multicenter prospective studies are needed not only to externally validate the PBA score but also to evaluate its impact on long-term survival, re-bleeding rates, and patient-reported quality of life to definitively establish its broad clinical utility.

CONCLUSION

The PBA score is a highly effective, validated, and easy-to-use prognostic tool for patients with spontaneous rHCC. It provides superior risk stratification relative to traditional liver function scores by integrating the primary quantitative morphological characteristic of PA with hepatic function indicators. Its application can enhance a personalized strategy to control this life-threatening condition and aid in clinical decision-making. Future prospective, multi-center validation studies are warranted to confirm the external validity of the PBA score across diverse patient populations and clinical settings.

ACKNOWLEDGEMENTS

The authors thank Mrs. Jirawan Jayuphan of the Epidemiology Unit, Faculty of Medicine, Prince of Songkla University, for her assistance with the statistical analysis.