Differentiation of dual- and non-dual-phenotype hepatocellular carcinoma based on contrast-enhanced computed tomography and patient characteristics.

Introduction
In 2020, primary liver cancer (PLC) was the sixth most commonly diagnosed cancer and the third most common cause of cancer-related death worldwide, accounting for 906,000 new cases and 830,000 deaths (1). Hepatocellular carcinoma (HCC) accounts for 75–85% of cases of PLC (1). Although the prognosis of HCC patients continues to improve through advances in therapies (2), many patients experience recurrence after potentially curative treatment, even when the disease is treated at an early stage (3,4).
These considerations highlight the importance of elucidating what makes some cases of HCC more likely to recur than others, which may help to channel limited medical resources to patients at greater risk. A subtype of HCC associated with higher risk of recurrence is dual-phenotype hepatocellular carcinoma (DPHCC) (5,6), which accounts for approximately 10% of HCC cases. Combined hepatocellular cholangiocarcinoma (CHC) is characterized as a PLC comprising the characteristics of both unequivocal hepatocytic and cholangiocytic differentiation with transitional features within the same tumor (7). Different from CHC, DPHCC shows typical HCC morphological features, but some of its cells (>15%) strongly coexpress both hepatocyte and cholangiocyte markers (6,8). The prognosis for DPHCC is worse compared with that for non-DPHCC, with a reported median overall survival (OS) and recurrence-free survival of 30.4 and 13.2 months, compared with 43.6 and 23.4 months for non-DPHCC, respectively (6). Research into this particularly aggressive form of HCC may identify morphological and molecular characteristics linked to recurrence, yet little is known about its clinicodemographic characteristics or imaging features (5,9,10).
The present study explored these aspects of DPHCC, and our imaging modality was contrast-enhanced computed tomography (CECT), which can detect focal liver lesions with high sensitivity and specificity (2,11,12). Our results suggest that this technique, when combined with certain clinicodemographic characteristics in HCC patients, may enable preoperative differentiation between DPHCC and non-DPHCC. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-24-1168/rc).

Methods

Patients
This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by ethics review board of Guangxi Medical University Cancer Hospital (No. KY2023847), and the requirement for informed consent was waived due to the retrospective nature of the design. Data and images were analyzed from a consecutive series of HCC patients who underwent CECT between January 2020 and December 2020 at our institution, followed within four weeks by potentially curative liver resection. In all patients, HCC was confirmed based on surgical pathology. Patients were excluded if they received anti-HCC treatment before imaging; if they were diagnosed with CHC based on surgical pathology, or if the results of immunohistochemistry against CK7/CK19 were unavailable. All patients were followed up to their date of death or the date this study was censored on 7 April 2025 after surgery.

Pathological criteria
Histopathologic sections showed typical HCC. Immunohistochemical staining was performed on hepatocyte markers (e.g., Hep Par 1, GPC-3, GS) and cholangiocyte markers (e.g., CK19, CK7). DPHCC was defined as the moderate or strong expression of hepatocyte markers and cholangiocyte markers in more than 15% of the tumor cells.

CECT and imaging analysis
The images analyzed in this study had been obtained under the following conditions. A total volume of 100 mL of non-ionic contrast medium iohexol (50 mL:15 g, Yangtze River Pharmaceutical Group, Jiangsu, China) was administered at 3 mL/s, then CECT was performed on 64-MDCT scanners, either a SOMATOM Sensation 64 (Siemens, Erlangen, Germany) or a Discovery CT750HD (GE Healthcare, Chicago, IL, USA), using z-axis modulation, spiral pitch of 1, section thickness of 5 mm, field of view of 300–350 mm, tube voltage of 120 kVp, and current of 230 mA. Patients were asked to hold their breath during scanning, which began 25–35 seconds after the end of contrast medium injection in order to image the hepatic arterial phase (AP), and 60–70 seconds after the end of injection in order to image the portal venous phase (PVP). The entire liver was scanned. Coronal and sagittal images 5 mm thick were reconstructed using the manufacturer-supplied post-processing algorithm with a gap of 2 mm.
Two radiologists, each of whom had more than 10 years of experience with CECT of the liver, retrospectively analyzed tomographic images according to the Liver Imaging Reporting and Data System (LI-RADS) 2018 [American College of Radiologists (ACR), Reston, VA, USA] in order to extract CECT features.
The CECT features were as follows: (I) tumor size: largest outer-edge-to-outer-edge dimension on transverse images; (II) non-smooth tumor margin: irregular tumors with budding portions at the periphery; (III) rim arterial phase hyperenhancement (APHE): spatially defined subtype of APHE in which AP enhancement is most pronounced in observation periphery; (IV) non-rim APHE: non-rim-like enhancement in AP unequivocally greater in whole or in part than liver; (V) arterial peritumoral enhancement: the detectable enhancement outside the tumor margin that broadly contacted the tumor border in AP, becoming isodense with background liver parenchyma in PVP; (VI) intratumoral arteries: tumor has internal arteries in AP; (VII) peripheral washout: spatially defined subtype of washout in which apparent washout is most pronounced in observation periphery; (VIII) non-peripheral washout: nonperipheral visually assessed reduction in enhancement in whole or in part relative to liver tissue from AP to PVP; (IX) delayed enhancement: area of progressive PVP enhancement; (X) mosaic architecture: presence of randomly distributed internal nodules or compartments; (XI) targetoid: hyperdense in observation periphery with hypodense in center in PVP; and (XII) enhancing capsule: a low-density rim around the tumor in AP but enhanced in PVP. In patients with multiple lesions, only the largest was analyzed in this study. The radiologists were blinded to patient data when they interpreted the images. Discrepancies in their assessments were resolved through discussion.

Statistical analysis
Statistical analyses were performed using the software SPSS 26.0 (IBM Corp., Armonk, NY, USA) and MedCalc 16.1 (MedCalc Software, Ostend, Belgium). Data were expressed as mean ± standard deviation if normally distributed, or as median (interquartile range) if skewed. Inter-group differences in continuous variables were assessed for significance using an independent-samples t-test or Mann-Whitney U-test. Differences in categorical variables were assessed using the χ2 or Fisher’s exact tests. Kaplan-Meier survival analysis was used to assess OS. Differences associated with P<0.05 were considered statistically significant.
Inter-observer agreement for CECT features were assessed by calculating an interclass correlation coefficient or kappa coefficient. The interclass correlation coefficient value was defined as follows: 0–0.39: poor; 0.40–0.59: fair; 0.60–0.74: good; and 0.75–1.00: excellent. The kappa value was defined as follows: 0–0.20: slight; 0.21–0.40: fair; 0.41–0.60: moderate; 0.61–0.80: substantial; and 0.81–1.0: almost perfect.
Univariate analysis was performed to identify clinicodemographic and CECT features that were associated with DPHCC and non-DPHCC. Variables that were associated with P<0.05 in this analysis were entered into multivariate logistic regression involving stepwise entry, a forward likelihood ratio model, and removal probability of 0.05. Results were reported in terms of odds ratio (OR) and associated 95% confidence interval (CI). Variables that emerged as significant from multivariate regression were assessed for their ability to predict DPHCC in our sample in terms of the area under a receiver operating characteristic curve (AUC). An imaging model based on CECT features and a combined model based on clinicodemographic characteristics and CECT features were constructed respectively. Delong’s test was used for comparison of AUC values between the two models.

Results
Of the 169 patients in the final analysis (Figure 1), 148 were men, the mean age across all participants was 52.01±12.14 years, and 29 (17.2%) had DPHCC. Although patients with DPHCC or non-DPHCC were similar in age and alpha-fetoprotein (AFP) level, those with DPHCC were significantly more likely to be female and significantly less likely to be infected with HBV (Table 1).
The inter-observer interclass correlation coefficient value was 0.989 (95% CI: 0.985–0.992) for tumor size, which indicated excellent agreement. The inter-observer kappa values were 0.443 to 0.694 for other CECT features, which indicated moderate and substantial agreement, respectively (Table 2).
Among all CECT features that we examined, rim APHE and peripheral washout were significantly more frequent among patients with DPHCC, whereas non-rim APHE and non-peripheral washout were significantly more frequent among patients with non-DPHCC (Table 3, Figures 2,3).
Multivariate logistic regression involving the variables of rim APHE, non-rim APHE, and peripheral or non-peripheral washout identified rim APHE as an independent predictor of DPHCC (OR 11.040, 95% CI: 1.98–63.532). The imaging model was constructed based on rim APHE-predicted DPHCC in our sample with an AUC of 0.562 (95% CI: 0.439–0.684).
Multivariate logistic regression involving the variables of female sex, HBV infection, rim APHE, non-rim APHE, and peripheral or non-peripheral washout identified three as independent predictors of DPHCC: female sex, HBV infection, and rim APHE (Table 4). The combined model was constructed based on three independent predictors predicted DPHCC in our sample with an AUC of 0.716 (95% CI: 0.603–0.829). Delong’s test showed that the AUC of the combined model was higher than that of the imaging model, and the difference was statistically significant (Z=3.207, P=0.0013) (Figure 4).
The OS rates of the patients in the DPHCC and non-DPHCC groups were 68.7% and 77.2%, respectively. Kaplan-Meier survival analysis showed no statistical difference in OS rates between groups (P=0.362) (Figure 5).

Discussion
Here, we identified female sex, infection with HBV, and some features of CECT as being significantly associated with DPHCC, and we found that the combination of female sex, HBV infection, and rim APHE may be able to predict this subtype of HCC. These findings may aid in early detection of DPHCC or at least in the screening of HCC patients for greater risk of recurrence.
In this study, we found that rim APHE was an independent risk factor for DPHCC; the result of the present study is similar to that reported by Liu et al. (13), but differ from Gu et al.’s analysis (14). It may be related to sample bias in these studies. Future work should verify the result with larger, multicenter samples. Our finding suggests that when a patient’s CECT exhibits rim APHE, they are 15-fold more likely to have DPHCC. Our finding of a strong association between rim APHE and DPHCC is consistent with the known association between peripheral rim-like enhancement and intra-hepatic cholangiocarcinoma (ICC) (15,16). DPHCC cells express both HCC and ICC markers, and the rim APHE may reflect the contribution of cells with a cholangiocytic phenotype peripheral hypercellularity within the DPHCC tumor. It would be interesting to explore whether the intensity of the rim APHE correlates with the proportion of tumor cells with a cholangiocytic phenotype. The inter-observer agreement value for rim APHE was 0.543, which is not so high. Although LI-RADS has detailed definitions for each feature, and the two radiologists had received LI-RADS training, in practical application, the evaluation of features may still be ambiguous, and the experience of the two radiologists and their familiarity with LI-RADS may have affected consistency. Further training of radiologists is needed in the future to improve experience and the ability to identify features to improve consistency.
Our finding of an association between female sex and DPHCC is similar to Gu et al.’s analysis (14) but contrasts with Huang et al.’s and Lu et al.’s (5,6) previous work that failed to detect a sex bias. Future work should explore whether the sex bias in our study is real and whether it reflects the contribution of cholangiocytic phenotype cells within the DPHCC tumor. Previous work has shown that although HCC affects more men than women, the prevalence of ICC is similar between the sexes; women with either liver condition are more likely to have ICC than HCC (17-19).
We found that HBV infection was less likely in DPHCC than in non-DPHCC. Infection with HBV or hepatitis C virus is the strongest risk factor for HCC (20); in fact, HBV infection is the most frequent risk factor for HCC in China (21). In contrast, the major risk factors for ICC differ from those for HCC and include intrahepatic lithiasis and primary sclerosing cholangitis (22,23). It is possible that the lower prevalence of HBV infection in DPHCC reflects the contribution of cholangiocytic phenotype cells within the DPHCC tumor.
In our study, an imaging model and a combined model were constructed respectively. The diagnostic value of the combined model was better than that of the imaging model. The combined model is helpful in diagnosing DPHCC.
We found that OS rate was numerically lower in the DPHCC groups compared with the non-DPHCC groups, but without a statistical difference; these results aligned with those of Huang et al. (5). A previous study demonstrated high recurrence and mortality rates in patients with DPHCC (6). It may be related to the number of patients enrolled and the duration of follow-up in these studies. Future work should verify the result with larger, multicenter samples. The OS rate was lower in patients with DPHCC, which suggests that DPHCC is more likely to cause death than non-DPHCC and active postoperative management is required.
Our findings should be interpreted with caution because of the small sample and retrospective study design, limitations which are difficult to avoid given the low incidence of DPHCC. Future studies should verify and extend our results with larger, preferably multi-site samples. Such work should also compare, in parallel, the clinicodemographic characteristics and imaging features of ICC, which is the second most frequent type of PLC after HCC.

Conclusions
The combined model established based on female sex, infection with HBV, and rim APHE by CECT can facilitate preoperative differentiation of DPHCC and non-DPHCC. DPHCC is more likely to cause death than non-DPHCC, suggesting that active postoperative management of patients with DPHCC is required.

Supplementary
The article’s supplementary files as