This retrospective study was approved by the institutional review boards of our hospital and in accordance with the Declaration of Helsinki. The requirement for written informed consent was waived by the institutional review boards due to the retrospective nature of the present study.

A total of 610 consecutive patients with unresectable HCC who initially underwent TACE at three hospitals between January 2015 and March 2020 were included. The inclusion criteria were as follows: patients had 1) Eastern Cooperative Oncology Group performance status 0; 2) compensated liver function (Child-Pugh class A or B); and 3) at least two consecutive TACE sessions, or although only one TACE session was performed, complete response (CR) was achieved after the TACE session. The exclusion criteria were as follows: patients had 1) portal venous tumor thrombus (n = 216); 2) distant metastasis (n = 109); 3) lost to follow-up (n = 23); 4) a time interval between the first and second TACE over 3 months (n = 21); 5) follow-up computed tomography (CT) or magnetic resonance (MR) imaging performed after TACE over 3 months (n = 17); 6) infiltrative HCC (n = 12); and 7) having been treated with a combination of TACE and other locoregional therapies such as ablation (n = 8). The flowchart of the study population is shown in Figure 1.

TACE procedures were discussed with the tumor board prior to administration for each patient. Celiac trunk and superior mesenteric arteriography, as well as indirect portography, were performed to visualize the variations in hepatic arterial anatomy and to evaluate the patency of the portal vein. Either a 2.7 French (Progreat, Terumo Medical Corporation) or a 2.2 French (Carnelian, Tokai Medical Products) coaxial microcatheter was placed into the tumor-feeding arteries with the assistance of cone beam computed tomography (CBCT) if needed. TACE was performed using either drug-eluting beads (DEB) (CalliSpheres Beads, Jiangsu Hengrui Medicine Co., Ltd.) loaded with epirubicin (Shandong New Time Pharmaceutical Co., Ltd.) or up to 20 ml emulsion of iodized oil (Lipiodol, Guerbet Asia Pacific Ltd.) mixed with epirubicin. The oil-epirubicin emulsion was created using the water-in-oil technique by mixing iodized oil with a distilled water solution containing a drug cocktail of dissolved epirubicin at a ratio of 3:1. The dosage of epirubicin in DEB-TACE ranged from 50 to 150 mg and the size of DEBs varied from 100 to 300 um and 300–500 um, while in conventional TACE the dosage of epirubicin was 50–75 mg/m² body surface area. In DEB-TACE, no additional embolization was performed after injected of 1–2 g DEB, while in conventional TACE, gelfoam slurries were injected to embolize the proximal tumor feeders after the oil-epirubicin emulsion was injected. The technical endpoint of TACE was defined as the reduction in arterial inflow to the tumor and tumor devascularization. Changes in embolic agents, chemotherapy drugs, or tumor-feeding artery reselection were conducted for the second TACE procedure when an insufficient response after the first TACE occurred.

Among 204 patients, 147 patients from hospital A and hospital B were assigned to a training set, and the remaining 57 patients from hospital C were assigned to a validation set (the training to validation ratio was approximately 7:3). The demographic, laboratory, and radiological data of patients were collected to assess the potential risk factors for early TACE refractoriness. The demographic and laboratory data included age, sex (male/female), Child-Pugh class (A/B), BCLC stage (0-A/B), underlying liver disease, history of resection, initial embolic agents (lipiodol/DEB), initial alpha-fetoprotein (AFP) level (≤400/>400 ug/L), and initial neutrophil to lymphocyte ratio (NLR). The radiological data included tumor distribution (unilobar/bilobar), number of tumors (solitary/2–3/>3), size of the largest tumor, and up-to-seven criteria (within/beyond). Patients who were beyond up-to-seven criteria was defined as: largest tumor diameter [cm] + number of tumors >7 (Mazzaferro et al., 2009; Koroki et al., 2020). Radiological data were independently reviewed by two radiologists with either 22 or 19 years of experience of abdominal imaging, respectively. Both of the radiologists were blinded to the clinical data and were not involved in the treatment. The final results of radiological data were made by the discussion between two radiologists.

Dynamic CT/MR imaging and laboratory variables were acquired before and after the first and the second TACE sessions. The treatment response of TACE was assessed by using dynamic CT/MR, and residual enhancement of nodules was measured with consideration of the 2019 version of Response Evaluation Criteria in Cancer of the Liver (RECICL) (Kudo et al., 2019).

The definition of TACE refractoriness was based on the JSH Consensus Guidelines as follows: 1) intrahepatic lesion: two or more consecutive ineffective responses was observed within the treated tumors (viable lesion>50%) or new lesion occurred in treated area, even after changing the chemotherapeutic agents or reanalysis of the feeding artery on response evaluation CT/MR after 1–3 months following adequate selective TACE; 2) AFP: continuous elevated levels of tumor markers right after TACE; 3) vascular invasion was observed; and 4) extrahepatic spread was observed.

The definition of early TACE refractoriness was that patients presented with TACE refractoriness after initial two consecutive TACE procedures.

The PFS was defined as the time interval between date of TACE procedure and death whatever the cause, tumor progression or last clinical follow-up. Tumor progression was assessed according to the 2019 version of RECICL criteria (Kudo et al., 2019), which was defined as tumor enlargement of ≥50%, excluding the area of treatment-induced necrosis in either target lesion or non-target lesion. However, new intrahepatic lesion occurred in non-treated area after TACE was not defined as tumor progression.

The data were shown as the mean with standard deviation (SD), median with interquartile range (IQR), or frequency. To evaluate the inter-reader agreement of radiological data between the two abdominal radiologists, either intraclass correlation coefficient (ICC) analysis (for numerical data) or Kappa test (for categorical data) was performed. Agreement was classified as poor (ICC or Kappa value, 0–0.40), fair to good (ICC or Kappa value, 0.40–0.75), and excellent (ICC or Kappa value, >0.75). In univariate analysis, Pearson’s chi-squared test or Fisher’s exact test was used to compare categorical variables, while the independent sample t-test or rank-sum (Mann-Whitney) test was used to compare numerical variables. In multivariate analysis, a forward stepwise logistic regression model and nomogram were used. Variables with a p-value less than 0.05 in the univariate analysis were included in the multivariate model, and all those variables were tested by Diagnosis of Collinearity with variance inflation factors less than 5 (VIF < 5). The discrimination of this predictive model was examined by the receiver operating characteristic (ROC) curve, and the goodness of fit was validated by the Hosmer-Lemeshow test, in which a p value > 0.05 indicated good performance. Based on the factors selected by forward stepwise logistic regression, a one-to-one propensity score matching (PSM) was conducted to compare the PFS between patients who were present or absent of early TACE refractoriness. PFS curve was estimated by Kaplan-Meier method and compared by log-rank test.

Statistical analyses were performed with SPSS statistical software (SPSS version 20, International Business Machines Corporation) and R software (version 3.4.2, http://www.R-project.org). A probability value of <0.05 was considered statistically significant.

Finally, a total of 204 patients were included (183 males and 21 females, with a mean age of 56.5 ± 11.6 years). All TACE procedures achieved technical success according to the Society of Interventional Radiology (SIR) guidelines (Gaba et al., 2016). The diagnosis of HCC was based on pathology (biopsy, n = 12) or on the American Association for the Study of Liver Practice Guidelines (n = 192). There were 181 (88.7%) patients with Child-Pugh class A and 23 patients with Child-Pugh class B (11.3%), 123 patients (60.3%) in BCLC stage 0-A and 81 patients (39.7%) in BCLC stage B. Patients with BCLC-0 or BCLC-A disease received DEB-TACE or conventional TACE for the following reasons: in cases beyond the Milan criteria, liver transplant was contraindicated; presence of portal hypertension or increased bilirubin, hepatectomy was contraindicated according to BCLC staging system; or for HCC lesions in unfavorable location, ablation was technically infeasible. Conventional chemoembolization was initially performed in 102 patients (102/204, 50.0%), and DEB-TACE was also initially performed in 102 patients (102/204, 50.0%). There were 127 patients with initial AFP ≤400 ug/L (62.3%) and 77 patients with AFP >400 ug/L (37.7%). The median of NLR was 2.95 (IQR 3.72).

The inter-reader agreements of radiological data between the two radiologists were all excellent, with Kappa values of 0.947 (tumor distribution) and 0.954 (number of tumors), and the ICC value was 0.838 (size of the largest tumor). Among all patients, the sizes of the largest tumors were ≤50 mm in 83 patients (40.7%), 50–100 mm in 75 patients (36.8%), and >100 mm (22.5%) in 46 patients. Seventy-two (35.3%) patients had tumors with bilobar involvement, and 132 (64.7%) had tumors with unilobar involvement. One hundred twenty-three patients (60.3%) had a single tumor, 48 patients (23.5%) had two or three tumors, and 33 patients (16.2%) had more than three tumors. Eighty-one patients (39.7%) were within up-to-seven criteria, and 123 patients (60.3%) were beyond the up-to-seven criteria. The detailed demographic, radiological and laboratorial characteristics are summarized in Table 1.

The patterns of early TACE refractoriness in patients with HCC are illustrated in Table 2. Totally, there were 73 patients presented with early TACE refractoriness (35.8%, 73/204). A typical patient with early TACE refractoriness is shown in Figure 2. In univariate analysis, early TACE refractoriness was associated with BCLC stage (p < 0.001), tumor distribution (p < 0.001), number of tumors (p < 0.001), size of the largest tumor (p = 0.021), initial embolic agents (p = 0.045), and within/beyond up-to-seven criteria (p < 0.001). There was no statistical relationship between early TACE refractoriness and age, gender, Child-Pugh class, underlying liver disease, history of resection, initial AFP level, and NLR level. Multivariate analysis was performed using the significant risk factors determined in the univariate analysis, and within/beyond up-to-seven criteria (p = 0.001; odds ratio = 3.640, 95%CI 1.686–7.859), tumor distribution (p = 0.002; odds ratio = 3.251, 95%CI 1.536–6.883) and number of tumors (p = 0.005; odds ratio = 1.894, 95%CI 1.212–2.961) were independent predictive factors associated with early TACE refractoriness. The results from the univariable analysis performed on the training data set were summarized in Table 3.

A predictive model and nomogram (Figure 3) were built on the training set for predicting early TACE refractoriness based on within/beyond up-to-seven criteria, tumor distribution and number of tumors, with an area under the curve (AUC) of 0.788 (95%CI, 0.707–0.868), a sensitivity of 74.4% and a specificity of 73.8% (Figure 4A). While in the validation set, the AUC was 0.706 (95%CI, 0.564–0.848), with a sensitivity of 78.1% and a specificity of 60.0% (Figure 4B). Moreover, satisfactory calibration was confirmed by the Hosmer-Lemeshow test, with p values of 0.236 and 0.539 in the training and validation cohorts.

Based on influencing factors selected by forward stepwise logistic regression including up-to-seven criteria, tumor distribution, and number of tumors, a PSM analysis was performed. After PSM, a total of 96 patients were enrolled, 48 of whom were present early TACE refractoriness, while 48 of whom were absent. There was no difference in baseline characteristics between two groups after PSM (Table 4). The median PFS in patients with or without early TACE refractoriness was 133 days (95% CI: 18.2–168.7) and 371 days (95% CI: 269.6–472.4), respectively. Patients who were absent of early TACE refractoriness had a significantly higher PFS rate than those of patients who were present (p < 0.001). The PFS curves of the two groups are shown in Figure 5.