Abstract
Purpose
Type 2 diabetes mellitus (DM) may influence invasive tumor phenotypes in hepatocellular carcinoma (HCC). We evaluated associations of pre-existing DM with tumor budding (TB), microvascular invasion (MVI), and postoperative outcomes.
Patients and Methods
This single-center retrospective cohort included 924 patients with pathologically confirmed HCC who underwent curative-intent hepatectomy from 2015 to 2020. Two blinded pathologists assessed TB and MVI. Logistic regression evaluated pathological associations, and Kaplan–Meier and Cox models evaluated recurrence-free survival (RFS) and overall survival (OS). Glycemic and treatment analyses in patients with DM were exploratory and used variable-specific available cases.
Results
Among 924 patients, 124 (13.4%) had DM, 314 (34.0%) were TB-positive, and 493 (53.4%) were MVI-positive. DM was associated with TB-positive status (adjusted odds ratio [OR], 3.508; 95% confidence interval [CI], 2.287–5.381; P < 0.001) and MVI-positive status (adjusted OR, 3.364; 95% CI, 2.083–5.434; P < 0.001). In the primary overlap-weighted PS analysis, DM was not associated with RFS (hazard ratio [HR], 1.453; 95% CI, 0.778–2.714; P = 0.240) or OS (HR, 1.127; 95% CI, 0.556–2.284; P = 0.740). TB remained associated with RFS but not OS in expanded Cox models, whereas MVI was associated with both outcomes.
Conclusion
Pre-existing DM was associated with TB-positive and MVI-positive status after hepatectomy for HCC. The primary PS analysis did not provide evidence of an independent DM survival association. Glycemic-control and treatment findings were exploratory and require prospective external validation.
Keywords: hepatocellular carcinoma;, hepatectomy, type 2 diabetes mellitus, tumor budding, microvascular invasion, postoperative recurrence, glycemic control
Graphical Abstract
Introduction
Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality worldwide. It commonly develops in the setting of chronic liver injury, including viral hepatitis, cirrhosis, metabolic dysfunction-associated steatotic liver disease, and other inflammatory liver disorders.1,2 As the global burden of metabolic disease increases, type 2 diabetes mellitus (DM) has emerged as an important risk factor for HCC incidence and cancer-related mortality.3–6 Beyond its role in carcinogenesis, DM may affect tumor biology through hyperglycemia, insulin resistance, chronic inflammation, altered lipid metabolism, and remodeling of the tumor microenvironment.
Tumor budding (TB) and microvascular invasion (MVI) represent distinct aspects of invasive tumor behavior. TB consists of isolated tumor cells or small clusters at the invasive front and is considered a histological manifestation of local invasion and epithelial–mesenchymal transition-like behavior.7,8 In the present study, HbA1c categories were defined with reference to contemporary diabetes guidance.9 Published HCC cohorts have used heterogeneous binary, threshold-based, and graded TB protocols, and no HCC-specific consensus assessment system has been validated.10 MVI reflects microscopic vascular dissemination of tumor cells and predicts early recurrence and poor survival after hepatectomy.11,12 Both features are therefore relevant to postoperative risk stratification in HCC.
Although DM, TB, and MVI have each been linked to aggressive HCC, whether metabolic comorbidity is associated with invasive pathological phenotypes remains uncertain. In particular, evidence is limited on the relationships of pre-existing DM, glycemic control, and antidiabetic treatment status with TB and MVI after resection. We therefore evaluated these associations in a resected HCC cohort and examined how TB and MVI related to postoperative RFS and OS.
Materials and Methods
Ethics
The study protocol was approved by the Ethics Review Committee of the First Affiliated Hospital of Xi’an Jiaotong University (approval no. XJTU1AF2015LSL-057). All patients were treated at the First Affiliated Hospital; patient identification, surgery, pathological specimen processing and review, clinical follow-up, source-data collection, and the original research work were performed at that institution. Given the retrospective nature of the study and the use of existing de-identified clinical and pathological data, the requirement for written informed consent was waived by the Ethics Committee. The study was conducted in accordance with the Declaration of Helsinki.
Patients and Tissue Preparation
We retrospectively screened patients with histopathologically confirmed primary HCC who underwent curative-intent hepatectomy at the First Affiliated Hospital of Xi’an Jiaotong University from 2015 to 2020. The final analysis included 924 eligible patients. Curative-intent hepatectomy was operationally defined as complete removal of all tumors identified by preoperative imaging and intraoperative exploration, with no macroscopic residual intrahepatic or extrahepatic disease. R0 resection was the intended surgical goal but was not confirmed for every patient; both anatomical and non-anatomical resections were eligible. The extent of hepatectomy was recorded by the number of Couinaud segments removed (≤2 vs >2 segments). The same cohort was used to evaluate TB, MVI, DM status, glycemic variables, and survival outcomes (Figure 1).
Figure 1.
Patient-selection flowchart. Among 1153 patients with primary hepatocellular carcinoma (HCC) who underwent curative-intent hepatectomy from 2015 to 2020, 229 were excluded, leaving 924 eligible patients. Tumor budding (TB) was classified as present or absent using the morphological definition proposed by the international tumor budding consensus conference (ITBCC), and microvascular invasion (MVI) was assessed as a binary pathological feature. The final cohort comprised 124 patients with type 2 diabetes mellitus (DM) and 800 without DM.
Eligible patients were 18 years or older and had pathologically confirmed primary HCC, curative hepatectomy, available clinicopathological and follow-up data, and hematoxylin and eosin (H&E)-stained sections available for TB and MVI assessment. Exclusion criteria were recurrent HCC, combined hepatocellular–cholangiocarcinoma, another concurrent malignancy, missing key clinicopathological variables required for the primary analyses, or unavailable pathological slides. Patients with missing key variables were excluded during cohort construction (Figure 1).
Definition of Type 2 DM and Glycemic Variables
Type 2 diabetes mellitus was defined as a documented pre-existing diagnosis recorded in the medical record before hepatectomy. The diagnosis had been established by the treating physician according to the applicable Chinese diagnostic criteria. Antidiabetic medication use was recorded separately as treatment status and was not used as the sole criterion for DM classification.
Among patients with DM, glycemic status was evaluated using available preoperative HbA1c values and preoperative fasting blood glucose. HbA1c was categorized as well controlled (≤7.0%), moderately controlled (>7.0% to <9.0%), or poorly controlled (≥9.0%).9,13 Preoperative fasting blood glucose was defined as the arithmetic mean of all eligible fasting venous plasma glucose measurements obtained during the preoperative hospitalization before hepatectomy. At least two measurements were required for each patient; the number and timing of measurements otherwise reflected routine clinical care. Postoperative, nonfasting, and random measurements were excluded. Antidiabetic treatment status was classified as treated or untreated; treated patients received oral glucose-lowering agents, insulin, or both before surgery. Reliable diabetes duration was unavailable for a substantial proportion of patients and was therefore not reconstructed, imputed, or analyzed. Treatment-related analyses were exploratory because detailed drug class, duration, adherence, and longitudinal glycemic trajectories were unavailable for many patients.
Pathological Evaluation of TB and MVI
The ITBCC system was developed for colorectal carcinoma. Because HCC currently lacks a validated consensus for hotspot selection, fixed-area counting, or low/intermediate/high thresholds, we adopted only the ITBCC morphological definition: an isolated tumor cell or a cluster of up to four tumor cells at the invasive front. TB was prespecified as absent (0 buds identified) or present (at least 1 bud identified), without applying colorectal quantitative thresholds or claiming an HCC-validated ITBCC grade.10 Existing HCC studies have used heterogeneous binary, threshold-based, and graded approaches; the binary operationalization was selected for this exploratory retrospective analysis to avoid imposing an unvalidated HCC-specific cutoff.
Two pathologists independently reviewed all slides while blinded to DM status, glycemic variables, treatment status, and clinical outcomes. Interobserver agreement for the initial independent binary TB ratings, before consensus review, was quantified using unweighted Cohen’s kappa, calculated as κ=(Pₒ−Pₑ)/(1−Pₑ), where Pₒ is observed agreement and Pₑ is agreement expected by chance. The Cohen’s kappa coefficient for TB was 0.80, indicating substantial agreement. Discrepant TB and MVI assessments were jointly reviewed until consensus was reached, and the consensus ratings were used for the final analytic dataset. MVI was defined as tumor-cell nests or microscopic tumor emboli within endothelial cell-lined vascular spaces. MVI was analyzed as a binary variable and was not further graded.
Outcomes
Postoperative surveillance was generally performed every 3 months during the first 2 years and every 6 months thereafter. Follow-up included clinical assessment, liver-function tests, serum alpha-fetoprotein, abdominal ultrasonography, and contrast-enhanced computed tomography or magnetic resonance imaging; chest imaging and additional examinations were obtained when clinically indicated. Recurrence was defined as a new intrahepatic or extrahepatic lesion showing imaging characteristics diagnostic of HCC on contrast-enhanced cross-sectional imaging, or pathological confirmation when imaging was indeterminate and tissue was available. The recurrence date was the first examination confirming recurrence. RFS was defined as the interval from hepatectomy to the first documented recurrence; death without documented recurrence was not treated as an RFS event and was censored at death, while patients alive without recurrence were censored at the last disease assessment. OS was measured from hepatectomy to death from any cause, with surviving patients censored at last contact.
Statistical Analysis
Categorical variables were summarized as frequencies and percentages. Continuous variables were summarized using the mean and standard deviation (SD) or the median and interquartile range (IQR), as appropriate. Groups were compared using the chi-square test or Fisher’s exact test for categorical variables and Student’s t-test or the Mann–Whitney U-test for continuous variables. The natural logarithm of the serum alpha-fetoprotein concentration (ln[AFP], ng/mL) was calculated. Expanded multivariable models incorporated age, sex, DM, TNM stage, Child-Pugh class, underlying liver disease, tumor differentiation, cirrhosis, other metabolic comorbidities, antiviral therapy among patients with viral hepatitis, tumor number, maximum tumor diameter, obesity, extent of hepatectomy, macrovascular invasion, and portal hypertension. Child-Pugh class and TNM stage were analyzed at the available binary resolution (A vs B/C and I/II vs III/IV). The sequential Cox Model 1 excluded TB and MVI, whereas Model 2 added TB and MVI. Adjusted models used the complete 924-patient analysis set; antiviral therapy was evaluated among patients with viral hepatitis only (N = 802). All estimates were interpreted as associations rather than causal effects.
Binary logistic regression models estimated odds ratios (ORs) and 95% confidence intervals (CIs) for associations with TB+ and MVI+ status. The fully adjusted logistic models used the expanded covariate set described above; results are shown in Figures 2 and 3. Among patients with DM, HbA1c category, treatment status, and preoperative fasting glucose were evaluated in separate variable-specific available-case logistic models. HbA1c analyses included 72 patients, treatment-status analyses included 102 patients, and fasting-glucose analyses included all 124 patients with DM. For these subgroup analyses, no missing-value imputation was performed; missing subgroup values were not coded as a separate category. Survival outcomes were analyzed using Kaplan–Meier curves, log-rank tests, and Cox proportional hazards models. The proportional hazards assumption was assessed using Schoenfeld residuals and, when appropriate, log-minus-log survival plots. All tests were two-sided, and P < 0.05 indicated statistical significance. Conventional analyses were performed using IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, USA); PS analyses were implemented in Python version 3.12 (Python Software Foundation).
Figure 2.
Multivariable logistic regression analysis of factors associated with tumor-budding (TB)-positive status in the overall cohort. Odds ratios (ORs) are shown with 95% confidence intervals (CIs). The expanded model included age, sex, type 2 diabetes mellitus (DM), tumor-node-metastasis (TNM) stage, Child-Pugh class, underlying liver disease, tumor differentiation, cirrhosis, other metabolic comorbidities, antiviral therapy among patients with viral hepatitis, tumor number, maximum tumor diameter, obesity, extent of hepatectomy, macrovascular invasion, and portal hypertension.
Figure 3.
Multivariable logistic regression analysis of factors associated with microvascular-invasion (MVI)-positive status in the overall cohort. Odds ratios (ORs) are shown with 95% confidence intervals (CIs). The expanded model included age, sex, type 2 diabetes mellitus (DM), tumor-node-metastasis (TNM) stage, Child-Pugh class, underlying liver disease, tumor differentiation, cirrhosis, other metabolic comorbidities, antiviral therapy among patients with viral hepatitis, tumor number, maximum tumor diameter, obesity, extent of hepatectomy, macrovascular invasion, and portal hypertension.
To address measured baseline imbalance between patients with and without DM, a PS for DM was estimated using multivariable logistic regression. The primary PS model included age, sex, tumor differentiation, Child-Pugh class, underlying liver disease, cirrhosis, other metabolic comorbidities, antiviral therapy, TNM stage, tumor number, maximum tumor diameter, ln(AFP), obesity (BMI ≥28 kg/m2), extent of hepatectomy, macrovascular invasion, and portal hypertension. TB and MVI were excluded from the primary PS model because they may represent downstream pathological features; glucose, HbA1c, and antidiabetic medication were also excluded because they characterize DM or were available only within the DM subgroup.
Overlap weighting was prespecified as the primary PS approach and estimated the average treatment effect in the overlap population (ATO), with weights of 1−PS for patients with DM and PS for those without DM. Stabilized inverse-probability-of-treatment weighting (IPTW), with the PS clipped at the 1st and 99th percentiles, and 1:1 nearest-neighbor matching without replacement using a caliper of 0.2 SD of the logit PS were sensitivity analyses. Covariate balance was assessed using standardized mean differences (SMDs), with an absolute SMD <0.10 considered acceptable. Weighted Cox models used robust sandwich standard errors; matched Cox models were stratified by matched pair with pair-cluster robust standard errors. A further conditional sensitivity PS model added TB and MVI. PS analyses included all 924 patients and were interpreted as adjusted associations rather than causal effects.
Results
Clinicopathological Characteristics According to DM, TB, and MVI Status
The analysis included 924 patients with pathologically confirmed HCC who underwent curative-intent hepatectomy; R0 resection was the intended surgical goal but was not confirmed for every patient. Most patients were male (745/924, 80.6%) and younger than 70 years (851/924, 92.1%). Overall, 124 patients (13.4%) had DM, 314 (34.0%) were TB+, and 493 (53.4%) were MVI+. Other metabolic comorbidities were more common in patients with DM than in those without DM (58.9% vs 22.9%; P < 0.001), and obesity was also more frequent (13.7% vs 6.6%; P = 0.006) (Table 1).
Table 1.
Baseline Characteristics According to Diabetes Mellitus Status
| Characteristics | Total Patients | Non-DM | DM | P value |
|---|---|---|---|---|
| Total number of patients (%) | 924 (100.0) | 800 (86.6) | 124 (13.4) | |
| Age, years | <0.001 | |||
| <70 | 851 (92.1) | 754 (94.3) | 97 (78.2) | |
| ≥70 | 73 (7.9) | 46 (5.8) | 27 (21.8) | |
| Sex | 0.015 | |||
| Female | 179 (19.4) | 145 (18.1) | 34 (27.4) | |
| Male | 745 (80.6) | 655 (81.9) | 90 (72.6) | |
| Tumor Budding, TB | <0.001 | |||
| Negative | 610 (66.0) | 559 (69.9) | 51 (41.1) | |
| Positive | 314 (34.0) | 241 (30.1) | 73 (58.9) | |
| Microvascular Invasion, MVI | <0.001 | |||
| Negative | 431 (46.6) | 399 (49.9) | 32 (25.8) | |
| Positive | 493 (53.4) | 401 (50.1) | 92 (74.2) | |
| TNM stage | 0.209 | |||
| I/II | 605 (65.5) | 530 (66.3) | 75 (60.5) | |
| III/IV | 319 (34.5) | 270 (33.8) | 49 (39.5) | |
| Child-Pugh | 0.514 | |||
| A | 649 (70.2) | 565 (70.6) | 84 (67.7) | |
| B/C | 275 (29.8) | 235 (29.4) | 40 (32.3) | |
| Differentiation | 0.461 | |||
| Well | 73 (7.9) | 60 (7.5) | 13 (10.5) | |
| Moderate | 683 (73.9) | 592 (74.0) | 91 (73.4) | |
| Poor | 168 (18.2) | 148 (18.5) | 20 (16.1) | |
| Underlying liver disease | <0.001 | |||
| None | 100 (10.8) | 77 (9.6) | 23 (18.5) | |
| Viral Hepatitis | 802 (86.8) | 709 (88.6) | 93 (75.0) | |
| Metabolic liver disease | 22 (2.4) | 14 (1.8) | 8 (6.5) | |
| Cirrhosis | 0.950 | |||
| None | 263 (28.5) | 228 (28.5) | 35 (28.2) | |
| Yes | 661 (71.5) | 572 (71.5) | 89 (71.8) | |
| Other metabolic comorbidities | <0.001 | |||
| None | 668 (72.3) | 617 (77.1) | 51 (41.1) | |
| Yes | 256 (27.7) | 183 (22.9) | 73 (58.9) | |
| Antiviral therapy | 0.722 | |||
| None | 402 (50.1) | 357 (50.4) | 45 (48.4) | |
| Yes | 400 (49.9) | 352 (49.6) | 48 (51.6) | |
| Tumor number | 0.383 | |||
| Single | 805 (87.1) | 700 (87.5) | 105 (84.7) | |
| Multiple | 119 (12.9) | 100 (12.5) | 19 (15.3) | |
| Obesity (BMI ≥28 kg/m2) | 0.006 | |||
| None | 854 (92.4) | 747 (93.4) | 107 (86.3) | |
| Yes | 70 (7.6) | 53 (6.6) | 17 (13.7) | |
| Extent of hepatectomy | 0.068 | |||
| ≤2 segments | 651 (70.5) | 555 (69.4) | 96 (77.4) | |
| >2 segments | 273 (29.5) | 245 (30.6) | 28 (22.6) | |
| Macrovascular invasion | 0.661 | |||
| None | 693 (75.0) | 602 (75.2) | 91 (73.4) | |
| Yes | 231 (25.0) | 198 (24.8) | 33 (26.6) | |
| Portal hypertension | 0.512 | |||
| None | 729 (78.9) | 634 (79.2) | 95 (76.6) | |
| Yes | 195 (21.1) | 166 (20.8) | 29 (23.4) | |
| Maximum tumor diameter, cm | 0.499 | |||
| Mean (SD) | 5.49 (3.73) | 5.57 (3.83) | 4.93 (2.89) | |
| Median (IQR) | 4.50 (2.90–7.00) | 4.50 (2.90–7.35) | 4.00 (3.00–6.25) | |
| ln(AFP) | 0.365 | |||
| Mean (SD) | 4.60 (3.04) | 4.63 (3.05) | 4.37 (2.94) | |
| Median (IQR) | 3.99 (1.80–6.89) | 4.02 (1.80–6.90) | 3.32 (1.71–6.81) | |
| Glycemic profiles among patients with DM | ||||
| HbA1c Stratification | 72 available | — | ||
| Well controlled | — | — | 24 (33.3) | |
| Moderately controlled | — | — | 31 (43.1) | |
| Poorly controlled | — | — | 17 (23.6) | |
| Antidiabetic treatment status | — | — | 102 available | — |
| Treated | — | — | 79 (77.5) | |
| Untreated | — | — | 23 (22.5) | |
| Preoperative fasting blood glucose, mmol/L | — | — | 124 available | — |
| Mean (SD) | — | — | 12.97 (4.36) | |
| Median (IQR) | — | — | 12.68 (9.70–15.99) |
Notes: Data are presented as n (%) unless otherwise specified. Continuous variables are summarized as mean (SD) or median (IQR). ln(AFP) was calculated as the natural logarithm of serum alpha-fetoprotein concentration (ng/mL). Antiviral therapy was analyzed only among patients with viral hepatitis (N = 802); the percentages and P value for this variable are based on these 802 patients rather than the full cohort of 924.
Abbreviations: AFP, alpha-fetoprotein; BMI, body mass index; DM, type 2 diabetes mellitus; HbA1c, glycated hemoglobin; IQR, interquartile range; MVI, microvascular invasion; SD, standard deviation; TB, tumor budding; TNM, tumor-node-metastasis.
Patients with DM were older than those without DM (age ≥70 years, 21.8% vs 5.8%; P < 0.001) and were more often female (27.4% vs 18.1%; P = 0.015). TB+ tumors were more frequent among patients with DM (58.9% vs 30.1%; P < 0.001), as were MVI+ tumors (74.2% vs 50.1%; P < 0.001). Tumor stage did not differ significantly by DM status (stage III/IV, 39.5% vs 33.8%; P = 0.209). Child-Pugh class, tumor differentiation, and ln (AFP) also did not differ significantly (median ln[AFP], 3.32 vs 4.02; P = 0.365). Among 124 patients with DM, HbA1c data were available for 72: 24 (33.3%) had well-controlled, 31 (43.1%) had moderately controlled, and 17 (23.6%) had poorly controlled glycemia. Treatment status was available for 102 patients, of whom 79 (77.5%) were treated and 23 (22.5%) were untreated. Fasting blood glucose was available for all 124 patients with DM. The mean was 12.97 mmol/L (SD, 4.36), and the median was 12.68 mmol/L (IQR, 9.70–15.99) (Table 1).
DM was more frequent in the TB+ group than in the TB− group (23.2% vs 8.4%; P < 0.001). Female sex was also more frequent in the TB+ group (23.9% vs 17.0%; P = 0.013). MVI+ was more frequent in TB+ than in TB− tumors (57.3% vs 51.3%), although the difference was not statistically significant (P = 0.083). Among patients with DM, untreated status was more frequent in the TB+ group (31.7% vs 9.5%; P = 0.008). HbA1c category and fasting blood glucose did not differ significantly by TB status (Table 1 and Supplementary Table S1).
MVI+ tumors were associated with more advanced stage (stage III/IV, 40.2% vs 28.1%; P < 0.001), poorer differentiation (P < 0.001), higher ln(AFP) (median, 4.38 vs 3.23; P < 0.001), and more extensive hepatectomy (>2 segments, 32.7% vs 26.0%; P = 0.027). Macrovascular invasion was also more frequent in MVI+ tumors (31.0% vs 18.1%; P < 0.001). Among patients with DM, treatment status differed by MVI status (P = 0.018), with more untreated patients in the MVI+ group. Preoperative fasting blood glucose was higher in patients with MVI+ than in those with MVI-negative tumors (median, 13.70 vs 10.26 mmol/L; P = 0.010) (Supplementary Table S2).
Independent Associations of DM with TB+ and MVI+ in the Overall Cohort
In the fully adjusted logistic model shown in Figure 2, DM was associated with higher odds of TB+ status (OR, 3.508; 95% CI, 2.287–5.381; P < 0.001). Male sex was associated with lower odds of TB+ status (OR, 0.654; 95% CI, 0.458–0.936; P = 0.020), whereas macrovascular invasion was associated with higher odds (OR, 2.685; 95% CI, 1.747–4.127; P < 0.001). The associations for age, stage III/IV disease, Child-Pugh class, underlying liver disease, tumor differentiation, cirrhosis, other metabolic comorbidities, antiviral therapy, tumor number, maximum tumor diameter, obesity, resection extent, portal hypertension, and ln(AFP) were not statistically significant in this expanded model.
In the fully adjusted logistic model shown in Figure 3, DM was associated with higher odds of MVI+ status (OR, 3.364; 95% CI, 2.083–5.434; P < 0.001). Moderate differentiation (OR, 3.729; 95% CI, 2.069–6.720; P < 0.001), poor differentiation (OR, 5.272; 95% CI, 2.701–10.291; P < 0.001), and macrovascular invasion (OR, 1.653; 95% CI, 1.077–2.535; P = 0.021) were also associated with MVI+ status. Stage III/IV disease and the other covariates were not statistically significant in this expanded model.
Prognostic Impact of TB, MVI, and DM in the Overall Cohort
The median follow-up was 17.5 months (IQR, 8.0–30.3 months). During follow-up, 310 patients (33.5%) experienced recurrence and 236 (25.5%) died. The conventional Cox models used the complete 924-patient analysis set; the PS analyses included all 924 patients, with 310 RFS events and 236 OS events.
Kaplan-Meier analyses characterized the unadjusted prognostic effects of TB and MVI. Patients with MVI+ had poorer OS and RFS than those with MVI− tumors (both log-rank P < 0.001). Patients with TB+ also had poorer OS (log-rank P = 0.005) and RFS (log-rank P < 0.001) than those with TB− tumors (Figure 4).
Figure 4.
Kaplan–Meier survival curves stratified by microvascular invasion (MVI) and tumor budding (TB) status in the overall cohort. (A) Recurrence-free survival (RFS) by MVI status. (B) Overall survival (OS) by MVI status. (C) RFS by TB status. (D) OS by TB status. P values were calculated using the log-rank test.
In univariable Cox analysis, TB+ status (HR, 1.456; 95% CI, 1.158–1.831; P = 0.001), DM (HR, 1.381; 95% CI, 1.022–1.865; P = 0.035), MVI+ status (HR, 1.897; 95% CI, 1.503–2.395; P < 0.001), stage III/IV disease, poor differentiation, and higher ln(AFP) were associated with poorer RFS. In the expanded Model 1, which excluded TB and MVI, DM remained associated with poorer RFS (HR, 1.473; 95% CI, 1.054–2.059; P = 0.023). After TB and MVI were added in Model 2, the DM estimate was attenuated and was not statistically significant (HR, 1.226; 95% CI, 0.867–1.735; P = 0.249). In Model 2, TB+ status remained associated with poorer RFS (HR, 1.296; 95% CI, 1.016–1.655; P = 0.037), as did MVI+ status (HR, 1.560; 95% CI, 1.218–1.999; P < 0.001) and higher ln(AFP) (HR, 1.070; 95% CI, 1.028–1.114; P < 0.001) (Table 2).
Table 2.
Expanded Cox Proportional Hazards Regression Models for Recurrence-Free Survival
| Characteristics | Univariate Analysis | Model 1: Excluding TB and MVI | Model 2: including TB and MVI | ||||||
|---|---|---|---|---|---|---|---|---|---|
| HR | 95% CI | P value | HR | 95% CI | P value | HR | 95% CI | P value | |
| Diabetes | |||||||||
| No | Reference | Reference | Reference | ||||||
| Yes | 1.381 | 1.022–1.865 | 0.035 | 1.473 | 1.054–2.059 | 0.023 | 1.226 | 0.867–1.735 | 0.249 |
| Age, years | |||||||||
| <70 | Reference | Reference | Reference | ||||||
| ≥70 | 0.739 | 0.474–1.151 | 0.181 | 0.654 | 0.403–1.062 | 0.086 | 0.650 | 0.400–1.055 | 0.082 |
| Sex | |||||||||
| Female | Reference | Reference | Reference | ||||||
| Male | 1.185 | 0.879–1.596 | 0.265 | 1.143 | 0.839–1.557 | 0.396 | 1.196 | 0.876–1.633 | 0.261 |
| TNM stage | |||||||||
| I/II | Reference | Reference | Reference | ||||||
| III/IV | 1.888 | 1.508–2.364 | <0.001 | 1.251 | 0.919–1.703 | 0.155 | 1.237 | 0.902–1.696 | 0.188 |
| Child-Pugh | |||||||||
| A | Reference | Reference | Reference | ||||||
| B/C | 0.948 | 0.738–1.219 | 0.677 | 0.779 | 0.601–1.011 | 0.060 | 0.789 | 0.608–1.025 | 0.075 |
| Underlying liver disease | |||||||||
| None | Reference | Reference | Reference | ||||||
| Viral Hepatitis | 1.369 | 0.909–2.061 | 0.133 | 1.257 | 0.795–1.987 | 0.329 | 1.189 | 0.748–1.888 | 0.464 |
| Metabolic liver disease | 1.549 | 0.723–3.319 | 0.261 | 1.468 | 0.670–3.217 | 0.338 | 1.480 | 0.674–3.251 | 0.329 |
| Differentiation | |||||||||
| Well | Reference | Reference | Reference | ||||||
| Moderate | 1.528 | 0.945–2.469 | 0.084 | 1.274 | 0.779–2.081 | 0.334 | 1.117 | 0.680–1.834 | 0.663 |
| Poor | 2.043 | 1.210–3.450 | 0.008 | 1.383 | 0.797–2.399 | 0.249 | 1.185 | 0.679–2.067 | 0.550 |
| Cirrhosis | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 1.046 | 0.816–1.342 | 0.722 | 1.006 | 0.752–1.348 | 0.965 | 1.004 | 0.749–1.347 | 0.977 |
| Other metabolic comorbidities | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 1.112 | 0.872–1.419 | 0.392 | 1.136 | 0.864–1.494 | 0.360 | 1.121 | 0.854–1.472 | 0.410 |
| Antiviral therapy | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 1.121 | 0.885–1.420 | 0.344 | 1.145 | 0.897–1.460 | 0.277 | 1.159 | 0.908–1.479 | 0.237 |
| Tumor number | |||||||||
| Single | Reference | Reference | Reference | ||||||
| Multiple | 1.519 | 1.118–2.062 | 0.007 | 1.409 | 1.027–1.933 | 0.034 | 1.392 | 1.013–1.911 | 0.041 |
| Maximum tumor diameter, cm | |||||||||
| Per 1-cm increase | 1.055 | 1.027–1.084 | <0.001 | 1.003 | 0.966–1.041 | 0.882 | 1.004 | 0.967–1.043 | 0.831 |
| Obesity (BMI ≥28 kg/m2) | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 0.999 | 0.653–1.528 | 0.997 | 1.140 | 0.716–1.816 | 0.582 | 1.112 | 0.697–1.773 | 0.657 |
| ln(AFP) | |||||||||
| Per 1-unit increase | 1.104 | 1.065–1.144 | <0.001 | 1.077 | 1.034–1.121 | <0.001 | 1.070 | 1.028–1.114 | <0.001 |
| Extent of hepatectomy | |||||||||
| ≤2 segments | Reference | Reference | Reference | ||||||
| >2 segments | 1.412 | 1.117–1.784 | 0.004 | 1.071 | 0.818–1.403 | 0.617 | 1.061 | 0.810–1.390 | 0.668 |
| Macrovascular invasion | |||||||||
| Negative | Reference | Reference | Reference | ||||||
| Positive | 2.361 | 1.869–2.983 | <0.001 | 1.971 | 1.427–2.722 | <0.001 | 1.784 | 1.280–2.487 | <0.001 |
| Portal hypertension | |||||||||
| Negative | Reference | Reference | Reference | ||||||
| Positive | 1.385 | 1.071–1.790 | 0.013 | 0.843 | 0.608–1.170 | 0.308 | 0.868 | 0.625–1.206 | 0.399 |
| TB | |||||||||
| Negative | Reference | Not included | Reference | ||||||
| Positive | 1.456 | 1.158–1.831 | 0.001 | 1.296 | 1.016–1.655 | 0.037 | |||
| MVI | |||||||||
| Negative | Reference | Not included | Reference | ||||||
| Positive | 1.897 | 1.503–2.395 | <0.001 | 1.560 | 1.218–1.999 | <0.001 | |||
Notes: Model 1 included the clinical covariates and excluded TB and MVI; Model 2 additionally included TB and MVI. Cox regression used Breslow handling of tied event times. Adjusted models used 924 patients; antiviral therapy was evaluated among patients with viral hepatitis only (N = 802). HRs for maximum tumor diameter and ln(AFP) are per 1-unit increase. All confidence intervals are Wald 95% CIs; P values are two-sided.
Abbreviations: AFP, alpha-fetoprotein; BMI, body mass index; CI, confidence interval; DM, type 2 diabetes mellitus; HR, hazard ratio; MVI, microvascular invasion; RFS, recurrence-free survival; TB, tumor budding.
For OS, TB+ status (HR, 1.454; 95% CI, 1.118–1.891; P = 0.005) and MVI+ status (HR, 2.368; 95% CI, 1.807–3.103; P < 0.001) were associated with poorer survival in univariable analyses. DM was not associated with OS in the univariable model (HR, 1.038; 95% CI, 0.704–1.531; P = 0.851). In the expanded Model 1, which excluded TB and MVI, DM was not statistically significant (HR, 1.431; 95% CI, 0.917–2.234; P = 0.115). After TB and MVI were added in Model 2, the DM estimate was further attenuated (HR, 1.132; 95% CI, 0.714–1.794; P = 0.599). In Model 2, TB+ status was not statistically significant (HR, 1.247; 95% CI, 0.929–1.674; P = 0.141), whereas MVI+ status remained associated with poorer OS (HR, 1.780; 95% CI, 1.333–2.376; P < 0.001) (Table 3).
Table 3.
Expanded Cox Proportional Hazards Regression Models for Overall Survival
| Characteristics | Univariate Analysis | Model 1: Excluding TB and MVI | Model 2: Including TB and MVI | ||||||
|---|---|---|---|---|---|---|---|---|---|
| HR | 95% CI | P value | HR | 95% CI | P value | HR | 95% CI | P value | |
| DM | |||||||||
| No | Reference | Reference | Reference | ||||||
| Yes | 1.038 | 0.704–1.531 | 0.851 | 1.431 | 0.917–2.234 | 0.115 | 1.132 | 0.714–1.794 | 0.599 |
| Age, years | |||||||||
| <70 | Reference | Reference | Reference | ||||||
| ≥70 | 0.703 | 0.423–1.168 | 0.174 | 0.505 | 0.288–0.888 | 0.018 | 0.461 | 0.260–0.815 | 0.008 |
| Sex | |||||||||
| Female | Reference | Reference | Reference | ||||||
| Male | 1.280 | 0.905–1.812 | 0.163 | 1.533 | 1.063–2.210 | 0.022 | 1.667 | 1.148–2.422 | 0.007 |
| TNM stage | |||||||||
| I/II | Reference | Reference | Reference | ||||||
| III/IV | 3.505 | 2.692–4.563 | <0.001 | 1.795 | 1.250–2.578 | 0.002 | 1.763 | 1.216–2.556 | 0.003 |
| Child-Pugh | |||||||||
| A | Reference | Reference | Reference | ||||||
| B/C | 1.414 | 1.079–1.852 | 0.012 | 0.985 | 0.744–1.304 | 0.918 | 1.006 | 0.759–1.334 | 0.966 |
| Underlying liver disease | |||||||||
| None | Reference | Reference | Reference | ||||||
| Viral Hepatitis | 0.831 | 0.566–1.219 | 0.343 | 0.702 | 0.447–1.102 | 0.124 | 0.636 | 0.401–1.007 | 0.054 |
| Metabolic liver disease | 0.651 | 0.253–1.678 | 0.374 | 0.614 | 0.232–1.624 | 0.326 | 0.609 | 0.230–1.610 | 0.317 |
| Differentiation | |||||||||
| Well | Reference | Reference | Reference | ||||||
| Moderate | 2.003 | 1.023–3.922 | 0.043 | 1.525 | 0.761–3.054 | 0.234 | 1.322 | 0.657–2.661 | 0.433 |
| Poor | 4.324 | 2.159–8.661 | <0.001 | 2.401 | 1.149–5.016 | 0.020 | 2.010 | 0.957–4.223 | 0.065 |
| Cirrhosis | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 1.030 | 0.778–1.362 | 0.838 | 1.247 | 0.880–1.765 | 0.214 | 1.318 | 0.925–1.878 | 0.126 |
| Other metabolic comorbidities | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 0.987 | 0.738–1.319 | 0.928 | 1.224 | 0.881–1.700 | 0.229 | 1.189 | 0.857–1.649 | 0.300 |
| Antiviral therapy | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 0.883 | 0.669–1.165 | 0.377 | 0.941 | 0.706–1.254 | 0.677 | 0.921 | 0.690–1.230 | 0.579 |
| Tumor number | |||||||||
| Single | Reference | Reference | Reference | ||||||
| Multiple | 1.558 | 1.100–2.206 | 0.013 | 1.398 | 0.969–2.015 | 0.073 | 1.358 | 0.938–1.965 | 0.105 |
| Maximum tumor diameter, cm | |||||||||
| Per 1-cm increase | 1.115 | 1.087–1.143 | <0.001 | 1.027 | 0.988–1.067 | 0.172 | 1.032 | 0.993–1.073 | 0.105 |
| Obesity (BMI ≥28 kg/m2) | |||||||||
| None | Reference | Reference | Reference | ||||||
| Yes | 0.925 | 0.564–1.515 | 0.756 | 1.366 | 0.791–2.360 | 0.263 | 1.384 | 0.801–2.392 | 0.244 |
| ln(AFP) | |||||||||
| Per 1-unit increase | 1.198 | 1.152–1.246 | <0.001 | 1.133 | 1.084–1.185 | <0.001 | 1.121 | 1.072–1.173 | <0.001 |
| Extent of hepatectomy | |||||||||
| ≤2 segments | Reference | Reference | Reference | ||||||
| >2 segments | 2.229 | 1.725–2.879 | <0.001 | 1.285 | 0.955–1.728 | 0.097 | 1.264 | 0.937–1.705 | 0.125 |
| Macrovascular invasion | |||||||||
| Negative | Reference | Reference | Reference | ||||||
| Positive | 3.704 | 2.866–4.786 | <0.001 | 1.879 | 1.308–2.699 | <0.001 | 1.668 | 1.141–2.437 | 0.008 |
| Portal hypertension | |||||||||
| Negative | Reference | Reference | Reference | ||||||
| Positive | 1.747 | 1.325–2.302 | <0.001 | 0.764 | 0.524–1.114 | 0.162 | 0.793 | 0.544–1.156 | 0.228 |
| TB | |||||||||
| Negative | Reference | Not included | Reference | ||||||
| Positive | 1.454 | 1.118–1.891 | 0.005 | 1.247 | 0.929–1.674 | 0.141 | |||
| MVI | |||||||||
| Negative | Reference | Not included | Reference | ||||||
| Positive | 2.368 | 1.807–3.103 | <0.001 | 1.780 | 1.333–2.376 | <0.001 | |||
Notes: Model 1 included the clinical covariates and excluded TB and MVI; Model 2 additionally included TB and MVI. Cox regression used Breslow handling of tied event times. Adjusted models used 924 patients; antiviral therapy was evaluated among patients with viral hepatitis only (N = 802). HRs for maximum tumor diameter and ln(AFP) are per 1-unit increase. All confidence intervals are Wald 95% CIs; P values are two-sided.
Abbreviations: AFP, alpha-fetoprotein; BMI, body mass index; CI, confidence interval; DM, type 2 diabetes mellitus; HR, hazard ratio; MVI, microvascular invasion; OS, overall survival; TB, tumor budding.
These nested Cox models showed model-dependent attenuation of the DM coefficient after TB and MVI were added. This attenuation indicates only that the conditional association between DM and outcome was smaller in Model 2; it does not establish mediation or a causal DM-to-TB/MVI-to-outcome pathway.
In the PS analyses, overlap weighting achieved balance across all variables in the primary PS model (maximum absolute SMD <0.001), and 1:1 matching retained 122 pairs with a maximum absolute SMD of 0.173 (Supplementary Table S3). In the primary overlap-weighted Cox analysis, DM was not significantly associated with RFS (HR, 1.453; 95% CI, 0.778–2.714; P = 0.240) or OS (HR, 1.127; 95% CI, 0.556–2.284; P = 0.740). The stabilized IPTW results were also nonsignificant for RFS (HR, 1.293; 95% CI, 0.856–1.954; P = 0.222) and OS (HR, 1.132; 95% CI, 0.694–1.846; P = 0.619). The 1:1 matched sensitivity analysis was not statistically significant for either RFS (HR, 1.310; 95% CI, 0.826–2.078; P = 0.251) or OS (HR, 1.188; 95% CI, 0.611–2.309; P = 0.613). In the conditional overlap-weighted sensitivity model that additionally included TB and MVI, DM was not associated with RFS (HR, 1.158; 95% CI, 0.626–2.144; P = 0.640) or OS (HR, 0.971; 95% CI, 0.478–1.974; P = 0.935) (Table 4).
Table 4.
Propensity-Score Adjusted Cox Analyses of Diabetes and Postoperative Survival
| Analysis | Estimand | RFS HR (95% CI) | P value | OS HR (95% CI) | P value | N (RFS/OS Events) |
|---|---|---|---|---|---|---|
| Overlap weighting (primary) | ATO | 1.453 (0.778–2.714) | 0.240 | 1.127 (0.556–2.284) | 0.740 | 924 (310/236) |
| Stabilized IPTW (sensitivity) | ATE | 1.293 (0.856–1.954) | 0.222 | 1.132 (0.694–1.846) | 0.619 | 924 (310/236) |
| 1:1 PS matching (sensitivity) | ATT | 1.310 (0.826–2.078) | 0.251 | 1.188 (0.611–2.309) | 0.613 | 244 (93/61) |
| Overlap weighting + TB/MVI (sensitivity) | ATO, conditional | 1.158 (0.626–2.144) | 0.640 | 0.971 (0.478–1.974) | 0.935 | 924 (310/236) |
Notes: HRs compare patients with diabetes mellitus vs those without diabetes mellitus. The primary analysis used overlap weights and estimated the average treatment effect in the overlap population (ATO). Stabilized inverse-probability weighting (ATE) and 1:1 nearest-neighbor matching (ATT in the matched sample) were sensitivity analyses. Weighted Cox models used robust sandwich standard errors; the matched Cox model was stratified by matched pair with pair-cluster robust standard errors. The primary PS model excluded TB, MVI, glucose, HbA1c, and antidiabetic medication. The TB/MVI sensitivity model additionally included TB and MVI. All 924 patients were retained and 122 matched pairs were retained.
Abbreviations: ATE, average treatment effect; ATO, average treatment effect in the overlap population; ATT, average treatment effect among the treated; BMI, body mass index; CI, confidence interval; Cox, Cox proportional hazards; DM, type 2 diabetes mellitus; HbA1c, glycated hemoglobin; HR, hazard ratio; IPTW, inverse-probability-of-treatment weighting; MVI, microvascular invasion; OS, overall survival; PS, propensity score; RFS, recurrence-free survival; TB, tumor budding.
Exploratory Analyses in the DM Subgroup
Among 124 patients with DM, 73 (58.9%) were TB+ and 92 (74.2%) were MVI+. HbA1c data were available for 72 patients (52 missing), treatment status for 102 (22 missing), and fasting glucose for all 124 (Table 1). Each exposure was analyzed using its available cases, without imputation. These subgroup analyses were exploratory because of limited sample size, incomplete HbA1c and treatment data, and sparse treatment-status cells.
For TB, neither poorer HbA1c category (OR per 1-level increase, 1.415; 95% CI, 0.742–2.699; P = 0.292) nor higher fasting blood glucose (OR per 1 mmol/L increase, 1.017; 95% CI, 0.936–1.105; P = 0.691) was significantly associated with TB+. Antidiabetic treatment was associated with lower odds of TB+ than untreated DM (OR, 0.227; 95% CI, 0.071–0.728; P = 0.013). Because the analysis was restricted to the DM subgroup and affected by incomplete treatment data and sparse cells, this association was considered exploratory and hypothesis-generating, not evidence of a direct treatment effect (Supplementary Figure S1).
For MVI, poorer HbA1c category was associated with higher odds of MVI+ (OR per 1-level increase, 2.738; 95% CI, 1.093–6.857; P = 0.032), as was higher fasting blood glucose (OR per 1 mmol/L increase, 1.149; 95% CI, 1.031–1.279; P = 0.012). Antidiabetic treatment was associated with lower odds of MVI+ than untreated DM (OR, 0.126; 95% CI, 0.016–0.990; P = 0.049). Given the limited subgroup size and sparse treatment-status distribution, particularly the small number of untreated patients without MVI, this association was considered exploratory and hypothesis-generating, not evidence of a direct treatment effect (Supplementary Figure S2).
In Kaplan-Meier analyses restricted to patients with DM, MVI+ and TB+ were associated with poorer RFS (log-rank P = 0.046 and P < 0.001, respectively). The corresponding OS differences were not statistically significant for MVI (log-rank P = 0.115) or TB (log-rank P = 0.220). Thus, within the DM subgroup, TB and MVI were more closely associated with postoperative recurrence than with OS. The limited sample size should be considered when interpreting these survival curves (Supplementary Figure S3).
Discussion
In this retrospective cohort of patients who underwent curative hepatectomy for HCC, pre-existing DM was associated with higher TB+ and MVI+ prevalence and with higher adjusted odds of both pathological features. TB was associated with RFS but not OS in the expanded Cox models, whereas MVI was associated with both endpoints. In contrast, the primary overlap-weighted PS analysis did not identify significant associations of DM with RFS or OS. Thus, the findings support an association between DM and adverse pathological phenotypes but do not provide consistent evidence that DM is an independent survival prognostic factor.
Our findings are consistent with recent clinical research on diabetes and HCC but address a different stage of disease. Large epidemiological cohorts have established diabetes as a risk factor for liver cancer incidence and mortality.5 In patients with nonalcoholic fatty liver disease (NAFLD) and diabetes, glycemic control and medication exposure may inform HCC risk stratification.6 A multicenter study of 1644 patients undergoing HCC resection associated coexisting obesity and diabetes with late recurrence and poorer prognosis.14 Our study extends this evidence by examining pathological risk markers, particularly TB and MVI, after resection. The associations between DM and these adverse features may help identify patients with a more invasive pathological phenotype, but the observational design does not determine the underlying mechanism.
This pathological focus is clinically relevant because postoperative HCC research increasingly uses histology to define high-risk disease. MVI is a reproducible marker of early recurrence after resection. In early-stage HCC, a recent study identified MVI and/or satellitosis as the only independent predictor of recurrence, aggressive recurrence, and OS after resection.15 Adjuvant trials such as IMbrave050 selected high-risk patients using vascular invasion, poor differentiation, and tumor burden. Updated results also suggest that pathological risk enrichment alone may be insufficient and that more refined biological stratification is needed.16,17 The association between DM and MVI observed here places diabetes within a familiar pathological risk framework. Metabolic status may also help identify patients who are more likely to harbor adverse pathology.
TB provides information complementary to MVI. Although TB is less established in HCC than in colorectal cancer, available studies support its prognostic relevance. Published HCC cohorts have used binary 0-versus-at-least-1-bud definitions, threshold-based groups, and multilevel bud-count categories, and higher counts may carry additional prognostic information.10,18 Sarelin et al associated TB+ with increased 5-year mortality in a Western surgical HCC cohort.18 A recent Chinese resection cohort also identified TB as a prognostic marker after curative hepatectomy.10 In the present analysis, TB was associated with RFS in the expanded Cox model but not with OS, whereas MVI remained associated with both endpoints. This difference may reflect the shorter follow-up, competing mortality, and the distinct biological information captured by local tumor-front budding vs vascular dissemination. Because TB was operationalized as a binary exploratory variable, the prognostic estimate should not be interpreted as a validated quantitative HCC score.
The nested Cox analyses showed model-dependent attenuation of the DM coefficient after TB and MVI were added. The primary overlap-weighted and stabilized IPTW analyses likewise did not identify significant associations of DM with RFS or OS. The 1:1 matched sensitivity analysis did not show a statistically significant association with either RFS or OS (RFS: HR, 1.310; 95% CI, 0.826–2.078; P = 0.251; OS: HR, 1.188; 95% CI, 0.611–2.309; P = 0.613). This analysis targeted the matched ATT and was not reproduced by the primary ATO or ATE analyses. Accordingly, the totality of the survival analyses does not provide consistent evidence that DM is an independent prognostic factor. The attenuation after TB/MVI adjustment is an association pattern only and does not establish mediation, a DM-to-TB/MVI-to-outcome causal chain, or a direct effect of DM. TB and MVI may be correlated manifestations of aggressive disease, confounded by unmeasured factors, or affected by selection; no formal mediation analysis was performed.
The DM subgroup findings were exploratory and should not be interpreted as treatment effects. Poor glycemic control and hyperglycemia have been associated with recurrence and survival after curative HCC treatment.19–21 In the present cohort, poorer HbA1c category and higher preoperative fasting glucose were associated with MVI+ status, but the available-case sample sizes were limited. Observational studies and a meta-analysis have reported associations between metformin use and improved HCC outcomes.22,23 Conversely, another resection cohort found no independent association between metformin use and recurrence and emphasized glycemic control.24 Accordingly, the lower odds of TB+ and MVI+ status among treated patients may reflect differences in diabetes severity, comorbidity, adherence, or access to care rather than a direct effect of therapy.
Several mechanisms could plausibly underlie these associations. Hyperglycemia, hyperinsulinemia, insulin resistance, chronic inflammation, altered lipid metabolism, and extracellular-matrix remodeling have been linked to tumor-cell motility, angiogenesis, immune–tumor interactions, and stromal invasion.25–32 Two recent studies offer complementary, although indirect, biological context for our observations. In a mouse model of type 2 diabetes combined with spinal cord injury, Xu et al found that activation of the RAGE–ROS–TXNIP/NLRP3 axis increased microglial pyroptosis, impaired microglial clearance of myelin debris, and hindered remyelination.33 Although conducted outside the liver, this work illustrates how diabetes may disrupt inflammatory-cell function and tissue repair. Using spatial multi-omics and experimental models in HCC, Ji et al identified DLL4–NOTCH3 signaling between capillary endothelial cells and matrix-producing cancer-associated fibroblasts, linking this interaction to extracellular-matrix remodeling and immune exclusion.34 Viewed together, the two reports inform complementary aspects of our discussion: diabetes-associated inflammatory dysfunction and HCC-specific stromal organization. Their relevance to our findings remains indirect, because neither study examined the relationship between DM and TB or MVI, and none of these pathways was measured in our cohort.
These findings may refine postoperative risk assessment but do not justify changes in treatment based on DM, TB, or observational treatment associations alone. In patients with HCC and DM, careful pathological assessment of MVI and TB may help characterize recurrence risk after hepatectomy. Prospective studies should integrate diabetes duration, longitudinal glycemic measurements, medication exposure, quantitative or standardized TB assessment, MVI, recurrence phenotype, and postoperative outcomes.
This study has several limitations. First, the retrospective, single-center design introduces the potential for selection bias, residual and unmeasured confounding, and precludes causal inference. The PS analyses can balance only measured covariates included in the treatment model and do not eliminate bias from unmeasured factors or model misspecification. Moreover, overlap weighting, stabilized IPTW, and matching target different estimands; the method-dependent RFS findings should therefore be interpreted as sensitivity results rather than interchangeable estimates. Second, the DM subgroup was modest; HbA1c and treatment data were incomplete, and the available-case subgroup analyses may have introduced selection bias and reduced statistical power. Third, treatment was classified only as treated or untreated, without reliable information on drug class, diabetes duration, treatment duration, adherence, or longitudinal glycemic trajectories. Fourth, TB was analyzed as a binary present/absent variable. This classification cannot represent the full tumor-bud count spectrum, dose–response information, or the potential effects of slide sampling and evaluable invasive-interface length. The prereconciliation Cohen’s kappa coefficient for TB indicated substantial agreement (κ=0.80), but external repetition, multislide quantitative assessment, and HCC-specific reproducibility validation were not performed. Fifth, the median follow-up of 17.5 months preferentially captures early postoperative events and may not adequately characterize late recurrence or long-term OS.35 Because TB and MVI were evaluated as invasive pathological features of the resected primary tumor, the present analysis may be more informative for early recurrence associated with primary-tumor pathology than for late de novo recurrence; longer-follow-up validation is required. Sixth, attenuation of the DM coefficient after TB/MVI adjustment is an association pattern, not evidence of mediation; no formal mediation analysis was performed. Seventh, the analysis-ready dataset provided Child-Pugh class and TNM stage only as combined B/C and III/IV categories; separate sensitivity analyses for Child-Pugh C or TNM IV require source-level case verification. The expanded models also included structurally related obesity variables, which may affect coefficient precision and interpretation. Finally, no direct molecular or biomechanical measurements were made, so the proposed biological pathways remain untested in this cohort.
Conclusions
Pre-existing DM was associated with higher TB+ and MVI+ prevalence and with higher adjusted odds of both pathological features in patients with resected HCC. However, the PS-adjusted analyses did not provide consistent evidence that DM was an independent prognostic factor for RFS or OS. The observational design cannot establish causal or mediating relationships among DM, TB/MVI, and postoperative outcomes. Glycemic control and treatment findings remain exploratory and require prospective external validation.
Funding Statement
This work was supported by the Noncommunicable Chronic Diseases–National Science and Technology Major Project (2025ZD0551500). The funders had no role in the study design; data collection, analysis, or interpretation; preparation of the manuscript; or the decision to submit the manuscript for publication.
Abbreviations
AFP, alpha-fetoprotein; ATE, average treatment effect; ATO, average treatment effect in the overlap population; ATT, average treatment effect among the treated; BMI, body mass index; CI, confidence interval; DM, type 2 diabetes mellitus; H&E, hematoxylin and eosin; HbA1c, glycated hemoglobin; HCC, hepatocellular carcinoma; HR, hazard ratio; IPTW, inverse-probability-of-treatment weighting; IQR, interquartile range; MVI, microvascular invasion; NAFLD, nonalcoholic fatty liver disease; OR, odds ratio; OS, overall survival; PS, propensity score; RFS, recurrence-free survival; SD, standard deviation; SMD, standardized mean difference; TB, tumor budding; TNM, tumor-node-metastasis.
Data Sharing Statement
The de-identified data supporting the findings of this study are not publicly available because of ethical and patient-privacy restrictions. Requests should be directed to Kai Qu (joanne8601@163.com), who will coordinate review of the data-access application with the First Affiliated Hospital of Xi’an Jiaotong University and the relevant ethics authority. Data may be made available to qualified researchers upon reasonable request, subject to institutional approval and any required data-use agreement.
Ethics Approval and Informed Consent
The study protocol was approved by the Ethics Review Committee of the First Affiliated Hospital of Xi’an Jiaotong University (approval no. XJTU1AF2015LSL-057). All patients were treated at the First Affiliated Hospital; patient identification, surgery, pathological specimen processing and review, clinical follow-up, source-data collection, and the original research work were performed at that institution. Given the retrospective nature of the study and the use of existing de-identified clinical and pathological data, the requirement for written informed consent was waived by the Ethics Committee. The study was conducted in accordance with the Declaration of Helsinki.
Consent for Publication
Not applicable. No identifiable individual data are included in this article.
Author Contributions
Z.T., X.Z., K.Y., and K.Q. conceived and designed the study. Z.T., F.J., S.X., J.G., H.R., Z.Y., K.W., and F.X. contributed to data acquisition, curation, and validation. Z.T., F.J., and S.X. performed the statistical analyses and data visualization. Z.T. drafted the manuscript. X.Z., K.Y., and K.Q. supervised the study and critically revised the manuscript for important intellectual content. All authors made a significant contribution to the work reported, whether in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; agreed on the journal to which the article was submitted; and agreed to be accountable for all aspects of the work.
Disclosure
The authors declare that they have no competing interests.
References
- 1.Zhang CH, Cheng Y, Zhang S, Fan J, Gao Q. Changing epidemiology of hepatocellular carcinoma in Asia. Liver Int. 2022;42(9):2029–17. doi: 10.1111/liv.15251 [DOI] [PubMed] [Google Scholar]
- 2.McGlynn KA, Petrick JL, El-Serag HB. Epidemiology of hepatocellular carcinoma. Hepatology. 2021;73(Suppl 1):4–13. doi: 10.1002/hep.31288 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Mao X, Zhang X, Kam L, et al. Synergistic association of sodium-glucose cotransporter-2 inhibitor and metformin on liver and non-liver complications in patients with type 2 diabetes mellitus and metabolic dysfunction-associated steatotic liver disease. Gut. 2024;73(12):2054–2061. doi: 10.1136/gutjnl-2024-332481 [DOI] [PubMed] [Google Scholar]
- 4.Fan W, Adebowale K, Váncza L, et al. Matrix viscoelasticity promotes liver cancer progression in the pre-cirrhotic liver. Nature. 2024;626(7999):635–642. doi: 10.1038/s41586-023-06991-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ho NT, Abe SK, Rahman MS, et al. Diabetes is associated with increased liver cancer incidence and mortality in adults: a report from Asia cohort consortium. Int J Cancer. 2024;155(5):854–870. [DOI] [PubMed] [Google Scholar]
- 6.Kramer JR, Natarajan Y, Dai J, et al. Effect of diabetes medications and glycemic control on risk of hepatocellular cancer in patients with nonalcoholic fatty liver disease. Hepatology. 2022;75(6):1420–1428. doi: 10.1002/hep.32244 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Lugli A, Zlobec I, Berger MD, Kirsch R, Nagtegaal ID. Tumour budding in solid cancers. Nat Rev Clin Oncol. 2021;18(2):101–115. doi: 10.1038/s41571-020-0422-y [DOI] [PubMed] [Google Scholar]
- 8.Lugli A, Kirsch R, Ajioka Y, et al. Recommendations for reporting tumor budding in colorectal cancer based on the international tumor budding consensus conference (ITBCC) 2016. Mod Pathol. 2017;30(9):1299–1311. [DOI] [PubMed] [Google Scholar]
- 9.American Diabetes Association Professional Practice Committee. 6. Glycemic goals, hypoglycemia, and hyperglycemic crises: standards of Care in diabetes-2026. Diabetes Care. 2026;49(Suppl 1):S132–S149. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Yang KB, Wu KJ, Lei ZT, et al. Tumor budding is a meaningful prognostic marker in patients with hepatocellular carcinoma after curative hepatectomy. Portal Hypertens Cirrhosis. 2024;3(2):86–95. [Google Scholar]
- 11.Li K, Zhang R, Wen F, et al. Single-cell dissection of the multicellular ecosystem and molecular features underlying microvascular invasion in HCC. Hepatology. 2024;79(6):1293–1309. doi: 10.1097/HEP.0000000000000673 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Zhang FF, Liu ZH, Shao CT, et al. Development of a new nomogram for predicting recurrence in HCC with MVI following curative hepatectomy. Int J Surg. 2026;112(2):3854–3864. [DOI] [PubMed] [Google Scholar]
- 13.Zhou X, Pang Z, Gao W, et al. Performance of an A1C and fasting capillary blood glucose test for screening newly diagnosed diabetes and pre-diabetes defined by an oral glucose tolerance test in Qingdao, China. Diabetes Care. 2010;33(3):545–550. doi: 10.2337/dc09-1410 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Shinkawa H, Kaibori M, Ueno M, et al. Impact of diabetes mellitus and obesity comorbidities on survival outcomes after hepatocellular carcinoma resection: a multicenter retrospective study. Liver Cancer. 2025;14(1):80–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Fuster-Anglada C, Mauro E, Ferrer-Fàbrega J, et al. Histological predictors of aggressive recurrence of hepatocellular carcinoma after liver resection. J Hepatol. 2024;81(6):995–1004. [DOI] [PubMed] [Google Scholar]
- 16.Qin S, Chen M, Cheng AL, et al. Atezolizumab plus bevacizumab versus active surveillance in patients with resected or ablated high-risk hepatocellular carcinoma (IMbrave050): a randomised, open-label, multicentre, phase 3 trial. Lancet. 2023;402(10415):1835–1847. [DOI] [PubMed] [Google Scholar]
- 17.Yopp A, Chen M, Cheng AL, et al. Updated data from IMbrave050: adjuvant atezolizumab plus bevacizumab for high-risk hepatocellular carcinoma. J Hepatol. 2026;84(6):1102–1111. [DOI] [PubMed] [Google Scholar]
- 18.Sarelin N, Kairaluoma V, Saarnio J, et al. Tumour budding and tumour-stroma ratio in hepatocellular carcinoma. Hum Pathol. 2025;165:105937. [DOI] [PubMed] [Google Scholar]
- 19.Kaneda K, Uenishi T, Takemura S, et al. The influence of postoperative glycemic control on recurrence after curative resection in diabetics with hepatitis C virus-related hepatocellular carcinoma. J Surg Oncol. 2012;105(6):606–611. [DOI] [PubMed] [Google Scholar]
- 20.Hosokawa T, Kurosaki M, Tsuchiya K, et al. Hyperglycemia is a significant prognostic factor of hepatocellular carcinoma after curative therapy. World J Gastroenterol. 2013;19(2):249–257. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Lee HJ, Choi MS, Song BG, et al. Glycemic burden and clinical outcomes of early stage hepatocellular carcinoma after curative treatment. Cancers. 2024;16(15):2652. doi: 10.3390/cancers16152652 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Seo YS, Kim YJ, Kim MS, et al. Association of metformin use with cancer-specific mortality in hepatocellular carcinoma after curative resection: a nationwide population-based study. Medicine. 2016;95(17):e 3527. doi: 10.1097/MD.0000000000003527 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Zhou J, Ke Y, Lei X, et al. Meta-analysis: the efficacy of metformin and other anti-hyperglycemic agents in prolonging the survival of hepatocellular carcinoma patients with type 2 diabetes. Ann Hepatol. 2020;19(3):320–328. [DOI] [PubMed] [Google Scholar]
- 24.Cho WR, Wang CC, Tsai MY, et al. Impact of metformin use on the recurrence of hepatocellular carcinoma after initial liver resection in diabetic patients. PLoS One. 2021;16(3):e0247231. doi: 10.1371/journal.pone.0247231 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Zhang J, Goliwas KF, Wang W, Taufalele PV, Bordeleau F, Reinhart-King CA. Energetic regulation of coordinated leader-follower dynamics during collective invasion of breast cancer cells. Proc Natl Acad Sci U S A. 2019;116(16):7867–7872. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Kunkel M, Reichert TE, Benz P, et al. Overexpression of glut-1 and increased glucose metabolism in tumors are associated with a poor prognosis in patients with oral squamous cell carcinoma. Cancer. 2003;97(4):1015–1024. doi: 10.1002/cncr.11159 [DOI] [PubMed] [Google Scholar]
- 27.Gatenby RA, Gillies RJ. Why do cancers have high aerobic glycolysis? Nat Rev Cancer. 2004;4(11):891–899. doi: 10.1038/nrc1478 [DOI] [PubMed] [Google Scholar]
- 28.Zanotelli MR, Rahman-Zaman A, VanderBurgh JA, et al. Energetic costs regulated by cell mechanics and confinement are predictive of migration path during decision-making. Nat Commun. 2019;10(1):4185. doi: 10.1038/s41467-019-12155-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Cox TR. The matrix in cancer. Nat Rev Cancer. 2021;21(4):217–238. doi: 10.1038/s41568-020-00329-7 [DOI] [PubMed] [Google Scholar]
- 30.Chatelin S, Oudry J, Périchon N, et al. In vivo liver tissue mechanical properties by transient elastography: comparison with dynamic mechanical analysis. Biorheology. 2011;48(2):75–88. doi: 10.3233/BIR-2011-0584 [DOI] [PubMed] [Google Scholar]
- 31.Rui L. Energy metabolism in the liver. Compr Physiol. 2014;4(1):177–197. doi: 10.1002/j.2040-4603.2014.tb00548.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Paul CD, Mistriotis P, Konstantopoulos K. Cancer cell motility: lessons from migration in confined spaces. Nat Rev Cancer. 2017;17(2):131–140. doi: 10.1038/nrc.2016.123 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Xu J, Fang L, Wei T, et al. Type 2 diabetes promotes the microglial pyroptosis by activating NLRP3 inflammasome to impede remyelination after spinal cord injury. Research. 2026;9:1237. doi: 10.34133/research.1237 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Ji F, Li H, Wang Q, et al. Spatial multi-omics identifies a NOTCH3-mediated capillary-mCAF crosstalk driving immune exclusion in hepatocellular carcinoma. iMeta. 2026;5(2):e70117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Nevola R, Ruocco R, Criscuolo L, et al. Predictors of early and late hepatocellular carcinoma recurrence. World J Gastroenterol. 2023;29(8):1243–1260. doi: 10.3748/wjg.v29.i8.1243 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The de-identified data supporting the findings of this study are not publicly available because of ethical and patient-privacy restrictions. Requests should be directed to Kai Qu (joanne8601@163.com), who will coordinate review of the data-access application with the First Affiliated Hospital of Xi’an Jiaotong University and the relevant ethics authority. Data may be made available to qualified researchers upon reasonable request, subject to institutional approval and any required data-use agreement.