An excellent nomogram predicts microvascular invasion that cannot independently stratify outcomes of small hepatocellular carcinoma.

BACKGROUND: Whether microvascular invasion is a prognosis factor for small hepatocellular carcinoma (sHCC) is controversial, and a preoperatively predictive model based on gadoxetate disodium (Gd-EOB-DTPA) MRI is clinically needed for MVI in sHCC.
METHODS: Between March 2012 and September 2020, 455 consecutive patients with pathologically confirmed HCC ≤3 cm who underwent hepatectomy and preoperative Gd-EOB-DTPA MRI were retrospectively enrolled. Univariate and multivariate logistic regression combined with cox regression were conducted to find the confounding factors in the cohorts. Propensity score matching (PSM) was employed to balance the biases between MVI and non-MVI groups. Nomogram with C-index visualized the predictive model of MVI.
RESULTS: Multivariate logistic regression identified that 5 characteristics (AFP, tumor size, tumor margin, peritumoral enhancement, radiologic capsule) were markedly associated with MVI of sHCC and incorporated into the nomogram with excellent predictive performance in the training (AUC/C-index: 0.884/0.874, n=288), validation (AUC/C-index: 0.845/0.828, n=123) and test cohorts (AUC/C-index: 0.903/0.954, n=44). Before PSM, histologic MVI independently affected tumor recurrence (hazard ratio: 1.555, 95% CI: 1.055-2.293, P=0.026). However, due to the confounder of tumor size, there was a significant bias between MVI-positive and MVI-negative groups (propensity score: 0.249±0.105 vs. 0.179±0.106, P<0.001). Meanwhile, the frequency of MVI significantly increased as tumor size growing (P<0.001). After PSM, 70 of 79 MVI cases matched with 171 non-MVI (total 332), and no biases were observed between the two groups (propensity score: 0.238±0.104 vs. 0.217±0.109, P=0.186). Although the median recurrence time in non-MVI sHCC was still longer than that in MVI group (74.3 vs. 43.0 months, P=0.063), MVI was not an independent risk factor for RFS in sHCC. Additionally, MVI was not independently vulnerable to mortality in our population.
CONCLUSIONS: A preoperative model, mainly based on the peritumoral hallmarks of Gd-EOB-DTPA MRI, showed an excellent performance to predict the occurrence of MVI. Nevertheless, MVI was a potential but not an independent risk factor for recurrence and mortality in sHCC ≤3 cm. 2021 Annals of Translational Medicine. All rights reserved.

Introduction

Hepatocellular carcinoma (HCC) is the sixth most prevalent neoplasm and the third leading cause of cancer death (1). The 5-year overall survival (OS) of HCC is approximately 10–20% (2-4), with 5-year recurrence rate after liver resection reaching 50–70% (5). Increasing evidence shows that microvascular invasion (MVI) impairs the surgical outcome (6,7), and tumor size is closely correlated to the incidence of MVI (7) and poor prognosis (8). This implies tumor size may be a potential confounder—a major threat to the retrospective study’s validity—for MVI in predicting prognosis. Meanwhile, early diagnosis of MVI in patients with HCC, especially small HCC (sHCC), will enable clinicians to better understand its pathobiological behaviour for optimizing therapeutic regimens and improving long-term survival.

The definition of sHCC varies greatly owing to its confusing criteria of the maximum tumor diameter from 2 to 5 cm (9-13). Recent studies have indicated that tumor growth over 3 cm is a critical turning point towards more aggressive behaviour (10,12,13). MVI presents in 15–57.1% surgical specimens of HCC and is only visible under microscopy (14). But a wide resection margin is recommended for MVI-positive patients (15). Hence, the effective prediction of histopathological MVI and the identification of key prognostic indicators before surgery can tailor the initial treatment plan to achieve the maximum benefit in sHCC patients.

Recently, some preoperative imaging traits (e.g., peritumoral enhancement) and radiomics features (e.g., texture signatures) have shown great discrimination in the diagnosis of MVI (6,14-16). But few studies focus on the tumor size within 3 cm. Compared with extracellular agents, gadoxetate disodium (Gd-EOB-DTPA) with hepatocyte-specific properties can not only reflect the vascularity of HCC lesions, but also can effectively identify small/early HCC (17-19). Additionally, few studies (20,21) have noticed the application of the Liver Imaging Reporting and Data System (LI-RADS) in MVI or outcome, especially for sHCC patients underwent Gd-EOB-DTPA MRI. Meanwhile, previous studies concerning MVI in HCC patients are retrospective analysis, which cannot avoid selective bias.

Therefore, this study aimed to explore the preoperative value of Gd-EOB-DTPA MRI based on LI-RARS for predicting MVI in patients with sHCC ≤3 cm, and to investigate the correlation between MVI and prognosis by propensity score matching (PSM). We present the following article in accordance with the TRIPOD reporting checklist (available at http://dx.doi.org/10.21037/atm-20-7952).

Methods

Study population

This retrospective, single-center, cohort study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved and informed consent was waived by the Ethical Review Committee of Zhongshan Hospital, Fudan University, Shanghai, China (approval No. B2018-236). Between March 2012 and September 2020, 455 consecutive patients with pathologically confirmed sHCC ≤3 cm underwent liver resection and preoperative Gd-EOB-DTPA MRI, who met the following inclusion criteria (): (I) without gross vascular invasion, bile duct tumor thrombosis or extrahepatic metastasis upon preoperative imaging; (II) without previous history of HCC-related treatments: radiotherapy, chemotherapy, transarterial chemoembolization (TACE), radiofrequency ablation (RFA), immunosuppressive therapy, liver resection or transplantation; (III) complete clinicoradiolgic data (i.e., the pathologic report and laboratory indexes); (IV) MRI with sufficient image quality scanned within 1 month before hepatectomy. Besides, the frequency of MVI stratified by tumor size is plotted in .

Figure 1

Flowchart of the study population. (A) the workflow of our study population before and after propensity score matching (PSM); (B) the frequency of MVI stratified by tumor size; (C) the distribution of propensity scores in MVI-negative and MVI-positive groups before and after PSM; (D) line plot of individual differences indicated that the standard difference of each covariate decreased significantly after matching; (E) histogram of standardized differences showed that the standard difference was concentrated around 0 after PSM, indicating that the variables were balanced and the matching effect was excellent; (F) dot plot of standardized mean differences (SMD) demonstrated an excellent balance in terms of propensity score and the confounder (tumor size), as result of the corresponding SMD reaching 0 after PSM.

Laboratory examinations and histopathology

Preoperative laboratory indexes () comprised serum alpha-fetoprotein (AFP), viral hepatitis type B (HBV), anti-hepatitis C virus, HBV-DNA loads, alanine aminotransferase, aspartate aminotransferase, total bilirubin, direct bilirubin, alkaline phosphatase, γ-glutamyl transpeptidase, albumin, prealbumin, and prothrombin time.

Table 1

Baseline characteristics for identifying MVI before and after PSM

Characteristics	Before PSM				After PSM
	MVI negative (n=332)	MVI positive (n=79)	P		MVI negative (n=171)	MVI positive (n=70)	P
Age (years), mean ± SD	54.57±11.48	54.58±9.72	0.994		55.67±11.30	54.90±9.56	0.615
Sex (female/male)	49 (11.9)/283 (68.9)	12 (2.9)/67 (16.3)	0.923		20 (8.3)/151 (62.7)	10 (4.1)/60 (24.9)	0.580
Size (cm), mean ± SD	1.57±0.63	2.02±0.57	<0.001‡		1.83±0.60	1.95±0.57	0.112‡
≤1/1–2/>2 cm	88/178/66	6/35/38	<0.001		22/97/52	6/35/29	0.224
ES (I–II/III–IV)	218 (53.0)/114 (27.7)	38 (9.2)/41 (10.0)	0.004		112 (46.5)/59 (24.5)	36 (14.9)/34 (14.1)	0.042
Cirrhosis†	118 (28.7)/214 (52.1)	24 (5.8)/55 (13.4)	0.386		66 (27.4)/105 (43.6)	22 (9.1)/48 (19.9)	0.294
Ki-67 (%)	23.68 (18.05)	32.27 (21.79)	0.001‡		24.90 (17.78)	32.56 (22.34)	0.032‡
HBV or HCV†	71 (17.3)/261 (63.5)	14 (3.4)/65 (15.8)	0.470		52 (21.6)/119 (49.4)	13 (5.4)/57 (23.7)	0.060
HBV-DNA (≤104/>104)	242 (58.9)/90 (21.9)	51 (12.4)/28 (6.8)	0.141		119 (49.4)/52 (21.6)	45 (18.7)/25 (10.4)	0.423
Multifocality†	286 (69.6)/46 (11.2)	62 (15.1)/17 (4.1)	0.114		145 (60.2)/26 (10.8)	55 (22.8)/15 (6.2)	0.243
Anatomical resection†	304 (74.0)/28 (6.8)	67 (16.3)/12 (2.9)	0.069		161 (66.8)/10 (4.1)	58 (24.1)/12 (5.0)	0.006
Alpha-fetoprotein (≤20/20–400/>400 ng/mL)	200/103/29	34/32/13	0.012		111/43/17	31/31/8	0.008
TBIL (≤20.4/>20.4 µmol/L)	285 (69.3)/47 (11.4)	63 (15.3)/16 (3.9)	0.176		150 (62.2)/21 (8.7)	57 (23.7)/13 (5.4)	0.203
DBIL (≤20.4/>20.4 µmol/L)	268 (65.2)/64 (15.6)	57 (13.9)/22 (5.4)	0.092		140 (58.1)/31 (12.9)	51 (21.2)/19 (7.9)	0.117
ALB (>35/≤35 g/L)	323 (78.6)/9 (2.2)	79 (19.2/0 (0.0)	0.139§		167 (69.3)/4 (1.7)	70 (29.0)/0 (0)	0.462§
ALT (≤50/>50 U/L)	275 (66.9)/57 (13.9)	68 (16.5)/11 (2.7)	0.485		139 (57.7)/32 (13.3)	61 (25.3)/9 (3.7)	0.272
AST (≤40/>40 U/L)	261 (63.5)/71 (17.3)	70 (17.0)/9 (2.2)	0.044		130 (53.9)/41 (17.0)	62 (25.7)/8 (3.3)	0.028
AKP (≤125/>125 U/L)	309 (75.2)/23 (5.6)	76 (18.5)/3 (0.7)	0.304		157 (65.1)/14 (5.8)	68 (28.2)/2 (0.8)	0.131
GGT (≤60/>60 U/L)	238 (57.9)/94 (22.9)	60 (14.6)/19 (4.6)	0.446		114 (47.3)/57 (23.7)	54 (22.4)/16 (6.6)	0.108
PALB (>250/≤250 mg/L)	79 (19.2)/253 (61.6)	14 (3.4)/65 (15.8)	0.246		38 (15.8)/133 (55.2)	12 (5.0)/58 (24.1)	0.377
PT (≤13.0/>13 s)	283 (68.9)/49 (11.9)	69 (16.8)/10 (2.4)	0.632		140 (58.1)/31 (12.9)	61 (25.3)/9 (3.7)	0.318
Child-Pugh (A/B)	316 (76.9)/16 (3.9)	75 (18.2)/4 (1.0)	0.928§		161 (66.8)/10 (4.1)	66 (27.4)/4 (1.7)	1.000§
Nodule in nodule†	313 (76.2)/19 (4.6)	76 (18.5)/3 (0.7)	0.494		160 (66.4)/11 (4.6)	67 (27.8)/3 (1.2)	0.731§
Fat deposition†	232 (56.4)/100 (24.3)	54 (13.1)/25 (6.1)	0.791		114 (47.3)/57 (23.7)	46 (19.1)/24 (10.0)	0.887
Necrosis†	283 (68.9)/49 (11.9)	57 (13.9)/22 (5.4)	0.006		139 (57.7)/32 (13.3)	54 (22.4)/16 (6.6)	0.465
Hemorrhage†	286 (69.6)/46 (11.2)	69 (16.8)/10 (2.4)	0.780		145 (60.2)/26 (10.8)	62 (25.7)/8 (3.3)	0.445
Ascites†	316 (76.9)/16 (3.9)	75 (18.22)/4 (1.0)	0.928§		162 (67.2)/9 (3.7)	66 (27.4)/4 (1.7)	1.000§
Nonim APHE†	71 (17.3)/261 (63.5)	26 (6.3)/53 (12.9)	0.030		38 (15.8)/133 (55.2)	22 (9.1)/48 (19.9)	0.133
Mosaic architecture†	327 (79.6)/5 (1.2)	77 (18.)/2 (0.5)	0.527§		169 (70.1)/2 (0.8)	68 (28.2)/2 (0.8)	0.707§
Targetoid†	304 (74.0)/28 (6.8)	74 (18.0)/5 (1.2)	0.536		153 (63.5)/18 (7.5)	66 (27.4)/4 (1.7)	0.239
HBP hypointensity†	9 (2.2)/323 (78.6)	3 (0.7)/76 (18.5)	0.606		3 (1.2)/168 (69.7)	3 (1.2)/67 (27.8)	0.490
Non-smooth margin†	226 (55.0)/106 (25.8)	15 (3.6)/64 (15.6)	<0.001		114 (47.3)/57 (23.7)	15 (6.2)/55 (22.8)	<0.001
Non-peripheral washout†	42 (10.2)/290 (70.6)	7 (1.7)/72 (17.5)	0.350		19 (7.9)/152 (63.1)	6 (2.5)/64 (26.6)	0.557
LI–RADS (3/4/5)	24/94/214	7/25/47	0.698		8/52/111	7/21/42	0.294
Enhancing capsule (intact/incomplete/absent)	236/40/56	15/33/31	<0.001		118/26/27	13/27/30	<0.001
Peritumoral enhancement†	269 (65.5)/63 (15.3)	31 (7.5)/48 (11.7)	<0.001		129 (53.5)/42 (17.4)	29 (12.0)/41 (17.0)	<0.001
Peritumoral hypointensity†	295 (71.8)/37 (9.0)	50 (12.2)/29 (7.1)	<0.001		147 (61.0)/24 (10.0)	44 (18.3)/26 (10.8)	<0.001
Propensity score	0.179±0.106	0.249±0.105	<0.001		0.217±0.109	0.238±0.104	0.186

Continuous and categorical variables were expressed as mean ± standard deviation and numbers (proportions), respectively. †, absent/present; ‡, Mann-Whitney U test; §, continuity correction. The other categorical variables, Chi-square test; the other numeric variables, independent samples t-test. MVI, microvascular invasion; PSM, propensity score matching; ES, Edmondson-Steiner grade; HBV, hepatitis B virus; HCV, hepatitis C virus; HBV-DNA, deoxyribonucleic acid of hepatitis B virus; ALT, alanine aminotransferase; AST, aspartate aminotransferase; TBIL, total bilirubin; DBIL, direct bilirubin; AKP, alkaline phosphatase; GGT, γ-glutamyl transpeptidase; ALB, albumin; PALB, prealbumin; PT, prothrombin time; APHE, arterial phase hyperenhancement; HBP, hepatobiliary phase; LI-RADS, Liver Imaging Reporting and Data System.

Histopathological characteristics included tumor size, tumor number, Edmondson-Steiner grade, MVI status, Ki-67 protein expression, liver fibrosis grade based on the Scheuer scoring system. MVI is defined as the cancerous thrombosis of portal vein, hepatic vein, or large capsular vessels lined with endothelium in the contiguous hepatic parenchyma (13-16). MVI grades are stratified as M0 (no MVI), M1 (invasion of micro-vessels up to five at peritumoral parenchyma within 1 cm from tumor surface) and M2 (MVI >5 sites or at >1 cm away from tumor surface) (13,16,22). These indexes were independently assessed by two experienced abdominal pathologists. In case of disagreement, a consensus was reached after discussion.

Meanwhile, the Barcelona Clinic Liver Cancer (BCLC) staging classification, the tumor-node-metastasis (TNM) system of the American Joint Committee on Cancer (8th Edition), the preoperative and postoperative Early Recurrence After Surgery for Liver tumor (ERASL-pre, ERASL-post) models (23) were enrolled into the prognostic study.

Gd-EOB-DTPA MRI

A 1.5T MRI scanner (Magnetom Aera, Siemens Healthcare, Erlangen, Germany) equipped with phased-array coils was performed on all eligible patients. Gd-EOB-DTPA (Primovist, Bayer Pharma, Berlin, Germany) was injected intravenously at a dosage of 0.025 mmol/kg at a rate of 1.5 mL/s, and immediately flushed by 20-mL saline. The conventional MRI sequences of upper abdomen were as below: free-breath diffusion-weighted imaging with b values of 0 and 500 s/mm2, axial T2-weighted imaging with fat suppression, in-phase and opposed-phase T1-weighted imaging (T1WI), pre- and post-contrast dynamic three-dimensional volumetric-interpolated breath-hold examination. When Gd-EOB-DTPA reached the ascending aorta (at 20–30 s after injection), the acquisition of arterial phase (AP) was automatically triggered. The subsequent portal venous phase (PVP), transitional phase (TP), and hepatobiliary phase (HBP) were collected at 60–70 s, 180 s and 20 min after injection, respectively. Detailed MRI parameters are shown in Table S1.

MRI hallmarks

Two radiologists with 8 and 15 years of abdominal MRI experience, who were blinded to MVI status, independently reviewed MRI morphologic hallmarks. Inconsistent diagnoses were discussed to make a final decision.

The major features of LI-RADS (version 2018) (24,25) were as follows: (I) tumor size, the longest axis diameter measured on HBP images; (II) enhancing capsule on PVP or TP images; (III) non-peripheral washout; (IV) non-rim arterial phase hyperenhancement (APHE). LI-RADS categories were calculated by these major features. The ancillary features of LI-RADS (24) were as below: (I) necrosis; (II) hemorrhage; (III) fat deposition; (IV) targetoid architecture; (V) nodule-in-nodule architecture; (VI) mosaic architecture; (VII) hypointensity of tumor on HBP images. Non-LI-RADS morphologic hallmarks comprised (I) non-smooth tumor margin, (II) peritumoral enhancement on AP images (26), (III) peritumoral hypointensity on HBP images (26), (VI) ascites. The definition of MRI hallmarks is shown in Supplementary methods.

Follow-up and data-grouping

Consistent follow-up with laboratory and imaging examinations was performed at intervals of 3 to 6 months after curative resection. Pathologic reports of recurrent/metastatic lesions were also included in the follow-up items, if applicable. The imaging follow-up was conducted by two radiologists, mainly based on the modified Response Evaluation Criteria in Solid Tumors (27) and the reliable diagnosis of HCC by “washin and washout” imaging pattern (1). Specifically, image-guided recurrent/metastatic lesion was determined by metabolic (i.e., maximum standardized uptake value beyond 2.5) or anatomic (e.g., altered lesion size, morphology, diffusivity, signal intensity, and contrast enhancement) pattern recognition of the de novo or original lesion. Notably, the abnormally elevated metabolism was regarded as an important auxiliary determinant of progression.

Patients were censored on September 26, 2020. The date of operation, recurrence, metastasis, death, and the last follow-up were recorded to calculate OS and recurrence-free survival (RFS). Among the 455 enrolled patients, 411 consecutive patients completed at least 6 months of follow-up, named as Pre-PSM cohort, and were used for MVI, PSM and prognosis analyses. In terms of MVI prediction, the 411 patients were divided into training and validation sets, according to the date of MRI examination at a ratio of 7:3. The remaining 44/455 population was classified as the test set to further verify the predictive performance of MVI model. Furthermore, the dataset of the eligible patients after PSM was denoted by post-PSM cohort.

Statistical analyses

Analyses were performed with IBM SPSS Statistics software (version 25.0; IBM Corporation, Armonk, NY, USA) or R (version 3.6.1, https://www.r-project.org) software. Continuous and categorical variables were expressed as mean ± standard deviation and numbers (proportions), respectively. Independent sample t-test, Mann Whitney U test, Chi-square test, Fisher’s exact test and logistic regression was applied to the univariate analysis, as appropriate. The intra-observer consistency of radiologic features was evaluated by McNemar’s and Kappa tests (κ statistics <0.40, poor; 0.40–0.59, fair; 0.60–0.80, good; and 0.80–1.00, excellent). P<0.05 was considered statistically significant.

Multivariate logistic regression (backward stepwise: ward) was used to identify the independent predictors of MVI. The discrimination of prediction model was quantified by the area under receiver operating characteristic curve (AUC) as well as Harrell’s C-index. The consistency between the predicted MVI model and the actual MVI state was evaluated by calibration curve analysis. With Kaplan-Meier analysis of two-sided log-rank test, survival curves were stratified by MVI status. The sample size was estimated by the reported median RFS of MVI and non-MVI cohorts (6). The detailed calculation is illustrated in Supplementary methods, and our sample sizes meets the statistical requirements. Multivariate Cox regression (forward stepwise: maximum like-hood ratio) determined the independent predictors of recurrence/mortality.

Multivariate logistic and Cox regressions were combined to identify confounding factors (28), and PSM was implemented to reduce confounders (28-30). Meanwhile, three R packages “PS MATCHING 3.03”, “SPSS Statistics R Essentials 22.0” and “R-2.15.3-win” were installed in SPSS for PSM analysis (29). The caliper and the optimal match ratio were set at 0.001 and 1:4, respectively. After PSM, the propensity score ensured that no significant differences were found in the baseline contexts of the matched cases, which was visualized in .

Results

eligible sHCC patients’ clinicoradiologic hallmarks for identifying MVI

Baseline characteristics of the training (n=288), validation (n=123) and test (n=44) cohorts for preoperative MVI prediction are shown in Table S2. The representative images of non-MVI and MVI cases are displayed in . Multivariate logistic regression indicated that AFP >20 ng/mL, larger tumor size, incomplete or absent capsule enhancement, peritumoral enhancement and non-smooth margin were significantly susceptible to MVI (P<0.05), both in training and Pre-PSM cohorts. These five independent predictors were visualized as a nomogram () that was further evaluated by calibration curves of training, validation and test cohorts (). The sensitivity, specificity, cut-off, AUC and C-index of the nomogram were as follows: 0.810, 0.848, 0.237, 0.884 (95% CI: 0.838–0.931; P<0.001; ) and 0.874 (95% CI: 0.826–0.921) in the training cohort; 0.762, 0.873, 0.255, 0.845 (95% CI: 0.734–0.956; P<0.001; ) and 0.828 (95% CI: 0.703–0.953) in the validation cohort; 1.000, 0.757, 0.182, 0.903 (95% CI: 0.813–0.994; P=0.001; ) and 0.954 (95% CI: 0.897–1.010) in the test cohort. The formula of the predictive MVI nomogram were illustrated as below:

Figure 2

Representative images of MVI and non-MVI cases. MVI case: Gd-EOB-DTPA MRI displayed an unsmooth lesion (A-G, blue arrows) in hepatic segment VI, with the following architectures: slight hyperintensity of T2-weighted images (A), restricted diffusion (B,C), wedge-shaped peritumoral enhancement on arterial phase images (E, red arrow), absent/incomplete capsule enhancement on transitional phase images (F, red arrow), and peritumoral hypointensity on hepatobiliary phase (G, red arrow). MVI was diagnosis by hematoxylin and eosin (HE, ×100) sample: multiple tumor thrombi of microvasculature (H, blue arrows) were detected in the triangle region between the liver membrane (the left upper corner), normal liver parenchyma (the right upper corner) and the infiltrating HCC lesion without tumor capsule (below). Non-MVI case: The lesion located at hepatic segment IV also detected slight hyperintensity on T2-weighted images (I) and restricted diffusion (J and K), but showed a well-circumscribed smooth tumor edge (L) without peritumoral enhancement and hypointensity (M/O, arrows), intact capsule enhancement (N, O, arrows). HE (100×) signified that no tumor thrombus was detected in microvascular system (P, arrows), which was distributed the normal liver tissue (the left side) that was separated from HCC lesion (the right side) by an intact tumor capsule.

Figure 3

The nomogram, calibration and receiver operating characteristic curves for predicting MVI. The predictive model for MVI in the training cohort was visualized as nomogram (A). The calibration and receiver operating characteristic curves of the training, validation and test cohorts for MVI prediction were plotted in (B) and (E), in (C) and (F), and in (D) and (G), respectively.

Meanwhile, the AUC and C-index yielded 0.875 (95% CI: 0.828–0.921; P<0.001) and 0.863 (95% CI: 0.815–0.911) in Pre-PSM cohort, and reached 0.824 (95% CI: 0.761–0.887; P<0.001) and 0.826 (95% CI: 0.764–0.887) in post-PSM cohort. Additionally, AFP >20 ng/mL, non-smooth margin, incomplete or absent capsule were robust independent risk factors for MVI after PSM.

sHCC patients’ clinicoradiologic hallmarks before PSM

Pathologically confirmed 79 (19.2%) MVI-positive and 332 (80.8%) MVI-negative patients in our hospital. Baseline characteristics of the two groups in Pre-PSM cohort are shown in . Owing to the conspicuous bias, the clinicoradiologic data for 11 of the 37 variables differed significantly (P<0.05), with an unmatched propensity score between MVI and non-MVI cases (0.249±0.105 vs. 0.179±0.106, P<0.001).

Confounders between MVI and prognosis

To investigate whether the existing confounders could lead to any bias, univariate logistic and Cox regressions were first conducted on the Pre-PSM dataset to filter out the variables related to MVI and recurrence/mortality (), respectively. Multivariate logistic regression showed that larger tumor size, AFP >20 ng/mL, non-smooth margin, peritumoral enhancement, incomplete or absent capsule enhancement were significantly positive associated with MVI (P<0.05, ). Meanwhile, larger tumor size, MVI and prealbumin <250 mg/L markedly suffered from postoperative recurrence by the multivariate cox analysis (P<0.05). Moreover, the older, HBV-DNA loads >104, alkaline phosphatase >125 U/L, peritumoral enhancement, and peritumoral hypointensity were independent risk factors for mortality (P<0.05).

Table 2

Univariate logistic and Cox regressions for confounders based on MVI status before and after PSM

Variables	Before PSM (univariate analysis, n=411)									After PSM (univariate analysis, n=241)
	Logistic regression (MVI)			Cox regression (RFS)			Cox regression (OS)			Logistic regression (MVI)			Cox regression (RFS)
	OR (95% CI)	P		HR (95% CI)	P		HR (95% CI)	P		OR (95% CI)	P		HR (95% CI)	P
MVI†	–	–		1.835 (1.264–2.662)	0.001		2.646 (1.155–6.063)	0.021		–	–		1.493 (0.976–2.283)	0.065
Age, mean (SD), years	1.000 (0.978–1.022)	0.994		1.020 (1.004–1.037)	0.012		1.054 (1.019–1.089)	0.002		0.993 (0.968–1.019)	0.614		1.018 (0.998–1.038)	0.075
Sex (male/female)	0.967 (0.487–1.918)	0.923		0.826 (0.523–1.303)	0.411		0,766 (0.262–2.244)	0.628		0.795 (0.351–1.797)	0.581		0.734 (0.422–1.277)	0.273
Size (cm)	2.952 (1.979–4.404)	<0.001		1.549 (1.197–2.006)	0.001		1.658 (0.906–3.034)	0.101		1.429 (0.891–2.294)	0.139		1.345 (0.954–1.897)	0.091
ES (III–IV/I–II)	2.063 (1.256–3.388)	0.004		1.478 (1.057–2.067)	0.022		1.680 (0.753–3.746)	0.205		1.793 (1.019–3.154)	0.043		1.463 (0.973–2.200)	0.068
Cirrhosis†	1.264 (0.744–2.146)	0.386		0.862 (0.609–1.221)	0.403		1.515 (0.601–3.818)	0.378		1.371 (0.759–2.477)	0.295		0.974 (0.636–1.492)	0.904
Ki-67 (%)	1.022 (1.010–1.034)	<0.001		1.007 (0.998–1.016)	0.112		1.020 (1.002–1.039)	0.034		1.020 (1.005–1.034)	0.007		1.002 (0.991–1.012)	0.762
HBV or HCV†	1.263 (0.670–2.362)	0.471		0.872 (0.555–1.370)	0.552		0.562 (0.208–1.520)	0.256		1.916 (0.966–3.800)	0.063		0.827 (0.494–1.386)	0.472
HBV-DNA (>104/≤104)	1.476 (0.877–2.485)	0.143		1.247 (0.869–1.787)	0.231		3.037 (1.360–6.784)	0.007		1.271 (0.707–2.288)	0.423		1.386 (0.903–2.127)	0.135
Multifocality†	1.705 (0.917–3.170)	0.092		1.603 (1.058–2.428)	0.026		0.039 (0–5.001)	0.190		1.521 (0.750–3.085)	0.245		1.677 (1.021–2.755)	0.041
Anatomical resection†	1.466 (0.705–3.046)	0.306		1.001 (0.584–1.716)	0.997		1.781 (0.777–4.081)	0.172		1.440 (0.608–3.409)	0.407		0.932 (0.510–1.700)	0.817
AFP, ng/mL		0.014			0.986			0.535			0.009			0.623
≤20	Reference			Reference			Reference			Reference			Reference
20–400	1.828 (1.067–3.130)	0.028		1.016 (0.705–1.464)	0.934		1.339 (0.555–3.232)	0.516		2.581 (1.403–4.750)	0.002		0.866 (0.551–1.361)	0.533
>400	2.637 (1.248–5.574)	0.011		0.966 (0.555–1.680)	0.902		1.878 (0.598–5.905)	0.281		1.685 (0.665–4.270)	0.271		0.737 (0.372–1.461)	0.382
TBIL (>20.4/≤20.4 µmol/L)	1.540 (0.821–2.890)	0.179		1.237 (0.768–1.993)	0.382		1.043 (0.310–3.511)	0.946		1.629 (0.765–3.469)	0.206		1.470 (0.829–2.604)	0.187
DBIL (>20.4/≤20.4 µmol/L)	1.616 (0.921–2.836)	0.094		1.226 (0.825–1.822)	0.313		0.955 (0.356–2.562)	0.928		1.682 (0.874–3.239)	0.119		1.304 (0.808–2.105)	0.278
ALB (>35/≤35 g/L)	0	0.999		1.721 (0.636–4.661)	0.285		2.037 (0.275–15.105)	0.486		0	0.999		1.161 (0.286–4.722)	0.835
ALT (>50/≤50 U/L)	0.780 (0.388–1.568)	0.486		1.237 (0.801–1.910)	0.337		2.171 (0.899–5.239)	0.085		0.641 (0.288–1.424)	0.275		1.193 (0.704–2.023)	0.512
AST (>40/≤40 U/L)	0.473 (0.225–0.993)	0.048		1.204 (0.795–1.825)	0.380		1.775 (0.736–4.284)	0.202		0.409 (0.181–0.925)	0.032		1.235 (0.752–2.028)	0.404
AKP (>125/≤125 U/L)	0.530 (0.155–1.813)	0.312		1.202 (0.611–2.364)	0.595		2.757 (0.941–8.074)	0.064		0.330 (0.073–1.491)	0.150		1.441 (0.696–2.981)	0.325
GGT (>60/≤60 U/L)	0.802 (0.454–1.416)	0.446		1.261 (0.881–1.804)	0.205		1.5333 (0.671–3.504)	0.311		0.593 (0.312–1.126)	0.110		1.432 (0.938–2.186)	0.096
PALB (≤250/>250 mg/L)	1.450 (0.772–2.723)	0.248		1.695 (1.083–2.655)	0.021		2.309 (0.688–7.752)	0.176		1.381 (0.673–2.833)	0.379		2.031 (1.107–3.728)	0.022
PT (>13/≤13 s)	0.837 (0.404–1.736)	0.633		1.018 (0.626–1.653)	0.944		0.849 (0.253–2.850)	0.791		0.666 (0.299–1.484)	0.320		1.296 (0.765–2.195)	0.335
Child-Pugh (B/A)	1.053 (0.342–3.242)	0.928		1.633 (0.830–3.213)	0.155		0.831 (0.112–6.154)	0.856		0.976 (0.296–3.222)	0.968		1.153 (0.504–2.640)	0.736
Nodule in nodule†	0.650 (0.188–2.254)	0.497		0.971 (0.454–2.079)	0.941		0.046 (0–139.683)	0.451		0.651 (0.176–2.409)	0.521		0.995 (0.403–2.453)	0.991
Fat deposition†	1.074 (0.633–1.823)	0.791		1.060 (0.741–1.518)	0.749		0.757 (0.301–1.909)	0.556		1.043 (0.580–1.877)	0.887		0.962 (0.623–1.485)	0.861
Necrosis†	2.229 (1.251–3.973)	0.007		0.849 (0.542–1.329)	0.474		0.841 (0.287–2.467)	0.752		1.287 (0.654–2.534)	0.465		0.626 (0.354–1.106)	0.107
Hemorrhage†	0.901 (0.433–1.875)	0.780		1.607 (1.042–2.480)	0.032		1.765 (0.658–4.731)	0.259		0.720 (0.309–1.678)	0.446		1.527 (0.902–2.585)	0.115
Ascites†	1.053 (0.342–3.242)	0.928		1.424 (0.724–2.802)	0.306		0.656 (0.154–2.795)	0.569		1.091 (0.325–3.666)	0.888		1.015 (0.443–2.326)	0.972
Nonim APHE†	0.555 (0.324–0.949)	0.032		0.783 (0.541–1.135)	0.197		0.624 (0.267–1.459)	0.276		0.623 (0.335–1.159)	0.135		1.087 (0.678–1.743)	0.728
Mosaic architecture†	1.699 (0.323–8.921)	0.531		0.047 (0.000–5.961)	0.216		0.048 (0–14212.712)	0.637		2.485 (0.343–18.002)	0.368		0.048 (0–29.068)	0.352
Targetoid†	0.734 (0.274–1.964)	0.538		1.086 (0.612–1.928)	0.778		1.501 (0.446–5.048)	0.511		0.515 (0.168–1.581)	0.246		0.979 (0.490–1.958)	0.953
HBP hypointensity†	1.417 (0.375–5.358)	0.608		0.612 (0.195–1.923)	0.401		0.047 (0–1017.030)	0.549		2.507 (0.494–12.736)	0.268		0.337 (0.047–2.432)	0.281
Non-smooth margin†	9.097 (4.954–16.704)	<0.001		1.420 (1.016–1.984)	0.040		2.509 (1.072–5.870)	0.034		7.333 (3.815–14.096)	<0.001		1.304 (0.868–1.960)	0.201
Non-peripheral washout†	1.490 (0.643–3.453)	0.353		0.843 (0.513–1.384)	0.499		1.357 (0.319–5.777)	0.679		1.333 (0.509–3.493)	0.558		1.025 (0.531–1.979)	0.941
LI-RADS		0.699			0.751			0.582			0.310			0.704
LR-3	Reference			Reference			Reference			Reference			Reference
LR-4	0.912 (0.352–2.359)	0.849		0.848 (0.445–1.616)	0.617		0.572 (0.148–2.217)	0.419		0.462 (0.149–1.434)	0.181		0.845 (0.347–2.058)	0.711
LR-5	0.753 (0.306–1.851)	0.536		0.797 (0.435–1.461)	0.463		0.515 (0.147–1.798)	0.298		0.432 (0.148–1.267)	0.126		1.028 (0.444–2.382)	0.949
Capsule enhancement		<0.001			0.009			0.032			<0.001			0.019
Incomplete/intact	12.980 (6.470–26.042)	<0.001		1.680 (1.104–2.556)	0.015		3.236 (1.214–8.627)	0.019		9.426 (4.294–20.690)	<0.001		1.628 (0.987–2.686)	0.057
Absent/intact	8.710 (4.404–17.223)	<0.001		1.696 (1.136–2.530)	0.010		2.990 (1.122–7.969)	0.029		10.085 (4.653–21.860)	<0.001		1.930 (1.187–3.138)	0.008
Peritumoral enhancement†	6.611 (3.898–11.214)	<0.001		1.653 (1.168–2.340)	0.005		3.543 (1.573–7.981)	0.002		4.342 (2.409–7.829)	<0.001		1.584 (1.050–2.389)	0.028
Peritumoral hypointensity†	4.624 (2.612–8.186)	<0.001		1.697 (1.143–2.521)	0.009		2.376 (1.016–5.557)	0.046		3.619 (1.891–6.926)	<0.001		1.442 (0.911–2.281)	0.118

†, present/absent. MVI, microvascular invasion; PSM, propensity score matching; ES, Edmondson-Steiner grade; HBV, hepatitis B virus; HCV, hepatitis C virus; HBV-DNA, deoxyribonucleic acid of hepatitis B virus; ALT, alanine aminotransferase; AST, aspartate aminotransferase; TBIL, total bilirubin; DBIL, direct bilirubin; AKP, alkaline phosphatase; GGT, γ-glutamyl transpeptidase; ALB, albumin; PALB, prealbumin; PT, prothrombin time; APHE, arterial phase hyperenhancement; HBP, hepatobiliary phase; LI-RADS, Liver Imaging Reporting and Data System.

Table 3

Multivariate logistic analysis for predicting MVI before and after PSM

Variables	Before PSM						After PSM (post-PSM cohort)
	Training cohort			Training & validation cohorts (pre-PSM cohort)
	OR (95% CI)	P		OR (95% CI)	P		OR (95% CI)	P
AFP, ng/mL		0.011			0.020			0.040
≤20	Reference			Reference			Reference
20–400	2.793 (1.249–6.246)	0.012		2.081 (1.050–4.122)	0.036		2.347 (1.139–4.835)	0.021
>400	4.342 (1.308–14.406)	0.016		3.580 (1.287–9.956)	0.015		2.623 (0.842–8.172)	0.096
DBIL (>20.4/≤20.4 µmol/L)	–	–		1.862 (0.909–3.813)	0.089		–	–
ES grade (III–IV/I–II)	–	–		1.758 (0.936–3.305)	0.080		–	–
Tumor size (cm)	2.235 (1.240–4.028)	0.007		2.187 (1.333–3.587)	0.002		–	–
Non-smooth margin†	3.443 (1.425–8.319)	0.011		3.429 (1.648–7.134)	0.001		3.667 (1.756–7.661)	0.001
Capsule enhancement		<0.001			<0.001			<0.001
Incomplete/intact	6.623 (2.490–17.618)	<0.001		4.901 (2.171–11.063)	<0.001		4.970 (2.101–11.756)	<0.001
Absent/intact	5.860 (2.148–15.987)	0.001		5.523 (2.456–12.423)	<0.001		7.288 (3.095–17.160)	<0.001
Peritumoral enhancement†	2.257 (1.044–4.879)	0.038		2.568 (1.320–4.993)	0.005		–	–

†, present/absent. MVI, microvascular invasion; PSM, propensity score matching; AFP, alpha-fetoprotein; DBIL, direct bilirubin; ES, Edmondson-Steiner grade; OR, odd ratio; CI, confidence interval.

Therefore, tumor size was selected out to be the only confounder of MVI for predicting recurrence in our study population. Additionally, the frequency of MVI dramatically increased as tumor size growing (, P<0.001): ≤1 cm, 7.5% (8/107); 1.1–2 cm, 16.6% (38/229); 2.1–3 cm, 34.2% (39/114); 3.1–4 cm, 49.0% (25/51); 4.1–5 cm, 66.7% (12/18); 5.1–9 cm, 78.3% (18/23); 9.1–12.6 cm, 100% (8/8).

matched sHCC patients’ clinicoradiologic hallmarks after PSM

Though reducing the confounder—tumor size, 70 of 79 MVI-positive cases were matched with 171 of 332 MVI-negative subjects. The distribution of propensity scores in MVI-negative and MVI-positive groups before and after PSM is shown in . After PSM, the standard difference of each covariate decreased significantly (), the standard difference was concentrated around 0 (), and the standardized mean difference of tumor size and propensity score decreased to 0 (), all of which indicated an excellent matching effect. Consequently, no bias was found for 37 clinicoradiologic characteristics in the matched MVI and non-MVI groups (propensity score: 0.238±0.104 vs. 0.217±0.109, P=0.186; ).

Prognostic value of MVI

Univariate Cox analysis demonstrated that MVI exerted an adverse impact on RFS of Pre-PSM (HR: 1.835, 95% CI: 1.264–2.662, P=0.001) and Post-PSM (HR: 1.493, 95% CI: 0.976–2.283, P=0.065) cohorts. Meanwhile, the Kaplan-Meier curve () signified that non-MVI patients benefited in a longer median RFS than that of MVI patients, whether before PSM (85.4 vs. 43.0 months, P=0.001) or after PSM (74.3 vs. 43.0 months, P=0.063). Baseline characteristics of recurrent prediction models before and after PSM are presented in Table S3.

Figure 4

The Kaplan-Meier curves in terms of MVI status and grade. Recurrence-free survival curves were stratified by MVI status (A) and MVI grade (B) in the Pre-PSM subgroup, and were classified by MVI status (C) and MVI grade (D) in the Post-PSM subgroup, respectively. In addition, the overall survival curves were divided by MVI status (E) and MVI grade (F), respectively.

Before PSM, MVI (HR 1.555, 95% CI: 1.055–2.293, P=0.026), prealbumin <250 mg/L (HR 1.625, 95% CI: 1.037–2.548, P=0.034) and tumor size (HR 1.412, 95% CI: 1.078–1.849, P=0.012) were independently adverse to RFS. The formulas are as below, with AUC of 0.662 (95% CI: 0.604–0.719, P<0.001). Notably, the clinicoradiogic model outperformed the BCLC, TNM, ERASL-pre and ERASL-post models in predicting RFS ().

Table 4

The prediction models for recurrence before and after PSM

Model	Pre-PSM cohort (n=411)			Post-PSM cohort (n=241)
	AUC (95% CI)	P		AUC (95% CI)	P
Clinicoradiologic model	0.662 (0.604–0.719)	<0.001		0.658 (0.587–0.730)	<0.001
BCLC stage	0.573 (0.514–0.632)	0.015		0.550 (0.476–0.625)	0.187
TNM stage	0.564 (0.504–0.624)	0.033		0.555 (0.480–0.631)	0.148
ERASL-post risk	0.511 (0.461–0.570)	0.723		0.503 (0.428–0.578)	0.933
ERASL-pre risk	0.509 (0.450–0.568)	0.766		0.504 (0.429–0.579)	0.917

BCLC, Barcelona clinic liver cancer; TNM, tumour-node-metastasis system of the American Joint Committee on Cancer (8th Edition); ERASL, early recurrence after surgery for liver tumour. ERASL-pre score =0.818 gender (0: female,1: male) + 0.447 Albumin-Bilirubin grade (0: Grade 1; 1: Grade 2 or 3) + 0.100 ln (serum AFP in ng/mL) + 0.580 ln (tumour size in cm) + 0.492 tumour number (0: single; 1: two or three; 2: four or more). Low risk: score ≤2.558; intermediate risk: score: 2.558–3.521; High risk: score >3.521 [5]. ERASL-post score =0.677 gender (0: female,1: male) + 0.458 Albumin-Bilirubingrade (0: Grade 1; 1: Grade 2 or 3) + 0.661 microvascular invasion (0: no,1: yes) + 0.082 ln (serum AFP in ng/mL) + 0.451 ln (tumour size in cm) + 0.379 tumour number (0: single; 1: two or three; 2: four or more). Low risk: score ≤2.332; Intermediate risk: score: 2.332–3.445; High risk: score >3.445 [6].

After PSM, prealbumin <250 mg/L (HR 1.980, 95% CI: 1.082–3.660, P=0.027), γ-glutamyl transpeptidase >60 U/L (HR 1.588, 95% CI: 1.036–2.434, P=0.034), incomplete or absent capsule enhancement (HR 1.804, 95% CI: 1.190–2.736, P=0.005) were independently induced to recurrence. Similarly, this clinicoradiogic algorithm (AUC 0.658, 95% CI: 0.587–0.730, P<0.001, ) was also superior to the BCLC, TNM, ERASL-pre and ERASL-post models for RFS prediction. However, MVI did not independently impair RFS in sHCC ≤3 cm after balancing the confounder—tumor size.

In the multivariable Cox regression of OS, the older (HR 1.045, 95% CI: 1.010–1.081, P=0.012), alkaline phosphatase >125 U/L (HR 3.143, 95% CI: 1.144–8.636, P=0.026), HBV–DNA load >104 (HR 2.045, 95% CI: 1.044–4.009, P=0.037), peritumoral enhancement (HR 2.319, 95% CI: 1.126–4.778, P=0.022), and peritumoral hypointensity (HR 2.377, 95% CI: 1.126–5.018, P=0.023) were independently associated with mortality before PSM:

The median OS of MVI-positive and MVI-negative patients were 90.533 and >103.067 months (P=0.064), respectively. Despite MVI had a significant effect on mortality in univariate Cox regression (HR: 2.646, 95% CI: 1.155–6.063, P=0.021, ), it failed to be an independent risk factor for mortality. Additionally, we further stratified MVI status into M0, M1, and M2, with median OS of >103.1, >81.7 and 90.5 months (P=0.005), respectively. However, MVI grade only significantly impaired OS in univariate Cox regression (P=0.011; M1, HR: 1.317, 95% CI: 0.539–3.220, P=0.546; M2, HR: 4.383, 95% CI: 1.671–11.493, P=0.003), it was also not independently prone to mortality. Accordingly, this MVI-related paper did not involve the Post-PSM analysis of overall survival. In addition, the comparison of the nomogram for predicting MVI between the published studies (6,14,15,31-36) and our research is summarized in .

Table 5

The nomograms for predicting MVI in our and the published studies

Previous studies	Sample size (mean tumor size)	Training cohort (AUC, 95% CI)	Validation cohort (AUC, 95% CI)	Model category	Outcome
Yan et al. (31)	268 (57.8% of patients >5 cm)	0.828 (0.767–0.886)	0.804 (0.707–0.901)	Clinical model	Present
Wang et al. (32)	454 (>4 cm)	0.79 (0.74–0.84)	0.81 (0.74–0.89)	Clinical model	Absent
Zhang et al. (33)	267 (>4 cm)	0.784 (0.719–0.840)	0.820 (0.713–0.900)	MRI radiomics	Absent
Ma et al. (37)	157 (>3.2 cm)	0.835 (0.760–0.910)	0.801 (0.672–0.930)	CT radiomics	Absent
Gao et al. (34)	89 (>5 cm)	0.851 (0.761–0.951)	0.861 (0.725–0.996)	Clinicoradiogic model of CT	Absent
Deng et al. (35)	513 (6.29 cm)	0.71 (0.66–0.75)	–	Clinicoradiogic model (CT/MR/US)	Absent
Peng et al. (36)	304 (>4.9 cm)	0.846 (0.787–0.905)	0.844 (0.774–0.915)	CT radiomics	Absent
Yang et al. (15)	208 (>1.7 cm)	0.943 (0.905–0.980)	0.861 (0.750–0.970)	Gd-EOB-DTPA MRI radiomics	Absent
Lei et al. (14)	707(>3.15 cm)	0.81 (0.78–0.85)	0.80 (0.75–0.86)	Clinicoradiogic model (CT/MR/US)	Present
Xu et al. (6)	495 (>4.9 cm)	0.909 (0.860–0.928)	0.889 (0.851–0.919)	CT radiomics	Present
Our study of HCC ≤3 cm	455 (>1.67 cm); Post-PSM: 241 (>1.83 cm)	0.884 (0.838–0.931)†; Pre-PSM: 0.875 (0.828–0.921)	0.845 (0.734–0.956)‡; 0.903 (0.813–0.994)§; Post-PSM: 0.824 (0.761–0.887)	Clinicoradiogic model based on Gd-EOB-DTPA MRI	Present

†, the training cohort (n=288); ‡, the validation cohort (n=123); §, the test cohort (n=44). Pre-PSM: Pre-PSM cohort (n=411); Post-PSM cohort (n=241). Gd-EOB-DTPA, gadoxetate disodium; US, ultrasound examination.

Discussion

Predictive and preventable strategies of MVI based on precision medicine will bring profound benefits (28). Compared with large HCC, sHCC within 3 cm has unique genetic changes (13,38), relatively homogeneous DNA stemline ploidy (39), benign biological behavior and favorable prognosis (10,12,13,39). Thus, this study aimed to investigate whether Gd-EOB-DTPA MRI is an efficient preoperative biomarker for MVI and whether MVI independently impairs prognosis in patients with sHCC ≤3 cm using a RCT-like method—PSM. We found that AFP >20 ng/mL, non-smooth margin, incomplete or absent capsule, peritumoral enhancement and larger tumor size were independent risk factors for histologic MVI. The frequency of MVI significantly elevated as tumor size growing (from 0 to 12.6 cm). Despite MVI independently impaired RFS before balancing the confounder—tumor size, it was ultimately identified as a potential but not an independent risk factor for recrudescence of sHCC ≤3 cm after PSM. Similarly, MVI was not independently detrimental to OS.

Elevated AFP level (14,15), non-smooth margin (6,7,15), incomplete/absent capsule enhancement (6,14), increased tumor size (14,40) and peritumoral enhancement (6,7,15) have been reported as independent risk factors for MVI, which also highly suitable for our patients with sHCC ≤3 cm. Furthermore, the first three predictors remained reliability and robustness in MVI prediction after PSM. Notably, peritumoral enhancement, peritumoral hypointensity, capsule and margin statuses imply peritumoral biological behaviour, reflecting an aggressive tendency to invade the tumor capsule and infiltrate into the peritumoral non-neoplastic parenchyma (41). Histologically, HCC’s capsule is composed of a fibrous-rich inner layer, and an outer layer containing portal venules (or sinusoids) and newly formed bile ducts (19). Thus, an intact capsule may serve as a border defenced against HCC’s dissemination and progression. Once cancerous cells break through the capsule, imaging signs of the incomplete/absent capsule enhancement and infiltrative border (non-smooth margin) will appear (42). These may interpret the excellent performance of capsule status in predicting MVI and even recurrence after PSM.

Pathologically, peritumoral parenchyma is the representative of tumor heterogeneity, rich in highly invasive cells (13), and is the first and most common susceptible region of MVI (26,41). Peritumoral enhancement potentiates the formation of MVI, which may stem from (I) compensatory arterial hyperperfusion, which often occurs in areas of decreased portal flow (e.g., minute portal venule occlusion caused by tumor thrombi) (15,42); (II) high expression of placental growth factor and vascular endothelial growth factor receptor (VEGFR) in peritumoral zones (43). Moreover, the tumor-host interaction between VEGFR and MVI has been reported to promote the progression of renal cell carcinoma (44). Hence, these may explain that peritumoral enhancement independently induce MVI and mortality in sHCC ≤3 cm. Analogically, peritumoral hypointensity independently shortened OS and is a potential risk indicator for MVI in our study. This imaging trait reflects the peritumoral reduction of Gd-EOB-DTPA uptake on HBP images, that is, a decreased expression of organic anionic transporting polypeptide (45). It may be induced by impaired hepatocyte function (45,46) or Kupffer cells injury in tumorous arterioportal shunts (46) because of portal branches obstruction by cancer embolus (i.e., MVI). Controversially, this radiologic feature is paucity of robustness for identifying MVI, RFS or OS in different studies (6,7,15,45,47-49). Thus, this equivocal finding needs further study.

According to the LI-RADS category, most patients are classified into the high-risk HCC of LR-5. Among the major features of LI-RADS, only tumor size and enhancing capsule were closely correlated with MVI and RFS. Similarly, in the majority of LI-RADS auxiliary features after univariate analysis, only non-rim APHE and necrosis were remarkably vulnerable to MVI, and only haemorrhage was markedly susceptible to recurrence. Generally, LI-RADS category and most LI-RADS features cannot independently predict MVI and prognosis. Our findings are basically agreement with the relationship between LI-RADS category/features and MVI in combined hepatocellular-cholangiocarcinoma (50), and postoperative recurrence in early HCC (21) or in LR-5 HCC (20,47). Previous study indicated (20) that mosaic architecture independently identified MVI and fat deposition independently predicted recurrence in LR-5 HCC after hepatectomy. The controversial results may originate from different grades of LI-RADS and thresholds of tumor size. Therefore, more homogeneous studies are needed to further explore these hallmarks for expanding the applicability of LI-RADS.

Compared with the published nomograms of MVI (6,14,15,31-36), only our study focused on sHCC ≤3 cm and the confounding factors of MVI for predicting prognosis. More precisely, two of ten published studies involved the prognostic analysis with the findings given below: the CT radiomics-related index of Xu et al. (6) and the clinicoradiologic indicators (HBV-DNA load, tumor size and number) of Lei et al. (14) were independent risk factors of MVI and recurrence/mortality. However, the two studies ignored the above-mentioned confounders between MVI and outcomes, and thus failed to recognize the actual role of MVI in prognosis to a certain extent. Besides, the study of Yan et al. (31) paid more attention to HCC lesions with tumor size >5 cm (155/268) and lymph node metastasis (41/268), and ignored the importance of imaging hallmarks in predicting MVI. Above all, while the Kaplan-Meier curves of Yan et al. (31) signified significance differences between MVI and non-MVI patients in RFS and OS, this study did not involve the multivariable COX analysis. Hence, this published paper (31) cannot interpret whether MVI independently predicts prognosis, and whether there are confounders between MVI and outcomes.

Controversially, MVI exerts an ambiguous effect on sHCC’s outcome. A non-PSM study of Du et al. (9) presented that MVI independently shortened OS and progression-free survival, and increasing tumor size facilitated higher incidences of MVI in solitary sHCC ≤3 cm. The results were highly consistent with our findings before PSM. However, the conclusion of Du et al. (9) neglected the bias of respective study and the potential confounder (e.g., tumor size) of MVI in prognostic analysis. Intriguingly, we identified that tumor size was the only confounder for MVI to predict RFS in our sHCC cases, confirming the hypothesis mentioned in the introduction. Furthermore, tumor size was significantly positive relevant to MVI incidence from 0 to 12.6 cm. This may interpret that unlike previous studies of larger tumor size (6,51), histologic MVI status or even MVI grades cannot independently induce to postoperative recurrence and mortality in our study. Contrary to our findings, the PSM study of Wang et al. (22) demonstrated that MVI engendered a negative influence on the RFS of solitary sHCC ≤2 cm. Nevertheless, this study only balanced the bias variables (P<0.1) between MVI and non-MVI groups, but neglected to reduce the confounder (alanine transaminase) which independently associated with both MVI and RFS. Meanwhile, this study displayed a higher percentage of MVI (pre-PSM cohort: 33.1%; post-PSM cohort: 50%) than that of our population with maximal size ≤2 cm (pre-PSM cohort: 13.7%; post-PSM cohort: 13.4%). In brief, our study had a lower but more robust incidence of MVI before and after PSM, and balanced the actual confounder between MVI and RFS, leading to an opposite conclusion of Wang et al. (22). Furthermore, Shindoh et al. (52) presented that the presence of MVI (26.5%) did not affect the long-term outcome of sHCC ≤2 cm, which was analogous with our findings. The prognostic discrepancy of MVI among sHCC studies may attribute to the mixture of different tumor sizes or stages, MVI incidences, underlying liver diseases, geographic heterogeneity, surgical procedure (e.g., wide or narrow resection margin, anatomical or non-anatomical resection, curative resection or liver transplantation) or postoperative treatment (e.g., preventive TACE, immunosuppression therapy). Consequently, the impact of MVI on sHCC’s prognosis remains a challenge, and more homogeneous studies need to be further verified.

At present, the radical treatment ratio of patients with sHCC ≤3 cm is markedly increased, posing an urgent and practical issue in hepatic surgery (12). Hence, we speculate that this random controlled trial (RCT)-like PSM study allows clinicians more time to optimize therapeutic schedule (I) with a reasonable surgical margin, because MVI status is independently irrespective of recurrence/mortality in sHCC ≤3 cm; (II) with a reasonable residual liver volume, for reducing the frequency of hepatic dysfunction/infection and improving the safety of liver resection (53). Furthermore, solitary HCC ≤3 cm can be ablated almost completely after RFA (54). Nowadays, the disparity of HCC patients’ survival between curative RFA and surgery is narrowing, especially for sHCC within 3 cm (55). Hence, this PSM study also provides additional evidence for clinicians to apply RFA for sHCC ≤3 cm, regardless of the histologic MVI status.

Several limitations should be noted. Firstly, this is a retrospective single-center study. Thus, we implemented PSM analysis to simulate a random controlled trial for minimizing selection bias. Meanwhile, the robustness of our results needs to be further verified by external and prospective dataset. Secondly, while BCLC, TNM, ERASL-pre, ERASL-post risk and the clinicoradiogic models have been used to predict prognosis, the AUCs of these algorithms were unsatisfactory. The results may stem from (I) the paucity of immunohistochemical biomarkers and postoperative characteristics; (II) the exclusion of well-established key risk factors for dismal outcomes (e.g., tumor size >3 cm, tumor thrombus in gross vessels) in our population; (III) the absence of radiomics that can high-throughput extract quantitative imaging signatures to improve diagnostic or prognostic accuracy (16,56). Hence, the above three points can be implemented to further investigate the relationship between MVI and prognosis of sHCC.

Conclusions

Mainly based on the peritumoral hallmarks rather than LI-RADS features of Gd-EOB-DTPA MRI, our preoperative nomogram can excellently distinguish the histopathological MVI status. However, MVI is a potential but not an independent risk factor for prognosis in patients with sHCC ≤3 cm.

The article’s supplementary files as