We retrospectively identified consecutive patients with HCC who received radical hepatectomy as their primary therapy between October 2013 and June 2020 at the Affiliated Hospital of Qingdao University from our prospective database, and a diagnosis of HCC was confirmed by pathological reports. The inclusion criteria were as follows: (1) diagnosis of HCC by pathologic criteria; (2) >18 years and ≤80 years of age; (3) histopathologically confirmed HCC with a high risk of recurrence after resection and not receive targeted therapy and immunotherapy before recurrence.; (4) Child−Pugh classification A or B; (5) Eastern Cooperative Oncology Group performance score (ECOG PS) ≤2; and (6) R0 resection. The exclusion criteria were as follows: (1) preoperative treatment, such as TACE, radiofrequency ablation, and antineoplastic agents; (2) hepatectomy of recurrent HCC; (3) a history of other malignancies; (4) invaded macrovasculature, such as portal or hepatic veins, or extrahepatic metastasis; (5) incomplete follow-up data; and (6) intrahepatic recurrence before PA-TACE, which made PA-TACE impossible. The training cohort included 378 patients, and 244 were excluded. We set up a cohort of 167 patients from the Affiliated Hospital of Qingdao University between July 2019 and June 2020 as external validation. According to the inclusion criteria and exclusion criteria, 106 patients were included as the validation cohort. Table 1 summarized the demographic and pathological characteristics of HCC patients. Figure 1 shows the flow chart of the entire process.

Anonymized or confidential patient data were maintained, and patient privacy was protected. We followed the 1964 Helsinki Declaration and its later amendments or comparable ethical standards in all procedures. The Affiliated Hospital of Qingdao University Research Ethics Committee approved the study protocol, protocol code QYFY WZLL 27141.

A variety of clinical information of patients was collected through preoperative imaging examinations (including abdominal B-ultrasound and contrast-enhanced computer tomography (CT) or magnetic resonance imaging (MRI)), laboratory examinations (including routine blood test, blood biochemical examination, coagulation tests, tumor marker examination and hepatitis serology tests), pathologic features, tumor recurrence, and details of the follow-up or date of death. Age, gender, body mass index (BMI), and history of HBV were included in the basic data collected from patients. We collected clinical characteristics, including alanine aminotransaminase (ALT), aspartate aminotransferase (AST), albumin (ALB), total bilirubin (TBIL), serum alpha-fetoprotein (AFP) and prothrombin time (PT), before surgery and PA-TACE. The indicators before PA-TACE were displayed as “T + indicators”, such as “TALT”. Since TALT, ALT and AFP are extremely skewed distribution, we converted TALT and ALT into binary variables with the upper limit of normal value of 40U/L, and AFP into binary variables with the boundary of 20ug/L (20, 21), and conduct subsequent analysis. Ki-67, tumor size, tumor number, microvascular invasion (MVI) and satellite lesions were confirmed based on imaging examinations and pathologic examination. PA-TACE time was defined as the number of days from hepatectomy to PA-TACE. The optimal cutoff value for PA-TACE time was obtained through the “survminer” package’s surv_cutpoint() function of R software (22, 23). The clinicopathologic characteristics of the patients in training cohort are summarized in Table 2 .

All specimens were sampled according to the “Evidence-based Practice Guidelines for Standardized Pathological Diagnosis of Primary Liver Cancer in China: 2015 Update” using a 7-point baseline sampling protocol (24). MVI is a condition in which tumor cells are visible on microscopy in a portal vein, hepatic vein, or large capsular vessel of the surrounding hepatic tissue (25). The maximum diameter of the pathology specimen was used to define tumor size. Tumor number was classified as 1, 2 and 3. The term satellite lesions refers to microscopic HCC nodules separated from the tumor by at least 2 cm of uninvolved liver parenchyma and not included in tumor counts. The surgical specimens were examined by two senior pathologists with more than 10 years of hepatic pathology experience. For discordant cases, consensus was reached through discussion.

Based on tumor characteristics identified by pathology reports, we evaluated the risk of recurrence for resection and included patients with high-risk recurrence factors. When a single tumor with MVI, two or three tumors, or a single tumor larger than 5 cm without MVI was present, patients were considered to have high-risk recurrence factors (8, 9, 18, 19).

Based on the Barcelona Clinic Liver Cancer (BCLC) staging system, we developed a treatment strategy for our patients. Child−Pugh classification and the indocyanine green (ICG) test were used to assess hepatectomy. Intraoperative sonography was used to determine the resection route. For inflow control of the liver during the operation, intermittent Pringle’s maneuver (15 min of clamping followed by 5 min of release) was applied in selected cases. We used an ultrasonic dissector or a pean-clamp for the transection of the liver parenchyma. The histologic examination showed that all patients had achieved R0 resection, which was defined as no residual tumor and a negative margin.

For all patients, the liver function, serum AFP level and contrast-enhanced CT or MRI of the abdomen were evaluated approximately one month after surgery. Following the exclusion of patients who were not suitable for PA-TACE, those with high-risk recurrence factors were recommended to undergo PA-TACE. Socioeconomic status and compliance with doctors played a major role in whether patients followed physicians’ recommendations. For this reason, we included patients up to 4 months after surgery in our study so we can study how the PA-TACE timing affected prognosis.

The Seldinger method was used to apply PA-TACE to the entire remnant liver. Any obvious tumor staining in the remnant liver was detected by hepatic angiography, computed tomography angiography, or both. An emulsion of lipiodol (5-10 mL) was applied after chemotherapeutic agents, including doxorubicin hydrochloride (10 mg), pirarubicin (THP), or pharmorubicin (20-40 mg), were administered slowly through the right and left hepatic arteries if tumor staining was not found. Based on body surface area and liver function, the dosage of lipiodol and doxorubicin was determined (26).

Suspicious imaging findings or biopsy-proven tumors were considered to be signs of recurrence (8). An evaluation of the therapeutic strategy was conducted once tumor recurrence was diagnosed based on tumor number, tumor location, liver function and general patient condition. Surgical reresection and ablation were used in the treatment with curative intent. The other treatment methods included TACE, targeted therapy, immunotherapy, etc.

Recurrence-free survival (RFS) was the primary endpoint of this study; overall survival (OS) and safety of PA-TACE were the secondary endpoints. In the first two years after surgery, patients were followed up once every 2 months and then once every 3 months thereafter. Each follow-up visit included liver function assessments, tumor markers, and abdominal ultrasounds. The patients were scheduled for contrast-enhanced CT or MRI once every 6 months or when recurrence/metastasis was suspected. RFS was defined as the time from the date of operation to the first documented disease recurrence through independent radiological evaluation or liver biopsy, and or death by any cause, whichever occured first. OS was defined as the time from date of surgery to date of death regardless of the cause of death. We recorded adverse events (AEs) from the day of PA-TACE to the last day of follow-up. Using the Common Terminology Criteria for Adverse Events (CTCAE) version 4.0, the safety of PA-TACE was evaluated (27). All follow-up data were summarized as of the end of January 2022.

External validation of the nomogram model was performed through the validation cohort. Exploratory subgroup analyses of RFS were performed in patients by age (≤65 years vs. >65 years), gender (male vs. female) and BMI (normal vs. abnormal).

SPSS software (version 25, IBM, New York, USA) was used for statistical analysis. The median (interquartile range) and the Mann−Whitney U test were performed for continuous variables with a skewed distribution, while the mean ± SD and t test were used for variables with a normal distribution. We compared categorical variables using the χ2 test or Fisher’s exact test and presented them as frequencies and proportions. The Kaplan−Meier method was used to plot survival curves, and the log-rank test was used to compare them. Univariate logistic regression analysis was performed. Then, to evaluate the risk factors affecting the PA-TACE timing, multivariate logistic regression analysis was conducted using the stepwise backward elimination procedure. The model fit was assessed with the Hosmer−Lemeshow goodness of fit test. Backward stepwise regression analysis was used to evaluate independent prognostic factors in univariate and multivariate Cox analyses.

We established a nomogram based on the results of multivariate Cox analysis using the package “rms” in R (version 4.1.2, Vienna, Austria). The receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) were used to quantify the discriminatory ability of the nomogram (28). With the Kaplan−Meier method, the calibration curve was depicted to assess whether the nomogram prediction was in agreement with observed real outcomes. Bootstraps with 1000 resamples were used for the validation of the nomogram and calibration curve construction. As described by Vickers and colleagues (29), the R package “rmda” was used to perform decision curve analysis (DCA) based on the net benefit, which was used to evaluate the performance of the established nomogram in clinical decision-making. The C-index was used for external validation, internal validation and comparison of the nomogram and conventional prognostic evaluation systems. Statistical significance was defined as a P value <0.05 for two-tailed tests.

The optimal cutoff value of the PA-TACE time was determined to be 48.63 days for RFS (Figure 2A), to confirm the effect of the PA-TACE timing on prognosis. We then divided the patients into early and later TACE groups based on the optimal cutoff value. The Kaplan−Meier analysis showed that the PA-TACE timing was a significant poor prognostic factor for RFS (P<0.05) ( Figure 2B ).

There was no significant difference in baseline characteristics between the training cohort and validation cohort ( Table 1 ). A total of 789 HCC patients were treated at our center with radical (R0) partial hepatectomy. 484 qualified patients were enrolled in the final study and were divided into training cohort (378) and validation cohort (106). In training cohort: The median age of the patients was 61 (54-68) years, and 82.28% of them were male. The median time of PA-TACE was 41.03 (35.05-49.99) days. Postoperative pathologic examination confirmed that 242 (64.02%) patients had MVI, the median tumor diameter was 4.2 (3-7) cm, and 198 (52.38%) patients had HBV infection. As shown in Table 2 , we divided the patients into early and later TACE groups based on the optimal cutoff value for RFS.

At the last follow-up in January 2022, the median RFS for all patients in training cohort was 2.6 (0.75-3.64) years, and the median OS was 3.4 (2.62-5.03) years. In the follow-up period, 232 (61.38%) patients relapsed, and 129 (34.13%) died, failing to reach the median survival. Compared to the late TACE group, the early TACE group had a significantly lower recurrence rate and mortality (57.72% vs. 70.75%, P<0.05; 29.41% vs. 46.23%, P<0.01). The RFS was 0.62 years longer in the early TACE group (2.66 years; 95% CI, 0.85-3.70 years) than in the later TACE group (2.04 years; 95% CI, 0.55-3.60 years). The OS was 0.15 years longer in the early TACE group (3.43 years; 95% CI, 2.66-5.03 years) than in the later TACE group (3.28 years; 95% CI, 2.29-5.04 years). The 1-, 3-, and 5-year RFS rates of the early TACE group were 72.1%, 48.8% and 37.7%, respectively; the 1-, 3-, and 5-year RFS rates of the later TACE group were 59.4%, 37.5% and 26.1%, respectively. In addition, the 1-, 3-, and 5-year OS rates of the early TACE group were 94.5%, 79.8% and 65.4%, respectively; the 1-, 3-, and 5-year OS rates of the later TACE group were 93.4%, 66.7% and 51.2%, respectively. In the validation cohort, the median RFS of HCC patients was 1.73 (0.93-2.07) years, and the median OS was 2.01 (1.71-2.22) years.

Logistic regression analysis was conducted to identify risk factors associated with the PA-TACE timing. Regarding RFS, univariate logistic regression analysis revealed that TALT, ALT, AFP, MVI, satellite lesions, age and history of HBV may be risk factors associated with the PA-TACE timing ( Figure 3A ). In the RFS, multivariate logistic regression analysis found that independent predictors of the PA-TACE timing included gender(OR=2.099,95% CI,1.021-4.312, P<0.05) and history of HBV (OR=2.886, 95%CI,1.723-4.835, P<0.001) (Figure 3B). The values of Nagelkerke’s R2 were 0.224, while the results of the Hosmer−Lemeshow test were 0.762 in RFS analysis. These results showed that the overall model fit was good, with a median effect size.

Cox regression analysis was used to explore the effect of the PA-TACE timing on the prognosis of HCC. Univariate Cox analysis showed that TAFP, TBIL, AFP, KI-67, tumor size, MVI and the PA-TACE timing were risk factors for RFS in HCC patients ( Figures 3C, D ). Then, multivariate Cox analysis showed that Ki-67(HR=1.014, 95%CI,1.007-1.020, P<0.001), tumor size(HR=1.056, 95%CI,1.021-1.092, P=0.001), MVI(HR=1.503, 95%CI,1.114-2.028, P<0.01) and the PA-TACE timing (HR=1.515, 95%CI,1.139-2.015, P<0.01) were independent prognostic factors for RFS in HCC patients.

To predict the prognosis of HCC patients, a nomogram was developed based on the training cohort that integrated the PA-TACE timing with significant clinical characteristics, such as Ki-67, tumor size and MVI, for RFS ( Figure 4A ). The ROC curve and AUC were calculated to evaluate and compare the discriminatory power of the nomogram model. The nomogram showed good predictive performance on the ROC curve. It was observed that the AUC values for 1-, 3- and 5-year RFS were 0.699, 0.685 and 0.700, respectively ( Figures 4B–D ). When comparing the constructed nomogram with the ideal model, the calibration plot showed good performance ( Figure 4E ). DCA also confirmed the predictive capacity of the nomogram ( Figure 4F ).

Internal validation was performed based on gender (male vs. female), age ( > 65 years vs. ≤65 years), and BMI (normal vs. abnormal) groupings to further validate the predictive power of the model for RFS. The results showed good predictive performance of the nomogram model for RFS in both the male and female subgroups ( Figures 5A, B ), the >65 and ≤65 age subgroups ( Figures 5C, D ), and the normal and abnormal BMI subgroups ( Figures 5E, F ). In the validation cohort, the 1-year and 2-year C-index of the nomogram were 0.698 and 0.697, respectively ( Figures 6A, B ). In addition, for the validation cohort, the 1-year and 2-year calibration plot of external validation of nomogram model performed well ( Figure 6C ). Furthermore, conventional prognostic evaluation systems, such as the Milan criteria (MC) (30), Albumin-Bilirubin (ALBI) grade and Glasgow Prognostic Score (GPS) (31, 32), were compared with the established nomogram to confirm which prognostic model was more reasonable and efficient. Our nomogram outperformed the conventional prognostic evaluation systems for RFS ( Figures 6D–F ) as measured by ROC curves of 1-, 3- and 5-year.

Table 3 summarizes the adverse events (AEs) related to PA-TACE in HCC patients. Overall, we found that most AEs were mild and manageable, and no toxicity-associated deaths occurred in this study. Nausea/vomiting (33.07%), pain (14.55%) and fever (13.49%) were the most common AEs. A few patients developed liver dysfunction (1.06%), leukopenia (0.53%) or thrombocytopenia (0.53%). No grade 3 or 4 AEs were observed based on the Common Terminology Criteria for Adverse Events (CTCAE) version 4.0.

The patients in our study received subsequent antitumor therapies after recurrence, including TACE (40.52%), locoregional ablation (13.36%), hepatectomy (3.02%), targeted therapy and immunotherapy (14.66%) ( Table 4 ). There were 34 patients who received targeted therapy and immunotherapy at the time of recurrence or suspected recurrence. We analyzed the effects of targeted therapy and immunotherapy on patients’ OS in Table 5 . The analysis results showed that the OS of patients with targeted and immunotherapy was better than that of patients without targeted and immunotherapy, although there was no statistical difference (P>0.05).