Pathological characteristics of hepatocellular carcinoma: A dual-center correlation analysis and validation

Abstract

BACKGROUND

High Ki-67 expression promotes epithelial-mesenchymal transition and angiogenesis, thereby increasing the risk of microvascular invasion (MVI) and satellite nodules (SN) in hepatocellular carcinoma (HCC). The synergistic interplay among these pathological characteristics is critical for prognostic assessment; however, robust multicenter validation remains lacking.

AIM

To validate the interrelationships within the “Edmondson grade-MVI-SN-Ki-67” aggressive pathological cluster and to evaluate the predictive value of Ki-67 for SN.

METHODS

A retrospective analysis was conducted on 223 patients with HCC who underwent surgical resection at two independent clinical centers. Spearman correlation and receiver operating characteristic curve analyses were performed.

RESULTS

The pathological features demonstrated strong intercorrelations, with MVI and SN exhibiting the strongest association (r_s = 0.637, odds ratio = 75.833). Ki-67 displayed optimal and consistent predictive performance for SN positivity across both centers (area under the curve = 0.792). A uniform Ki-67 cutoff value of 32.5% was identified, yielding high specificity (0.813). The predictive performance of Ki-67 for SN showed no significant difference between centers (area under the curve = 0.780 vs 0.835, P = 0.372). In multivariate analysis, Ki-67 remained an independent predictor of SN after adjusting for study center (odds ratio = 1.058, P < 0.001).

CONCLUSION

Key pathological characteristics of HCC constitute a tightly linked aggressive cluster. Ki-67 serves as a reproducible predictive biomarker for SN positivity, providing validated multicenter pathological evidence for postoperative risk stratification.

Key Words: Hepatocellular carcinoma; Edmondson-Steiner grading; Ki-67; Microvascular invasion; Satellite nodule; Biomarker

Core Tip: This dual-center retrospective study validates a tightly linked aggressive pathological cluster (Edmondson grade, microvascular invasion, satellite nodules, Ki-67) in hepatocellular carcinoma. We identify Ki-67 as a core, reproducible biomarker, with a validated cutoff of 32.5% demonstrating high specificity (0.813) for predicting satellite nodules positivity across two independent centers. These findings provide multicenter pathological evidence to inform postoperative risk stratification and adjuvant therapy decisions.

INTRODUCTION

Hepatocellular carcinoma (HCC) ranks as the sixth most common malignancy globally and the third leading cause of cancer-related mortality, accounting for over 900000 new cases and 830000 deaths annually[1]. A 55% increase in incidence is projected between 2020 and 2040[2]. Eastern Asia, including China, represents the region with the highest disease burden, underscoring the urgent need for continued research advancement.

Comprising 75%-85% of primary liver cancers[3], HCC exhibits substantial biological heterogeneity that complicates treatment planning. Although clinical diagnosis relies primarily on imaging[4], approximately 30% of atypical cases require pathological confirmation[5,6]. Despite the availability of multimodal treatments, the 5-year recurrence rate remains as high as 70%[7], highlighting the critical importance of precise pathological evaluation for risk assessment[8].

Tumor differentiation, microvascular invasion (MVI), satellite nodules (SN), and Ki-67 promote HCC progression through shared signaling pathways: Poorly differentiated tumors activate the Wnt/β-catenin pathway to enhance cell proliferation (reflected in high Ki-67 expression)[6,9], while processes such as endothelial-to-mesenchymal transition and angiogenesis facilitate the development of MVI and SN[9,10].

Current research in this field is limited by three major shortcomings[11,12]. First, correlation analyses remain fragmentary: Studies have predominantly examined Edmondson-Steiner grading, MVI, and Ki-67 in isolation[8], neglecting their synergistic relationships. Second, non-invasive prediction models have inherent limitations; although triphasic computed tomography radiomics demonstrate promising results for predicting Edmondson grading, MVI, and SN status [area under the curve (AUC) = 0.810-0.895][13], these models rely on complex feature extraction and overlook biological markers such as Ki-67. Third, single-center bias poses a significant challenge: Regional etiological variations (e.g., hepatitis B virus-associated HCC prevalence exceeding 80% in southern China[14]) may result in overestimated pathological correlations, thereby limiting generalizability.

This dual-center retrospective study analyzed 223 patients to validate, for the first time, the cross-center consistency of the “Edmondson-Steiner grade-MVI-SN-Ki-67” aggressive pathological cluster. Through Spearman correlation, logistic regression, receiver operating characteristic (ROC) curve analysis, and stratified statistics, this study quantified pathological feature associations and Ki-67 predictive performance, providing multicenter evidence for postoperative stratified treatment strategies in HCC.

MATERIALS AND METHODS

Study population

A retrospective analysis was conducted on 223 patients with pathologically confirmed HCC who underwent surgical resection between January 2018 and July 2025 at the Third Affiliated Hospital of Guangxi Medical University (Center A) and the First People’s Hospital of Qinzhou (Center B). Inclusion criteria were as follows: (1) Postoperative pathological confirmation of HCC according to the World Health Organization 2019 classification; (2) Curative hepatic resection (R0); and (3) Complete clinicopathological data, including Edmondson-Steiner grading, MVI, SN, and Ki-67 status. Exclusion criteria included: (1) Combined hepatocellular-cholangiocarcinoma or mixed carcinoma; (2) Receipt of any preoperative anticancer therapy (e.g., transcatheter arterial chemoembolization, radiotherapy, or targeted therapy); (3) Poor-quality pathological sections; and (4) Incomplete clinical follow-up data. The final study cohort comprised 223 patients with HCC after excluding ineligible cases (Figure 1).

Figure 1 Flowchart of patient selection and cohort formation. HCC: Hepatocellular carcinoma.

Ethical approval

This study was conducted in compliance with the Declaration of Helsinki and was approved by the Medical Ethics Committee of the Medical Ethics Committee of Nanning Second People’s Hospital (the Third Affiliated Hospital of Guangxi Medical University), approval No. Y2025341. The requirement for informed consent was waived owing to the retrospective design and the use of anonymized data.

Definition of pathological characteristics

Ki-67 expression was classified using a 20% cutoff: > 20% was designated as high expression and ≤ 20% as low expression[15,16]. The 20% cutoff was based on previous research widely accepted for assessing proliferative activity in HCC[11]. Additionally, ROC analysis identified an optimal cutoff of 32.5% for SN prediction, which offers higher specificity (0.813). The prognostic value of specific Ki-67 thresholds in HCC has been demonstrated in recent studies[17]. Edmondson-Steiner grading was dichotomized as follows: 0 = grade I + II (well/moderately differentiated) and 1 = grade III + IV (poorly differentiated/undifferentiated).

MVI grading followed the Chinese Expert Consensus on Pathological Diagnosis of HCC MVI[18]: M0, absence of MVI; M1, presence of ≤ 5 tumor emboli within peritumoral liver tissue (≤ 1 cm from the tumor border); M2, either > 5 tumor emboli or vascular invasion detected > 1 cm beyond the tumor margin. This grading system is consistent with the World Health Organization 2019 Classification of Digestive System Tumors[19]. MVI status was defined as positive (M1/M2) or negative (M0). SN status was defined as positive (presence of SN) or negative (absence).

Pathological sections were processed using standardized hematoxylin and eosin staining and Ki-67 immunohistochemistry protocols. Ki-67 scoring was performed manually using the “hotspot counting method”. For each specimen, two independent pathologists, blinded to clinical data, selected five representative high-power fields (× 400 magnification) within the areas of highest Ki-67 positivity (hotspots). The percentage of positively stained tumor cell nuclei among the total number of tumor cells was recorded for each field, and the average percentage across the five fields was used as the final Ki-67 index. This approach ensures consistency with international guidelines[20] and minimizes sampling bias. Interobserver agreement, measured by Kappa values, reached 0.870 for Edmondson-Steiner grading, 0.890 for MVI, and 0.850 for SN status (all P < 0.001), meeting the criterion of excellent agreement (kappa ≥ 0.81).

Statistical analysis

Data analysis was performed using SPSS 26.0 (IBM Corp., United States) and R 4.2.0 (R Foundation for Statistical Computing). Continuous variables with non-normal distributions were expressed as median (interquartile range) [media (P25, P75)] and compared using the Mann-Whitney U test. Categorical variables were presented as n (%) and compared using the χ² test or Fisher’s exact test, as appropriate. Spearman’s rank correlation analysis assessed relationships between pathological characteristics, with correlation strength interpreted as follows: |r_s| < 0.3 (weak), 0.3 ≤ |r_s| < 0.5 (moderate), and |r_s| ≥ 0.5 (strong)[21]. Multivariable logistic regression models, adjusted for study center, assessed the independent predictive value of Ki-67 for SN positivity, poor Edmondson-Steiner differentiation (grade III-IV), and MVI positivity. ROC analysis was employed to evaluate the predictive performance of Ki-67, with the AUC calculated. The optimal cutoff value was determined by maximizing Youden’s index (sensitivity + specificity - 1). Center interactions were assessed using likelihood ratio tests. A two-sided P value < 0.05 was considered statistically significant.

Sample size consideration

As a retrospective dual-center validation study, all eligible patients during the specified study period were included (a convenience sample). The precision of our estimates, particularly for the primary finding, is reflected in the narrow 95% confidence intervals (CIs) (e.g., for the AUC of Ki-67 predicting SN: 0.733-0.852). Although the sample size provided adequate power for detecting moderate correlations in the pooled analysis, the uneven distribution between centers (Center A: n = 157; Center B: n = 66) may affect the precision of center-specific estimates, particularly for subgroup analyses.

RESULTS

Baseline characteristics and inter-center comparison

A total of 223 patients with HCC were enrolled: 157 (70.4%) from Center A and 66 (29.6%) from Center B. The overall cohort had a median age of 55.1 years (interquartile range: 48.0-65.0), with a male predominance (188 males, 84.3%). The median tumor diameter was 4.0 cm (interquartile range: 3.0-6.2). Comparative analysis revealed no significant differences between the two centers in terms of age, sex, tumor diameter, or key pathological characteristics, including Edmondson-Steiner grading, MVI grading, SN status, and Ki-67 expression (all P ≥ 0.05; Table 1), indicating well-balanced baseline characteristics.

Table 1 Baseline clinicopathological characteristics by center, n (%).

Characteristics	Center A (n = 157)	Center B (n = 66)	Test statistic	P value
Age (year), mean ± SD	56.0 ± 12.1	53.9 ± 10.0	t = 1.233	0.219
Sex (male/female)	136/21	52/14	χ2 = 2.156	0.142
Edmondson grading (low/high)	66/91	31/35	χ2 = 0.460	0.498
MVI positive	111 (70.7)	41 (62.1)	χ2 = 1.576	0.209
SN present	69 (44.0)	31 (46.9)	χ2 = 1.171	0.679
Ki-67 high expression	83 (52.9)	31 (46.9)	Z = -1.639	0.104

The Mann-Whitney U test was used to compare Ki-67 expression (non-normally distributed) between centers. MVI: Microvascular invasion; SN: Satellite nodule.

Correlation analysis of pathological characteristics

Association between Edmondson-Steiner grading and MVI/SN status: MVI positivity was significantly higher in Edmondson grade III-IV tumors than in grade I-II tumors (89.6% vs 51.6%, χ² = 36.667, P < 0.001). Spearman correlation analysis confirmed a moderate positive correlation between Edmondson-Steiner grading and MVI grading (r_s = 0.492, P < 0.001). Similarly, the SN positivity rate was significantly elevated in the poorly differentiated/undifferentiated group (73.2% vs 23.0%, χ² = 55.796, P < 0.001), with a moderate correlation (r_s = 0.500, P < 0.001).

Association between MVI and SN: A highly significant association was observed between MVI grading and SN status (χ² = 89.436, P < 0.001). The SN positivity rate was 87.5% (63/72) in M2 patients, 38.6% (31/80) in M1 patients, and 8.5% (6/71) in M0 patients. Using M0 as the reference, the odds ratio (OR) for SN positivity in M2 patients was 75.833 (95%CI: 25.508-225.45). Spearman correlation demonstrated a strong positive correlation (r_s = 0.637, P < 0.001) (Table 2).

Table 2 Associations between aggressive pathological features.

Variables	χ2 (df)	P value	Effect size	OR (95%CI)
Ed grading × MVI status	36.667 (1)	< 0.001	Cramer’s V = 0.405	8.165 (3.888-17.145)
Ed grading × SN status	55.796	< 0.001	Cramer’s V = 0.500	9.134 (4.956-16.835)
Ed grading × Ki-67	37.347	< 0.001	Cramer’s V = 0.409	6.833 (3.557-13.129)
MVI status × SN status	55.776	< 0.001	Cramer’s V = 0.500	17.557 (7.152-43.099)
MVI status × Ki-67	31.416	< 0.001	Cramer’s V = 0.375	5.331 (2.903-9.792)
SN status × Ki-67	34.274	< 0.001	Cramer’s V = 0.392	6.036 (3.211-11.349)

Ed: Edmondson-Steiner; MVI: Microvascular invasion; SN: Satellite nodule; OR: Odds ratio; CI: Confidence interval.

Association of Ki-67 with other pathological features: The rate of high Ki-67 expression was significantly elevated in poorly differentiated tumors (84.5% vs 44.4%, χ² = 37.347, P < 0.001; OR = 6.833), in MVI-positive patients (74.3% vs 35.2%, χ² = 31.416, P < 0.001), and in SN-positive patients (83.0% vs 44.7%, χ² = 34.274, P < 0.001). Correlations were moderate to strong (r_s = 0.474, 0.479, and 0.507, respectively; all P < 0.001).

Predictive performance of Ki-67 for aggressive pathological characteristics

ROC analysis demonstrated that Ki-67 exhibited excellent predictive performance for SN positivity (AUC = 0.792, 95%CI: 0.733-0.852, P < 0.001), with an optimal cutoff of 32.5% (sensitivity 0.640, specificity 0.813). For MVI positivity, the AUC was 0.778 (cutoff 27.5%, sensitivity 0.730, specificity 0.718). For poor Edmondson-Steiner differentiation (grade III-IV), the AUC was 0.774 (cutoff 32.5%, sensitivity 0.629, specificity 0.794) (Table 3).

Table 3 Predictive performance of Ki-67 expression.

Outcome	AUC	Sensitivity	Specificity	Cutoff	Youden index
SN positive	0.792	0.640	0.813	32.5	0.453
MVI positive	0.778	0.730	0.718	27.5	0.449
Ed grading (III-IV)	0.774	0.629	0.794	32.5	0.423

AUC: Area under the curve; SN: Satellite nodule; MVI: Microvascular invasion; Ed: Edmondson-Steiner.

Validation of predictive performance across two centers

DeLong’s test was used to compare the predictive performance of Ki-67 between the two centers (Figure 2). For SN positivity, no significant difference in AUC was observed (0.780 vs 0.835, P = 0.372), and the optimal cutoff remained consistent at 32.5% in both centers. For MVI positivity, the AUC difference (0.763 vs 0.871) did not reach statistical significance (P = 0.098). For poor Edmondson-Steiner differentiation, a significant difference was detected (0.731 vs 0.895, P = 0.002), with divergent optimal cutoffs (27.5% in Center A and 32.5% in Center B).

Figure 2 Receiver operating characteristic curves for Ki-67 predicting. A: Satellite nodule positivity; B: Microvascular invasion positivity; C: Poor Edmondson-Steiner differentiation; D: Center-stratified receiver operating characteristic curve comparison. ROC: Receiver operating characteristic; SN: Satellite nodule; AUC: Area under the curve; CI: Confidence interval; MVI: Microvascular invasion; Ed: Edmondson-Steiner.

Multivariate logistic regression analysis

A multivariate logistic regression model was constructed with SN positivity as the dependent variable. After incorporating Ki-67 (as a continuous variable) and study center (as a categorical variable) as covariates, and adjusting for the potential confounding effect of study center, Ki-67 remained a strong and independent predictor of SN positivity (OR = 1.058, 95%CI: 1.032-1.085, P < 0.001).

DISCUSSION

This dual-center study validates the strong interrelationships within a key aggressive pathological cluster in HCC and identifies Ki-67 as a reproducible predictive biomarker for SN. The exceptionally strong association between MVI and SN (r_s = 0.637, OR = 75.833) underscores their likely shared underlying metastatic mechanisms, potentially involving processes such as endothelial-to-mesenchymal transition and angiogenesis, which promote tumor cell dissemination and MVI[8-10,22].

Ki-67 as a core predictive biomarker with cross-center consistency

The central finding of this study is the validation of Ki-67 as a predictive marker for SN. Ki-67 showed strong correlations with SN status (r_s = 0.507) and demonstrated excellent predictive performance (AUC = 0.792) with a high-specificity cutoff of 32.5%. Critically, its predictive performance for SN exhibited remarkable consistency across the two independent centers, with no significant difference in AUC (P = 0.372) and an identical optimal cutoff value (32.5%). This cross-center reproducibility represents a major strength of this study, substantially enhancing the generalizability and potential clinical utility of these findings. Multivariate analysis further confirmed Ki-67 as an independent predictor of SN after adjusting for study center (OR = 1.058, 95%CI: 1.032-1.085, P < 0.001). As Ki-67 was treated as a continuous variable, an OR of 1.058 indicates that for every 1% increase in Ki-67 expression, the odds of SN positivity increase by 5.8%, highlighting the significant clinical impact of Ki-67 on SN prediction[23].

The strong MVI-SN association suggests a biological cascade: MVI may act as an upstream event, providing a pathway for Ki-67-high tumor cells to disseminate and form SN[9,20]. This interpretation aligns with previous research demonstrating that MVI is a critical precursor to intrahepatic metastasis in HCC[22,24]. The synergistic interplay between MVI and Ki-67 supports a potential “MVI-Ki-67-SN” cascade.

Beyond mechanistic insights, Ki-67 demonstrates superior clinical utility compared with existing diagnostic approaches. Its discriminatory power for SN (AUC = 0.792) surpasses that of traditional serum biomarkers such as alpha-fetoprotein (reported AUCs typically ranging from 0.60 to 0.70)[23] and is comparable to advanced MRI-based radiomic models designed to predict Ki-67 expression itself (AUC range 0.75-0.85)[6,25]. Furthermore, this performance is in line with that of ultrasound-based models for predicting MVI (AUC range 0.75-0.80) reported in recent meta-analyses[26], suggesting that immunohistochemistry offers a cost-effective and readily available alternative with similar discriminatory power. Crucially, Ki-67 assessment offers distinct practical advantages: It follows standardized “hotspot counting” methodologies that have been independently validated for reliability[27], and this study exhibited excellent interobserver concordance (kappa = 0.89). This contrasts with the greater technical variability and computational complexity of radiomic feature extraction[13,14], while the cost-effectiveness of immunohistochemistry further supports its viability in diverse clinical settings[23].

Addressing context-dependent performance for tumor grading

A notable secondary finding is that while Ki-67 also correlated with and predicted poor Edmondson-Steiner differentiation, this relationship was not consistent across centers. The predictive performance (AUC) differed significantly (P = 0.002), and the optimal cutoff varied (27.5% vs 32.5%). This discrepancy suggests that the link between the proliferation index and histological grade may be more susceptible to inter-center assessment variability or etiological heterogeneity. It underscores that the most robust and actionable finding from our multicenter design is the Ki-67 cutoff for SN prediction rather than for grading. This observation aligns with and extends previous research that has primarily focused on predicting grade or MVI in isolation, often using single-center data[11,12,16].

Clinical implications and practical utility

The proposed Ki-67 cutoff of 32.5% offers a practical, high-specificity tool for identifying patients with HCC who have a high likelihood of SN positivity - a feature strongly linked to postoperative recurrence[7,28]. In clinical practice, a Ki-67 index exceeding this threshold could serve as an adjunctive criterion to guide adjuvant therapy decisions, such as recommending postoperative transarterial chemoembolization or enrolling patients in clinical trials for targeted agents, particularly when the presence of MVI is equivocal on histopathology. It may also prompt a strategy of more intensive postoperative surveillance. Integrating a “Ki-67 + MVI + SN” triple assessment into standard pathological reporting could significantly refine postoperative risk stratification, especially within high-risk subgroups defined by current consensus guidelines[18]. From a health economics perspective, Ki-67 immunohistochemistry is a cost-effective test, enhancing its feasibility for widespread clinical adoption. Furthermore, a Ki-67 index > 32.5% may indicate tumor biology conducive to micrometastasis, suggesting potential benefit from adjuvant therapies such as transarterial chemoembolization or targeted agents[3,28], and could inform strategies to overcome potential resistance to immunotherapies[3].

Innovations and limitations

While the individual associations among Edmondson grade, MVI, and SN have been previously described, the principal and distinguishing contribution of this work is the first dual-center validation of the robustness and cross-center consistency of these interrelationships. More importantly, we have, for the first time, identified and validated a reproducible Ki-67 cutoff (32.5%) for predicting SN positivity that demonstrates remarkable consistency across two independent cohorts. This finding substantially enhances the generalizability and potential clinical applicability of Ki-67 as a practical biomarker for postoperative risk stratification.

However, several limitations must be acknowledged. First, both participating centers are located in southern China, where HCC etiology is predominantly hepatitis B virus-related. This represents a significant limitation, as the findings may not be directly generalizable to patient populations with different etiological backgrounds, such as those with hepatitis C virus-associated HCC, alcohol-related liver disease, or metabolic dysfunction-associated steatotic liver disease. Validation in geographically and etiologically diverse cohorts is therefore essential. Second, the sample size for certain subgroups (e.g., M2, Center B) was limited, affecting the precision of some estimates. The uneven distribution between centers (Center A: n = 157; Center B: n = 66) may have contributed to the observed discrepancies in predicting Edmondson-Steiner grading across centers, and this should be considered when interpreting the robustness of center-specific analyses. Third, a major limitation of this study is the absence of direct clinical outcome data, such as recurrence-free survival or overall survival. While we establish Ki-67 as a robust predictor of aggressive pathological features - which are well-established surrogate endpoints for poor prognosis - we acknowledge that this does not substitute for direct validation against survival outcomes. Therefore, the ultimate clinical utility of this 32.5% cutoff will be further solidified by prospective studies with long-term survival follow-up. Such validation would pave the way for its integration into routine clinical decision-making and postoperative risk stratification algorithms.

Future directions

Future research should focus on: (1) Correlating the validated Ki-67 cutoff (32.5%) with survival outcomes to establish its definitive prognostic role; (2) Developing integrated models that combine this readily available pathological marker with preoperative imaging predictors - such as the triphasic computed tomography radiomics model from our previous work (AUC for SN = 0.829)[13] or MRI-based predictors of Ki-67[6,25] - to create robust non-invasive preoperative risk assessment tools; (3) Exploring the molecular mechanisms linking high Ki-67 expression to SN formation, such as the Wnt/β-catenin and VEGF signaling pathways[9,10]; and (4) Conducting larger, geographically diverse multicenter studies to validate the universal applicability of this cutoff.

CONCLUSION

In conclusion, this dual-center study validates the strong intercorrelations among Edmondson grade, MVI, SN, and Ki-67 in HCC, confirming them as an aggressive pathological cluster. Ki-67 is a reproducible and independent predictor of SN, with a consistent cutoff of 32.5% across centers. These findings provide multicenter pathological evidence to support postoperative risk stratification and guide adjuvant therapy decisions in HCC.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge Xin-Chi Liu and Qin-Ping Zhao for their assistance in clinical data collection and screening, and Fei He for verifying computed tomography scanning parameters. These collaborators do not meet the ICMJE criteria for authorship. The authors are solely responsible for the content of this manuscript.